Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
General Editors
Werner Abraham
Christer Platzack
Guglielmo Cinque
Ian Roberts
Gnther Grewendorf
Liliane Haegeman
Sten Vikner
Hubert Haider
C. Jan-Wouter Zwart
Harvard University
University of Venice
University of Lund
Cambridge University
McGill University
University of Aarhus
University of Groningen
Volume 137
Grammar as Processor. A Distributed Morphology account of spontaneous
speech errors
by Roland Pfau
Grammar as Processor
A Distributed Morphology account
of spontaneous speech errors
Roland Pfau
University of Amsterdam
TM
2008044246
Table of contents
Acknowledgments
List of abbreviations
chapter 1
Introduction
1.1 Motivations for investigating speech errors 2
1.2 The speech error corpus 7
1.3 A first look at the speech error data 10
1.3.1 Semantic anticipation and perseveration 10
1.3.2 Feature mismatch 11
1.3.3 Feature stranding and feature shift 16
1.3.4 Accommodation 18
1.4 Outline of the book 20
chapter 2
Grammar in use
2.1 On mentalism and psychological reality 25
2.2 The processing of grammatical structure 30
2.2.1 Phonological structure 30
2.2.1.1 Segments and segment clusters 31
2.2.1.2 The manipulation of subsegmental features 34
2.2.1.3 Suprasegmentals 36
2.2.2 Morphological complexity 38
2.2.2.1 Stranding 39
2.2.2.2 Non-concatenative morphology 41
2.2.3 Syntactic transformations
2.2.3.1 The Derivational Theory of Complexity 47
2.2.3.2 The psychological reality of empty elements 48
2.2.3.3 Transformational errors 50
2.3 Conclusion 52
xi
xiii
25
45
Grammar as Processor
chapter 3
Theoretical background
55
3.1 Multi-level models of language production 55
3.1.1 Processing levels 56
3.1.2 Lexical retrieval 58
3.1.3 Flow of information 60
3.1.4 Summary 62
3.2 Distributed Morphology: A sketch of the framework 62
3.2.1 The structure of the grammar 64
3.2.2 Morphological Structure 68
3.2.2.1 Morpheme types and local licensing 68
3.2.2.2 Morphological operations: Merger, insertion, and fusion 72
3.2.3 Phonological Form 76
3.2.3.1 Vocabulary insertion 76
3.2.3.2 Phonological readjustment 79
3.2.4 Summary 81
3.3 Conclusion 81
chapter 4
Semantic features in language production
4.1 Non-random insertion: Distinguishing cats from dogs 83
4.2 Semantic features in speech errors 86
4.2.1 Semantic substitutions 86
4.2.2 Semantic anticipations and perseverations 91
4.2.3 Summary 93
4.3 A note on compositional semantics 93
4.3.1 The count/mass distinction 94
4.3.2 Semantic (natural) gender 97
4.3.3 Semantic negation 101
4.3.4 Summary 101
4.4 Conclusion 102
chapter 5
Morphosyntactic features in language production
5.1 The processing of grammatical gender 106
5.1.1 Definition and assignment of gender 107
5.1.2 Underspecification of grammatical gender 109
5.1.2.1 Gender accommodation 111
83
105
5.2
5.3
5.4
5.5
Table of contents
Grammar as Processor
chapter 6
223
Rethinking accommodation
6.1 A typology of accommodations 224
6.1.1 Phonological accommodation 225
6.1.2 Morphophonological accommodation 225
6.1.3 Morphological accommodation 226
6.1.4 Morphosyntactic accommodation 228
6.1.5 Summary 230
6.2 Feature copy 230
6.2.1 Gender agreement 231
6.2.2 Subject-verb agreement 232
6.2.3 Summary 233
6.3 Feature stranding 234
6.3.1 Spell-out of feature bundles 235
6.3.2 Phonological readjustment 236
6.3.3 Context-sensitive spell-out of features 237
6.3.4 Summary 239
6.4 Local licensing 240
6.4.1 Phonological readjustment and suppletion 242
6.4.2 Morpheme insertion 246
6.4.3 Competing nominalizations and DP-internal structure 250
6.4.4 Accounting for categorial identity 252
6.4.4.1 The role of licensing in root exchanges 254
6.4.4.2 The role of adjacency in root exchanges 256
6.4.5 Summary 262
6.4.6 An alternative account: Minimize Exponence 263
6.5 Action!: Two complex cases 267
6.5.1 Error #1: Morpheme insertion, feature copy & readjustment 267
6.5.2 Error #2: Case assignment, morpheme insertion & feature copy 270
6.6 Against repair strategies 273
6.6.1 Reconsidering (morpho)phonological accommodation 274
6.6.2 An exception: Lexical construal 278
6.6.3 A possible surface filter 280
6.6.4 Summary 285
6.7 Repairs: Two further issues 286
6.7.1 Too late for repair 286
6.7.1.1 Morpheme stranding 287
6.7.1.2 Feature mismatch within DP 287
6.7.1.3 Subcategorization errors 291
6.7.1.4 Summary 292
Table of contents
301
315
353
369
Acknowledgments
To be sure, this book could have been written without the input, feedback, and
support from linguists and friends but it would have been much harder and less
inspiring, and the result would certainly look different. I am therefore grateful to
all the people who helped me shape and develop my ideas by providing invaluable
feedback, be it through electronic or personal correspondence.
I am particularly indebted to Helen Leuninger for her inspiration and for
teaching me to appreciate the value (and beauty) of speech errors. Moreover, I
would like to thank Rajesh Bhatt, Jason Brown, Brendan Costello, Heidi Harley,
Annette Hohenberger, Donald MacKay, Pamela Perniss, Martin Salzmann, Dan
Siddiqi, and Eva Waleschkowski for stimulating discussions and miscellaneous
valuable contributions. My drive to work on this book experienced a boost following my participation in the LSA 2005 Summer Institute workshop on The
state of the art in speech error research. I thank Carson Schtze and Vic Ferreira
for organizing the workshop and for inviting me. I am also very grateful to Adam
Albright for his insightful commentary on my presentation. In addition, I wish to
thank the workshop participants, in particular, Thomas Berg, Julie Franck, Merrill
Garrett, Teri Jaeger, and Stephanie Shattuck-Hufnagel, for the helpful feedback
they provided.
The positive effect of the support I received from colleagues at the University
of Amsterdam while working on this book cannot be overestimated. In particular, I wish to thank the members of the parts-of-speech circle Jan Don, Marian
Erkelens, Kees Hengeveld, and Eva van Lier for inspiring discussions and commentary on parts of this book. Moreover, I am grateful to Enoch Aboh, Anne
Baker, Dik Bakker, Hans den Besten, Robert Cirillo, and Josep Quer for input and
encouragement.
A separate paragraph has to be devoted to my dear friends Katharina Hartmann and Markus Steinbach. I owe them very special thanks for their continuous
support and advice in both linguistic and non-linguistic matters, and, most of all,
for their friendship.
On the editorial side, I very much appreciated the help of the series editors
Werner Abraham and Elly van Gelderen. Special thanks to Werner for his encouragement, his patience, and his invaluable comments on the manuscript. Moreover,
Grammar as Processor
I wish to thank Kees Vaes for his professional assistance and the very pleasant
cooperation during the production process.
A substantial part of this book was written on a sort of work retreat in the
beautiful German Alps. While being there, Christl and Richard Altenried and Lilli
and Manfred Ulbert would occasionally tend to my physical well-being, thereby
contributing in a significant way to the end product (the only qualification being
Plenus venter non studet libenter). I thank them for being such lovely and caring
hosts.
Most of all, I thank Federico Lopez for inspiration and support beyond lingu
istics. For everything.
Last but not least, I want to express my deep gratitude to my wonderful family,
my parents Renate and Klaus and my brothers Mathias and Jochen, for their love
and support. This book is dedicated to Anna, Lieselotte, and Karl the next generation to produce intriguing speech errors and, who knows, maybe the next generation to study them.
List of abbreviations
abs
acc
AgrS
comp
dat
dim
DM
du
erg
expl
f
foc
fut
gen
imp
inf
loc
absolutive case
accusative case
subject agreement
comparative
dative case
diminutive
Distributed Morphology
dual
ergative case
expletive
feminine gender
focus
future tense
genitive case
imperfect tense
infinitive
locative case
m
masculine gender
mod.part modal particle
MS
Morphological Structure
n
neuter gender
neg
negation (particle or affix)
nmlz
nominalizing suffix
nom
nominative case
part
participle
pass
passive
plural
pl
poss
possessive
pres
present tense
rec
reciprocal pronoun
reflexive pronoun
refl
rel
relative pronoun
sg
singular
SVA
subject-verb agreement
chapter 1
Introduction
A word fitlyspoken
is like apples of gold in a setting ofsilver.
(Proverbs25,11)
Our spontaneous speech is far from being perfect. Rather, it is interspersed with
irregularities and errors of various kinds. Speakers may begin an utterance but
never bring it to an end at least not to the end they originally intended because,
while speaking, they realize that what they were going to say is inadequate,
imprecise, or simply wrong. In such cases, a speaker may consciously decide to
break off and to repair the utterance. More often than not, however, the irregularities that occur are of an unconscious nature. Just switch on the TV and watch
a random fragment of some interview or talk show. Most likely, you will notice
that the participants frequently produce strings of words that, when listened to
closely, resemble gibberish: wild parentheses, missing verbs, and stranded ideas.
Still, the conversational partner and the audience usually manage to make sense
of this verbal chaos, and what is more, they may not even be aware of the flaws
itcontains.
On top of all of these idiosyncrasies, utterances are further distorted by spontaneous speech errors, slips of the tongue. Typically, and in contrast to the irregularities mentioned previously, in slips, two linguistic elements interact, such as,
for instance, segments, morphemes, or words. Two segments may change place
in a sound exchange and two synonymous words may fuse into one in a blend,
to give just two examples. Slips of the tongue are unconscious errors, too, and for
the most part, they are neither noticed by the speaker nor by the hearer. Occasionally, a speaker may correct her/his utterance, but even such a self-repair may
gounnoticed.
Fortunately, some slips of the tongue are noticed and find their way into error
collections compiled by psycholinguists. Actually, it is often the researcher herself/
himself, equipped with pen and paper, who zealously notes down an error when
hearing one. For the psycholinguist, spontaneous speech errors are a highly welcome and exciting phenomenon, since due to their regular (that is, non-random)
character speech errors are taken as valuable evidence for mental representations
Grammar as Processor
and processes involved in speech production.1 In this book, I take a closer look at
what insight speech errors can provide concerning grammatical representations
and processes. In particular, I investigate in how far the observed error patterns
can be accounted for in a model of grammar, the Distributed Morphology model,
and in how far the grammar model can be mapped onto the psycholinguistic language productionmodel.
In the remainder of this chapter, I will first give a brief historical overview of
the motivations that brought researchers to investigate speech errors. In Section 1.2,
I will introduce the speech error corpus that is the basis of this study as well as
the distribution of the slip types it contains. Speech errors come in many different types; not all of the attested types, however, are relevant for the present study.
In Section 1.3, I will familiarize the reader with the error types that will play an
important role throughout this book and that will be subject to more extensive
analysis and discussion in subsequent chapters. In Section 1.4, I sketch the outline of thebook.
Introduction
b. Ich
I
. Note that only parts of the chapter on speech errors (Das Versprechen) are from 1901;
some parts were added by Freud in later editions of the book. The error in (1a), for instance,
was added to the original chapter only in 1917. In some English versions (e.g., Freud 2005),
the parts that were added later are missing. Moreover, in the English version, linguistic difficulties often made it necessary to modify or substitute examples given by Freud in order
to make them comprehensible to the English readership. For that reason, example (1b)
although it is contained in the original (1901) version of the chapter did not make it into
the before-mentioned English version.
Grammar as Processor
However, not all Freudian slips need arise from repressed sexual thoughts.
Some errors call for much more complex explanations which are to be looked for
in the biography of the respective speaker. For instance, in order to understand the
deeper meaning of the error in (1b), Freud argues, one must thoroughly consider
the personal history of the speaker, one of his female patients, who owns a guesthouse (Freud 1901/2000:124). She herself claims to know what caused the slip.
That morning she waited for the streetcar at a station called Hasenauerstrae.
While waiting, it occurred to her that if she were French, she would pronounce the
name of that street without the /h/ in the onset. Upon further enquiry of the therapist, she remembers that as a 14 year old, she had played a French girl in a play,
on that occasion deliberately speaking German with a French accent. According
to Freud, the fact that a few days earlier, a French guest had arrived at her guesthouse, evoked this hidden memory. Consequently, the phoneme exchange in (1b)
has to be analyzed as the result of an interfering unconscious thought from a completely differentcontext.
This causal chain is by far not the most complex one called in by Freud in order
to explain a spontaneous speech error. Obviously, his analyses do not say anything
about the actual processes and grammatical entities involved in language production. All he is interested in is how the subconscious may influence various actions we
perform (see Grnbaum (1984) for a philosophical critique of Freuds approach).3
However, Freud was not the first one to study speech errors. A few years earlier,
Meringer & Mayer (1895), a linguist and a psychiatrist from Austria, provided what
is now considered to be a more traditional analysis.4 On the basis of a corpus of slips
of the tongue that Rudolf Meringer had collected, they arrived at the conviction:
. It should be noted that influence of the subconscious does not necessarily imply
hidden desires or emotions. In so-called environmental contaminations (T. Harley 1990),
for instance, visual or auditive material present in the speakers environment, but irrelevant
to the speakers intended utterance, may be incorporated into speech. Consider, for instance,
the following object name source contamination (T. Harley 1990:48; error element in bold
face).
(i) You havent got a computer, have you? You havent got a screwdriver, have you?
In these errors, the name of an object present in the visual environment intrudes into the utterance. In (i), the speaker was entering a room where the addressee was working at a computer.
. Freud was familiar with the study by Meringer and Mayer. On the first page of his chapter
on speech errors, he points out that he does not agree with their viewpoints, in particular
with their attempt to provide purely linguistic explanations for these errors. He also notes that
Meringer and Mayer themselves have not overlooked the possibility of speech disturbances
Introduction
dass man sich nicht regellos verspricht, sondern dass die hufigeren Arten
sich zu versprechen auf gewisse Formeln gebracht werden knnen. Mit der
Regelmigkeit der Sprechfehler [] gewinnen dieselben an Bedeutung, sie
mssen durch konstante psychische Krfte bedingt sein, und so werden sie zu
einem Untersuchungsgebiet fr Naturforscher und Sprachforscher, die von ihnen
Licht fr den psychischen Sprechmechanismus erwarten drfen. (Meringer &
Mayer 1895:9)5
The above quote can be seen as an early formulation of the third motivation for
doing speech error research: the effort to investigate linguistic rules and processes that are active in language production. In fact, spontaneous errors played
an increasingly important role in psycholinguistic attempts to construct linguistic performance models (Lashley 1951; Bierwisch 1970/1982; Fromkin 1971;
Garrett 1975, 1976, 1980a, 1988, 1990, 1993; Shattuck-Hufnagel 1979; Butterworth
1982; Leuninger 1987). The crucial question motivating this endeavour is: What
kinds of (possibly ordered) processes mediate between a communicative intention and the articulation of an utterance? A closely related question concerns the
role that grammatical units and rules play in the generation of an utterance. It is
this latter question that tackles the issue of the relationship between competence
andperformance.
Fromkin (1971) was the first one to develop a performance model on the
basis of slips of the tongue, thereby giving the impetus to a new field of research
within psycholinguistics. On the one hand, Fromkins aim was to furnish proof
of the psychological reality of theoretical linguistic entities (see Chapter 2).
On the other hand, however, she also wanted to demonstrate that characteristic properties of various types of spontaneous errors can be related to certain
planning mechanisms and processing stages in a performance model (see
Section3.1).
In the following, I shall only be concerned with the third of the abovementioned possible motivations for doing speech error research. That is, I will
focus on what grammar theory can tell us about the nature of speech errors
and vice versa what speech errors can tell us about the nature of grammar.
through complicated psychic influences, that is, through elements outside of the same word
or sentence or the same sequence of words (Freud 2005:27).
. that one does not randomly produce slips, but that the more frequent types of errors can
be reduced to certain formulas. With increasing frequency, speech errors gain in importance;
they must be caused by constant psychological forces and therefore, they become a field of
investigation for natural scientists as well as linguists who may expect them to shed light on
the psychological mechanism involved in speaking. (my translation).
Grammar as Processor
Consequently, slips of the tongue (as well as other behavioral data, such as, for
instance, acquisition data and data from impaired speakers) are of interest to
linguists who (implicitly or explicitly) accept the assumption that the rules of
grammar enter into the processing mechanism such that evidence concerning
production, recognition, recall, and language use in general can [] have bearing
on the investigation of rules of grammar (Chomsky 1980:200f). This, in turn,
implies that meaningful psycholinguistic analyses of error data can only be made
against the background of significant hypotheses concerning the structure, that is,
the grammar, of the language inquestion.
With that in mind, let us have another look at the two Freudian slips in (1).6
In a psycholinguistic model such as the one proposed by Fromkin (1971), these
two errors would simply be treated as a form-based verb substitution and a phoneme exchange, respectively. In (1a), the intended element dekoriert (decorated)
and the intruder dekolletiert (low-cut) are phonologically very similar: they have
the same number of syllables and share the first syllable, the onset and nucleus of
the second syllable, and the rhyme of the third syllable. Therefore, they are tightly
linked to each other in a form-based lexical network. At the point in the production process when a lexeme is selected for insertion, the processor erroneously
picks a wrong element. Coincidentally, this leads to a somewhat suggestive utterance but in principle, it could be argued that the process is no different from, say,
the substitution of fingiert (faked) for fixiert (to have afixation).7
. The study of Ellis (1980) is an attempt to reanalyze Freuds original collection of speech
errors in terms of the more modern and process-oriented models of language production
outlined in Section 3.1.
. The actual slip I am referring to is given in (i). The context of the utterance was such that it
was very unlikely that the speaker unconsciously referred to an incidence in which she herself,
the addressee, or one of the persons referred to in the utterance had actually faked something.
Introduction
In (1b), two word-initial phonemes, the nasal /n/ and the glottal stop //, are
exchanged; the result are two non-existing but phonologically possible words.
Phonological errors that result in existing words are more readily interpreted in a
psychoanalytic fashion, of course. Still, as we have seen, Freud manages to come
up with a rather tedious explanation for the first error element Ase. Note, however,
that no attempt is made to explain the second word that results from the error,
namely natmen. In fact, Freud simply neglects this element. In other words: only
the elements that fit the post-hoc analysis areconsidered.
(i) er ist so auf seinen Vater fingiert, h, fixiert
he is so on his
father faked, er, have.fixation.on
He has such a fixation on his father.
Still, it is, of course, quite possible that, if one only digs deep enough, one could also come
up with a psychoanalytic interpretation for this error. But this is exactly one of Grnbaums
(1984) points of critique: if one searches long enough, one can always come up with an explanation which involves subconscious processes of some type, the only limit for this endeavour
being the length of the therapy session.
Grammar as Processor
. This is scanty information when compared to the information Rudolf Meringer recorded for
each slip he heard someone produce. Meringer also included the birthdate of the speaker, the
educational background, the time of the day, the state of health and tiredness, etc. Fromkin (1971)
points out that Meringer thus became the most unpopular man at the University of Vienna. It is a
well-known fact, though, that even without including all of these personal details, interrupting a
speaker by exclaiming Oh, you said a funny thing or Wait a second, I have to grab a pen is not
conducive to ones reputation. Obviously, these are the risks one has to take for scientific advance.
. At the moment of completing this book, the Frankfurt corpus of spontaneous speech
errors contains 5595 slips. Around 1995, the Frankfurt group also began to collect spontaneous slips of the hand. Up to now, approximately 200 spontaneous slips of the hand have
been collected. Moreover, in the context of a comparative study on language production,
signed and spoken errors were elicited in an experimental setting. In this setting, adult signers
and speakers were asked to describe picture stories under various cognitive stress conditions.
The elicitation task yielded in total 944 spoken and 640 signed errors (see Leuninger et al.
(2004) for extensive discussion and comparison of the spoken and signed data).
. Non-deliberate bias is probably observed in all speech error collections for the simple
reason that some slips are perceptually more salient than others. For instance, errors that affect
whole words are noticed more often than errors that affect features, segments, or syllables
(Tent & Clark 1980) and anticipations have been claimed to be more salient than perseverations (Nooteboom 1980). On the role of perceptual bias and on the validity of speech error
data see Cutler (1981) and Ferber (1995).
Introduction
extracted all relevant errors from the Frankfurt corpus. On the other hand, over
the past thirteen years, I have also collected a considerable amount of slips myself.
At present, my corpus contains 829 slips. In contrast to the organization of the
Frankfurt corpus, the errors in my collection are not grouped together according
to the type of error (for example, blend, substitution, anticipation, etc.) but rather
according to what property makes them interesting for myinvestigation.
Roughly, the errors contained in my corpus are of four types (to be elaborated
on in the next section). Firstly, there are errors that involve the anticipation or
perseveration of a semantic competitor. Secondly, the corpus contains errors that
involve some kind of feature mismatch, be it a feature mismach between a subject
and a verb, a mismatch within the determiner phrase (DP), or a mismatch between
a case-assigning element and an argument DP (subcategorization errors). The
third group comprises those errors in which an abstract morphosyntactic feature is
either stranded or shifted. Finally, errors of the fourth type all involve some kind of
post-error adaptation process, often referred to as accommodation, where adaptation may affect the error element itself, may have an influence on the post-error
context in which the error element appears, or may involve morpheme insertion.
The distribution of errors from my corpus is given in Table (2). Note that error distributions for a given error type will be specified in more detail in various summary
sections throughout the book (also see Appendix II for a more detailedoverview).
semantic anticipations
semanticperseverations
49
18
31
subject-verb agreementerrors
mismatch on pronominalelement
mismatch withinDP
subcategorizationerrors
406
219
45
96
46
featurestranding
featureshift
133
87
46
accommodation of errorelement
accommodation ofcontext
morphemeinsertion
lexicalconstrual
Total
241
63
129
34
15
829
Grammar as Processor
Before introducing the different error types, two remarks concerning the classification of errors have to be made. First, it should be kept in mind that there are
few perfect speech errors (Cutler 1988). This implies that frequently, a particular
error allows for more than one interpretation, and therefore classification. Occasionally, for instance, it is not clear whether a segmental error should be classified
as a perseveration or an anticipation, when the relevant segment precedes and
follows the error locus in the utterance. Similarly, a segmental error that results in
an existing word might also be classified as a form-based substitution. I have made
an effort to classify errors as meticulously as possible, sometimes also pointing
out alternative interpretations. The possibility remains, however, that some of the
errors aremisclassified.
Secondly, in my corpus, a slip is listed twice when it combines two of the above
properties. In some exchange errors, for instance, accommodation may affect not
only one of the error elements (for example, stem change in the new environment)
but also the context (for example, choice of appropriate determiner). In other words:
the total number of errors reflects the total number of relevant phenomena attested in
the data, not the number of error tokens. Throughout the book, whenever there is no
source given for an error, this error comes from my own corpus of speecherrors.
Introduction
in which this slip was produced (an inquiry about a visit to the opera), the two
nouns Platz (place) and Sitz (seat) are synonymous. In the utterance, the onset of
Platz combines with the rhyme of its competitor Sitz (the notational conventions
are explained in AppendixI).
(3) a. ich brauch-e dafr
auch zwei Kartoffel-n,
I
need-1.sg for.that also two potato-pl,
I also need two onions for that (dish).
h, Zwiebel-n
er, onion-pl
b. Hatte-st du
ein-en
gut-en
Plitz
had-2.sg you(sg) a-m.acc good-m.acc (error)
ein-en gut-en
Platz
// ein-en
gut-en
Sitz
a-m.acc good-m.acc place(m) // a-m.acc good-m.acc seat(m)
Semantic anticipations and perseverations constitute a third, yet less well known
error type in which semantic features play a role. In these errors, a semantic property of an element within the intended utterance mostly a noun triggers the
appearance of an erroneous lexeme that shares this semantic property. In the
semantic perseveration (4a), for instance, the appearance of bellen (to bark) can
be explained by its semantic relation to the preceding noun Hund(dog).
(4) a. ich woll-te
den Hund anbell-en
anbind-en
I
want-past the dog
bark.at-inf tie.up-inf
I wanted to tie up the dog.
b. schreib-st du
mir bitte,
h, gib-st
write-2.sg you(sg) me please, er, give-2.sg
du
mir bitte ein-en
Stift
you(sg) me please a-m.acc pen(m)
The same line of reasoning can be applied to the slip in (4b), a semantic anticipation. Obviously, the intruding verb schreiben (to write) bears a semantic relation
to the noun Stift (pen). Semantic anticipations and perseverations are informative
because they constitute further evidence for the assumption that semantic features
do play a role in language production; they will be subject to more detailed discussion in Section4.2.2.
1.3.2 Feature mismatch
As we are going to see, various types of slips of the tongue exchanges, perseverations, anticipations, substitutions, and blends may result in a feature mismatch.
The four types of feature mismatch to be introduced in this section are mismatch
Grammar as Processor
between subject and verb, mismatch on pronouns, mismatch within DP, and
subcategorizationerrors.11
Let us first have a look at subject-verb agreement errors (SVA-errors). In
German, a verb agrees with its subject with regard to the morphosyntactic features person and number. Correct spell-out of these features on the verb, however, may be thwarted by various influences. SVA-errors come in two types, as
local and long-distance agreement errors. Consider the representative examples
in (5).
In (5a), the head of the subject DP ein Ende (an end) is singular but is followed by the plural genitive modifier der Unruhen (of the disturbances). Due to
the number specification of the more local noun, the copula verb, which immediately follows the complex subject DP (indicated by square brackets), appears
in its plural form. The same phenomenon is observed in (5b); the structural
conditions, however, are different. In this slip, the local plural DP verschiedene
Erklrungen (different explanations) is not part of the subject DP but rather a
direct object preceding the verb in the embedded clause. Again, the verb surfaces
in its pluralform.
. In earlier work, I referred to feature mismatch between subject and verb as anti-agreement (Pfau 2000, 2003), following the use of this term in the syntax literature. In these studies,
the label anti-agreement is usually applied to cases in which a verb does not (fully) agree
with its subject under certain, well-defined conditions. For the sake of illustration consider
the Arabic examples in (i) and (ii). In Arabic, in the presence of an auxiliary verb, both the
auxiliary and the main verb agree with the subject, as can be seen in (i). However, this kind of
double agreement is not observed in all configurations. In particular, when the subject follows
the auxiliary but precedes the main verb, as in (ii), number agreement is only realized on the
main verb (Bahloul & Harbert (1993:16); also see Mohammad (2000); Bakker (2005)).
(i) Al-bint-aani kaan-ataa ta-ktub-aani darsa-humaa
the-girl-3.du was-3.f.du 3.f-write-du lesson-f.du
(ii) Kaan-at
al-bint-aani ta-ktub-aani darsa-humaa
was-3.f.sg the-girl-3.du 3.f-write-du lesson-f.du
The two girls were writing their lesson.
Similarly, in Berber, the verb appears in a special participle form after wh-extraction of the
subject (Ouhalla 1993). See Ouhalla (1993) for discussion of further anti-agreement phenomena in Breton and Turkish.
Apparently, in all these cases, the (auxiliary) verb does not agree with (some features of)
the subject. It should be emphasized, however, that in these languages, the anti-agreement
effect is restricted to certain constructions and that frequently, the non-agreeing verb appears
in a fixed (default) form. As we will see in Sections 5.2 and 5.3, both these restrictions do not
hold for the spontaneous subject-verb agreement errors. Therefore, in the present study, I use
the more neutral term feature mismatch instead of anti-agreement.
Introduction
Erklrung-en
ermglich-en
explanation-pl make.possible-3.pl
c.
ich wiss-t,
h, ich wei,
[dass ihr
Profis
seid]
I know-2.pl, er, I
know.1.sg that you(pl) professionals be.2.pl
I know that you are professionals.
My corpus also contains intriguing SVA-errors in which the error source is clearly
more distant to the verb than the verbs subject although these are much rarer
than the local agreement errors in (5a) and (5b). One such case is given in (5c). In
this slip, the matrix verb wissen (to know) agrees in person and number with the
2nd person plural subject pronoun of the complement clause. Note that we are not
dealing with the anticipation of an agreement suffix, since the verb in the intended
matrix clause also undergoes stemchange.
Local agreement is not only observed in SVA-errors but also in pronominal
agreement; in this case, the relevant features are gender and number. Just as in
local SVA-errors, the relevant features are sometimes copied from an intervening,
that is, more local noun. Most frequently, this kind of erroneous co-reference is
observed in relative clause constructions. In (6), for instance, the relative pronoun
(rel) agrees in gender with the masculine noun Kragen (collar) which is the head
of the local compound Pelzkragen (fur collar). This compound is contained in a
prepositional phrase modifying the feminine head noun Jacke(jacket).
(6) mein-st
du
die Jacke
mit dem
Pelz-kragen, den
mean-2.sg you(sg) the.f jacket(f) with the.m.dat fur-collar(m), rel.m.acc
ich, h, die
I, er, rel.f.acc
Do you mean the jacket with the fur collar that I bought last year?
Grammar as Processor
The error in (7a) is a blend of the two DPs ein spannendes Finale (an exciting
final) and ein spannender Wettkampf (an exciting competition) which, at least in
the context of the utterance, convey roughly the same meaning. The nouns contained in the competing DPs are of different gender: Finale is neuter, Wettkampf (or
rather, the head -kampf (fight) of the compound) is masculine. In the resulting slip,
Wettkampf takes the place of Finale but the adjective spannend is marked for neuter
gender. Note that in this particular case, the mismatch is not visible on the indefinite article, since in the nominative, the neuter and masculine indefinite article are
isomorphous. (7b) is an exchange, presumably the exchange of two definite determiners, which gives rise to two gender mismatches.12 In the intended utterance,
both DPs receive accusative case. At the first error site, the masculine determiner
den combines with the feminine noun Hand (hand) while at the second error site,
the feminine determiner die accompanies the masculine noun Mund(mouth).
(7) a. das wird ein
ganz spannend-es Wettkampf,
h, Finale
that be.fut a.m/n very exciting-n competition(m), er, final(n)
Mund,
h, die
Hand
vor
den
Mund
mouth(m), er, the.f.acc hand(f) in.front.of the.m.acc mouth(m)
ge-halt-en
part-hold-part
auf jed-en
Fall
// auf alle
Fll-e
in every-sg.m.acc case(m) // in all.pl.m.acc case(m)-pl
In (7a) and (7b), the mismatch only concerns the gender feature. Actually, in my
corpus, the cases in which a gender conflict alone causes the error have the highest share in the DP-internal agreement errors. Still, a mismatch may also be due
to other features or a combination of features, as is illustrated by the blend in
. Alternatively, one could analyze this slip as an instance of a feature exchange. See
Section 5.4.4 for further discussion.
Introduction
(7c). This slip is quite peculiar, since the two competing prepositional phrases contain the same noun, Fall (case), but with different number specification. Consequently, a gender conflict cannot come to fruition. The one and only reason for the
mismatch in this example is the number feature: the quantifying expression jeden
(every) is singular while the noun Flle (cases) isplural.
Subcategorization errors constitute the fourth and final type of feature mismatch. In these errors, the relevant morphosyntactic feature is the case feature. For
the most part, these errors occur in blends. The two slips in (8) are of this type but
exemplify subcategorization errors of different nature. In German, the two verbs
imponieren and beeindrucken both convey the meaning to impress but assign different case to the patient argument which, in (8a), is a second person singular
pronoun. In the error, imponieren combines with the accusative pronoun of the
alternative sentence frame instead of with the required dativepronoun.
(8) a. wer hat dich
am
meisten imponier-t
who has 2.sg.acc at.the most
impress-part
b. das lohnt
sich den
Aufwand nicht
that is.worth refl the.m.acc effort(m) not
das lohnt
sich nicht // das lohnt
den
Aufwand nicht
that is.worth refl not // that is.worth the.m.acc effort(m) not
In contrast, in (8b), the verb lohnen (to be worth it) does not combine with a
wrongly case-marked argument but rather surfaces with a superfluous argument.
Here, the two alternative realizations are: (i) a reflexive construction containing
the reflexive pronoun sich versus (ii) a structure involving the accusative argument
Aufwand (effort). In the actual utterance, no decision between the alternatives is
made and the verb appears with the reflexive pronoun and the accusativeDP.13
. The opposite, that is, errors in which a required argument is missing, is also occasionally
observed. In (i), one of the possible realizations involves the verb sein (to be) in combination
with the nominative demonstrative das. The alternative realization is somewhat more complex
since the verb sagen (to say) requires a nominative as well as an accusative argument. In the
error, however, sagen is only accompanied by the demonstrative pronoun (note that the demonstrative has the same form in the nominative and the accusative).
(i) das kann sag-en das
kann sein // das
kann man sag-en
that can say-inf that.nom can be.inf // that.acc can one say-inf
Thats possible//You can say that.
Grammar as Processor
This concludes the survey of errors involving feature mismatch. Note that
SVA-errors both local and long-distance errors as well as feature mismatches
on pronouns will be subject to further discussion in Sections 5.2 and 5.3. DPinternal feature mismatches will be reconsidered in Section 5.1.3. Together with
subcategorization errors, they will make yet another appearance in Section6.7.1.
1.3.3 Feature stranding and featureshift
Some of the above examples have already made clear that morphosyntactic features play a central role in the derivation of an utterance, no matter whether the
result is grammatically correct or grammatically deviant, as, for instance, in the
examples in (5) and (6) which involve defective copy of agreementfeatures.
Besides being copied from a controller onto a target, morphosyntactic features may also be subject to stranding or shift in speech errors. Turning to feature
stranding first, it is noteworthy that in errors involving nouns, it is commonly the
case that nominal roots are shifted while the number specification is left behind.14
In (9a), a noun exchange, Bank (bench), which is specified for singular in the
intended utterance, appears in its plural form while the opposite is true for Kissen
(pillow). Note that Kissen forms its plural by means of a zero suffix. The fact that,
in the error, we are dealing with the singular form is evidenced by the singular
definite determiner (which, by the way, accommodates to the gender feature of
Kissen). Moreover, this error illustrates in an impressive way that we are dealing
with number stranding and not with suffix stranding, since after the exchange has
taken place, the appropriate plural allomorph is chosen for Bank and, moreover,
umlaut formation is triggered (accommodation sites areunderlined).15
. The slip in (i) is an exception to number stranding in a noun exchange. In this error,
the two nouns Ohren (ears) and Nacht (night) are exchanged together with their respective number specification. Otherwise, the resulting utterance would have been das Ohr um
die Nchte schlagen assuming that accommodation of the determiner and the plural-suffix
would have taken place. In (i), accommodation is not evident since the definite plural determiner and the definite feminine singular determiner have the same phonological form (see
Section 5.4.1 for further discussion).
Introduction
(9) a.
bist
du
so nett und leg-st
die
Bnk-e
auf
be.2.sg you(sg) so kind and put-2.sg the.pl bench-pl on
In (9b), we observe stranding of the tense feature. Again, this is a clear case of
feature stranding, since the anticipated verb kommen (to come) has an irregular
past tense form, in contrast to the intended verb versuchen (to try) which takes
the regular past tense suffix-te.
In my corpus, errors involving feature shift are much rarer than those involving feature stranding; two of the few cases are given in (10). (10a) is an instance of
a number feature perseveration. Within the DP fnf Punkte Vorsprung (five points
in the lead), the plural feature associated with Punkte (points) is perseverated
onto Vorsprung (lead). While both nouns take the same plural suffix, Vorsprung
also undergoes umlaut in its plural form. In contrast, (10b) is best analyzed as an
exchange of case features. Assuming that pronouns are the spell-out of feature
bundles, it is clear that the pronouns surfacing in the error differ from the intended
pronouns only in their case specification. Note that therefore, the error does not
only exemplify feature exchange but also feature stranding, namely stranding of
person, number, and gender (for the third person pronoun)features.
(10) a. er hatte fnf Punkt-e Vorsprng-e fnf Punkt-e Vorsprung
he had five point-pl lead-pl
five point-pl lead
He was five points in the lead.
b. ich mcht-e dir
ihn
wirklich vorstell-en,
I
want-1.sg 2.sg.dat 3.sg.m.acc really
introduce-inf
h, dich
ihm
er, 2.sg.acc 3.sg.m.dat
for instance, for the noun exchange in (i) since both Woche (week) and Stunde (hour) take
the plural suffix -n.
(i) sie arbeitet nur zehn Woche-n die Stunde zehn Stunde-n die Woche
she works only ten week-pl the.f hour(f) ten hour-pl the.f week(f)
She only works ten hours a week.
Grammar as Processor
More errors involving feature shift and feature stranding will be discussed in
Section 5.4. Besides number, tense, and case, I will also address the features
negation, and gender. Furthermore, feature stranding will figure prominently
in Section 6.3, where I show that many of the cases that have traditionally been
subsumed under the label error accommodation (see next section) are actually
better analyzed as cases of feature stranding. Note finally that all of the above
examples exemplify the stranding or shift of morphosyntactic features. In my
corpus, there are also a few cases which seem to involve the stranding/shift of
compositional semantic features; these errors will be discussed in Section4.3.
1.3.4 Accommodation
As was shown in Section 1.3.2, speech errors may give rise to a feature mismatch,
that is, to an ungrammatical utterance. There are also intriguing cases, however,
in which an ungrammatical string that might have resulted from an error is made
good by means of an accommodation that brings the utterance in line with some
grammatical constraint. Garrett (1980b:263) defines accommodations as errors
in which the phonetic shape of elements involved in errors accommodates to the
error-induced environment. Traditionally, it has been assumed that accommodations are post-error processes that come in two types, as error accommodations
and context accommodations (Berg1987).
Let me illustrate this distinction by means of two pronoun exchanges. In
an error accommodation, it is the shifted (that is, anticipated, perseverated, or
exchanged) element itself that undergoes accommodation. This is what we observe
in (11a), where both of the exchanged pronouns could be argued to accommodate to the case specification at their respective landing sites (remember that the
accommodated elements areunderlined).
(11) a. ich
soll-te
doch
ihn,
h,
1.sg.nom shall-past mod.part 3.sg.m.acc, er,
er
soll-te
doch
mich
anruf-en
3.sg.m.nom shall-past mod.part 1.sg.acc call-inf
b. sie war
21, als
ich ge-storb-en
bin
she be.past 21 when I
part-die-part be.1.sg
ich war
21, als
sie ge-storb-en
ist
I
be.past 21 when she part-die-part be.3.sg
In contrast, in a context accommodation, the accommodation affects the environment of a shifted element. Note that in (11b), there is no need for the exchanged
Introduction
pronouns to accommodate since both slots are assigned nominative case. In this
error, it is the context of the error element, namely the verb in the embedded
clause, that accommodates to the person feature of the subjectpronoun.
Frequently, context accommodations involve the adaptation of material within
DP with respect to the gender feature of a shifted noun. In the noun exchange
(12a), for instance, the involved nouns Seite (side) and Sprung (jump) are of different gender and trigger gender accommodation in both positions, that is, on
the possessive pronoun sein (his) as well as on the definite determiner, which
cliticizes to the preposition zu (to) (note that the full form of the definite dative
masculine determiner isdem).
(12) a. sein-e Seite
zum
Sprung
ist schief-ge-gang-en
his-f side(f) to.the.m.dat jump(m) is wrong-part-go-part
sein Sprung
zur
Seite
his.m jump(m) to.the.f.dat side(f)
Finally, (12b) illustrates that error and context accommodation may co-occur in
one error. First, the anticipated verb stem sprech (to speak) undergoes ablaut in
its nominalized form Spruch (saying, motto). Secondly, the indefinite determiner
and the quantifying expression halb (half ) accommodate to the masculine gender feature of Spruch. Note that (12b) represents a type of error that Stemberger
(1989) terms incomplete: due to the self-repair right after the error item Spruch,
it cannot be decided whether we are dealing with an anticipation or an exchange
(in the latter case, the complete utterance would have been sie wird mindestens
einen halben Spruchstunden).
The last type of accommodation I wish to introduce are accommodations
involving morpheme insertion. Strictly speaking, these errors are a special type
of context accommodation, the context, however, not being a word separate from
the error element (for instance, a determiner or verb) but rather a morphological
part of the error element. That is, the error element surfaces in a position where
it is accompanied by a morpheme that would not have been part of the intended
utterance. Consider, for instance, (13a). In this slip, the root schn (beautiful)
is anticipated and takes the position of the noun Frisur (hairdo). In this position, the anticipated root combines with the nominalizing suffix -heit, thereby
deriving the noun Schnheit (beauty). (13b) is quite similar but involves the
Grammar as Processor
ihr-e Schn-heit,
h, ihr-e Frisur
ist total schn
her-f beautiful-nmlz(f), er, her-f hairdo(f) is very beautiful
Her hairdo is very beautiful.
b. i ch fands
ein besonders fest-lich-es,
h, gelungen-es Fest
I foundit a.n particularly party-adj-n, er, successful-n party(n)
I think it was a particularly successful party.
Introduction
Grammar as Processor
Introduction
rules in certain licensing environments. In the same section, I also point out that
problematic cases, that is, errors in which the category of the involved elements
seems to play a role, can be accounted for when we assume that the licensing
environment of roots constrains their interaction in an error. In order to further
clarify and summarize the theoretical claims brought forward in the previous
sections, I present a detailed analysis of two particularly complex, and therefore
particularly interesting, speech errors in Section 6.5. Based on the previous discussion, I conclude in Section 6.6 that accommodations (or repairs) in the true
sense do not exist. That is, once we take advantage of the tools made available by
DM, we need not postulate the application of costly post-error repair strategies.
Rather, all accommodatory processes are in fact due to mechanisms that apply
in the course of the (morpho)syntactic derivation anyway. In this section, I will
also point out possible exceptions to this generalization. Some remaining issues
concerning the (non-)availability of repairs are discussed in Section6.7.
In Chapter 7, I conclude that the morphosyntactic theory of Distributed
Morphology and the multi-level processing model can and should be related to
each other. Most importantly, the speech errors which are discussed throughout
this book (as well as other ones) receive a straightforward explanation within the
DM framework. That is, with DM, we have at our disposal a formal model of grammar which corresponds very well to the psycholinguistic model, which has explanatory power, and which therefore makes for a psychologically plausiblemodel.
chapter 2
Grammar in use
As mentioned in the previous chapter, one of the main objectives of this book is to
investigate the psychological reality of the theory of Distributed Morphology. In
essence, this boils down to the question of whether and, if yes, how the theoretical constructs and operations as assumed in this theoretical framework can help
us in explaining performance data. In this chapter, I shall have a look at some of
the aspects that are related to the question of how a particular grammar is put to
use. In Section 2.1, I will be concerned with a mentalistic perspective on linguistic
theory and with the notion of psychological reality of grammar. Results from
a number of psycholinguistic studies that investigate the psychological reality of
grammatical entities (such as features, segments, and morphemes) and operations
(such as syntactic transformations) will be presented in Section2.2.
Grammar as Processor
Obviously, the best kind of theory is one which systematizes the widest range of
facts. In the spirit of Chomskys criticisms of taxonomic theories, Katz (1964) tries
to show that a mentalistic theory can account for everything that a taxonomic
theory accounts for and, in addition, for many things that a taxonomic theory
must fail to account for. In particular, he demonstrates that only a theory that
makes reference to mental states and psychological factors is capable of providing
answers to the three fundamental questions with which a synchronic description
of a particular language deals (Katz 1964:130). These are: (1) What is known by a
speaker who is fluent in a natural language? (2) How is such linguistic knowledge
put into operation to achieve communication?, and (3) How do speakers come to
acquire thisability?1
Traditionally, linguistic theory investigates the first question, while psycholinguistics is concerned with the latter two. According to Chomsky (1980), however,
a strict division of the two disciplines does not make much sense, since knowledge
of language, language use, and language acquisition are in fact intimately related
to eachother.
Delineation of disciplines may be useful for administering universities or
organizing professional societies, but apart from that, it is an undertaking of
limited merit. A person who happens to be interested in underlying competence
will naturally be delighted to exploit whatever understanding may be forthcoming
about process models that incorporate one or another set of assumptions about
linguistic knowledge. Furthermore, it seems evident that investigation of
performance will rely, to whatever extent it can, on what is learned about the
systems of knowledge that are put to use. (Chomsky 1980:202)
. Compare the three basic questions with respect to knowledge of language, as formulated
in Chomsky (1986:3): (i) What constitutes knowledge of language? (ii) How is knowledge of
language acquired?, and (iii) How is knowledge of language put to use?
Grammar in use
The distinction between question (1) on the one hand and questions (2) and (3)
on the other also reflects the well-known distinction between linguistic competence and linguistic performance. Competence concerns only the knowledge of a
speaker/hearer of his language. Therefore, Chomsky (1965) excludes from consideration as data for a linguistic theory ofcompetence
such grammatically irrelevant conditions as memory limitations, distractions,
shifts of attention and interest, and errors (random or characteristic) in applying
his [the speaker/listeners; RP] knowledge of the language in actual performance.
(Chomsky 1965:3)
Grammar as Processor
find the position taken by Chomsky and many other generativists, a position
termed naive optimism by Linell (1979).2 These scholars contend that speakers have at their disposal a highly internalized mental grammar and that the
abstract rule formulations the theoreticians arrive at are actually rules of the
speakers internal grammar. That is, a psychologically real grammar is supposed to be a theory of covert psychological abilities underlying a speakers
linguisticpractice.
But in what sense can a formal grammar reflect underlying knowledge? Once
again, there are different opinions as to what degree a grammar can or should be a
representational model of psychological realities. According to a strong position,
every aspect or detail of the theory is assumed to be isomorphic to some psychological (or neurological) counterpart, while according to a weak position, the
relationship between the theoretical grammar model and the speakers internalized knowledge is more indirect (Linell 1979:11). This opposition is what Ringen
(1975) refers to as strong versus weakmentalism.
Chomsky himself adopts the weak position. He points out that it is only
at an appropriate level of abstraction that we may expect a psychologically
real theory to describe properties alleged to be true of whatever the real
elements of the world may be when considered at that level of description
(Chomsky1980:104).
It is often argued that the theory of Universal Grammar whatever its
merit has not been convincingly shown to be psychologically real (see Fodor,
Bever & Garrett (1974) and the discussion in Section 2.2.3 below). But what
exactly is psychological reality supposed to mean? Chomsky suggests that the
term is to be understood on the model of physical reality. He points out that
in the natural sciences, however, one is not accustomed to ask whether the best
theory we can devise in some idealized domain has the property of physical
reality. Actually, the question when sensibly put is the same in both cases: is
the theory accurate for a certain area of investigation, that is, does it explain the
available evidence and does it make correct predictions? Chomsky illustrates his
argument with an illuminatinganalogy.
. In between these two extremes lie pessimism and moderate realism. The former attitude corresponds approximately to what was called a taxonomic view above: speakers may
very well possess knowledge of their language but it is taken to be impossible to find out
exactly what properties these psychological structures have. Under the latter attitude, speakers
are assumed to possess specific structural knowledge of their language. Still, an investigation
of psychological reality cannot be pursued with purely linguistic methods, but also needs to
take into account many types of external evidence.
Grammar in use
Obviously, it would be desirable to place a laboratory inside the sun to obtain more
direct evidence, but being unable to do so, we must test and confirm our theory indirectly. Similarly, a cognitive theory which, at an abstract level of description, aims at
characterizing properties of the language faculty can only be tested indirectly, that is,
by its success in providing explanations for selected phenomena. In principle, an effort
to relate this theory to other levels of description for instance, neurophysiological
and biological systems does not pose more problems than an effort to obtain access
to the thermonuclear reactions inside the sun in a more direct way than by measuring
emitted light (Grewendorf 1995). Chomsky concludes that the question of psychological reality is no more and no less sensible in principle than the question of the
physical reality of the physicists theoretical constructions (Chomsky1980:192).
Note that the psychological reality of linguistic structures and rules has nothing
to do with their introspective accessibility or intuitive plausibility. As a matter of fact,
many generativists doubt the relevance and reliability of introspectiveevidence.
Any interesting generative grammar will be dealing, for the most part, with mental
processes that are far beyond the level of actual or even potential consciousness;
furthermore, it is quite apparent that a speakers reports and viewpoints about his
behavior and his competence may be in error. (Chomsky 1965:8)
Put differently, the fact that a speaker has internalized a generative grammar that
expresses his knowledge of language does not imply that the structures and rules
assumed in this grammar must be conscious to the individual. But clearly, the
actual data of linguistic performance and this, of course, also comprises error
data may provide evidence for determining the correctness of certain hypotheses
about linguistic structures and rules.3 Some evidence of this kind (obtained in psycholinguistic studies in the last four decades) will be presented in Section2.2.
. For further discussion of the psychological reality and the mental representation of grammars see Fodor et al. (1974), Bresnan (1978), Stabler (1983), Berwick & Weinberg (1984),
Bresnan & Kaplan (1984), and Matthews (1991).
Grammar as Processor
Note that in the present study, I adopt the assumptions of weak mentalism. In
particular, I am going to investigate whether a particular theory of grammar the
Distributed Morphology framework (to be introduced in Section 3.2) is successful in providing elegant explanations for a certain kind of psycholinguistic
data, namely spontaneous speech error data. Crucially, I am not going to claim
that every detail of the theory theoretical constructs like, for example, V-to-Tns
movement or fusion of terminal nodes must be isomorphic to some psychological counterpart. Rather, I am going to demonstrate that Distributed Morphology
makes for a psychologically real theory of grammar in the sense that it is accurate
for the data underinvestigation.
Grammar in use
In (1a), the segment /f/ is anticipated from coffee, in (1b), the segment /p/ is perseverated within the word proliferation, and in (1c), the segments /r/ and /w/ change
place. The exchange in (1d) illustrates that vowels may also be affected. Data such
as these show that the processor must have access to the segmental structure of a
word at some point in the generation of an utterance, thereby verifying the psychological reality of segments in on-lineprocessing.
The same argument has been made for sign languages. Just like spoken words,
signs are not holistic, undividable units. Rather they are made up of sublexical
units sometimes also called phonological parameters; these include handshapes,
locations, and movements as first noted by Stokoe (1960). A crucial difference to
the structure of spoken words is that much of the phonological structure of signs is
organized simultaneously. That is, except for locations and movements which may
sequentially follow each other (Liddell 1984), the phonological parameters are
not articulated in a sequential fashion like consonants and vowels. A handshape,
. I will leave aside the issue of syllables and syllable-internal structure. For discussion of speech
error evidence that is informative with respect to the psychological reality of syllabic structure,
see Laubstein (1987, 1999), Treiman (1987), Sevald, Dell & Cole (1995), and Berg (2007).
Grammar as Processor
(error)
parents
his/her
parents
Further justification for the assumption that individual segments are relevant units
in language production comes from the observation that in some errors, only one
segment of a consonant cluster is involved, that is, the cluster is split up in the
error, as illustrated by the examples in (3) (Fromkin1971:31f).
(3) a. blake fruid
b. at the Broadway spores the prices are
brake fluid
at the Broadway stores
In (3a), the two segments /r/ and /l/, both of which take the second position in
a consonant cluster (a complex syllable onset) are exchanged, while in (3b), the
segment /p/ which is the first part of the cluster /pr/ is anticipated and takes the
second position in thecluster /st/.5
A particularly interesting issue is the status of diphthongs. In many theories, diphthongs are treated as single, though complex segments. Under this view
. A fair number of errors reveals the transposition of whole clusters. Fromkin (1971),
however, points out that this does not evidence the fact that such clusters are inseparable units.
Rather, movement of whole clusters (as in (i)) constitutes further evidence for the assumption
that a syllable is not an unstructured unit either and that syllable parts (in this case onsets)
may also be affected in speech errors. Moreover, CV- or VC-sequences which constitute part
of a syllable can also be involved in errors (see (ii)).
Grammar in use
und dass du
dann um [dri], drei dort bist
and that you(sg) then at (error) three there be.2.sg
and that youll be there at three o clock.
On the basis of German slips of the tongue, Berg (1986) challenges the monophonematic view of diphthongs. In his corpus, there are a number of instances
in which diphthongs are split apart in an error. Two of his examples are given in
(4b) and (4c). In (4b), two diphthongs are involved. The glide part of the diphthong /i/ in Euphorie (euphoria) is anticipated and takes the place of the glide
in the diphthong /au/ in Kauf (buying), thereby giving rise to the vowel+glide
sequence /ai/. In (4c), the // of dort (there) is anticipated and substituted for the
vowel part of the diphthong /ai/ in drei (three) (Berg1986:201).6
Making use of her speech error data, Fromkin (1971) also examines the status
of affricates and of the velar nasal []. With respect to affricates, she observes that
not a single example in her corpus shows a splitting of an affricate into sequences
of stop plus fricative. With respect to the velar nasal, however, some of her data
indicate that at some level of performance, [] may be derived from an underlying
sequence /ng/ (Chomsky & Halle 1968). One such case is the slip in (5). Obviously,
. Further albeit somewhat anecdotical evidence for the status of diphthongs comes
from talking backwards. People who are able to talk backwards fluently usually operate
on phonemes, that is, they reverse the order of phonemes. Cowan, Braine & Leavitt (1985)
observe that eight out of ten English speaking subjects left the diphthongs intact when talking
backwards, while the other two tried to break them up. Interestingly, two German speakers
who were also tested consistently reversed the structure of the German diphthongs.
Grammar as Processor
in this error, only the nasal part of [] in string is anticipated and takes the postvocalic position incut.
In sum, it appears that single segments can be accessed separately in speech errors.
It is, however, a well-known fact that segments are not indivisible units either but
rather bundles of features. As it turns out, a fair number of phonological speech
errors reported in the literature may be analyzed as involving single features rather
than entiresegments.
2.2.1.2The manipulation of subsegmental features
Many phonological errors are ambiguous as to whether they involve whole segments or segmental features. This is true, for example, for the error given in (6a)
which may either be analyzed as a reversal of the segments /k/ and /t/ or as a
reversal of the features [coronal] and [+coronal] (Fromkin 1971:31). As a matter
of fact, Van den Broecke & Goldstein (1980) report that the relation between consonants interacting in segmental errors is not a random one but is dependent on
their phonological similarity. They point out that [t]ypically, one-feature errors
occur more often than two-feature errors, which again occur more often than
three-feature errors, etc. (Van den Broecke & Goldstein1980:48).
(6) a. teep a cape keep a tape
b. mity the due teacher, I mean, nity the poor teacher, no pity the new teacher
c. pig and vat big and fat
d. ich hr die Flhe hsten die Flhe husten
I
hear the fleas (error) the fleas cough
I imagine things.
In contrast to the ambiguous case in (6a), the other three slips in (6) clearly evidence the independence of phonological features. In (6b), for instance, only the
feature [nasal] is exchanged in the first error step (with the redundant voicing
accommodation). Consequently, the /p/ of pity which is [nasal] becomes /m/
and the /n/ of new which is [+nasal] changes to /d/ (Fromkin 1971:35). Interestingly, in a first attempt to correct the error, the whole segments /p/ and /n/ are
exchanged; moreover, we observe lexical construal which turns the resulting nonword /pju:/ into the existing lexeme poor (also see Section6.6.2).
The example in (6c) exemplifies a change of value only for the feature [voiced].
Consequently, the /b/ of big which is [+voiced] becomes a voiceless /p/ , while the
/f/ in fat which is [voiced] is changed to a voiced /v/ (Fromkin 1971:36). Finally,
the German error in (6d) is best analyzed as a perseveration of the feature [back]
Grammar in use
of the segment // in Flhe (fleas). It is this very feature that turns the /u:/ of
husten (to cough) intoan /:/.7
Therefore, the above data indicate that distinctive features are psychologically
real in that they can be independently manipulated in speech errors in the same
way as segments (or feature complexes).8 Fromkin (1971:38), however, points out
that the claim that all distinctive features (as proposed by Chomsky and Halle)
are identical with phonetic properties that can in principle be independently controlled in speech is not borne out by the data of speech errors. She concludes that
in actual speech performance, only some of the features can be accessed independently (for example, [nasal], [voiced], and place features), while other features are
highly dependent on the existence of other properties of thesegment.
A more radical view is taken by Roberts (1975). He argues that all phonological errors should be treated as feature errors. That is, contrary to the generally
held position, he claims that there is no evidence for the syllable or segment as
independent linguistic units. Here, I only want to briefly mention one counterargument against Roberts proposal. Stemberger (1982b) points out that there
appear to be no word blends in which a new segment arises that is not part of one
of the intended words, but which would combine features of segments from the
two blended words. For instance, a blend of the two verbs cook and boil could very
well result in coil but not in *goil where the velarity of the /k/ in cook combines
with the voicedness of the /b/ in boil (Stemberger 1982b:241). Obviously and
contra Roberts in blends, all segments come as whole units from one of the two
target words. That is, features of a segment from one word never combine with
features of a segment from the competing word to produce a thirdsegment.
Brown (2002, 2004), too, argues against the dominant viewpoint that segments
are the units responsible for phonologically driven speech errors. His view, however, is less radical than that of Roberts since he claims that in sublexical errors,
there is positive evidence not only for distinctive features but also for prosodic
(syllabic) constituents. Still, he maintains that there is never any positive evidence
. Newkirk et al. (1980:184) report two American Sign Language slips which involve the
anticipation of a handshape feature. In one error of their corpus, for instance, it is not a handshape that is anticipated but rather the feature [+bent] of the handshape of the second sign.
That is, the selected finger of the first sign (the index finger) remain the same but its position
changes from extended to bent.
. A different view is taken by Klatt (1981). He argues that most errors can be interpreted as
segmental errors. He points out that, as a matter of fact, only three out of 6000 errors in the
MIT corpus can be unambiguously analyzed as featural errors.
Grammar as Processor
for segments and thus concludes that there is no need for a segmental level of
analysis inphonology.
Mowrey & MacKay (1990) go one step further by challenging the assumption
that features (and segments) are transposed in speech errors. In their study, they
report on the laboratory elicitation of sublexical speech errors by means of tongue
twisters such as She sells seashells by the seashore. In order to observe speech motor
activity in phonological errors, electromyography (EMG) was used. This method
implies that electrodes were inserted into the lower lip and into the tongue of subjects participating in the experiment and that subsequently, motor activity was
recorded and made visible by oscillographic traces. The obtained data show conclusive evidence of subphonemic errors, many of which would not have been noted
using the standard technique of transcribing errors. Most importantly, many of
the errors clearly indicate that subfeatural components, that is, individual muscular components of articulatory gestures, may be transposed in speech errors. For
instance, an [s] may be articulated with a strong labial component (typical of [])
but may still sound like [s]. Consequently, the change in muscular activity is only
revealed by the EMGrecording.
Based on their findings, Mowrey and MacKay criticize that little or no effort
has been made to address the problem of identifying the precise nature of the
errors accurately. Their results indicate that the problem of error characterization
is so pervasive as to render the significance of traditionally collected data corpora questionable. They conclude that it appears that errors which have been
consigned to the phonemic, segmental, or feature levels could be reinterpreted as
errors at the motor output level (Mowrey & MacKay1990:1311).
2.2.1.3Suprasegmentals
Finally, I want to briefly address the role of suprasegmentals in language production, such as stress, intonation, and tone. In early generative phonology (Chomsky
& Halle 1968), it was assumed that all phonological features including
suprasegmentals are features of segments (for example, [stress], [high tone]).
More recent phonological theories such as autosegmental phonology (Halle &
Vergnaud 1980; McCarthy 1981) and CV phonology (Clements & Keyser 1983),
however, posit separate tiers or levels for different types of suprasegmental features. On the basis of such theories, we predict that suprasegmentals are relatively
independent of the segments they are associated with and that segments, syllables, or words may be transposed without any effect on the associated suprasegmental features (Stemberger1984).
This prediction is in fact borne out. On the basis of speech error data, Fromkin
(1971) and Garrett (1975) both show that phrasal stress must be independent of
segments or even words, since there is no change in the stress pattern (or intonation
Grammar in use
contour) of the sentence when target units are disordered (also see Becker (1979),
who specifically argues in favour of an autosegmental representation of stress on
the basis of slip data). This phenomenon is illustrated by the examples in (7a) and
(7b). In (7a), two vowels are exchanged but phrasal stress remains in place (Garrett
1975:147), while in (7b), the stress pattern remains unaffected in a word exchange
(Fromkin 1973b:255).
(7) a. avoid the tre prening the tre prning
b. in her dll pper bx in her pper dll bx
c. kin khaaw lw kin khaaw lw
[] []
[]
[] []
[]
(Ive) already had dinner.
. Besides that, Stemberger (1984) presents evidence for the autosegmental treatment of
length (McCarthy 1981). Based on an investigation of German (from Meringer & Mayer
1895), Swedish (from Sderpalm 1979), and English speech errors, he shows that in a large
majority of the German (80%) and Swedish (83.3%) errors, in which long and short vowels
interact, the misordered vowel does not retain its original length, but takes over the length of
the vowel it replaces. This is illustrated by the German slip in (i), where the anticipated [y] is
short in its original location but long at its landing site.
(i) so sht [zy:t], sieht [zi:t] das Glck [glyk] der
Ehe
aus!
so (error) looks
the happiness
of.the marriage particle
Thats what the happiness of marriage looks like!
Things are different in English, however, where in 41 out of 42 errors, in which a long and a short
vowel interact, the vowel retains its original length, as is shown in the perseveration in (ii) and
the exchange in (iii) (Stemberger 1984:905); also see the German slip (2a) in Section 6.1.1).
Grammar as Processor
In summary of this subsection, we may note the following: there is a considerable amount of speech error evidence that proves that from a processing point of view the speech signal is not a continuous, indivisible sequence
but rather is divided into disrete units. It is, of course, intuitively clear that the
speech signal is made up of words and that words are made up of segments (and
possibly syllables). The error data, however, verify that at least some of these
units are not only convenient theoretical constructs but are actually processed
in on-line language production. The same is true for abstract entities such as
segmental and suprasegmental features. We may therefore conclude that these
units are psychologicallyreal.
2.2.2 Morphological complexity
Besides phonological structure, speech errors may also shed light on the processing
of morphological structure. In the psycholinguistic literature, it is a matter of debate
whether morphological information plays any role in producing or recognizing
words and whether a morphological level of processing is needed. In fact, it has
been argued that effects that are interpreted as the result of morphological relatedness might as well be attributed to semantic and/or phonological relatedness, since
most morphologically related words have both formal and semanticoverlap.
However, in many studies focusing on spontaneous speech errors, the psychological reality of morphological structure is taken to be evidenced by morphological speech errors, the most important type being the so-called stranding errors.10
The defining property of these errors is that only word stems are affected, while
other morphological material, such as derivational or inflectional affixes, remains
in its original position, that is, it is stranded in that position. Stranding errors
(as well as other morphological errors such as morpheme shifts) show that
morphological information can be accessed during processing and that single
morphemes can be detached and manipulatedindependently.11
Stemberger argues that length is suprasegmental and can therefore be dissociated from the
vowel to which it is linked. However, dissociability also depends on how cohesive sequences
in the language are; if a vowel and its associated structure are very cohesive, as in English, they
will rarely be dissociated in errors (Stemberger 1984:911).
. See MacKay (1979) for the discusson of various types of errors involving morphological
structure and abstract morphosyntactic features, such as affix substitutions, stem alternations,
misagreement errors, and regularization errors.
. Further evidence for an explicit level of representation for morphemes comes from experimentally induced speech errors (using the Word Order Competition technique). Pillon
Grammar in use
In this section, I will have a closer look at errors in which the morphological
structure of words plays a crucial role. I will start in Section 2.2.2.1 by considering
classical stranding errors in which different types of concatenative morphemes
are stranded in the error, also taking into account some language-specific phenomena. Stranding errors that involve the manipulation of non-concatenative
morphological structure, such as templatic morphology and reduplication, will be
subject to discussion in Section2.2.2.2.
2.2.2.1Stranding
As mentioned before, in a stranding error only part of a morphologically complex word, the word stem, is manipulated, while grammatical morphemes be
they derivational or inflectional remain unaffected. The German examples in (8)
illustrate some of the potential combinatory possibilities. Note that alle of the slips
in (8) are exchanges; however, stranding is also attested in anticipations, perseverations, and even blends. Moreover, it has been pointed out in the literature that
morphological errors involving derivational morphemes are less frequent than
inflectional errors (Fromkin 1973a; Garrett1980a).
In (8a), a derived adjective interacts with a compound. In the error, the word
stems Schreck (shock) and Schmerz (pain) are exchanged, while the derivational suffix -lich, the head of the compound Schmerzanfall (pain attack) and
the two plural suffixes remain in their original positions. Another combination
is illustrated by the slip in (8b). Again, two stems are exchanged the noun stem
Baum (tree) and the verb stem sterb (die) leaving behind one derivational
suffix (the non-category changing diminutive suffix -chen) and one inflectional
suffix (3rd person singular-t).12
(1998) and Melinger (2003) show for French and English, respectively, that morphological
errors are more frequent than phonological or syllable level errors in laboratory-induced settings. Their results support the inclusion of morpheme boundaries in the lexical representation of complex words and are best accounted for by models of speech production that
postulate a level of representation for morphemes.
. Note that in (8b), both the affected stems have been subject to a stem-internal change
(i.e., to a phonological readjustment rule) in their original positions: the diminutive triggers
umlaut in Baum (tree), while sterb (die) is subject to an ablaut rule in the 3rd person singular (see Section 3.2.3.2). Interestingly, the readjusted stems are exchanged and not their
base forms which would have given rise to mein Sterb-chen baum-t (see Section 6.7.1.1 for
further discussion of such stranding errors).
Grammar as Processor
ber die
Teile sprech-en, die ich kenn-e
about those parts speak-inf that I
know-1.sg
. Occasionally, however, a bound morpheme is shifted, as, for instance, the adverbial suffix
-ly in the English slip in (i) (Garrett 1980b:263) or the genitive -s in the German slip in (ii).
der Schrecken-s-herrschaft
the terror-gen-reign
Grammar in use
(Foldvik 1979:119). In the Turkish noun exchange (9b), the 1st person singular
possessive suffix as well as the locative suffix remain in their original positions.14
Finally, a possessive suffix, namely -ha (indicating 3rd person singular), is also
stranded in the Arabic utterance in (9c) (Abd-El-Jawad & Abu-Salim1987:153).
(9) a. ikke ei sol for
sky-a
ei sky for
sol-a
not one sun in.front.of cloud-art one cloud in.front.of sun-art
(There is) not a single cloud in front of the sun.
b. gne-im
yz-de
yz-m
gne-de
sun-poss.1.sg face-loc face-poss.1.sg sun-loc
My face is in the sun.
c. uxt zoo-ha
zoo
uxt-ha
sister husband-poss.3.sg husband sister-poss.3.sg
her sisters husband
While the suffixes involved in the above errors (definiteness, possessive, and
locative suffixes) may be language-specific, the type of morphological operation
involved, sequential affixation, is still the same as in the German examples in (8).
In the following section, I will turn to errors in which non-concatenative morphological processes play arole.
2.2.2.2Non-concatenative morphology
Even in languages that are predominantly characterized by sequential (concatenative) morphology, such as German and English, non-concatenative morphological
processes may be attested, sometimes in combination with sequential affixation
(see, for instance, the German example(8b)).
Waleschkowski (2004) and Hohenberger & Waleschkowski (2005) point out
that languages of the concatenative morphological type show a higher rate of
morphological errors than languages of the non-concatenative type (for example,
fusional languages or languages with simultaneous morphological processes, such
as, for instance, German Sign Language). In order to evaluate to what extent nonconcatenative morphemes are decomposable in on-line processing, they designed
an experimental task employing a repeat/reverse paradigm (Humphreys 2002).
Simplifying somewhat, in the experiment, subjects (n = 26) were asked to learn
short pairs of phrases by heart that contained singular and plural nouns, such
. Moreover, in this example, morphophonological accommodation of the stranded material plays a crucial role. For further discussion of morphophonological accommodation see
Section 6.1.2.
Grammar as Processor
as the ones given in (10).15 Note that in (10a), the plural is realized by a suffix
(concatenative morphology), while in (10b), it is realized by a stem-internal change
(umlaut, a non-concatenative morphological process). Following the target pair,
the subjects were presented two phrase pairs that primed the number pattern of
the target pair (for (10)singular-plural).
(10) a. das Auto reparier-en; die Lok-s
reinig-en
the car repair-inf; the locomotive-pl clean-inf
repair the car; clean the locomotives
b. den Vater ehr-en;
die Mtter
lieb-en
the father honour-inf; the mother.pl love-inf
honour the father; love the mothers
Under one experimental condition, subjects were asked to reverse the nouns of
the target pair. Hohenberger and Waleschkowski were primarily interested in the
distribution of root exchanges (that is, stranding of the number information) versus word exchanges. Both possible output patterns are given for the pair involving
concatenative plural morphology in (11a) and (11a) and for the pair involving
non-concatenative plural morphology in (11b) and(11b).
(11) a. die Lok-s
reparier-en; das Auto reinig-en
the locomotive-pl repair-inf; the car clean-inf
[word exchange]
a. die Lok
reparier-en; die Auto-s reinig-en
the locomotive repair-inf; the car-pl clean-inf
[root exchange]
b. die Mtter
ehr-en;
den Vater lieb-en
the mother.pl honour-inf; the father love-inf
[word exchange]
[root exchange]
Most of the output pairs in the reverse condition involved word exchanges
(67.2%); root exchanges were observed in only 10% of the output pairs.16 For the
. Actually, the experimental material comprised different morphological processes such
as inflection (e.g., tense, number), derivation (e.g., diminutive, nominalization), and compounding. Here, I only report on the part of the experiment involving plural morphology.
. Other output pairs involved word repetition (that is, cases in which the required exchange was not performed), affix exchanges, omissions, and others.
See Hohenberger & Waleschkowski (2005) for the results of a similar experiment involving German Sign Language pairs that include signs undergoing various concatenative and
non-concatenative morphological processes.
Grammar in use
root exchanges, a clear result was obtained with respect to the different types of
morphology: the regular concatenative plural inflection -s was stranded twice as
often as the irregular non-concatenative plural inflection. In other words: output
(11a) was more likely than output(11b).
The authors take this as evidence for the claim that regular morphology is processed differently from irregular morphology. In particular, they take their result
to support the dual-route model of morphology which assumes that regular forms
undergo decomposition in on-line processing, while irregular forms have separate lexical entries and are therefore not decomposed (see, for example, Pinker &
Prince (1994)). Still, the data show that occasionally, a non-concatenative morpheme can be detached from the root, too. This observation receives further support from spontaneous speech errors in which abstract morphemes are stranded
or shifted (see Sections 5.4 and 6.3 for furtherdiscussion).
Non-concatenative morphology does not always imply stem-internal
alternations (as in (10b) above). Particularly intriguing morphological error
patterns in which non-concatenative morphemes are affected have been
reported for Arabic. As is well-known, word formation processes in Arabic are
fundamentally different from word formation in the Germanic and Romance
languages. For a large part, morphological processes in Arabic are of a templatic (discontinuous) nature, that is, word stems are formed on the basis of
a triliteral root, a fixed sequence of three consonants, between which vowels
are inserted. The sequence of C- and V-positions (the CV-skeleton or prosodic template) is always fixed for a given derivation.17 The existence of such
a morphological pattern raises the question whether such discontinuous morphemes (transfixes) may also be subject to stranding in speech errors. The slip
in (12a) verifies that stranding of discontinuous morphemes is indeed attested.
What is exchanged in this example are merely the consonantal roots k-l-m and
s-- 1 and 4 in the illustration in (12b), while the stem-forming vowels
(2 and 5) are stranded (Abd-El-Jawad & Abu-Salim 1987:149). Moreover, we
observe stranding of the 3rd person singular possessive suffix (3) just as in
(9c) above (for discussion of Arabic slips also see Safi-Stagni (1990) and Berg &
Abd-El-Jawad(1996)).
. There are 15 different derivational classes (binyanim) for the triliteral roots which determine the arrangement of the root consonants with respect to the vowel positions. The precise
nature of these derivations is quite intricate and the reader is referred to McCarthy (1981) and
Spencer (1991:134ff) for details.
Grammar as Processor
(12) a.
b.
saaa-a
kaliim kalaam-ha
saii
(error)-poss.3.sg (error) talk-poss.3.sg correct
What she says is correct.
1
k a l aa m - ha
s a ii
The last type of non-concatenative morphological process I want to consider is reduplication. Reduplication, while clearly being a sequential operation, is still usually
considered non-concatenative since it does not involve the addition of a fixed suffix but rather (exact or inexact) copy of stem material (Marantz 1982). Stemberger
& Lewis (1986:151) take reduplication to be [o]ne of the more extreme forms of
phonologically dependent allomorphy. In a small-scale study, they address the role
of reduplication in experimentally elicited speech errors in Ewe, a Kwa language
spoken in Ghana and Togo. In Ewe, many inflectional operations are realized by
reduplication, one of these being the formation of verbal participles which involve
prefixation of a copy of the first stem consonant and stem vowel to the stem, as in, for
example, si (escape) sisi (escape.part) and dzra (sell) dzadzra(sell.part).
In particular, Stemberger & Lewis (1986) investigate whether in elicited consonant exchanges, the exchanged consonant occurs in the base and the reduplicant of the participle, only in the base, or only in the reduplicant. To that end,
they presented pairs of words to the subjects. All words were existing monosyllabic (CV) words in Ewe and every first element in a pair was a verb. As in the
Hohenberger & Waleschkowski (2005) study discussed above, a certain structural pattern was primed by three word pairs that preceded the target pair. In this
case, the pattern was a sequence of consonants. In (13), for instance, the primed
pattern is h-f. Subjects then saw a target pair with reverse pattern (f-h) and they
were asked to repeat the pair at the same time transforming the first member of
the pair into the participleform.
(13) a. priming pairs:
target pair:
correct response:
he
ho
he
fa
fafa
fu
fe
fi
ho
ho
Grammar in use
or anticipated consonant appeared in the base and the reduplicant of the first element (erroneous response for the list in (13): haha fo). In the other four cases, the
exchanged consonant appeared only in the reduplicated syllable, not in the stem of
the first element (erroneous response for (13): hafa fo). This is taken as evidence
for the fact that, for the most part, the error takes place at an early stage of processing, that is, before the stem is reduplicated.18 In other words: reduplication, just like
sequential affixation and other non-concatenative processes is psychologicallyreal.
Taken together, the above examples neatly illustrate that the processor has
access to morphological structure during language production. Inflectional as well
as derivational affixes may strand in an error and, moreover, language-specific word
formation rules influence the observed error patterns. Note that in later chapters
of this book, the notion of morpheme stranding will be refined (see Sections 5.4
and 6.3). In particular, adopting Distributed Morphology ideas, I will argue that
what is stranded is either an abstract morpheme (derivational morphology) or some
morphosyntactic feature (inflectional morphology). On the basis of this argument,
I will revise the claim that the language processor must have access to morphological structure. Moreover, stranding can occur before or after Vocabulary insertion
has taken place. Only in the former case, a grammatical outcome is guaranteed
(see Section 6.7.1.1 for furtherdiscussion).
2.2.3 Syntactic transformations
As pointed out above, Katz (1964) argues that the grammar plays a decisive role
in on-line language processing, that is, in parsing and language production. The
. The second most common type of error, the one which has the erroneous consonant only
in the reduplicant (hafa fo), is assumed to be the result of a late error occurring after reduplication has taken place. Stemberger & Lewis (1986) point out that the third type of conceivable
error, the one in which the anticipated consonant would appear only in the base but not in the
reduplicant (faha fo), is unlikely to occur because the reduplicant is not identical to the base.
It is worth pointing out, however, that this last type of error should be possible if it took place
after reduplication of the base. Just like hafa fo, it would constitute a late phonological error,
the only difference being that in hafa fo, the interacting consonants occupy a structurally parallel position, the word-initial position. In fact, it is quite likely that this structural parallelism
favours hafa fo over faha fo (remember that the correct response is fafa ho), since it is wellknow that in spontaneous speech errors, too, an error like the one in (i) is much more likely
than the (hypothetical) error in (ii).
(i) eine telbe
Gasche eine gelbe Tasche
a
(error) (error) a
yellow bag
(ii) eine gelte
Basche eine gelbe Tasche
a
(error) (error) a
yellow bag
Grammar as Processor
crucial role of the grammar is, of course, not restricted to phonological and morphological structure. Rather, the syntactic make-up of a sentence may also have an
important impact on its processing. In this context, one central question that arises
is: what kinds of syntactic constituents are actually processed? Not surprisingly, it
was demonstrated in a number of psycholinguistic studies that the sentence as well
as the clause that is, a part of a sentence that has a subject and a predicate are
major processing units. Jarvella (1971), for instance, was able to show that there is
a clause boundary effect in recallingwords.
One particularly controversial technique that was used to explore the size of
syntactic units in parsing is the so-called click displacement technique (Fodor &
Bever 1965; Garrett, Bever & Fodor 1966). The basic idea of this technique is that
major processing units resist interruption. The experimental setting was as follows
(see T.Harley 1995). Subjects heard speech over headphones in one ear that was
interrupted by extraneous clicks in the other ear at certain points in the sentence.
It was predicted that even if the click appears in the middle of a constituent, it
should be perceived as falling at a constituent boundary. This prediction was borne
out. For example, a click presented at point * in the sentence That he was* happy
was evident from the way he smiled, was reported by the listeners to appear after
the word happy. In other words, the click was displaced by the listeners in order to
maintain the integrity of the syntactic constituent. This result was taken to verify
that the clause is a major unit of syntacticprocessing.
Note that click experiments are post-perceptual. This, however, is not true
for spontaneous speech errors which therefore do not show a memory bias. Certain speech errors point to the psychological reality of phrasal units such as DPs
and VPs in language production. Consider, for instance, the two exchanges in
(14a) and(14b).
(14) a. I got into [this guy] with [a discussion] into a discussion with this guy
b. He [facilitated what he was doing] to [remove the barricade]
he removed the barricade to facilitate what he was doing
c. theres an [island] on the [small restaurant] a small restaurant on
the island
In (14a), two DPs are exchanged (Garrett 1980a: 192), while in (14b), two VPs
change place (Garrett 1980a:188). Note that in example (14b), the tense features
are stranded. Exchanges of intermediate projections are less frequent. Fromkin
(1988: 129) reports one such case, given in (14c), where small restaurant is an
N-constituent. Interestingly, in this error, both determiners accommodate to the
exchangedelements.
However, psycholinguistic investigations were not only concerned with the
psychological reality of syntactic constituents, but also with the influence of
Grammar in use
Sentence (15a) in which according to early generative theories no transformational rule has applied served as a baseline. It is possible to derive increasingly
complex sentences from this kernel sentence. Sentence (15d), for example, is
derived from (15a) by the application of three transformations: passivization, negativization, and question formation. Miller and McKean observed that the time
Grammar as Processor
needed to process sentences such as (15b), (15c), and (15d), that is, to detransform
them back to the underlying sentence (15a), was linearly related to the number of
transformations that are involved. This finding was interpreted as supporting the
psychological reality of transformational rules, since apparently transformations
have an impact on the processingcomplexity.19
In further investigations, however, these conclusions were called into question. In an experiment similar to that of Miller and McKean, Slobin (1966) examined the processing of reversible (16a) and irreversible (16b) passive sentences.
He found that Miller and McKeans results could only be obtained for reversible
passives. That is, the time needed to process related irreversible active (16c) and
passive (16b) sentences is approximately thesame.
(16) a. The ghost was chased by the robot.
b. The flowers were watered by the robot.
c. The robot watered the flowers.
This result implies that the DTC is not always true and that other factors than the
transformational complexity may also have an impact on the processingtime.20
Moreover, Fodor & Garrett (1967), were able to demonstrate that the processing of some transformationally complex sentences did not require the predicted
time. For example, sentences with double self-embedding as, for instance, The first
shot the tired soldier the mosquito bit fired missed are particularly difficult to process.
But obviously, the transformational derivation cannot be held responsible for this
difficulty, since it turned out that these sentences are easier to understand when
they are made grammatically more complex by the application of further transformational rules as, for instance, by the double application of a passive transformation, as in The first shot fired by the tired soldier bitten by the mosquitomissed.
2.2.3.2The psychological reality of empty elements
In later versions of the generative framework (Chomsky 1973, 1981), it is assumed
that in the transition from a deep structure to a surface structure representation, elements are moved leaving behind a trace in their original position. A
moved DP and its trace are said to be coreferential. Before turning to the processing of traces, I shall say a few words about the processing of pronouns in
on-linecomprehension.
. In other studies, it was shown that transformational distance between sentences also
predicts confusability between them in memory and that the ease of memorizing sentences
is predicted by the number of transformations that have applied to them (Mehler 1963).
. Also see Forster & Olbrei (1973) and Marslen-Wilson & Tyler (1980) for the interaction
of semantic/pragmatic and syntactic factors in language processing.
Grammar in use
Amongst other things, Cloitre & Bever (1988) examine whether the existence
of an anaphoric relation between a DP and a coreferential pronoun facilitates the
retrieval of an adjective that is part of the DP. In their experiment, participants
were presented with sequences of sentences like the one given in(17).
(17) a. [DP The skinny bellboy]i did a softshoe routine in the lobby.
b. The hotel guests were amazed by himi.
b. The hotel guests could not believe their eyes.
In one test version, participants were presented with the sentences (17a) and (17b)
in succession, in the other, with the sentences (17a) and (17b). In both cases, the
probe word was the adjective skinny that is part of the subject DP of the first sentence. That is, the participants were asked to decide as quickly as possible whether
the probe word was contained in the text. It turned out that the decision latencies
were shorter after presentation of sentence (17b) which contains the anaphoric
pronoun him. From this, Cloitre & Bever (1988) conclude that the implicit repetition of the DP containing the probe word by means of a pronoun allows for a faster
recognition of the probeword.
This observation leads to the question whether phonetically empty coreferential elements (as, for instance, traces) facilitate the retrieval of an antecedent
DP in the same way as anaphorical pronouns do. This question was investigated
in a study by Bever & McElree (1988). In this study, participants were to read sentences of the following types (note that no participant was presented with both
versions of onesentence).
(18) a. [DP The astute lawyer who faced the female judge]i was certain [tDP]I to
argue during the trial.
b. [DP The astute lawyer who faced the female judge] hated the long speech
during the trial.
In the raising construction (18a), the DP is moved from its underlying position to
the beginning of the sentence leaving behind a trace ([tDP]), while in the control sentence (18b), the same DP is not moved. As in Cloitre & Bever (1988), the probe word
was the adjective contained in the subject DP (astute in the above example). Bever &
McElree (1988) found that compared to the control sentence the trace in the raising construction did indeed facilitate the recognition of the probe word. Obviously,
coreferential empty categories behave in a way similar to anaphoric pronouns. The
authors take this to be strong evidence for the psycholinguistic reality of linguistically defined traces: when the trace is detected in on-line processing, its antecedent is
retrieved and the elements contained in the antecedent become moreaccessible.
Bever & McElree (1988) also investigated PRO constructions, such as the
one given in (19a). The results show that PRO constructions also elicited faster
Grammar as Processor
probe recognition times than the control sentences. The reaction time difference
between PRO and raising constructions was only marginally significant. In yet
another experiment, the authors considered passive constructions with a trace in
object position, such as the one given in (19b). Again, probe adjective recognition
times were faster following the passive than following the control sentence without
trace in (19c).21
(19) a. [The astute lawyer who faced the female judge]i strongly hoped PROi to
argue during the trial.
b. [The astute lawyer who faced the female judge]i was suspected ti by the boys.
c. [The astute lawyer who faced the female judge] had spoken to the boys.
In sum, the results from the comprehension experiments indicate that highly
abstract syntactic elements, such as empty elements, are psychologically really.
Just like overt anaphoric elements they prime (information contained in)
theirantecedents.
2.2.3.3Transformational errors
All of the results reported above were obtained in studies dealing with language
comprehension. But what about syntactic operations in language production? It
is clear that the process of language production is not as readily controlled in an
experimental setting, for instance, by means of reaction timeexperiments.
In this respect, too, spontaneous speech errors may possibly provide important
insights. If we think of a speech production device as applying transformations to
an underlying (deep) structure, we can make predictions about possible malfunctions of this device. In two studies, Fay (1980a,b) examines slips of the tongue that
he takes to result from the wrong or non-application of a transformational rule.
On the basis of his data, he hypothesizes that speech production involves a direct
realization of a transformational grammar (in the style of Chomsky(1965)).
Fay points out that in general, a transformation consists of three parts: a
structural analysis, a structural change, and conditions on the application of the
rule. He shows that each of these aspects is error-prone. Turning first to the structural analysis, he points out that this part of a rule may go wrong in three ways
(Fay 1980b). First, a phrase marker may be misanalyzed in a way such that the
rule applies when it should, but applies incorrectly. Second, a phrase marker may
be misanalyzed in a way such that an obligatory rule does not apply at all. Third,
the transformational device may misanalyze a phrase marker so as to allow a rule
. See MacDonald (1989), McElree & Bever (1989), and Bever, Straub, Shenkman, Kim &
Carrithers (1990) for further experiments dealing with this issue.
Grammar in use
to apply when it should not. A particularly illuminating example from his corpus
for the first type of error is the one given in(20a).
(20) a. Why do you be an oaf sometimes? Why are you an oaf sometimes?
b. deep structure:
wh-fronting:
*subject-aux inversion:
do-support:
morphophonemics:
I shall not discuss the third type of structural analysis error Fay suggests (that is,
application of a rule when it should not apply) and turn to the structural change
part of a transformation. Even when the production device has correctly analyzed a
phrase marker and performed a transformation, errors may still be made in carrying out a follow-up transformation. In particle movement, for instance, an element
is first copied into a new position and then the original is deleted. Fay assumes that
the error in (22) is due to an omission of the obligatory deletion operation. Consequently, the particle on appears twice in the erroneousutterance.
(22) Do I have to put on my seat belt on? Do I have to put on my seat belt?
. Cutler (1980a) points out that this error may also have a conceptual cause. She shows
that the intonation contour of the error is that of a yes/no question such as Linda, do you talk
on the telephone with your right ear? (the prosody of the error had been phonetically transcribed). It is therefore possible that (21) is a sentence blend, not a transformational error.
Grammar as Processor
Note that due to the fact that the rule of particle movement is optional, the slip in
(22) can also be analyzed as a blend of the two competing sentence frames Do I
have to put on my seat belt and Do I have to put my seat belton.
Particle movement, however, is obligatory when the object DP is a pronoun,
and this brings us to the third part of a transformation, namely the conditions on
transformations. Errors may also result from violations of such conditions. In the
following example, for instance, the condition on particle movement is not satisfied. According to Fay, the movement rule is treated as being optional (as in (22))
and therefore, the particle is not moved. In (23), however, the application of the
rule is obligatory due to the presence of a pronominal object (note that a sentence
like His secretary types up the manuscript is fullygrammatical).
(23) His secretary types up it His secretary types it up
In view of the data, Fay concludes that transformations, like other mental operations, may be subject to malfunction. Hence, the data support the claim that transformations are carried out as mental operations in speechproduction.
It should be emphasized, however, that for many of the errors discussed by Fay,
other non-transformational explanations are available (some of them might, for
example, be analyzed as blends or shift errors). Stemberger (1982a) criticizes Fays
analyses and tries to account for the properties of naturally occurring syntactic speech
errors in a model that does not assume transformational rules. In particular, he discusses syntactic errors in terms of the selection of phrase structures in an interactive
activation model. If a related phrase structure becomes overactivated, it may inhibit
the target structure and replace it, even if it is not the appropriate one. According
to Stemberger (1982a), this kind of phrasal substitution accounts for the erroneous
application of a transformation as well as for application failures (as in (20), (21),
and (23)). In contrast to that, phrasal blends can account for the apparent failure of a
part of a transformation (as in (22)). Stemberger therefore concludes that theories of
syntax that make no use of transformations but generate surface structure directly
also have the potential to be psychologically real models of syntacticprocessing.
2.3 Conclusion
In this chapter, I introduced some of the basic assumptions of a mentalistic linguistic theory. I pointed out that a mentalistic theory is not content with investigating
the observable linguistic behavior but rather takes into account mental processes
involved in the generation of an utterance. Most importantly, it is assumed that
a speaker has internalized a generative grammar that expresses his knowledge
of language and, moreover, determines and constrains his linguistic behavior.
Grammar in use
Therefore, the actual performance data may provide valuable evidence on the
basis of which hypotheses about linguistic structures and rules can betested.
In order to illustrate possible relations between grammatical constructs and
performance data, I presented results from a number of studies that investigate
the psychological reality of such linguistic structures and rules. As it turns out,
speech error data supply valuable evidence for the on-line processing of phonological units such as abstract features, segments, and suprasegmentals. Moreover,
they provide evidence for the psychological reality of morphological units no
matter whether they are concatenative or non-concatenative in nature and
phrasal units. The situation, however, is not as clear with respect to syntactic
operations. While results from comprehension experiments indicate that empty
categories (traces) are in fact processed in on-line comprehension, most of the
available production data furnish only weak evidence for the psychological reality
of syntactictransformations.
chapter 3
Theoretical background
Having established that spontaneous speech errors constitute valuable evidence
when it comes to testing hypotheses concerning the psychological reality of linguistic representations and processes, I shall now introduce the basic properties of
the psycholinguistic and the grammar model I adopt, that is, the multi-level model
of language production (Section 3.1) and Distributed Morphology (Section 3.2),
respectively. In both sections, I will focus on characteristics of the models that will
turn out to be important for the analysis of error data from my corpus in subsequent
chapters, while neglecting those details that are not directly relevant in the present
context. Remember that ultimately, it is my goal to evaluate to what extent the time
course of the (morpho)syntactic derivation as assumed in Distributed Morphology
can be mapped onto the time course of processing as assumed in multi-level models of language production or viceversa.
Grammar as Processor
MESSAGE SOURCE
(conceptualisation)
potential levels of
psychological representation
for sentences
translation
processes
(formulation)
?
instructions to articulators
ARTICULATORY SYSTEM
(utterance of a sentence)
This sequence of processing stages from intention to articulation that is, thought
before utterance is probably intuitively plausible. However, in a series of papers,
Garrett (1975, 1976, 1980a, 2000) argues that the formulation (translation) process,
too, can be subdivided into a series of distinct levels of processing (the potential
levels of psychological representation in (1)). Crucially, processing in the model
occurs in a strict temporal order: the sequence of levels is fixed and the flow
of information is strictly top-down. Therefore, the operations executed at each
level cannot influence each other, that is, there is no interaction between levels
in the sense of an exchange of information. Moreover, every level uses its own
Theoretical background
seltenen Flle
rare
cases
Garrett (1980a) notes that these two properties phrasal membership and grammatical category are clearly related to each other: the likelihood of correspondence
of grammatical category is affected by whether the error is phrase-internal ornot.
The distributional features of exchange errors suggest an interesting interpretation, namely that they arise at different processing levels. On the one hand, word
exchanges must occur at a point at which the syntactic category of units is part of
the processing vocabulary.1 Moreover, the fact that the exchanged elements may
appear at some distance from each other suggests that the processing domain is
. See Section 6.4.4 for a different account which in the spirit of DM does not involve syntactic categories. Crucially, in DM, categorial membership cannot be a determining factor since
the elements that are manipulated in the syntax are acategorial in nature (see Section 3.2.2).
Grammar as Processor
the whole sentence. Garrett calls this stage in the processing of an utterance the
functional level. On the other hand, since sound exchanges are not subject to
the syntactic category constraint, they are assumed to occur at a point at which
the syntactic category of elements is not part of the processing vocabulary. Rather,
phonological (and morphological) properties of words are processed phrase by
phrase, which also explains why, for the most part, segment exchanges occur
phrase-internally. In Garretts sentence production model, this processing level is
called positionallevel.2
3.1.2 Lexical retrieval
Another central property of the model concerns the lexical retrieval of items in
production. The retrieval of lexical items is the process in speech production
whereby a speaker turns thoughts/concepts into sounds. With respect to lexical
retrieval, two important questions arise. First, how many stages are involved in
lexical retrieval? And second, are these stages independent, or do they interact
with eachother?
On the basis of characteristic properties of whole word substitutions, many
psycholinguists have argued for a two-stage theory of lexical retrieval, where
the first stage is semantically organized, while the second one is based on phonological properties (for example, Fromkin 1971; Garrett 1975, 1980a; Dell &
Reich 1981; Butterworth 1989; Levelt 1989, 1992). The different types of substitutions are illustrated by the noun substitutions in (3) two each from German
and English. In (3a), the intended noun Kalorien (calories) and the erroneous noun Alkohol (alcohol) are semantically related; the same is true for the
planned item collar and the intruder belt in (3b) (Fromkin 1973b:262). In contrast, the nouns Phrase/Phase (3c) and apartment/appointment in (3d) (Garrett
1980a:207) share phonological (word onset, number of syllables) and possibly
morphologicalproperties.
. Further evidence for the division of separate levels within the translation process comes
from stranding errors like the one in (i) (Fromkin 1973b:259).
Obviously, in this error, the morpheme maniac has been accessed independently of its plural
affix, that is, the plural information is stranded (see Section 5.4.1). Furthermore, weekend
appears with the appropriate plural allomorph [z], not with [s], as would be appropriate in
the intended environment. This morphophonological accommodation (see Section 6.1.2) to
the post-error environment suggests that the exchange takes place before the plural suffix is
phonologically specified.
Theoretical background
so viel
Alkohol
so much alcohol
haben
have
so viele Kalorien
so many calories
FORMULATOR
grammatical encoding
(functional level)
phonological encoding
(positional level)
ARTICULATOR
LEXICON
lemmas
forms
Grammar as Processor
With the help of the double retrieval theory, the different kinds of word substitutions are readily explained: meaning-based errors arise in lemma selection,
while form-based errors arise in lexeme selection.3 However, as will become
clear in the next section, the assumption of a strict division of tasks has not
remainedunchallenged.
3.1.3 Flow of information
Remember that Garrett assumes a strict separation of processing levels. That
is, the phonological specification of an utterance (at the positional level) only
begins after lemma retrieval (at the functional level) has been completed.
Therefore, in his model, phonological processes cannot have any influence
upon lemma selection and consequently, semantic and phonological processes
should not overlap in lexical retrieval. As a consequence, we expect to find
meaning-based substitutions and form-based substitutions (as in (3)) but not
mixed errors in which the intruding element is both semantically and phonologically related to the target. Various authors, however, have argued that
semantic substitutions are facilitated by phonological similarity (Dell & Reich
1981; Butterworth 1982; T.Harley 1984, 1993; Dell 1986; Cutting & Ferreira
1999; Harley & MacAndrew 2001). Possible examples are given in (5). Note
that the phonologically related nouns in (5a) both refer to time concepts, while
the ones in (5b) are both related to music (Fromkin 1973b:262).4
(5) a. am
nchsten Monat, h, Montag
on.the next
month, er, Monday
next Monday
b. were playing the art of the flute of the fugue
Theoretical background
result in existing words more often than is expected by chance (for instance, the
existing word reel in the sound exchange reel feally bad feel really bad; also see
some of the phonological errors in Section 2.2.1above).
The authors conclude that the processing levels cannot be completely independent of each other and that therefore, the independent level hypothesis has to
be rejected. The simplest solution, they claim, is to allow interaction between the
levels. Feedback between the phonological level and the lemma level explains the
tendency of a combined influence on substitution errors as well as the lexical bias
in sound errors. Production models which endorse feedback between the processing levels are called interactive activation or spreading activation models
(Stemberger 1985; Dell 1986, 1988; Berg1988).
Interactive activation models, too, discriminate different processing levels,
but the independence of these levels is undermined by ubiquitous feedback. Still,
the flow of activation in these models is not completely unconstrained, that is,
there is no absolute but only relative interactivity. Hence, the planning process
is organized in a weakly heterarchical fashion. The flow of information may very
well proceed in both directions but obviously, the flow towards the articulator
has the greatest share in speech production (Schade1992).5
In the following, I do not intend to argue for or against one or the other
type of model, since the question of strictly serial versus interactive processing is not at issue here. What I take to be particularly important in the present
context is that both the modular as well as the spreading activation architectures
endorse processing at several levels with distinct information being available at
eachlevel.
Grammar as Processor
3.1.4 Summary
Most language production models no matter whether they adopt serial or interactive processing postulate specialized processing levels that mediate between a
message intention and the articulation of an utterance. In addition, it is assumed
that the lexicon is accessed twice during production: first to retrieve lemmas, then
to retrieve phonologically specified word forms. Models differ from each other,
however, with respect to the type of information flow they allow. While serial
models endorse strict top-down processing, interactive models allow for feedback
between processing levels. Garrett (1980a:190) notes that the terms which find
use in his production model (that is, functional level and positional level) are
studiously neutral with respect to their correspondence to levels of description
in a formal grammar. Still, if we take the idea of a psychologically real grammar
(as decribed in Section 2.1) seriously, then it is of course desirable to have at ones
disposal a formal model of grammar that can be mapped onto the psycholinguistic
processingmodel.
As will become clear in the next section, DM also supports a separationistic
view. In particular, advocates of DM assume that the mechanisms which are
responsible for producing the form of syntactic expressions are distinct from
the mechanisms which produce the form of the corresponding phonological
expressions. I therefore take it to be a promising endeavour to try to relate
the psycholinguistic theory of multi-stage sentence processing to a multi-level
theory ofgrammar.
Theoretical background
Grammar as Processor
in Andersons theory, any parallel between the layering of syntax and the layering of phonology is merely accidental. Therefore, rather than to eliminate all
affixes from morphology, they choose to redefine the notion ofmorpheme.
On the other hand, Halle and Marantz also criticize Chomskys (1995) suggestion that affixation takes place in the lexicon prior to lexical insertion and
that the interface between a verbs internal morphological structure and the
syntax involves a system of feature checking. According to Chomsky (1995),
and many others scholars working within the Minimalist Program, the verb
raises to various functional heads in the syntax and checks its features with
the features of the functional heads to which it adjoins. Affixation and checking, however, are independent of each other. Most importantly, affixation in the
lexicon does not impose any particular structure on the organization of features and consequently, no particular order is imposed on the checking operations. Therefore, in Chomskys framework, the connection between the internal
phonological structure of the verb and its inflectional features (to be checked
off in the course of the derivation) has no consequences for the rest of
thegrammar.
In order to overcome the shortcomings of the affixless and the checking theory
sketched above, Halle and Marantz propose DM as an alternative account. In the
remainder of this section, I will introduce some of the basic assumptions of DM,
highlighting those aspects of the theory which will be relevant for the discussion
of the speech error data in Chapters 4 to 6, while neglecting details which are not
of immediate importance in the present context. I start in Section 3.2.1 by looking at the structure of the grammar and the role of the lexicon as assumed in DM.
In Section 3.2.2, I turn to the post-syntactic level of Morphological Structure and
discuss the morphological and morphosyntactic operations that are taken to apply
at this level. The processes that are at work at the level of Phonological Form are
introduced in Section3.2.3.
3.2.1 The structure of the grammar
DM is separationistic in nature in that it adopts the idea that the mechanisms which
produce the form of syntactically and semantically complex expressions are separate
from the mechanisms which generate the form of the corresponding phonological expressions. One of the core assumptions of DM is that syntax proper does not
manipulate anything resembling lexical items, but rather generates structures by
manipulating and combining morphosyntactic features by means of various syntactic operations (such as merger and movement). The model of grammar as adopted
in Halle & Marantz (1993) and subsequent work (see, for example, Harley & Noyer
(2003:465)) is sketched in Figure(6).
Theoretical background
(6)
List 1
syntactic operations
(e.g. merger, movement)
MORPHOLOGICAL
STRUCTURE (MS)
LOGICAL
FORM (LF)
PHONOLOGICAL
FORM (PF)
Conceptual
interface
(Meaning)
Phonetic interface
morphological operations,
(e.g. merger, fusion),
adjunction of Agr nodes,
morpheme insertion
insertion of Vocabulary
items (spell-out),
readjustment rules,
phonological rules
List 2
List 3
Halle & Marantz (1993:111f) state that the name Distributed Morphology was
chosen in order to emphasize the fact that the machinery of what traditionally has
been called morphology is not concentrated in a single component of the grammar, but rather is distributed among several different components. Consequently,
word formation may take place at any level of grammar through different processes such as head movement and/or merger of adjacentheads.
At the level of Morphological Structure (MS), which is taken to be the interface between syntax and phonology, various operations may change the structure
and/or number of terminal nodes (see Section 3.2.2). It is only after syntax at the
level of Phonological Form (PF) that phonological expressions, called Vocabulary
items, are inserted in a process called spell-out. Vocabulary insertion is therefore
referred to as late insertion.6 For a given Vocabulary item to be inserted in some
. This is somewhat different from a proposal made in Halle (1990, 1994). Halle distinguishes between concrete morphemes and abstract morphemes. Concrete morphemes are
those that have a single fixed underlying phonological representation (identifying index):
morphemes such as the stems dog, eat, and red, but also bound affixes like, for instance, un- or
-ness. In contrast, morphemes that do not have a fixed phonological shape are referred to as
abstract morphemes. They differ from concrete morphemes in that they lack a phonological
Grammar as Processor
terminal node at PF, none of its morphosyntactic features may conflict with a morphosyntactic feature present in that node. Again, certain operations may apply, at
this time altering the phonological shape of already inserted Vocabulary items (see
Section3.2.3).
In DM, there is no lexicon in the sense familiar from earlier generative models. Rather, there is a number of distributed lists which take over the jobs assigned
to the lexicon component in earlier theories (Marantz 1997). In the syntax, the
terminal nodes are purely abstract; they consist only of abstract roots (root) and
features that actually play a role in the syntactic computation.7 Hence, features
that are only relevant at the phonological or semantic interface are not part of the
terminal nodes in the syntax. Morphosyntactic features which are relevant to the
computational system, such as, for instance, [definite], [plural], and [1st],
are drawn from a feature set which is made available by Universal Grammar. In a
sense, List 1 in (6) most directly replaces the traditional lexicon in that it provides
the units on which the syntax operates: abstract roots as well as bundles of grammatical features. List 1 (which Marantz refers to as the pure or narrow lexicon)
is generative in nature in that sets of grammatical features may be freely formed,
subject only to combinatorial constraints about featureco-occurrences.8
List 2, the Vocabulary, provides the phonological forms for the terminal nodes
from the syntax. Thus, the Vocabulary establishes a connection between sets of grammatical features and phonological features. The Vocabulary is non-generative in
nature but expandable. Actually, the Vocabulary items come very close to matching
usual notions of morphemes. Note that Vocabulary items may be underspecified with
Theoretical background
respect to the features of the syntactic position in which they can be inserted. Moreover, various items may compete for insertion at a given terminal node, with the most
highly specified item that does not conflict with the feature specification of this terminal node winning the competition. Vocabulary items connect the paradigmatic space
defined by the features of a given terminal node to phonological form by mentioning
some of these features and by describing how they relate to phonologicalfeatures.
The third and final list in (6) is List 3 or the Encyclopedia. The Encyclopedia
lists the special meanings of particular roots. In other words, it relates Vocabulary
items to meanings. In DM, the Encyclopedia is taken to be the list of idioms of
a given language. Note that the term idiom embraces more than the conventional use of the term. It is used to refer to any expression (even a single word or
subpart of a word) whose meaning is not wholly predictable from its morphosyntactic structural description (Marantz 1997, 1998). Just like the Vocabulary, the
Encyclopedia is non-generative but expandable. Two examples one single word
idiom and one phrasal idiom are given in(7).
(7) a. cat:
b. (rain) cats and dogs:
Grammar as Processor
with linguistic and non-linguistic lexical knowledge stored in different lists. In the
following sections, I am going to have a closer look at the operations which apply
at the post-syntactic levels of Morphological Structure and PhonologicalForm.
3.2.2 Morphological Structure
In DM, every expression acquires at least two structural descriptions in the course
of its derivation. In a morphosyntactic description, the morphemes that are part
of an expression and their arrangement are indicated (using elements drawn from
List 1); the morphosyntactic description of, for example, dogs would look like
[root [+pl]]. In a morphophonological description, the phonological pieces of
an expression (its Vocabulary items taken from List 2) areindicated.
The morphosyntactic structure of an expression is generated by several
mechanisms. In the syntax, for instance, head movement plays a crucial role
in constructing complex morphosyntactic structures (for instance, movement
of a verbal root through various functional heads). At the post-syntactic level
of Morphological Structure (MS), several additional structure-changing mechanisms may apply; these mechanisms will be subject to closer examination in the
following twosections.
3.2.2.1 Morpheme types and local licensing
Harley & Noyer (2003:467) note that in DM, the term morpheme properly refers
to a syntactic (or morphological) terminal node and its content, not to the phonological expression of that terminal, which is provided as part of a Vocabulary
Item. As described in detail before, the content of a morpheme which is active
in the syntax consists of morphosyntactic (and compositional semantic) features
drawn from a set made available by Universal Grammar. In this section, I will
consider possible distinctions between morpheme types which may be relevant
for the computational system and/or Vocabulary insertion in order to yield a correct (or desired)output.
Traditionally, many (morpho)syntactic theories rely on the assumption that
syntactic categories such as N, V, and A each possibly being a combination of
features of a more abstract kind (Chomsky 1970) are labels associated with
lexical items and that these categories are also essential in building syntactic tree
structures following the X-bar scheme (Jackendoff 1977). That is, nouns which are
inserted into a terminal node project a noun phrase, verbs project a verb phrase,
and so on. DM departs from this view. Harley & Noyer (1998a), for instance, suggest that morphemes are of two basic kinds: f-morphemes and l-morphemes, a
distinction that corresponds approximately to the well-known division between
Theoretical background
b. Die
Kind-er rger-te-n
ihre
Mtter
the.pl.nom child-pl tease-past-pl poss.pl.acc mother.pl
The children teased their mothers.
Grammar as Processor
a.
vP
(agent) DP
v
v
[]
b.
LP
l-node
licensing
DP
LP
D
[+]
DP
licensing
l-node
(PP)
Since l-nodes lack a categorial specification, the phrase they project is labeled LP.9
In both structures in (9), the l-node hosts the same root. In (9a), the verbal status of
brech (break) is the result of inserting a Vocabulary item into a terminal l-node
which is governed by v (note that the licensing relation is indicated by an arrow).
In (9b), the nominalization of the same root is the result of inserting the same
Vocabulary item into an l-node which is governed by D.10 Moreover, in (9b), the
Vocabulary item is subject to a stem-internal change (ablaut); see Section 3.2.3.2
for furtherdiscussion.11
. In (9a), a split-VP approach like the one assumed in, for example, Chomsky (1995) is
adopted. In this approach, (agentive) external arguments are taken to be base-generated in the
specifier of a light verb phrase (vP) which is projected separately from a lower basic VP.
. See Section 6.4 for the licensing of roots that surface as adjectives.
. The idea of acategorial roots is also adopted in Marantz (2001); his implementation,
however, is different. According to Marantz, the construction in which a root occurs is assigned a category through merger with a category node (a head) called little x, in which
x can be verbal (little v), nominal (little n), or adjectival (little a). Little x determines the
edge of a cyclic domain at which a derivation is shipped off to PF and LF. In accounting
for the speech errors in which local licensing comes to fruition (Section 6.4), I will follow
Theoretical background
Let me add a few words about the content of the light verb head. Actually,
the light verb head is a functional head with a very limited inventory of meanings. H.Harley (1995), for instance, maintains that v may only have three different specifications, namely be (stative), cause, and become. In (9a), the l-node
combines with the cause morpheme to yield the transitive verb brech (break).
The role of the light verb in the derivation can be neatly illustrated by the German
verb pair senken (to lower) versus sinken (to drop, to sink), where the first one
is transitive and the latter one intransitive and unaccusative, as illustrated by the
examples in(10).
(10) a. Der Hndler senk-t
die Preise
the dealer
lower-3.sg the prices
The dealer lowers the prices.
b. Die Preise sink-en
the prices drop-pl
The prices drop.
In (10a), as in (9a) above, the l-node is licensed by a cause morpheme in the head
of vP and a transitive verb is produced; the agentive argument dealer occupies the
specifier position of vP. A different relation between v and the l-node holds in the
intransitive sentence in (10b). Following Harleys (1995) proposal, we may assume
that in this example, the light verb heading vP must be become, with no agentive
external argument in SpecvP. Presumably, the internal argument is base-generated
within VP. Bracketed structures for the two examples are given in(11).
(11) a. [vP [SpecvP der Hndler] [v cause [LP gesenkt [DP die Preise]]]]
b. [vP [SpecvP ] [v become [LP gesenkt [DP die Preise]]]]
In the transitive as well as in the intransitive case, the same l-node, designating the
resultant state (gesenkt (lowered)), combines with the little v morpheme in order
to produce the final verbal form. Thus, from a given numeration of initial bundles
of features and roots, the syntax creates legitimate structures, which are then filled
with appropriate Vocabulary items. As we have seen, the same Vocabulary item may
surface in different morphological categories depending on the syntactic context
in which the corresponding l-morpheme (or root) appears, that is, depending on
the f-morpheme by which it is licensed (see Section 6.4 for furtherdiscussion).
the ideas as formulated in Harley & Noyer (1998), implicitly assuming that the data could
as well be accounted for in Marantzs little x theory. I will, however, come back to Marantz
proposal in Section 6.4.2, where I consider morpheme insertion, and in Section 6.4.6,
where I discuss modifications concerning phonological readjustment and licensing as suggested in Siddiqi (2006).
Grammar as Processor
Theoretical background
Hence, merger exchanges a structural relation between two elements at one level
of representation for a different structural relation at a subsequent level. The structures in (12) illustrate the application of thisoperation.
(12) a.
XP
X
XP
b.
YP
YP
Y
ZP
Y
Y
ZP
X
In order to further clarify the mechanism, let us consider an English example. For
English, it has been claimed that main verbs do not overtly raise to Tense. Still,
verbs do inflect for tense in sentences like John opened the letter. The joining of
Tense with the main verb has sometimes been attributed to a lowering operation
(Pollock 1989). However, it is argued in Halle and Marantz (1993) that this joining
is in fact an example of merger under structural adjacency. The result of merger
Tense being affixed to the l-node containing open is shown in(13b).12
(13) a.
TnsP
Tns
TnsP
b.
LP
LP
[+] L
DP
DP
Tns
[+]
Another source of the mentioned lack of isomorphism between PF and the syntax
is the fact that additional morpheme nodes may be inserted at MS. For instance, in
. Given that English main verbs are incapable of raising to Tns, it is also impossible for
them to further raise to C in yes/no-questions; cf. the ungrammaticality of *Opened John the
letter?. In this case, the dummy element do will be inserted under Tns and will raise to a [+Q]marked complementizer, resulting in Did John open the letter?. In contrast to English, German
main verbs may raise and adjoin to Tns and in a second step to C.Hence, the inflected verb
appears in sentence-initial position in German yes/no-questions. Another important difference to English is that in German, Tns and AgrS will not fuse, since person/number distinctions are maintained in the past tense.
Grammar as Processor
(some versions of) DM, it is assumed that in many languages, subject-verb agreement is implemented by adjoining an agreement morpheme to the tense node
(Halle & Marantz 1993; Embick 2000). Subsequently, features of the subject DP
are copied onto this agreement node. In a similar fashion, case-number-gender
concord within German DPs is implemented by supplying appropriate agreement
suffixes to adjective and determiner nodes and by copying features associated with
the nominal head of the DP onto them. By inserting these morphemes only at
MS, that is, after syntax but before spell-out, their lack of effect in the syntax is
accounted for (see Marantz (1991) for the determination and realization of case
features at MS). The insertion of the AgrS morpheme onto which the appropriate
features of the subject have been copied transforms the relevant part of the tree in
(13b) into the tree in(14b).
(14) a.
b.
Tns
[+]
Tns
Tns
Agr
[+]
[3.]
At this stage of the derivation, a third structure-changing operation may come into
play. This is the operation fusion which takes two terminal nodes that are sisters
under a single category node and fuses them into a single terminal node. After
fusion has applied, only one Vocabulary item may be inserted and this item must
have a subset of the morphosyntactic features of the fused node (which includes
features from both input terminal nodes). Hence, unlike merger, fusion reduces
the number of independent morphemes in a syntactic structure. For English, for
instance, it is assumed that Tns and Agr fuse into a single morpheme at MS (indicated by the broken circle in (14b)), since there are no number/person distinctions
in the pasttense.13
. Two further mechanisms that contribute to the noted lack of isomorphism between the
syntax and PF are impoverishment and morpheme fission. Impoverishment (Bonet 1991) is
an operation on the content of morphemes which involves the deletion of morphosyntactic
features in certain contexts. In contrast, fission is capable of splitting a single morphosyntactic
terminal node into two (or more) phonological pieces to account for situations in which a
single morpheme may correspond to two separate Vocabulary items (see Halle (1997) and
Noyer (1997, 1998) for details).
Theoretical background
Grammar as Processor
Insertion takes place. The insertion of Vocabulary items will be discussed in the
followingsection.
3.2.3 Phonological Form
At the level of Phonological Form, Vocabulary items are inserted into abstract morphosyntactic structures. In addition to that, phonological readjustment rules may
alter the phonological structure of Vocabulary items in the context of certain features.
The process of Vocabulary insertion will be sketched in Section 3.2.3.1, while the role
of phonological readjustment rules will be subject to discussion in Section3.2.3.2.
3.2.3.1 Vocabulary insertion
A Vocabulary item is not merely a phonological string; rather, it also contains
information about where that particular string may be inserted. At the level of
PF, Vocabulary insertion (spell-out) inserts Vocabulary items into terminal nodes
which contain abstract roots and/or morphosyntactic features. In the unmarked
case, the relation between Vocabulary items and morphemes is one-to-one. However, as discussed above, several factors may disrupt thisrelation.
Regardless of the type of morpheme l-morpheme or f-morpheme spellout involves the association of phonological pieces with abstract morphemes. In
the case of f-morphemes, sets of Vocabulary items compete for insertion into a
given node. For insertion to take place, only a subset of the features specified in
the terminal node must be matched by features of the corresponding Vocabulary
item. Insertion of phonological strings into abstract morphemes is governed by
the Subset Principle (Halle1997:428).
Subset Principle
The phonological exponent of a Vocabulary item is inserted into a morpheme
in the terminal string if the item matches all or a subset of the grammatical
features specified in the terminal morpheme. Insertion does not take place if the
Vocabulary item contains features not present in the morpheme. Where several
Vocabulary items meet the conditions for insertion, the item matching the greatest
number of features specified in the terminal morpheme must be chosen.
The Subset Principle determines which Vocabulary item wins the competition for
insertion: items that match more features take precedence over items that match
fewer features. That is, the application of a given spell-out rule bleeds the application of all spell-out rules ordered belowit.
With this in mind, let us now come back to the English example in (8a):
The men sold a house. The relevant morphosyntactic features in this example
are [+past], [def], and [+pl]. In specific environments, that is, with certain
stems, [+past] is spelled out as zero, as indicated in (16a). However, in the
Theoretical background
remaining (or elsewhere) environments, the default affix /-d/, which does not
require a specific context, will be inserted (16b). Hence, as far as [+past] is
concerned, the insertion of the zero affix bleeds the insertion of /-d/, a property
which accounts for the fact that in English doubly marked past tense forms
areungrammatical.15
(16) a. [+past] / X + ____
(where X = drive, sing, beat, )
b. [+past] /-d/
c. [+pl]
/ Y + ____
(where Y = man, sheep, moose, )
d. [+pl]
/-()z/
The insertion of the plural suffix proceeds in a similar fashion: in a specific environment, for instance, with the noun man, a zero suffix will be inserted (16c), in other
environments other suffixes may be chosen, and in the remaining cases, the default
suffix /-()z/ will be inserted (16d). Finally, the insertion of the definite article
// and the indefinite article // in terminal nodes containing the feature [def] is
straightforward because no competition amongst Vocabulary items isinvolved.
In the German example (8b), Die Kinder rgerten ihre Mtter (The children
teased their mothers), the morphosyntactic features which need to be spelled
out by f-morphemes are [+pl] (on nouns and verbs), [+def], [+past], [+poss],
as well as case features. Remember that the plural feature on the nouns (Kinder,
Mtter) is chosen from List 1, while the plural feature on the verb is copied onto
the AgrS node at MS. For German nouns, the choice of plural suffixes is considerably larger than in English. In (17), I give the Vocabulary items for three
of the available plural allomorphs: two suffixes which require a specific environment in (17a) and (17b) and the one which many researchers (for example,
Janda 1990; Marcus, Brinkmann, Clahsen, Pinker & Wiese 1995; Wiese 1996)
take to be the default plural marker in (17c).16 As in English, for certain German
. According to Halle & Marantz (1993), the Tense node contains the features [+past] and
[part]. This implies that the Vocabulary items in (16a) and (16b) are underspecified with
respect to the terminal node in which they are inserted, since they are not specified for the
feature [part].
. Note that for German common nouns, -s is a very infrequent plural marker. It is suffixed to
a few consonant-final nouns, e.g., Park-s (parks), and to a slightly larger group of nouns ending
in a vowel, e.g., Bro-s (offices), Kino-s (cinemas). Moreover, personal names, e.g., Mller-s,
Katharina-s and abbreviations, e.g., CD-s, LKW-s (trucks), take -s as the plural marker.
Wiese (1996) points out that the importance of this suffix lies precisely in the fact that
it acts as a default suffix in spite of its low frequency. It is, for instance, overgeneralized in the
Grammar as Processor
verbs, the past tense suffix is zero (17d). The Vocabulary item for the default past
tense morpheme is given in (17e). Note that some of the suffixes given in (17)
are accompanied by stem-internal phonological changes: umlaut or ablaut. Such
vowel changes will be discussed in the followingsection.
(17) a. [+pl]
/ X + ____
(where X = Mutter, Vater (father), )
b. [+pl]
/-r/ / Y + ____
(where Y = Kind, Haus (house), )
c. [+pl]
/-z/
d. [+past]
/ X + ____
(where X = singen (sing), geben (give), )
e. [+past]
/-t/
f. [+poss]
/i:r/
[+pl][acc]
For the possessive pronoun ihre (their) to be spelled out correctly, further operations must have taken place prior to Vocabulary insertion. Firstly, the number
feature of the noun must have been copied onto D; secondly, a case (accusative)
feature must have been assigned (17f). I will leave open the question whether the
pronoun is possibly underspecified for one of these features (as it is, for instance,
for the genderfeature).
As opposed to the f-morphemes discussed above, for l-morphemes there is
a choice regarding the Vocabulary item to be inserted. In an appropriate local
(licensing) relation to a determiner, a root morpheme might, for example, be filled
by cat, girl, table, or any other Vocabulary item which we would call a noun. These
Vocabulary items are not in competition; rather, they may freely be inserted at
the point of spell-out (see Section 4.1 for discussion). Two exemplary Vocabulary
items for l-morphemes are given in(18).
(18) a. house /has/
b. kind /kInt/
acquisition of German plurals by children and it is also used by adults when asked to produce
the plural of a nonsense word.
Theoretical background
into the l-node. However, Marantz argues that senken is only a morphophonological variant of the basic intransitive root. Consequently, this apparent l-morpheme
alternation is not determined by competition of Vocabulary items but should
rather be seen as the product of post-insertion readjustment rules which will be
discussed in the nextsection.
3.2.3.2 Phonological readjustment
In DM, two sorts of readjustment rules are distinguished. Morphosyntactic readjustment rules, on the one hand, are capable of manipulating morphosyntactic
features in the context of other such features. As already mentioned in Footnote
13, these rules are called impoverishment rules in case they delete a certain feature.17 Clearly, morphosyntactic readjustment rules must apply prior to Vocabulary insertion, since the morphosyntactic features of Vocabulary items which are
inserted at PF are required to be non-distinct from those contained in the (already
readjusted) terminal nodes. Phonological readjustment rules, on the other hand,
change the phonological form of already inserted Vocabulary items and logically
follow Vocabulary insertion. In this section, I will only be concerned with the latter type of readjustmentrules.
In many cases, the information contained in the Vocabulary entries is not
sufficient to guarantee the generation of the correct phonological output. It is,
for instance, not at all unusual for affixation to be accompanied by stem-internal
modifications. As suggested in Halle (1990, 1994), the remaining part of the information concerning the phonological form of morphemes is provided by a set of
phonological readjustmentrules.
Again, we use the examples in (8) to illustrate the mechanism. In the English
sentence (8a), the verb sell undergoes regular past tense suffixation (see (16b)). At
the same time, however, a phonological readjustment rule comes into play which
causes ablaut within the stem. Note that ablaut is also observed in some of the
verbs which take a zero past tense suffix (for example, sing-sang). The readjustment rule which alters the vowel quality of sell in the context of [+past] is given in
(19a) (Halle & Marantz1993:128).
(19) a. V
[+back] / W ___ U [+past]
[+round] (where WVU = sell, tell)
b. [V1]X [back]
/ X + [+pl]
(where X = Mutter, Vater, Not (need), Land (country), )
Grammar as Processor
Another readjustment rule affects the stem vowel of German Mutter (mother)
when combined with a plural suffix, resulting in umlaut formation (Mtter).
In this particular case, the readjustment rule is triggered by an empty affix. The
other Vocabulary items mentioned in (19b), however, indicate that in German
plurals, umlaut is not restricted to empty affixes: Not (need), for example, takes
a /-/-suffix in the plural (Nte), while the plural of Land (country) is formed
bysuffixing /-r/(Lnder).18
Phonological readjustment rules may not only become active in the context
of certain morphosyntactic features, as has been illustrated in (19); they may also
be triggered by the licensing environment in which a root appears. In (9b) above,
for instance, brech (break) will be spelled out by the Vocabulary item /br7X/.19
This Vocabulary item undergoes phonological readjustment when licensed by a
determiner. The relevant readjustment rule is given in(20a).
(20) a. /7/ //
/ X [+d]
(where X = brechen, sprechen (speak))
b. V
[high] / X [cause]
(where X = sinken, trinken (to drink))
Last but not least, I want to present the readjustment rule that is active in the
sinken/senken alternation that has been illustrated in (10). As mentioned earlier, I
follow Marantz in assuming that there is only one common root for both verbs. In
case this root is licensed by a cause morpheme in the head of vP, the readjustment
rule in (20b) will apply and change the quality of the stemvowel.20
A few other German verbs are capable of causativization by means of umlaut formation,
for instance, fallen/fllen (to fall/to fell) or saugen/sugen (to suck/to suckle). For those
verbs, the mechanism is exactly the same, only the relevant readjustment rule will look
somewhat different.
Theoretical background
3.2.4 Summary
In this chapter, I have presented the major theoretical assumptions of Distributed
Morphology. DM is separationistic in nature in that it endorses a strict separation
of terminal elements which are manipulated in the syntax from the phonological
realization of these elements. Using English and German examples, I have shown
how and at what point in the derivation certain morphosyntactic and phonological mechanismsapply.
We have seen that in DM, roots (l-nodes) are assumed to combine with
inflectional features, bundled in terminal nodes, through various operations that
are either syntactic in nature or rely on syntactic structure; for instance, head
movement and adjunction, morphological merger under structural adjacency,
fusion of sister nodes, and the addition of morphemes at MS. All these structural
manipulations operate on terminal nodes which are hierarchically organized and
yield modified terminal nodes which, again, are hierarchically organized. At PF,
all terminal nodes f-nodes and l-nodes, those present in the syntax and those
added at MS are subject to Vocabulary insertion in exactly the sameway.
3.3 Conclusion
One particularly intriguing property of multi-level models of language production
is that they endorse a division of labour amongst several processing levels no
matter whether the flow of information in a given model is assumed to be stricly
top-down or whether interaction between levels is allowed. Crucially, grammatical encoding at the functional level precedes phonological encoding at the
positionallevel.
The exposition in the preceding section has made clear that in DM, we find
a similar allocation of tasks: the mechanisms that produce a syntactically complex expression are taken to be strictly separate from the mechanisms that supply the corresponding phonological expressions. It is that very property of
the grammar model which suggests to relate it to psycholinguistic models of
languageproduction.
In the following three chapters, I will investigate how the spontaneous speech
errors from my corpus can be accounted for within the DM framework. My
investigation will roughly follow the time course of processing as assumed in the
models: from semantic planning and selection of items from List 1 (Chapter 4)
via grammatical encoding and manipulation of morphosyntactic features (Chapter 5) towards morphological processes, Vocabulary insertion, and phonological
readjustment (Chapter 6). On the one hand, I am going to show which of the
Grammar as Processor
chapter 4
Grammar as Processor
example, does not have an effect on any syntactic principle, rule, or constraint.
Consequently, the computational system does not need to know whether a certain terminal node contains dog or cat. It has therefore been suggested that
List 1 does not include different types of roots, such as, for example, dog, cat,
love, blue, etc. Rather, it has been argued that List 1 only contains one syntactic element root (or simply ). Harley & Noyer (2003) suggest that [Root] is
just another morphosyntactic feature that enters the derivation. Clearly, a terminal
node containing root (or [Root]) has to be spelled out by some sort of content
morpheme but the syntax would be indifferent to which content morpheme eventually realizes root. As pointed out by Harley & Noyer (2003:473), a Root morpheme in an appropriately local relation to a Determiner might be filled by cat,
dog, house, table or any other Vocabulary item we would normally call anoun.
Here, I want to depart from the assumption that roots come in only one variant. Rather, I suggest that roots are individually distinguished from the beginning
of the syntactic computation because a full feature description of a Vocabulary
item in terms of general syntactic and compositional semantic features does not
suffice to unambiguously individuate it (also see Siddiqi (2006)). Obviously, at the
point of spell-out, the PF component must know which root a particular terminal node contains; otherwise Vocabulary insertion would proceed at random. In
a sentence like The cat is eating, for instance, the slot of cat might as well be
taken by dog, fish, or porcupine etc. since at the point of Vocabulary insertion,
there is no distinction whatsoever between these items with respect to the features
which determine the insertion of one or the other item. This dilemma is illustrated
in Figure (1). Note that all competing Vocabulary items match the information
root plus compositional semantic features contained in the terminalnode.
(1)
DP
D0
LP
[+]
[animate], [count]
/dfg/
/fi/
/pf:kj~pain/ ...
[anim], [count]
[anim], [count]
[anim], [count]
[anim], [count]
Moreover, the Encyclopedia (List 3) which relates Vocabulary items to meanings must also have access to information about which element appears in a given
node. That is, the use of the Encyclopedia must involve knowledge of Vocabulary
insertion at PF, where a choice between competing Vocabulary items must have
been made. Consequently, the choice of one Vocabulary item over the other (and,
of course, as opposed to all the other items that might be inserted into an l-node
containing root, [animate], and [count noun]) must be registered in order to
allow for interpretation of that item with the help of theEncyclopedia.
In the DM framework, the distinction between the competing items is made
solely on the basis of semantic features in the Encyclopedia. But how can the
appropriate Vocabulary item be inserted when at PF, the terminal node contains
only information like root and features like [pl], [animate], etc.? Looking up
further information in the Encyclopedia is not a solution because based on these
features alone, it is not even clear where to look. Even if distinguishing semantic features (like, for example, [miaouws] or [furry]) are linked to the concept
cat in the Encyclopedia, it remains unclear how the processor, at the point of
semantic interpretation, should know that it is that very concept that it is looking
for. Obviously, the search for a concept cannot be carried out on the basis of the
features present in the computational system. Consistently, Marantz (1995:401)
acknowledgesthat
[f]or an N node that has the features, count noun and animate (among perhaps
others), presumably the Vocabulary entries for cat and dog would be equally
specified with the relevant syntactic and compositional semantic features and
either might be inserted at that node.
Since the difference between cat and dog is a matter of encyclopedic knowledge,
the use of the Encyclopedia to interpret sentence elements must involve knowledge of Vocabulary insertion at PF. But once again, we are caught in the same
trap. Remember that Encyclopedia entries connect the output of the grammar to
non-compositional meanings. That is, the correct interpretation of a sentence in
which the Vocabulary item /kt/ has been inserted at PF (in a node in which
/dg/ could just as well have been inserted) is guaranteed. But still, the particular
choice of one Vocabulary item over the other in the course of the derivation is in
principle a randomone.
I therefore suggest that at least from a processing point of view we need to
assume that semantic/conceptual features are available at a very early point in the
derivation in order to guide the choice of elements from List 1. That is, List 1 does
not only contain morphosyntactic features to be manipulated by the syntax, but
also the formal concepts that will later be interpreted by the Encyclopedia. This
implies that there is not just one root but rather a large number of contentful
Grammar as Processor
roots that are activated and selected on the basis of the concepts that make up
the abstract preverbal message. Thus, the Vocabulary item for cat might look as
indicated in (2a). This Vocabulary item can only be inserted into a terminal node
which contains the specific root cat and it competes for insertion with the item
in (2b), amongst many others, just like Vocabulary items that spell out feature
bundles compete against each other for insertion. In both cases, the item that best
matches the content of the target node will win the competition. Clearly, (2a) will
win the competition for a node containing cat, while (2b) will win the competition for a node containingdog.
(2) a. cat /kt/
b. dog /dg/
. The errors in (3) are from the Frankfurt corpus of speech errors. Substitution errors, be they
meaning- or form-based, are only included in my corpus when they give rise to a feature mismatch
or are followed by a gender accommodation. Note that in all substitutions in (3), the intended and
the intruding noun are of identical gender (see Section 5.1.2.2 for further discussion).
The nouns involved in the slip in (3a) Radiergummi (eraser) and Spitzer
(pencil sharpener) are cohyponyms; they are both members of the class of writing implements (also see example (3a) in Chapter 1). In example (3b), the intruder
Hhe (height) is the antonym of the target noun Tiefe (depth). The error in (3c)
exemplifies a case where target and intruder stand in a part-whole relationship: a
branch (Zweig) is part of a tree (Baum). Finally, in (3d), the involved nouns Tafel
(blackboard) and Kreide (chalk) have a somewhat more loose associative connection with eachother.2
(3) a.
hast du
einen Radiergummi da
einen Spitzer
have you(sg) an.m eraser(m)
there a.m pencil.sharpener(m)
Do you have a pencil sharpener?
grnen Zweig
green.m branch(m)
With that youll never get anywhere.
. Hotopf (1980) claims that words are never replaced by synonyms or sub- and superordinate expressions, respectively. However, this observation is probably an artifact, since we may
safely assume that, for the most part, such substitutions would go unnoticed. Imagine, for
instance, the substitution of town for city or of dog for poodle.
Grammar as Processor
By definition, meaning-related items are close to each other in a semantic network; activation spreads from one to the other and therefore, a wrong item may
beselected.3
A somewhat different view is taken by Levelt et al. (1999). These authors also
conceive of language production as a staged and strictly feed-forward process,
leading from a preverbal message to the initiation of articulation. In contrast to
Garrett (and others), however, they assign greater importance to the stage of conceptual preparation (the conceptualizer module in Figure (4) in Chapter 3).4
Conceptual preparation is the process which leads to the activation of lexical concepts, the terminal vocabulary of the speakers message construction (Levelt
et al. 1999:8); it is triggered by a speakers communicative intention. Apart from
possible pragmatic causes of lexical concept activation, the theory emphasizes
semantic causes of activation. Within a conceptual network, concept nodes may
be linked to other concept nodes. These conceptual links are labeled in order to
express the nature of the connection, for instance, X is a Y, X is the opposite
of Y, X is part of Y, and so on. In that sense, the concept cat which represents
the meaning of the word cat will be linked to numerous other concepts such as
tomcat, dog, animal, and miaow, and it will spread activation via the links to
the semantically relatedconcepts.
Moreover, lexical concepts which are active at the conceptual stratum spread
activation to corresponding lemma nodes at the lemma stratum. Levelt et al.
(1999) assume that lemma selection is essentially a statistical mechanism which
favors the selection of the highest activated lemma. Crucially, in this conception
of speech production, lemma nodes are not semantically but only syntactically
specified; their selection, however, is triggered by semantic factors. The figure in
(4) illustrates the lexical network underlying lexical access and its components
(adapted from Levelt et al.(1999:4)).
. Harley & MacAndrew (2001) discuss further factors that may have an influence on word
substitution errors. They argue, for instance, that in semantic substitutions, words tend to be
replaced by more imageable competitors, while there is no such imageability effect in phonological substitutions.
. Levelt et al. (1999) do not base their theory on speech error evidence. Rather, they have
developed their model almost exclusively on the basis of evidence from reaction time experiments. They argue that the ultimate test for models of speech production cannot lie in how
they account for infrequent derailments of the process but rather must lie in how they account
for the normal process. Still, their theory is not neutral with respect to speech errors. Ultimately, the model should be able to account for error patterns as well as for the results from
the reaction time experiments.
(4)
Conceptual
Stratum
cohyponym of
is an
Lemma
Stratum
lex.cat.
cat
number
gender
count
/kt/
Form
Stratum
onset
nucleus
coda
/k/
//
/t/
Note that the division of lemma stratum and form stratum is maintained in this
model. That is, the phonological form of a given lemma is accessed only at a later
stage of the derivation. Consequently, in the model, a lexical entry is not a unique
entity but rather consists of a lexical concept, a lemma, and a corresponding phonological form (including morphological, segmental, and metricalstructure).
It is that very property of the model, the separation of semantic, syntactic, and
phonological properties of a given word, which brings us into the vicinity of DM
assumptions. Remember that DM does not assume the existence of a lexicon in the
familiar sense. Rather, there are distributed lists which take over the jobs assigned
to the lexicon component in earlier theories: List 1 (the lexicon) which contains
roots and those (morpho)syntactic features which are relevant to the principles of
syntax, List 2 (the Vocabulary) which contains phonologically specified Vocabulary items, and List 3 (the Encyclopedia) which connects the output of the grammar to non-compositionalmeanings.
My claim is that, with respect to this division of tasks, the processing model
and the grammar theory are readily mapped onto each other. On the one hand,
we may think of List 1 as being the lemma stratum in the processing model,
containing only roots and features drawn from a universal set. On the other
Grammar as Processor
hand, List 2 can be seen as the form stratum containing phonologically specified elements which are to be inserted into terminal nodes according to their
featuralspecification.5
Processing factors, however, suggest a non-trivial modification or rather
enrichment of the DM conception with respect to the selection of items from
List 1. DM advocates are usually indifferent with respect to the question on what
grounds items are selected from this list and for good reason since the syntax
does not care about the meaning of the elements it manipulates. Still, it must be
implicitly assumed that items are retrieved on the basis of what message a speaker
wishes to convey. In a processing model, we need to make this assumption explicit.
That is, during conceptual preparation, concepts are activated according to a preverbal message intention. Concept nodes are linked to meaning components
(properties, semantic relations, etc.) and we may therefore think of the conceptual
network as being identical to List 3, which contains basically the same information. This implies that the Encyclopedia is not only responsible for interpreting
the output of the computational system but also for determining which elements
enter the computational system.6 Activation from the conceptual network is fed
into List 1, the lemma stratum, where the most highly activated elements roots as
well as non-compositional semantic and morphosyntactic features are selected
and enter thederivation.
In line with the reasoning in the previous section, I therefore propose that
the roots which are selected from List 1 bear indices according to what lexical
concept they refer to, a fact which, of course, makes List 1 much more extensive.
Therefore, as argued before, what is retrieved from List 1 is not simply root but
rather a more specific element such as, for example, cat. I wish to emphasize
. Of course, the Vocabulary is not an unstructured list either. Like the conceptual stratum, it
is a multi-linked network where activation flows from target items to phonological neighbors.
Form-based substitutions arise whenever a phonological competitor happens to receive more
activation than the target element.
. Alternatively, one might maintain that the conceptual network and the Encyclopedia are
independent modules where activation in the former sets off the derivation, while the latter
interprets the output of the derivation. Clearly, the two modules contain similar information.
In both, a concept like, for example, cat would be linked to property nodes like miaows,
furry, domestic animal, etc. and to semantically related nodes like dog, mammal, etc.
I leave it to future research to investigate in how far the two modules might also contain
diverging information. It might, for instance, be the case that only the conceptual network
contains concepts like multiple(x) or definite (which would give rise to the activation of
the morphosyntactic features [+pl] and [+def], respectively), while the Encyclopedia does
not require such concepts for its interpretive task.
that this does not imply that roots have any semantic features. In order to account
for semantic substitutions, we must assume that roots are activated on the basis of
their meaning. The selection mechanism, however, only sees activation but is blind
towards the semantics of the roots it selects. I maintain, with DM, that semantic
features do not enter the computational system because they do not play any role
within thatsystem.7
4.2.2 Semantic anticipations and perseverations
The second error type that provides evidence for the early availability of semantic
features are semantic anticipations and perseverations. Just as semantic substitutions, these errors manifest the intrusion of a semantic competitor. In contrast to
the substitutions, however, target and intruder do not compete for the same slot.
Rather, a target concept activates a meaning-related concept whose corresponding lemma then takes another slot in the sentence a fact which indicates that
indeed the slots themselves are not semantically specified. Semantic anticipations
and perseverations come in two slightly different shapes: true semantic anticipations/perseverations and shift-blends. Let us first consider the latter type which is
illustrated by the examples in(5).
In example (5a), the two nouns Toten (dead) and Leichen (corpses) are obviously semantically related. Both concepts are activated at the conceptual level but
no decision is made. Consequently, both corresponding roots are selected from
List 1 and enter the computation. Frequently, an error like this results in a word
blend, that is, a phonological fusion of the corresponding Vocabulary items at
PF. In the present example, however, the result was not a word blend like Teichen
(comparable to English dorpses as a result of blending dead and corpses; also see
(3b) in Chapter 1). Rather, the semantic competitor leiche takes the place of
another root in the utterance, the root sarg (coffin) which is also licensed by a
determiner. That is, the competititor is perseverated into another slot. I therefore
refer to this type of error as ashift-blend.
. Alternatively, we may conceive of the computational system and the semantics as working
in parallel (I am indebted to Heidi Harley for pointing out this possibility to me). That is, a
speaker forms the intention to communicate, for instance, the message The dog chased the
cat. This intention activates two separate modules: (i) the computational system is instructed
to build an intransitive frame for a state verb and to access the necessary (morpho)syntactic
features and the appropriate number of roots; (ii) at the same time, the Encyclopedia is
consulted to inquire what the roots are that are needed to talk about the concepts dog, cat,
and chase, and the Encyclopedia will make available the roots dog, cat and chase. When
the frame has been constructed, the roots that the Encyclopedia has selected are inserted into
the appropriate slots, and the whole thing is sent off to PF for Vocabulary insertion.
Grammar as Processor
(5) a.
in
die Srg-e
into the coffin-pl
A similar analysis applies to (5b). In this error, the competing roots that both
find their way into the utterance are sprech (speak) and red (talk). (5b) differs from (5a) in two respects: first, the competing root red is anticipated into
another slot and second, red appears in a different licensing environment in the
erroneous utterance. There are 27 shift-blends in mycorpus.
A competing root analysis is not available for the class of true semantic
anticipations and perseverations. Crucially, in the examples in (6), it is hard if
not impossible to conceive of a competing sentence frame that would contain the
anticipated/perseverated element. Hence, the intruding root does not compete
with the target root for the same slot; the only reason for it to appear in the utterance is its semantic relation to the target root. In (6a), for instance, the appearance
of finger (finger) in the utterance is due to the activation of the concept finger
by the semantically related concept zeh (toe) at the conceptual level. In the error,
finger takes the position of the intended root tisch(table).
(6) a.
h, am Tisch angestossen
er, at.the table knock.on.part
And then I knocked my little toe on the table.
b. ich htte
meine Bohne mit Karotte-n und Erbse-n
I
have.cond my
bean with carrot-pl and pea-pl
The same line of reasoning can be applied to the other two errors in (6). In (6b),
the intruding concept bohne (bean) has received activation from the semantically related concepts karotte (carot) and erbse (pea). bohne, however,
does not appear in the slot of one of the respective roots but rather takes the slot
of dose (can). There are 22 slips in my corpus that I consider true semantic
anticipation/perseveration.8 Finally, an example of an English semantic anticipation is given in (6c). In this error, the appearance of fly is most probably
caused by the close semantic relationship between the concepts wing and fly
(T. Harley1984:201).
4.2.3 Summary
Clearly, the data in (5) and (6) can only be accounted for when we assume that
specific roots are selected from List 1 based on the activation they receive from the
conceptual level. The distribution of semantic anticipations and perseverations in
my corpus is given in Table(7).
(7) Distribution of semantic anticipations and perseverations (n =49)
semantic anticipation
true semanticanticipation
7
shift-blend
11
18
semanticperseveration
true semanticperseveration
15
shift-blend
16
31
49
Total
In this section, I have investigated the possibility of bringing into accord the model
of lexical access during speech production with the idea of a distributed lexicon as
postulated in the DM framework. My claim is that, in this respect, the psycholinguistic model and the grammar theory are very well compatible with each other,
provided we supplement the DM-model with a pre-syntactic conceptual level.
Activation from this level determines the selection of items roots and features
from List 1. Without the inclusion of lexical concepts and labeled roots, it is neither warranted that the appropriate Vocabulary item will be inserted into a given
terminal node (random insertion; see Figure (1) above) nor can speech errors
involving semantically related items be accounted for in a straightforwardway.
4.3 A note on compositional semantics
In psycholinguistic production models, lemma nodes are are taken to be directly
connected not only to a set of syntactic properties (including the grammatical
. Admittedly, the distinction between the two types of semantic anticipations/perseverations is not always clear-cut. Hence, the classification of some of the semantic anticipations
and perseverations may be somewhat intuitive.
Grammar as Processor
category of a word) but also to certain (morpho)syntactic features; the latter are
referred to as diacritic parameters by Levelt (1989). Diacritic parameters are of
two kinds: they can either be selected on the basis of the speakers intention as is
true, for instance, for the number feature or they can be lexically specified, that
is, inherent to a given lemma as is true for the gender feature in many languages
(see Section 5.1 for further discusson). Besides that, lemmata are also connected
to compositional semanticfeatures.
While non-compositional semantic features which trigger the selection of
roots from List 1 do not play any role in the syntax as argued in the previous
section compositional semantic features do enter the computational system. We
may therefore expect them to be subject to (mis)manipulation in speech errors
(but see Fodor, Fodor & Garrett 1975). In this section, I will investigate whether
this expectation is borne out in the data. The compositional semantic features to
be considered are [count] (Section 4.3.1), semantic gender (Section 4.3.2), and
semantic negation (Section4.3.3).
4.3.1 The count/mass distinction
In DM, the compositional semantic feature [count] is assumed to be amongst the
features drawn from List 1.9 Consequently, just like morphosyntactic features, this
feature is available at an early stage in the derivation and plays a role within the
computational system. The root cat, for instance, will come with information
which specifies that the root refers to a count noun; therefore, cat can either be
combined with a singular or with a plural number feature. In contrast, water is
linked to a feature which specifies that it refers to a mass noun ([count]) and may
therefore not bepluralized.10
Convincing experimental evidence for the assumption that the feature
[count] is available during language production comes from tip-of-the-tongue
. In the present context, the term compositionality is not used in its Fregean sense. According to Freges Principle of Compositionality, the meaning of a sentence is computed on
the basis of the meaning of its well-formed parts and the syntactic relations of these parts
to each other. For the (non-syntactic) compositional semantics of concepts see, for example,
Katz (1972).
. Experimental results suggest that these features are selected only when actually needed
in the local syntactic environment of the noun. That is, they are selected only when the speaker
has to produce a phrase but not when he has to produce a bare noun (Schriefers 1993; van
Berkum 1997). Consequently, Roelofs, Meyer & Levelt (1998) distinguish between the activation and the selection of a feature. The gender of a noun, for example, is selected when needed
to choose the correct definite determiner, but in producing a bare noun, the gender information will only be activated but not selected.
In most of the cases, subjects were able to choose the appropriate context even
though they were not able to produce the phonological form of the target word.
This experimental result is taken as evidence for the early availability of compositional semanticfeatures.
Unfortunately, convincing evidence from spontaneous speech errors that
would support this assumption is sparse. Noun substitutions involving mass and
count nouns might be a touchstone for the processing of the feature [count] in
production. Depending on where that feature is specified and processed, we might
expect the interacting nouns to match with respect to that feature. We must, however, take into account that count nouns are much more frequent in German (as
well as in English). Consequently, there are only very few errors involving mass
nouns in the Frankfurt corpus (only 12 out of 554 noun substitutions), three of
which are given in(9).11
In the meaning-based substitution in (9a), two mass nouns interact: the mass
noun Kaffee (coffee) is substituted for the mass noun Milch (milk). In (9b), the
mass noun Wsche (linen) replaces the count noun Bett (bed) (note that Wsche
may also mean laundry; therefore, the error could also be analyzed as a blend of
to do the laundry and to strip the bed). In (9c), we observe the opposite pattern:
the count noun Brtchen (roll) appears in the position of the intended mass noun
Kaffee(coffee).
. Here, I only consider meaning-based substitutions, since, presumably, compositional semantic features do not play a role when Vocabulary items are selected from List 2, the point in
the derivation at which form-based substitutions may occur.
Grammar as Processor
Possibly, the error in (9c) is somewhat more informative than the other ones.
Interestingly, the noun Brtchen appears in its plural form, as is indicated by the
plural article die (Brtchen takes a zero plural suffix). However, the slot in which
it intrudes is either marked for [pl] or, assuming that mass nouns are unspecified for number, not marked for number at all.12 Where, then, does the plural
feature come from? A possible, albeit speculative, explanation might be that the
[count] specification of Kaffee is stranded. Clearly, the featural specification of
the intruder, the count noun Brtchen, is incompatible with the stranded feature.
Possibly, the processor tries to resolve this conflict by introducing the plural feature, thereby bringing the intruding noun as close to a mass interpretation as possible although, even in the plural, Brtchen remains countable, ofcourse.
In my corpus, there are six root exchanges with similar characteristics; two
of these are given in (10). In (10a), reis (rice) changes place with topf (pot),
that is, a mass noun changes place with a singular count noun which subsequently
surfaces in its plural form. Again, it seems reasonable to assume that the [count]
feature of reis is to be held responsible for the appearance of the plural form of
topf. Note, however, that in (10a), the syntactic context requires a plural form.
In the intended utterance, reis is not accompanied by a determiner. In the singular, however, this option is only available for mass nouns. In other words: if
topf had not surfaced in its plural form, the utterance (Topf in den Reis) would
have been ungrammatical. Consequently, for this slip, it is not clear which factor is
decisive in triggering the plural form of the count noun. Possibly, both factors, the
[count] feature and the syntactic frame, join forces in determining the morphosyntactic properties of the erroneousutterance.13
. As a matter of fact, the noun substitution in (9c) is the only one in the Frankfurt corpus
in which a singular noun is substituted by a plural noun.
. The same is true for two more of the errors from my corpus that might be analyzed as
involving stranding of [count]. That is, for three out of a total of only seven slips of that type,
the syntactic context might be of influence.
Reis in
den Topf
rice into the pot
b. fll-st du
bitte die
Flasche-n in
den Sand,
fill-2.sg you(sg) please the.pl bottle-pl into the.m sand(m),
h, den Sand
in
die
Flasche
er, the.m sand(m) into the.f bottle(f)
The exchange in (10b) is quite similar to (10a): after having been exchanged with
the mass noun Sand (sand), the count noun Flasche (bottle) surfaces in the plural.
Still, in this example, the appearance of the plural form cannot be attributed to the
syntactic context in which the count noun appears. Crucially, in (10b), both roots
are accompanied by determiners. Hence, in contrast to (10a), the structure resulting from the exchange would be fully grammatical even if flasche had been
spelled out in the singular, that is, if the resulting utterance had been die Flasche in
den Sand. Again, it seems reasonable to assume that the [count] feature is to be
held responsible for the appearance of the plural form of the countnoun.14
4.3.2 Semantic (natural) gender
As is well known, in German, but not in English, morphosyntactic gender plays an
important role in the derivation of an utterance. All nouns are inherently specified
for one of three genders (feminine, masculine, or neuter) and their gender feature
In the root exchange in (i), the appearance of a plural form is even more surprising. In this
error, both of the involved nouns, Obst (fruit) and Saft (juice) are mass nouns. Still, saft
surfaces in the plural. The reason for this unexpected morphosyntactic modification might be
that Saft, in contrast to Obst, does have a plural form when the intended meaning is types of
juice. The pluralization option alone, however, does not explain the surfacing plural form.
(i) Obst aus Sft-en schmeckt oft
besser als Obst selbst,
fruit from juice-pl tastes
often better than fruit itself,
. Note that in (10b), there are also various accommodation processes at work: the first determiner accommodates to the (newly introduced) plural feature of Flaschen, while the second
determiner accommodates to the masculine gender of Sand.
Grammar as Processor
will be copied onto other material within DP before spell-out. Besides that, certain
nouns may also have semantic (that is, biological or natural) gender. A German noun
like, for instance, Gabel (fork) is morphosyntactically specified for [feminine],
while a noun like Mutter (mother) comes with a morphosyntactic and a semantic
feminine gender feature. In the following, I will only be concerned with the possible
role of semantic gender in speech errors; the role of grammatical gender in language
production will be considered in some detail in Sections 5.1 and5.4.4.
Again, there are only a few relevant errors in my corpus, most of which
involve kinship terms. Consider first the error in (11a). One way to account for
the appearance of the intruder Schwester (sister) might be to claim that the concept brother is a combination of the conceptual features sibling and male.
In the error, the semantic gender of Frau (wife) is perseverated and combines
with sibling, thereby yielding a combination that will be spelled out as Schwester.
However, an arguably simpler way to explain (11a) might be to analyze it as a
semantic substitution, followed by morphosyntactic accommodation of the possessive pronoun (note that the speaker does not have asister).
(11) a. das Gleiche gilt
fr die
Frau
mein-er Schwester,
the same
holds for the.f wife(f) my-f.gen sister(f)
h, mein-es
Bruder-s
er, my-m.gen brother(m)-gen
h, den Vater
mein-er Schwester, nee, noch mal,
er, the.m father(m) my-f.gen sister(f), no, once again,
A substitution account is not available for the impressive and complex error in
(11b). This slip involves an unsuccessful repair attempt followed by the correct
utterance (such gradual approximations are sometimes referred to as conduite
dapproche). Let us neglect the repair, which can be analyzed as a root exchange,
and focus on the first part of the error. As suggested before, one may analyze the concept schwester (sister) as a combination of sibling and female and the concept
father (father) as a combination of parent and male. Following this line of argumentation, the error receives a straightforward explanation when we assume that
the atomic concepts sibling and parent are exchanged, while the semantic gender
features are stranded. This will give rise to the combinations {parent, female} and
{sibling, male} which will be spelled out as Mutter (mother) and Bruder (brother),
respectively.15 In (12), I give the relevant part of the syntactic structure for (11b),
neglecting morphosyntactic processes like case assignment and feature copy onto
theD-positions.
DP
(11)
D
[+]
LP
L
DP
D
LP
[+]
[1.]
exchange
of concepts
The exchange takes place prior to Vocabulary insertion. At PF, the Vocabulary items
that best fit the content of the terminal nodes will be selected from List 2 for insertion.
As far as the l-nodes in (12) are concerned, these are the Vocabulary items in(13).
(13) a. {parent, female}
b. {sibling, male}
/mt/
/bru:d/
Actually, (11b) is the only example from my corpus that I consider unambiguous evidence for the manipulation of semantic gender in language production. Yet
another phenomenon is illustrated in (14a). In this example, the possessive pronoun agrees with respect to grammatical gender with the noun Mutter (mother),
. Cutler (1980b:692) reports the strikingly similar French exchange error in (i). Just like
(11b), this error is best analyzed as stranding of semantic gender features. If we treat mere
(mother) as a combination of parent plus female and mari (husband) as a combination of
spouse and male, then the error is simply the result of exchanging spouse and parent. In its
new slot, the concept spouse combines with female and the matching Vocabulary item femme
(wife) is inserted. The concept parent combines with male and voil! pre (father) will
be selected for insertion.
Grammar as Processor
bei sein-er
Mutter
with 3.sg.m.poss-dat.f mother(f)
h, ihm
ein-e Postkarte
schreib-en kann,
kann-st
er, 3.sg.m.dat a-f postcard(f) write-inf can.3.sg, can-2.sg
He has given you his address so that you can write him a postcard.
The results show that speakers are three times more likely to produce agreement errors when
the notional gender of the subject and indirect object mismatched (as in (i)) than when the
genders matched (as in (ii)). Thus, just like other agreement processes, notional gender agreement appears to be susceptible to attraction by other elements.
blieb
glcklicherweise aus
fail.to.appear.past fortunately
particle
The same line of reasoning can be applied to (15b). As before, the verb ausbleiben
(to fail to appear) is semantically negative; the meaning it expresses can be paraphrased as to not happen. During the derivation, the negative feature associated
with the verb is anticipated but, in contrast to (15a), it does not project but rather
combines with a root, the root vorhersag (predict). In this context, the negative feature will be spelled out by the negativeprefix /un-/.
4.3.4 Summary
The question whether conceptual vocabulary is just a set of lexical concepts or,
rather, a set of primitive conceptual features that make up these lexical concepts
Grammar as Processor
is, of course, a classical and controversial issue (Fodor et al. 1975, 1980). In the
theory of Levelt et al. (1999), for instance, lexical concepts are taken to be noncompositional in nature, that is, they are not represented by sets of semantic features (also see Roelofs1997).
In this section, I have investigated whether there is any evidence from speech
errors for the existence of such primitive conceptual features. Interesting as the
examples discussed in this section may be, we still have to conclude that the very
small number of relevant errors and the somewhat speculative character of their
interpretation do not allow for any safe conclusions about the processing of compositional semantic features (see Section 5.2.2.8 for discussion of the compositional semantic feature [animate]). The distribution of (possibly) relevant errors
from my corpus is given in(16).
(16) Distribution of errors involving compositional semantic features (n =17)
feature stranding
[count]
7
semantic (natural)gender 1
featureshift
semanticnegation
semantic (natural)gender
Total
4
5
17
4.4 Conclusion
In this chapter, I have taken a closer look at the role of non-compositional and
compositional semantic features in language production. First, I have argued that
non-compositional semantic features must be available at the outset of the derivation and that these features guide the choice of elements from List 1. Crucially,
List 1 does not only contain one primitive root but rather labeled roots (such
as dog and cat) which are selected on the basis of activation at the conceptual
level. I have shown that otherwise, the insertion of the appropriate Vocabulary
items at PF cannot be guaranteed and moreover, speech errors that involve the
manipulation of semantically related elements cannot be accounted for. The early
availability of semantic features, however, does not imply that these features enter
the computational system together with roots and features. Rather, I have argued
that the processor selects roots from List 1 on basis of their activation but is blind
to the semantics of the activatedroots.
chapter 5
Morphosyntactic features
in language production
While there are only few errors that furnish proof that compositional semantic features are processed in the course of language production, there is overwhelming
error evidence for the assumption that the processor has access to the morphosyntactic features drawn from List 1 when generating an utterance. The morphosyntactic features that will be of interest in the discussion of the German speech errors
are number, person, gender, tense and case features as well as the negation feature.
The first three feature types participate in agreement processes with number being
the only of the three that plays a role in both subject-verb agreement and DPinternal agreement. No matter whether one assumes the adjunction of agreement
nodes at MS followed by feature copy or whether one endorses the insertion of
fully inflected forms (drawn from the numeration) followed by a process of feature
checking, it is clear that, at some point during the derivation, these features must
be visible to the processor visibility being a prerequisite for manipulation and
manipulation being error-prone. Gender is different from the other feature types
because it is an inherent feature of roots. It will be shown that the special status
of gender features limits their possibility to participate in speech errors. Case features, on the other hand, are the only features that are not drawn from List 1. In
German, these features are introduced by virtue of case-assignments properties
of verbs and prepositions. Finally, tense and negation are commonly assumed to
head functional projections of their own, while the same is not necessarily the case
for the other features although suggestions along these lines have been made in
theliterature.
This section will provide evidence for the psychological reality of the beforementioned feature types. I will start out in Section 5.1 by discussing the role of
grammatical gender in noun substitutions. In particular, I will consider the possibility of gender underspecification and will test predictions concerning gender
accommodation. Moreover, I will present errors involving gender mismatch. A
different type of feature mismatch, subject-verb agreement errors, will be subject
to discussion in Sections 5.2 and 5.3. Remember that in DM, person and number
features of a subject DP are assumed to be copied onto the AgrS node at the level
Grammar as Processor
of MS (see Section 3.2.2). Occasionally, this copy process may be defective, that is,
a wrong DP may be chosen as the agreement controller. Subject-verb agreement
errors come in two types, as local agreement errors (Section 5.2) and as longdistance agreement errors (Section 5.3), with the former type being much more
common. Section 5.2 also includes a subsection on errors involving local gender
agreement. The manipulation of morphosyntactic features will be subject to further investigation in Section 5.4. In this section, I will discuss various spontaneous errors in which a morphosyntactic feature is either left behind in its original
position (that is, stranded) or displaced (that is, shifted, exchanged, anticipated, or
perseverated). Section 5.5 concludes the survey of the manipulation of morphosyntactic features in speechproduction.
gen
acc
dat
masc.
sg
pl
fem.
sg
pl
neuter sg
pl
In this section, I will only be concerned with gender agreement within DP and
specifically with feature copy (and lack thereof) in noun substitution errors. Other
types of errors that give rise to gender mismatch within DP will be discussed in
Sections 5.4.4 and 6.7.1.2. Gender agreement between nouns and pronouns (erroneous feature copy) has already been briefly mentioned in Section 4.3.2 (see example (14b)) and will be subject to further discussion in Section 5.2.5 where I analyze
speech errors which manifest local genderagreement.
Schriefers & Jescheniak (1999) point out several considerations which motivate
the psycholinguists interest in the investigation of grammatical gender. Two of these
are of major importance in the present context: Firstly, grammatical gender is a lexical, that is, inherent, property of nouns. Since both theories of language production
and Distributed Morphology make clear predictions about the storage and retrieval
of such properties, we may investigate whether they converge with respect to gender
processing. Secondly, grammatical gender is also a paradigmatic case for studying the
actual use of morphosyntactic information in grammatical encoding. Consequently,
speech errors may provide clues about the establishment of agreementrelations.
In the present context, a first question concerns the storage versus computation of gender. That is, we need to ask whether grammatical gender in German
is simply stored as a syntactic property of nouns or whether it is computed on
the basis of certain semantic, morphological, or phonological characteristics of
a given noun. This question will be dealt with in Section 5.1.1. If it turns out that
gender in German is in fact stored information, then a second question emerges,
namely where and how it is stored. In Section 5.1.2, I will address this question. In
particular, since DM allows for the underspecification of certain features (or feature values), I will consider the possibility of gender underspecification. A closer
look at meaning- and form-based noun substitutions suggests that the gender
feature must be specified throughout. Interestingly, based on the course of the
syntactic derivation as assumed in DM, we can make clear predictions about the
(im)possibility of gender accommodation following both types of noun substitutions. These predictions will be tested in Section 5.1.3. Again, I am going to show
how the morphosyntactic theory can be matched with the assumptions made in
psycholinguistic models of languageproduction.
5.1.1 Definition and assignment of gender
Following Matthews (1997:248), grammatical gender may be defined as a system
in which the class to which a noun is assigned is reflected in the forms that are
taken by other elements syntactically related to that noun. In German, for instance,
in the nominative case, the masculine nouns Mann (man) and Lffel (spoon)
require the masculine form der of the definite article, while the feminine nouns
Grammar as Processor
Frau (woman) and Gabel (fork) require the feminine form die, and neuter nouns
like Kind (child) and Messer (knife) require the neuter form das (see Table (1)).
Thus, agreement in gender between a noun and other related items is crucial to the
concept of grammaticalgender.
As has already been pointed out in Section 4.3.2, it is important to distinguish
grammatical gender from the related concept of natural gender, or sex. Although
gender systems exist in which grammatical gender is closely correlated with sex
(see below), in many others, we observe striking mismatches between gender and
sex. In his brilliant essay on The awful German language (1878), Mark Twain gives
some particularly dramatic examples for such mismatches in German:
Every noun has a gender, and there is no sense or system in the distribution; so
the gender of each must be learned separately and by heart. [] In German, a
young lady has no sex, while a turnip has. Think what overwrought reverence that
shows for the turnip, and what callous disrespect for the girl. [] In the German
it is true that by some oversight of the inventor of the language, a Woman is a
female; but a Wife (Weib) is not, which is unfortunate. (Twain 2000:24ff)
gender predictable. To give just two examples: nouns that bear one of the diminutive suffixes -chen and -lein are always neuter (irrespective of the gender of the
base noun), while nouns derived by the nominalizing suffix -ung are always of
femininegender.
In German, however, the gender of only a very small percentage of nouns can
be predicted on the basis of semantic and phonological properties of the noun.
As Mark Twain notes, there is no sense or system in the distribution, that is, the
German gender system is largely arbitrary. With the exclusion of the relatively few
cases in which gender is also a semantic feature of the respective concept (such as
Mutter (mother(f)), Onkel (uncle(m)), and Kuh (cow(f)), there is no obvious
semantic basis for the gender taken by a noun. This fact is neatly illustrated by the
category silverware (see above): the German nouns for knife, fork, and spoon are
all of different gender.1 The lack of a semantic basis for gender assignment is further corroborated by the fact that the gender of a noun may vary across languages:
flower and flute, for instance, are masculine in Italian but feminine in French and
inGerman.
5.1.2 Underspecification of grammatical gender
The above discussion suggests that grammatical gender in German is not computed anew on the basis of a nouns semantic or phonological properties each time
it is needed. Rather, a nouns gender is autonomously specified with respect to its
semantic features and its phonological form. That is, grammatical gender is stored
as an inherent syntactic property of nouns. Moreover, certain derivational affixes
must also be inherently specified for a genderfeature.
If grammatical gender is in fact stored information, then the question of how
and where it is stored emerges. In Levelts model of language production (Levelt
1989; Levelt et al. 1999; see Section 3.1), all nouns of a given grammatical gender
are taken to be linked to one gender node specifiying this grammatical gender.
That is, instead of specifying each nouns grammatical gender separately in the
corresponding lexical entry, there is only one abstract node for each grammatical
gender. To be more precise, it is not the noun which is linked to the gender node
. The assumption of an essentially arbitrary relation between a German noun and its grammatical gender has not remained unchallenged. Complex rule systems have been proposed
that may guide the assignment of a nouns gender once certain phonological, morphological,
and semantic information about that noun are taken into account (cf. Zubin & Kpcke 1981,
1986; MacWhinney, Leinbach, Taraban & McDonald 1989; Konishi 1993; Schwichtenberg &
Schiller 2004). However, in some cases, the principles that are adduced are of such complexity
that one may question their validity or at least their explanatory power.
Grammar as Processor
but rather the nouns lemma, that is, an abstract lexicon entry which is not specified
for phonological features but only connects to nodes representing the words syntactic properties (syntactic category and morphosyntactic features; see Figure (4)
in Chapter 4). The phonological form of a target word becomes activated only
after the corresponding lemma has beenselected.
This conception of language production, however, does not facilitate any predictions about the precise nature of the connection between a lemma node and
its gender feature. Rather, it is simply assumed that a language-specific number of
gender nodes exists and that lemmas are always linked to a gender node. Clearly,
this is not the only conceivable option. As mentioned above, within the theoretical framework I adopt, we need to investigate the possibility that there are gaps
with respect to these links. In principle, there are two options for the underspecification of a given feature, both of which will be subject to investigation in the
followingsubsections.
First, a feature may be underspecified within the computational system, which
implies that it is not among those features that are relevant to that system and is
therefore not drawn from the universal feature set (List 1). Consequently, such
a feature can neither fulfill a function within the computational system nor can
it have an influence on the insertion of one Vocabulary item over another. Gender is a potential candidate for this type of underspecification, one might argue,
because it does not figure prominently within the computational system (it does,
for instance, not trigger any movement operations). Therefore, one may hypothesize that it need not be present in thissystem.
A second possibility is the underspecification of Vocabulary items (List 2)
with respect to some feature. As pointed out in Section 3.2.3, the Vocabulary items
which are inserted at PF may be underspecified for the morphosyntactic feature
complexes that they realize. For instance, as noted by Halle & Marantz (1993), the
English past participle ending /-d/ in a sentence like I had played tennis all day
will correspond only to the feature [+past] in its Vocabulary entry although, in
this example, it will be inserted at a node that contains the feature [+participle]
in addition to the feature [+past]. Remember that, for a Vocabulary item to be
inserted, it is only important that it does not conflict with any of the features contained in the terminal node. Hence, in case a Vocabulary item is underspecified for
gender, it may in principle be inserted at a terminal node no matter what gender
specification that nodehas.
In the following, I am going to argue that a closer examination of speech error
data strongly suggests that, at least in German, neither of the two options for gender underspecification sketched above is available. Rather, gender features must be
present, that is, specified, throughout. The evidence I discuss comes from errors
involving gender accommodation (Section 5.1.2.1) and from the identical gender
effect (Sections 5.1.2.2 and5.1.2.3).
SomehowImtyingmytongueinknotstoday.
von Frage-satz
und normal-em
Satz
of question-clause and normal-m.dat sentence(m)
thedifferencebetweeninterrogativeclauseandnormalclause
c. ob
dein
Irrtum genauso ausfllt wie mein-er
whether your.m error(m) exactly turn.out like mine-m
ob
dein Urteil
genauso ausfllt wie mein-es
whether your.n judgement(n) exactly turn.out like mine-n
whetheryourjudgementwillturnouttobeexactlylikemine
. See Mller (2004) for further arguments in favour of the assumption that gender features
are provided to the syntax and Albright (2007) for critical evaluation of this assumption.
Grammar as Processor
The slip in (2c) without doubt, a gem for every disciple of Freud exemplifies
an instance of a meaning-based substitution: the masculine noun Irrtum (error)
replaces Urteil (judgement) which is of neuter gender. Here, gender accommodation is not observed within DP (in the nominative case, the possessive pronoun
dein (your) is the same for masculine and neuter nouns). Rather, it is the coreferential possessive pronoun in the elliptic comparative phrase that is adjusted
according to the masculine feature of the intrudingnoun.
In (3), I supply an exemplary syntactic structure for the error (2a). The tree
in (3) represents only the relevant part of the utterance (the object DP) after the
root exchange has taken place. Note that the case feature which according to
DM assumptions is implemented at MS also plays a crucial role: accusative case
is assigned to the DP eine Zunge (a tongue) by the verb haben (to have), while
dative case is assigned to the DP dem Knoten (the knot) by the preposition in; both
case specifications influence the surface forms of the respectivedeterminers.3
(3)
DPACC
LP
D
[]
PP
DPDAT
[][]
feature
copy
LP
[+]
[in]
[][]
feature
copy
At MS, the morphosyntactic features of the roots will be copied onto the respective
determiner nodes. The copy mechanism does not only target the gender feature
but also the number feature, since spell-out of the determiner also depends on the
number feature in D (see Table (1)). In (3), I specify the number feature as [pl].
Note, however, that I will argue in Section 5.2.2.5 that [pl] is a default value and
is not amongst the features drawn from List1.
. As is well-known, the status of prepositions is a matter of debate. In the present context, I
shall therefore leave open the question whether P constitutes an l-node or an f-node, or, to put
it differently, whether P is occupied by a root or by some feature (or feature combination).
Crucially, within the DM framework, we cannot assume that the gender feature
is copied from the noun the respective Vocabulary items being specified for
that feature onto D after the noun has been inserted. This is due to the fact
that all operations which alter the arrangement and content of terminal nodes
(for instance, merger, fusion, feature copy, and morphosyntactic readjustment) are
taken to apply before Vocabulary insertion. This constraint will play an important
role in the discussion of gender accommodation in Section5.1.3.
5.1.2.2 The identical gender effect
The data presented in the previous section show that gender features must be available before Vocabulary insertion takes place in order to facilitate insertion of the
appropriate Vocabulary items. Besides that, there is another good reason for assuming that gender features must be specified within the computationalsystem.
If gender was not specified, then one would predict that in noun substitutions,
the gender of the intruding noun should not have any influence on the probability
of its insertion. However, researchers studying noun substitution errors in languages
with grammatical gender systems have reported that there is a strong tendency for
the intended and the intruding noun to be of the same grammatical gender. This phenomenon is called the identical gender effect. Berg (1992) reports that in his corpus
of German slips of the tongue, the involved nouns are of identical gender in 118 out
of 175 non-contextual noun substitutions, that is, in 67.4% of the cases (30 meaningbased errors, 79 form-based errors, and 9 situation-based errors). The identical
gender effect is also observed by Marx (1999). In her collection of German errors,
206 out of 260 noun substitutions (79.2%) obey the identical gender constraint (135
meaning-based errors and 71 form-based errors). An even stronger identical gender
effect has been reported for Spanish noun substitutions by Vigliocco et al. (1999).
. Note that the Vocabulary item for the indefinite article need not be specified for number.
Also, in (3), (4), and below, I give gender features as unary features. Alternatively, one may conceive of gender values as a combination of binary features: masculine would then be specified
as [+m;f], feminine as [m; +f], and neuter as [m;f].
Grammar as Processor
They note that in a corpus of Spanish speech errors (collected by Del Viso, Igoa, and
Garcia-Albea), 171 out of 180 noun substitutions, that is, 95% obey the identical gender constraint (see Arnaud (1999) for French nounsubstitutions).
In order to further evaluate the identical gender effect, I checked all noun
substitutions from the Frankfurt corpus with respect to gender features. At the
time of checking, that corpus contained 728 slips that were classified as noun substitutions. It turned out, however, that for a fair number of these substitutions,
the analysis was not unambiguous; some of them might, for instance, as well have
been analyzed as phonological errors or phrasal blends. All of these ambiguous
cases, as well as a few errors involving proper names, were crossed out from the
list. Following that procedure, 554 noun substitutionsremained.
In order to get a non-biased picture, further errors had to be removed from
the list. First of all, all errors involving plural nouns were put onto a separate list.
This decision was motivated by the fact that in the plural, irrespective of gender, all
nouns appear with the same determiner, namely die in the nominative and accusative, der in the genitive, and den in the dative (see Table (1)). Consequently, it
cannot be decided whether the language processor treats an inherently masculine
noun as such even when it combines with a plural feature. Rather, it is quite possible that plural nouns are not specified for gender at all. For that reason, the two
meaning-based noun substitutions in (5), both of which involve nouns of different
gender, were not included in the count. In (5a), the plural masculine noun Monate
(months) is substituted for the plural feminine noun Wochen (weeks). The error
in (5b) is different in that a plural feminine noun (Augen (eyes)) interacts with
a singular masculine noun (Magen (stomach)). Still, if one assumes that plural
nouns are not specified for gender, this error, too, is not informative when it comes
to testing the identical gender effect (note that I will come back to noun substitutions involving plural nouns in the nextsection).5
(5) a. er hatte schon fnf Monat-e
Urlaub, nee, fnf Woche-n
he had already five month(m)-pl vacation, no, five week(f)-pl
Healreadyhadfivemonthsvacation,no,fiveweeks.
b. aus
dem
Magen,
aus
dem
Sinn
out.of the.m.dat stomach(m), out.of the.m.dat mind(m)
aus
den
Auge-n
out.of the.pl.n.dat eye(n)-pl
Outofsight,outofmind.
. Both errors in (5) involve accommodation. In (5a), the intruding noun Monat is spelled
out with the appropriate plural allomorph; in (5b), the definite determiner accommodates to
the masculine gender feature of Magen.
Secondly, noun substitutions involving compounds were not taken into consideration whenever it could not be decided whether the whole compound was substituted for or only its first part, that is, the modifying element. Clearly, if it was
the whole compound that was replaced in the error, then the gender of target and
intruding noun would necessarily be identical since the gender of a compound
is always determined by the gender specification of its head. The examples in (6)
illustrate the procedure. Since Dosenmilch (canned milk) is an existing German
compound, it is quite likely that in (6a), the whole compound Sonnenmilch (suntan lotion) was substituted for and not just its non-head part Sonne (sun) (note
that Dose (can) as well as Sonne arefeminine).
(6) a. Dosen-milch Sonnen-milch
can(f)-milk(f) sun(f)-milk(f)
cannedmilk
suntanlotion
b. Material-wasser,
h, Mineral-wasser
material(n)-water(n), er, mineral(n)-water(n)
. Berg (1992) did not exclude derived nouns from his count. In his corpus, he observes an
unusually high rate of form-based errors in which both interactants are of feminine gender
(49 instances in contrast to only 8 instances of meaning-based F F substitutions). He states
that, while many rules of gender assignment based on the formal characteristics of nouns are
of a probabilistic nature, some extremely frequent derivational suffixes, like, for instance, -ion
and -ung, assign feminine gender to the derived noun. According to Berg, this fact may help to
explain the high rate of form-based F F substitutions.
Grammar as Processor
(7) a.
hereinbricht der
Tour-ismus
descends
the.m tour-nmlz(m)
Thisisasmallvillageuponwhichtourismdescendssometimes.
Closer inspection of the 554 unambiguous noun substitutions from the Frankfurt
corpus revealed that 114 of them involve a target and/or intruding noun that is
specified for plural, 35 involve ambiguous compounds, and 54 involve derived
nouns. Consequently, 203 noun substitutions were removed from the list. The
351 errors which remained after applying that procedure show the following distribution with respect to the interaction of genderfeatures.
Total
gender
Semantic
ff
mm
nn
45
48
29
52
48
38
97
96
67
fm
fn
mn
17
14
18
15
10
17
32
24
35
171
180
351
Total
The numbers in (8) clearly indicate that the identical gender effect is also observed
in the noun substitutions in the Frankfurt corpus. For 260 out of 351 noun substitutions (the first three rows in (8)), that is, for 74.1% of the total number, it is true
that the intended noun and the intruding noun are specified for the same gender
feature. This percentage lies between the one reported by Berg (1992) for his corpus
(67.4%) and the one given by Marx (1999) for her collection of slips (79.2%). Interestingly, while Berg reports a higher percentage of form-based and Marx a higher
percentage of meaning-based substitutions within the set of identical gender errors,
in the Frankfurt corpus, the difference between meaning- and form-based errors is
quite small:122 out of 171 meaning-based substitutions (71.3%) and 138 out of 180
form-based substitutions (76.6%) show the identical gender effect. For the sake of
illustration, in (9), I give two examples each of noun substitutions in which nouns
of identical gender interact and of noun substitutions in which target and intruding
noun have different genderspecifications.
The identical gender effect is observed in (9a) and (9b). All three nouns participating in the meaning-based substitution in (9a) a conduite dapproche
are of neuter gender, while in the form-based error in (9b), both target noun and
intruder are specified for [masculine]. In contrast, the slips in (9c) and (9d) do
not obey the identical gender constraint. In (9c), the semantically related nouns
Hunger (hunger) and Klte (cold) are masculine and feminine, respectively,
while in the form-based slip in (9d), the intruder Pech (bad luck) is neuter and
the target Pest (pestilence) isfeminine.
(9) a. das Brett,
h, das Tablett, h, das Blech
ist
the.n board(n), er, the.n tray(n), er, the.n baking.tray(n) is
sowieso voll
anyway full
Theboard,er,thetray,er,thebakingtrayisfullanyway.
Pest
und Cholera
pestilence(f) and cholera(f)
thechoicebetweenpestilenceandcholera
What does the above distribution of noun substitutions tell us about the specification of the gender feature? Obviously, the identical gender effect is observed
in meaning-based and form-based noun substitutions. As far as the former are
concerned, this implies that the roots contained in List 1 must be specified for
gender because semantic substitutions occur when roots are selected from List 1
(based on their activation from the conceptual level, as argued in Section 4.2.1).
Otherwise, the identical gender effect in semantic substitutions (9a) could not be
accounted for.7 As mentioned before, gender features associated with roots enter the
. Further evidence for the early availability of gender features comes from the tip-of-the-tongue
(TOT) phenomenon (see Section 4.3.1 for TOT-evidence concerning the compositional
semantic feature [count]). As has been demonstrated in various studies, speakers in a TOT
state can provide correct information about the gender of a noun the phonological form of
which they are unable to access (Caramazza & Miozzo 1997; Miozzo & Caramazza 1997;
Vigliocco, Antonini & Garrett 1997; Lopez Cutrin & Vigliocco 2007). There is, however, an
Grammar as Processor
computational system since they are a necessary prerequisite for establishing agreement relations by means of feature copy.8 The same line of reasoning excludes the
second option for underspecification, that is, underspecification of Vocabulary
items with respect to the gender feature.9 Form-based substitutions occur when
Vocabulary items are drawn from List 2 for insertion. The fact that List 2 is organized according to phonological principles may give rise to the occasional misselection of an item that is phonologically similar to the target. The numbers in
Table (8) indicate that Vocabulary items cannot be underspecified with respect to
gender. Otherwise, the insertion of (phonologically related) Vocabulary items into
ongoing controversy on whether wordform access occurs only upon successful gender retrieval
or not (cf. Caramazza & Miozzo 1997; Roelofs et al. 1998; Caramazza & Miozzo 1998).
. Results from naming experiments conducted by Vigliocco, Vinson, Indefrey, Levelt &
Hellwig (2004) indicate that the identical gender effect (gender preservation) in spontaneous
errors is related to the production of gender-marked phrases. Crucially, the identical gender
effect was much weaker when participants were asked to produce bare noun (in naming pictures of animals and body parts under time pressure) than when naming involved the production of a definite determiner plus noun phrase. In the former condition, 106 out of 229 errors
showed an identical gender effect (46.3%); in the latter condition, gender preservation was
observed in 183 out of 225 errors (81.3%).
. Lumsden (1992) claims that this kind of underspecification is found in, for example,
Romanian, which also has a three-way distinction of gender. With the help of his underspecification analysis, Lumsden accounts for certain gender disagreement facts in Romanian.
In Romanian, adjectives agree in number and gender with the nouns they modify and
demonstrative pronouns agree in number and gender with their antecedent. In case a demonstrative pronoun refers to an event, it appears in its feminine form, see (i). However, when a
demonstrative pronoun referring to an event appears in a construction with a predicate adjective, the pronoun behaves as if it had a feminine antecedent, whereas the adjective behaves
as if it had a masculine or neuter antecedent. In other words, although a subject DP usually
agrees in gender with a predicative adjective, in (ii) this is not the case (Farkas 1990; cited in
Lumsden 1992:474f).
. Lumsden (1992), for instance, does not assume full underspecification of positions and
morphological signals. He claims that positions (i.e., terminal nodes) become fully specified
with respect to the binary features [f] and [n] by means of redundancy rules. The morphological signals (i.e., the Vocabulary items), however, are underspecified with respect to certain
feature values. For example, the masculine and neuter singular are specified as [f] only, while
the feminine singular is completely unspecified.
. This bias towards neuter/masculine interactions is also observed in Bergs (1992)
corpus. In Marxs (1999) error sample, however, feminine/masculine interactions make up
the highest percentage.
Grammar as Processor
If such an impoverishment rule was in fact active at MS, then we would predict that the identical gender effect is not observed in form-based noun substitutions involving plural nouns. Due to gender impoverishment at MS, the
gender feature would no longer be present at the point of Vocabulary insertion and should therefore not have any influence on the selection of Vocabulary items that spell out plural nouns. A first look at Table (11) might lead us
to conclude that this prediction is borne out. Adding up the numbers in the
first three rows of the table, we see that there are 43 semantic same gender substitutions but only 31 form-based same gender substitutions. We must take
into account, however, that the total number of phonological substitutions
involving plural nouns is also much smaller and this fact cannot be related
to the availability or non-availablility of the gender feature (note that in singular noun substitutions, the form-based substitutions had a somewhat higher
share than the meaning based ones; see Table (8)). When calculating percentages, it turns out that 88.6% of the form-based plural noun substitutions
(31 out of 35) but only 60.6% of the semantic plural noun substitutions (43 out
of 71) obey the identical gender effect. That is, contrary to the above prediction,
phonologically related nouns that interact in a substitution are more likely to
share the same gender feature than semantically related nouns. In view of this
distribution, it seems highly unlikely that an impoverishment rule like the one in
(10) alters the featural content of terminal nodes at MS inGerman.
(11) Grammatical gender in plural-noun substitutions (n =106)
noun substitutions (plural)
gender
Semantic
Form-based
Total
ff
mm
nn
21
16
6
14
12
5
35
28
11
fm
fn
mn
10
9
9
2
2
0
12
11
9
Total
71
35
106
On the basis of the above noun substitution data, I conclude that German nouns
are specified for their gender feature throughout. That is, the roots drawn from
List 1 as well as the Vocabulary items drawn from List 2 are linked to a gender feature and two roots or Vocabulary items that are linked to the same gender feature
are more likely to interact in form- and meaning-based substitutions. Since this
observation also holds for plural nouns, we cannot assume that an impoverishment rule deletes gender features in the context of a plural feature. In the next
section, I will have a closer look at what predictions the DM-model makes concerning genderaccommodation.
5.1.3 The limits of gender accommodation
The fact that not only meaning-based but also form-based noun substitutions show
an identical gender effect is somewhat problematic for language production models
which postulate a strict division between the lemma (functional) level and the word
form (positional) level (for instance, Garrett 1975, 1980a, 1990; Levelt 1989; Levelt
et al. 1999). Remember that in these models, it is assumed that only lemma nodes
connect to a words syntactic and morphosyntactic properties, including gender,
and that the flow of activation from lemma to phonological form is unidirectional.
Therefore, once the processing of a sentence has reached the word form level, information about (morpho)syntactic features of a word should be no longeravailable.
On the other hand, production models which allow for feedback between
processing levels (for instance, Dell & Reich 1981; Dell 1986; Berg 1988; Dell &
Grammar as Processor
OSeaghdha 1992) can account for the identical gender effect in form-based
substitutions in a straightforward way, since in these models it is assumed that
morphosyntactic information is still available at the word form level. Berg
(1992), for instance, claims that formal similarity need not stop short of phonological criteria but might also encompass morphosyntactic criteria. However,
Marx (1999), who argues for a modular two-step retrieval model, points out that
these models, too, are capable of explaining the characteristics of form-based
noun substitutions. In line with Berg, she claims that gender also constitutes a
formal property of nouns in German and that one may therefore assume that
nouns of the same gender are lexical neighbors not only at the lemma level but
also at the lexeme level. Still, she stresses that at the latter level, gender is just an
organizational criterion; it is neither activated nor processed at thatlevel.
DM assumptions allow for a similar line of reasoning. In that theory, Vocabulary
items (that is, word forms) are selected from the Vocabulary along with certain morphosyntactic features. Since these features are checked against the featural content of
terminal nodes, they are essential for selecting the appropriate Vocabulary item for
a given node. As mentioned before, a Vocabulary item may be underspecified with
respect to the features contained in a terminal node but it may not clash with one
of these features. Therefore, in DM, it is quite natural to assume that gender has an
influence not only on semantic substitutions but also on form-basedsubstitutions.
Still, there is one important difference between meaning- and form-based noun
substitutions. This difference concerns the establishment of agreement relations in
the course of language production. As argued above, semantic substitutions take
place when roots are selected from List 1, that is, before the level of MS. In contrast,
form-based substitutions occur when Vocabulary items are inserted into terminal
nodes, that is, after the level of MS. According to DM assumptions, both the implementation of agreement nodes and agreement feature copy take place at MS. Based
on these derivational characteristics, we can make an interesting prediction: determiners (as well as other material within DP and co-referential pronouns) should
only be able to accommodate to the gender of the erroneous noun after meaningbased noun substitutions. In contrast, following form-based substitutions, accommodation should not be observed, since at this point of the derivation, agreement
feature copy has already been executed. An erroneous form-related noun may be
selected for insertion but it cannot pass on its gender feature to otherelements.
From Table (8) above, we can infer that there are 49 meaning-based and 42
form-based singular noun substitutions in the Frankfurt corpus in which target
and intruding noun do not share the same gender feature, that is, 91 instances in
which accommodation could in principle be observed. However, a fair number
of these substitution errors is not informative in the present context for one of
the following three reasons: there is no gender cue in the environment, there is
an ambiguous gender cue in the environment, or the error occurs within a compound. The slips in (12) illustrate these threephenomena.
The meaning-based substitution in (12a) as well as the form-based substitution in (12b) involve bare nouns. That is, within DP, there are no agreeing elements
whatsoever and consequently, the gender features from the intruding nouns Wunder (miracle) and Erotik (eroticism), respectively, need not be copied (note that
the same is true for the errors in (9c) and (9d)above).
(12) a.
b. berall
gibts Erotik
Aerobic
everywhere giveit eroticism(f) aerobic(n)
Thereareaerobicclasseseverywhere.
c. aufgrund von Verzgerungen der
Stadt-planung
because of delays
the.f.gen city(f)-planning(f)
der
Satz-planung
the.f.gen sentence(m)-planning(f)
becauseofdelaysinsentenceplanning
d. eher
geht ein
Kanal,
h, ein
Kamel
more.likely goes a.m.nom canal(m), er, a.n.nom camel(n)
durchs
Nadelhr
through.the eye.of.a.needle
Itismorelikelyforacameltogothroughtheeyeofaneedle.
In (12c), a compound is involved in the error. Even if we assume that only the
specifier of the compound is affected by the substitution (see the discussion of the
compound errors in (6)), that is, that Satz (sentence) has been substituted for by
the form-related noun Stadt (city), it is clear that there is no need to accommodate the genitive determiner since the gender-determining head of the compound,
Planung (planning), is the same. In German, the indefinite determiners for masculine and neuter nouns happen to be the same in the nominative, genitive, and
dative case. Consequently, in (12d), where the masculine noun Kanal (canal) takes
the place of the neuter noun Kamel (camel) in a slot that is assigned nominative
case, we cannot decide whether feature copy from the intruder onto the indefinite
determiner has in fact taken place. Given that this substitution is phonological in
nature, I assume that feature copy cannot have taken place and that the fact that a
grammatical utterance surfaces is acoincidence.
Once we remove the non-informative errors from the set of 91 noun substitution errors in which nouns of different gender interact, we are left with only 34
Grammar as Processor
noun substitutions that unambiguously involve the accommodation or nonaccommodation of the syntactic context. These 34 slips from the Frankfurt corpus
are included in my corpus. The addition of two further errors that fulfill these
criteria but are not included in the Frankfurt corpus yields a total of 36 noun
substitutions. Still, the picture that emerges from the comparison of form- and
meaning-based substitutions is pretty clear, as shown in Table(13).
(13) Distribution of (non-)accommodation after noun substitutions (n =36)
accommodation?
noun substitution
Yes
No
meaning-based
form-based
21
2
1
12
. I am not implying that the error is the result of a DP-blend, that is, a blend of the DPs der
Bruder (the brother) and die Schwester (the sister). From a semantic point of view, it seems
highly unlikely that these two DPs are in competition with each other. What I am implying
is that two more complex constructions might have been competing with each other, as illustrated in (i).
(i)
dass die
N.
sein-e Schwester ist
that the.f.dat N.(f) his-f sister(f) is
(14) a.
du
muss-t
die Tr
dann festhalt-en,
you(sg) must-2.sg the.f door(f) then hold-inf,
Youllhavetoholdthewindowthen.
der
Bruder
von der
N.
ist
the.m brother(m) of the.f.dat N.(f) is
ThenithasturnedoutthathesthebrotherofN.
Form-based substitutions show the reverse pattern. As mentioned before, gender accommodation is not expected following a form-based substitution because
these errors take place after gender features have been copied. In fact, 12 out of 14
phonological substitutions result in a gender feature mismatch. A representative
example is given in (15a). In this error, the neuter noun Chaos (chaos), which takes
the slot of the masculine noun Kasus (case), is accompanied by the masculine
definite determiner. The error in (15b) is one of the two clear instances in which
a form-based substitution triggers accommodation. Clearly, it is hard to think
of a possible semantic relationship between Kalender (calendar) and Gelnder
(railing).13 This error is peculiar in the sense that it cannot be accounted for in a
straightforward way. With respect to the DM-model, one solution that comes to
mind is to allow feedback from spell-out to MS. But once we allow feedback, that
is, feature copy after Vocabulary insertion, it becomes entirely unclear why not
more or all of the phonological substitutions are subject toaccommodation.
(15) a. das ist immer der
gleiche Chaos,
h, Kasus
that is always the.m same
chaos(n), er, case(m)
Thatsalwaysthesamechaos,er,case.
b. wo
sie ber den Kalender
guckt ber das Gelnder
where she over the.m calendar(m) looks over the.n railing(n)
whereshelooksovertherailing
. Moreover, the Frankurt corpus contains the unclear case in (i). Note that in Standard
German, the intruder Kompost (compost) is masculine, while the form-related target noun
Kompott (compote) is neuter, as indicated in (i).
(i)
gekocht-er Kompost
gekocht-es
cooked-m compost(m) cooked-n
cooked compote
Kompott
compote(n)
In some German dialects, however, Kompott is of masculine gender. Since it is unclear whether
the speaker used the Standard German or the dialect variant, it cannot be decided whether we
are in fact dealing with a gender accommodation following a form-based substitution.
Grammar as Processor
Still, the general picture confirms the prediction made above: only meaning-based
substitutions are capable of triggering accommodatory processes because only these
substitutions occur before the level of Morphological Structure, that is, at a stage of
the derivation at which agreement relations have not yet been established. At the
point of Vocabulary insertion, all feature copy operations have been executed. Consequently, form-based substitutions cannot be followed by accommodation, since
feature copy is a necessary prerequisite for accommodation to takeplace.14
Note finally that within the DM framework, one cannot assume that whole
DPs are replaced in an error. If, for instance, we analyzed example (14a) as a DPsubstitution that is, the DP das Fenster (the window) was substituted for by the
DP die Tr (the door) then accommodation would be unnecessary. In DM,
however, this kind of replacement is not an option. Remember that Vocabulary
insertion targets terminal nodes. Obviously, there is neither a terminal node nor
a Vocabulary item with the relevant features corresponding to a constituent like,
for instance, [DP die Tr]. Still, there are some very few exceptions, all of which,
however, involve idiomatic expressions. Consider, for instance, the examples given
in (16). In (16a), rote Bete (beetroot), the name of a vegetable (Bete alone is not
attested), is substituted for grne Soe (green sauce), the name of a typical dish
from the Frankfurt area.15 Therefore, we may safely assume that the internal structure of both these DPs is not computed in on-line languageproduction.
(16) a. rote Bete grne Soe
red beet green sauce
greensauce
b. inouracademicivyleague
academicivorytower
I assume that the same is true for the English slip in (16b), in which the DP ivy
league is substituted for ivory tower (Fromkin 1973b:262). Again, both DPs are fixed
constructions. Therefore, the slips in (16) do not constitute counterevidence to the
generalization that only words but not phrases are subject to substitutionerrors.16
. The same observation is made by Marx (1999). The data from her corpus show that
both types of substitutions behave unambiguously complementarily: whereas accommodation never occurs in the case of form-related noun substitutions, it always takes place after
meaning-related noun substitutions (with one single exception).
. The greenish color of this delicious dish comes from several herbs (borage, cress, parsley,
sorrel, chives, and chervil) which are combined with egg yolks, oil, sour cream, and possibly
mayonnaise (see Scherenberg & Stier 1990:18). Interestingly, there are two slips in the corpus
involving the replacement of Grne Soe. But after all, its the Frankfurt corpus, isnt it?
. Remember, however, that DPs and other phrasal constituents can be exchanged, anticipated, and perseverated; see, for instance, example (14a) in Chapter 2.
5.1.4 Summary
I have started my investigation of the processing of morphosyntactic features in
language production by looking at the gender feature. Based on specific properties of noun substitution errors (the identical gender effect), I have argued that in
German, the gender feature must be specified throughout the derivation. That is,
roots are linked to a gender feature in List 1 and they bring along that gender feature when they are merged into a syntactic structure. At MS, the gender feature will
be copied onto other material within DP as well as onto co-referential pronominal
elements. In addition, the Vocabulary items contained in List 2 are also linked to
a gender feature. At the point of Vocabulary insertion, the feature associated with
the Vocabulary item must match the feature contained in the terminalnode.
Furthermore, the analysis of accommodation patterns shows that only semantic
but not formal substitutions are subject to gender accommodation. In other words,
the few form-based substitutions that do not obey the identical gender effect give
rise to a feature mismatch within DP. This is the expected pattern given that gender
agreement is assumed to be instantiated before Vocabulary insertion takes place
(see Albright (2007) for further discussion of the treatment of gender inDM).
In the following sections, I will discuss the manipulation of other morphosyntactic features such as person, number, and tense. However, the gender feature
will also make another appearance, namely in Section 5.2.5, where I consider local
gender agreement errors on pronouns and in Section 5.4.4, where I discuss possible instances of gender shift and genderstranding.
Grammar as Processor
section. Secondly, it is possible that the error source is more distant to the target
than the correct controller; this option, termed long-distance agreement, will
only be subject to discussion in Section5.3.
Feature copy processes such as number/person marking on verbs and gender/
number marking on pronouns are an essential part of encoding grammatical relations in sentences. They are used to signal that linguistic constituents which are
specified for the same feature(s) are linked regardless of whether they appear adjacent in the utterance or in some distance from each other. Bock, Nicol & Cutting
(1999:330) illustrate the importance of such marking by the different interpretations of the minimally different sentences in(17).
(17) a. Descriptionsofthemassacrethatwerediscoveredyesterday
b. Descriptionsofthemassacrethatwasdiscoveredyesterday
Obviously, the two sentences mean different things: while example (17a) is concerned with the discovery of descriptions, example (17b) describes the discovery
of a massacre. This contrast in interpretation is due to the distinct number specification of the verb in the relative clause (see Section 5.2.5 for discussion of feature
mismatch in relativeclauses).
In Section 3.2.2, we have already seen that feature copy operations play a
prominent role in the Distributed Morphology framework; these copy operations
take place at MS.17 However, there is an important difference to other generative
theories. In some accounts, subject-verb agreement is taken to involve featuretransmission from a noun onto Agr and subsequent movement of the verb to Agr
(Chomsky 1981; Baker 1985; Pollock 1989). In other, more recent accounts, agreement is seen as the result of a checking operation. The verb is inserted fully inflected
and its agreement features must at some point in the derivation be checked against
features of a DP under Spec-head agreement (Chomsky 1995, 2000, 2001). In contrast, in DM, at least in Halle & Marantz (1993) original proposal, verbs are not
drawn from List 1 in their inflected form. Moreover, verb movement to Agr is
impossible in DM, since agreement nodes are not assumed to be present in the
syntax. Remember that Halle & Marantz suggest that agreement nodes are only
inserted at MS; AgrS, for example, being implemented as sister of the Tns node.
Just like gender features (see (3) above), relevant features of a subject DP must
be copied onto this newly inserted node before Vocabulary insertion takes place
. Some psycholinguistic models also endorse feature copy. In Kempen & Hoenkamps
(1987) computational model of language production, for instance, agreement is assumed to
involve a feature copy mechanism whereby the features of one constituent in a syntactic tree
are copied onto another constituent. In the case of subject-verb agreement, the features person
and number are copied from the subject DP onto the verb.
DP
D
LP
LP
[+]
[+]
[] [+]
tL
Tns
L
feature copy
Tns
Tns
AgrS
[+]
[+]
In the syntax, the verb (or rather lach) moves and adjoins to Tns. At MS, AgrS
is inserted as sister of the Tns node and the relevant features of the subject DP are
subsequently copied onto AgrS. In this case, the only relevant feature is the number
feature since, except for the 2nd person, German plural DPs are not specified for
person. Note that the plural feature is also copied onto the determinerposition.
As far as defective feature copy is concerned, the main focus in this section
will be on defective subject-verb agreement (henceforth: SVA). SVA is the classical
case of a syntactic dependency where information which influences the surface
form of a certain element in a sentence may be separate from this element. In
particular, a phrase, a clause, or even more material may intervene between the
subject DP and the verb.18 In such contexts, SVA-errors like the one in (19) are
. Here, intervene should be understood as linear intervention. The structural relation
between the subject DP and the verb usually seen as a relation of Spec-head agreement is,
of course, always the same, no matter how much phrasal material intervenes.
Grammar as Processor
most likely to occur (Bock & Miller (1991:46); complex subject DP in brackets; in
the English examples, non-agreeing elements are given initalics).
(19) [thereadinessofourconventionalforces]areatanall-timelow
thereadinessisatanall-timelow
. For ease of description, in the following, I will refer to elements such as readiness in (19)
as the nominal head of the subject DP, being aware of the fact that actually readiness is the
head of an NP which is sister to the highest D0 element of the subject DP.
The experiment revealed that the probability of inserting a plural form into the
verb slot increased with the number of nominal plural forms preceding the verb,
even though the subject DP did not force such a decision. Consequently, insertion
of the plural form were was more likely for (20b) than for (20a) because (20b)
contains the plural noun scientists. Based on these results, Mann proposed that the
plural noun(s) preceding the verb give rise to what he calls a plural atmosphere,
which influences the selection of the correspondingverb.
Manns study is not informative regarding the question whether the capacity
of the memory (mental energy in the above quote from Jespersen) influences
the selection of a particular verb form since in his experimental setting the presentation of printed material the subject DP was continuously available (off-line
condition). However, that very question was the center of attention in the first
experiment conducted by Bock & Miller (1991). In this experiment, sentence preambles were acoustically presented to the participants. The participants were asked
to repeat and to complete the preambles all of which included a complex subject
Grammar as Processor
DP. In particular, the nominal head of the subject DP was followed by another
DP whose number specification differed from the one of the nominal head (mismatch condition). For half of the sentence beginnings, the nominal head was plural, while the local noun (that is, the noun linearly closer to the verb) was singular;
the other half showed the reverse number pattern. Moreover, a control condition
was added in which the number features of the nominal head and the local noun
were the same. The length of the modifiers following the nominal head varied.
In (21), I give some representative examples for the sentence preambles used in
this experiment. All of these examples involve a number mismatch between the
head noun (key/boy) and the noun contained in the modifying phrase (cabinet/
snake), a prepositional phrase in the first two examples, a relative clause in the latter two examples. Note that the material within brackets was included only in the
version with a more complex postnominalmodifier.
(21) a. Thekeytothe(ornateVictorian)cabinets
b. Thekeystothe(ornateVictorian)cabinet
c. Theboythatlikedthe(colorfulgarter)snakes
d. Theboysthatlikedthe(colorfulgarter)snake
First of all, the results showed that the large part of agreement errors (more than
90%) occurred in the mismatch condition. The complexity of the constituent containing the mismatching local noun, however, did not have any influence on the
error rate a fact which contradicts the assumption that the limited capacity of
the memory can be held reponsible for the errors. In addition, it turned out that
more errors were made after prepositional phrases than after relative clauses. A particularly interesting error pattern emerged when singular subjects were compared
to plural subjects in the mismatch condition: agreement errors almost exclusively
occurred in the experimental condition with a singular nominal head and a local
plural DP, that is, after sentence preambles like (21a) and (21c). This pattern clearly
indicates that the errors cannot be reduced to a problem in correctly identifying the
nominal head within the complex subject DP because if that had been the case, errors
in the condition with a local singular noun should have occurred asfrequently.20
. All of the experiments discussed in this section were concerned with the processing of
the number feature in production. Other experiments used a similar error induction technique to investigate the computation of gender agreement on predicative adjectives in French,
Italian, and Spanish (Vigliocco & Franck 1999, 2001; Vigliocco & Zilli 1999; Antn-Mndez,
Nicol & Garrett 2002). For instance, in the study by Vigliocco & Franck (1999), the participants
first saw an Italian adjective in its feminine and masculine form (for example, ross-o (red-m)
and ross-a (red-f)). Then a sentence preamble was presented which contained a masculine or
feminine head noun followed by a local noun of different gender. The participants were asked
On the basis of further refined experiments, Nicol (1995) assumes that the
specification of a verbs number is effected while a (possibly complex) subject DP
is produced. At that time, several nouns may be simultaneously activated, namely
the nominal head of the subject DP and DPs contained in postnominal modifiers.
DPs, however, which occur in a sentence after a critical point of time cannot interfer with the number specification on the verb. In an earlier experiment, sentence
preambles like the ones in (22) were subject to investigation; in these preambles,
the distance between the singular nominal head (helicopter) and the mismatch
(flights vs. canyons)varied.
(22) a. Thehelicopterfortheflightsoverthecanyon
b. Thehelicopterfortheflightoverthecanyons
It turned out that following sentence beginnings of type (22a) with the mismatching DP the flights being structurally closer to the nominal head of the subject DP
three times as many SVA-errors were produced than following sentence beginnings
of type (22b). Obviously, it is not only the distance between the nominal head of the
subject DP and the verb which has an impact on the agreement process but also the
distance between the nominal head and the mismatching DP that follows it. Interestingly, if an SVA-error occurs in cases like (22a), the verb does not agree with the
linearly more local DP the canyon but rather with the semi-local DP theflights.
But why should the syntactic distance have such an impact? Nicol (1995)
assumes that in a syntactic structure, a plural feature percolates from a DP to the
verb, while simultaneously, the post-nominal modifier is constructed. Possibly, the
more local plural DP (in (22b) the canyons) is simply processed too late to have an
effect on the specification of the number on the verb; that is, its plural feature does
not have enough time to percolate. The error rate after sentence preambles like
(22b) was also shown to be lower than the error rate after preambles of the type the
editor of the history books (as used in the preceding experiments), although in both
to repeat the sentence preamble and to complete it using the previously specified adjective.
One of the Italian sentence preambles used in the experiment is given in (i) with the correct
form of the predicative adjective in brackets.
(i)
il
cero
in ciesa
(e ross-o)
the.m candle(m) in church(f) (is red-m)
The candle in the church (is red).
As in the experiments dealing with number mismatch, it turned out that agreement errors
were induced by the local nouns grammatical gender. That is, subjects tended to produce
errors like il cero in ciesa e rossa, in which the adjective agrees in gender with the local feminine noun ciesa (church). Note that all the nouns used in the experiments were singular.
Therefore, the only feature available for copy was the gender feature.
Grammar as Processor
cases the mismatching DP is equally close (that is, adjacent) to the verb. According to Nicol, this is due to the fact that the modifying PP of the history books is
less complex than the PP for the flight over the canyons. Consequently, in the latter
modifying PP, the plural feature of the mismatching DP canyons has a longer way
to take (also see Franck et al.(2002)).21
In one of their experiments, Bock & Miller (1991) also considered the role
of the semantic factor animacy in the production of number agreement. Consideration of this factor is motivated by the well-known fact that a feature like animacy is correlated with subject status to a high degree (see, for instance, Clark
& Begun 1971). Consequently, one might expect that the construction of SVA is
more error-prone when the nominal head of the subject DP lacks that prototypical subject property, while another preverbal noun possesses that property.
According to this hypothesis, an agreement error like the one in (23a) is expected
to be more likely than the one in (23b) since only in the former, the local noun
(babies) is[+animate].
(23) a. Theblanketonthebabiesweresmall.
b. Thebabyontheblanketsweresmall.
In the experiment, however, this expectation was not borne out: animacy of the
local noun did not have any influence on the error rate. This outcome supports
the assumption that a conceptual feature like animacy may very well play a role in
the selection of a sentential subject, that the same feature, however, does not influence the implementation of subject-verb agreement (at least not in English). The
authors interpret this as proof for a modular conception of language production
according to which different information is available at different processing levels
(see Section 3.1). At the level where syntactic relations are established, semantic
and conceptual features are no longer available and can therefore not have any
influence on the occurrence oferrors.
. In another experiment, Nicol (1995) investigated the role of the syntactic distance. In
that experiment, sentence preambles like the ones in (i) and (ii) were presented.
(i)
(ii)
The syntactic path which the plural feature of realtors has to take is shorter in (i) than it is
in (ii), since the relative clause in (i) stands in a hierarchically higher position (high attachment) than the one in (ii) (low attachment). If the syntactic distance had an influence on
the error rate then due to the shorter path more SVA-errors should occur following sentence beginnings like (i). Unfortunately, Nicols results were not significant. Altogether, there
were not enough errors in order to corroborate an influence of the syntactic distance (the
attachment site).
Barker, Nicol & Garrett (2001) point out that Bock & Miller (1991) may have
missed certain animacy effects since they only looked at a subset of the possible
combinations of animacy (inanimate-animate; animate-inanimate). In line with
Bock and Miller, Barker et al. find that the animacy of the local noun did not have
an impact on the error rate. Besides that, however, they show that the animacy of
the subject noun did influence the error rate: local agreement was more likely with
inanimate subjects (for example, the blackboard behind the desks ) than with
animate subjects (for example, the girl behind the desks ).22
To sum up, let me repeat the most important results from the experiments
dealing with proximity concord. In all studies, the experimental setting was
designed in a way to elicit erroneous agreement on the verb. In particular, the
results show that the verb occasionally fails to agree with the nominal head of the
subject DP when the subject DP is complex and contains a noun with different
number specification that intervenes between the nominal head and the verb. By
far the highest number of errors occurred when the head noun was singular and
the local noun plural (see Section 5.2.2.5 for further discussion of this phenomenon). Moreover, the results indicate that the nature of the modifying constituent
also has an impact: after PP complements, more errors were observed than after
sentential complements. Finally, a higher rate of SVA-errors was observed when
the head noun wasanimate.23
In the following section, spontaneous errors from my corpus shall be compared
to the experimental data. On the one hand, it is worthwhile investigating to what
extent the two groups of errors show similar characteristics. On the other hand, we
also need to consider the possibility that other influences may come to fruition in
. Barker et al. (2001) investigated the role of two further semantic factors: semantic overlap
between head and local noun and plausibility of modification by the sentence predicate. As
for the first factor, it turned out that degree of semantic overlap can affect the error rate. SVAerrors were almost twice as likely when the two nouns had a high degree of overlap, as in (i),
than when they had a low degree of overlap, as in (ii).
(i)
(ii)
With respect to the second factor, it turned out that the agreement process was not affected by
the degree to which each of the nouns within the complex subject DP bears a plausible relation to the sentence predicate. To give one example: in the canoe by the cabins was damaged the
predicate might plausibly refer to either of the nouns, while the same is not true for the girl
behind the desks is smart (also see Footnote 44).
. In further experiments, Nicol, Forster & Veres (1997) and Pearlmutter, Garnsey & Bock
(1999) were able to show that the by now familiar asymmetric interference from plurals does not
only arise during sentence production but is also observed in sentence comprehension tasks.
Grammar as Processor
the German slips. As far as experimentally induced errors are concerned, various
studies have reported that speakers of different languages may have at their disposal
different devices for constructing subject-verb agreement (Vigliocco, Butterworth
& Semenza 1995; Vigliocco, Butterworth & Garrett 1996; Vigliocco, Hartsuiker,
Jarema & Kolk 1996). Apparently, the results of Bock and colleagues are not easily generalized, since English with its rather poor inflectional system may not
be the best candidate for investigating this phenomenon. In a series of experiments (analogous to those conducted by Bock & Miller (1991)), in which Italian,
Spanish, French, and Dutch sentence preambles were to be completed by the subjects, Vigliocco et al. (1995, 1996) were able to detect semantic influences that did
not play a role in control experiments conducted with English speakers. Note that
throughout Section 5.2.2, the discussion of German spontaneous errors from my corpus will be supplemented by further experimental results wheneverappropriate.
5.2.2 Local subject-verb agreement in speech errors
In some of the studies the results of which I have reported in the preceding section
as well as in a study on proximity concord by Francis (1986), one may also come
across some scattered spontaneous errors like, for instance, the one cited in (19)
above and the two errors given in (24) ((24a) is from Francis (1986:318), (24b)
from Bock & Cutting(1992:99)).24
(24) a. [thefullimpactofthecuts]haventhithardasyet
thefullimpacthasnthitashardyet
b. [theonlygeneralizationIwoulddaretomakeaboutourcustomers]are
thattheyrepierced theonlygeneralizationisthattheyrepierced
. I was taking care to only cite spontaneous spoken errors. Consequently, errors for which
a letter or a newspaper was given as a source were not taken into consideration (e.g., some
data in Francis (1986)), even though these certainly share important characteristics with slips
of the tongue (as well as with the experimental data), such as, for instance, a stronger influence
of a local plural DP. For written SVA-errors see Strang (1966), Francis (1986), Fayol, Largy &
Lemaire (1994), and Chanquoy & Negro (1996).
. Other types of SVA-errors shall not be considered here. Amongst these are shifts of
agreement markers as in (i) (Garrett 1975:163) and omissions of agreement markers as in (ii)
(Stemberger 1983a:578).
(i)
(ii)
is more local to the verb than the nominal head of the subject DP (in 24 of them,
the local noun is adjacent to the verb). Moreover, in 22 cases, the local noun with
which the verb agrees is plural. For the most part, it is contained within a prepositional phrase modifying the head (17 cases; see, for instance (24a)). More rarely,
the local noun is part of a (reduced) relative clause (3 cases; see, for instance
(24b)), of an infinitival complement, or another constituent (6 cases). Clearly, as
far as error type and error source are concerned, the similarities between the few
spontaneous English slips and the errors elicited in the experiments arestriking.
With that in mind, I will now turn to the spontaneous local SVA-errors from
my corpus. As is well-known, the German agreement system is richer than that
of English. German verbs agree with their subject with regard to the features
person and number. Table (25) illustrates the agreement paradigm for the regular
German verb lachen (to laugh). It will turn out, however, that for the most part,
the German SVA-errors are due only to the number feature. That is, just as in the
experimental and spontaneous English errors, the conflicting specifications are
3rd person singular and 3rd personplural.
(25) Subject-verb agreement inGerman
1st
2nd
3rd
singular
plural
ich lach-e
du lach-st
er /sie/ es lach-t
wir lach-en
ihr lach-t
sie lach-en
Grammar as Processor
a complex subject DP. In this group, two different types of erroneous agreement
have to be distinguished: agreement with a noun contained in a genitive modifier and agreement with a noun from within a prepositional phrase. For illustration, consider the two representative examples in (26). In (26a), the nominal
head of the subject DP Einfhrung (introduction) is followed by a genitive
DP containing the plural noun Mnzen (coins) and the verb verlaufen (to go
off ) agrees in number with that local noun. In example (26b), a PP intervenes
between the nominal head Abstimmung (voting) and the verb stattfinden (to
take place) and the verb agrees with the plural noun Gesetze (laws) contained
in thisPP.
(26) a. [die Einfhrung der
neu-en Mnze-n] verlief-en
the introduction the.gen.pl new-pl coin-pl
go.off.past-3.pl
die Einfhrung verlief
reibungslos
the introduction go.off.past.3.sg smoothly
reibungslos
smoothly
Theintroductionofthenewcoinswentoffsmoothly.
b. weil
morgen [die Abstimmung ber die
neu-en Gesetz-e]
because tomorrow the voting
about the.pl new-pl law-pl
weil
morgen
die Abstimmung
because tomorrow the voting
stattfind-en
take.place-3.pl
stattfind-et
take.place-3.sg
becausethevotingaboutthenewlawswilltakeplacetomorrow.
Clearly, this error pattern is reminiscent of the one described above for the
English SVA-errors, be they elicited or spontaneous. In all cases, the verb erroneously agrees with a noun intervening between the nominal head of the subject DP
and the verb. Moreover, and also similar to the English data, the verb agrees with
an intervening plural noun in 70 of the 76 cases (92.1%). The six errors not conforming to this pattern will be discussed in Section5.2.2.5.
In (27), I give a simplified structure for the slip in (26b). In the structure,
I only indicate the feature copy process that is relevant for the error, that is,
feature copy onto AgrS. Of course, number and gender features are also copied
from the roots onto other nodes within DP. In this and the following structures,
the arrow labeled feature copy highlights the agreement controller and the
agreement target (as well as their linear proximity), while the arrows along
the branches of the tree structure mark the syntactic path the plural feature
of the local noun has to take. In addition, the number feature of the nominal
head of the subject DP, that is, the feature that should have been copied, is indicated by a brokencircle.
(27)
TnsP
Tns
DP
LP
D
[+]
PP
P
[] []
[ber]
Tns
LP
Tns
Tns
DP
[+]
tL
[]
D
LP
LP
[] [+]
AgrS
[+]
feature copy
The structure in (27) illustrates that the erroneous agreement controller gesetz,
while being linearly closer to the verb (and AgrS) than the nominal head, is syntactically more distant from the verb. Put differently, the distance the number
feature of gesetz has to travel through the syntactic structure to reach AgrS
is considerably longer than the way the number feature of the nominal head
abstimm would have to take. It might therefore be tempting to assume that
linear distance, not syntactic distance, is the decisive factor inSVA-errors.
Remember, however, that Nicol (1995) was able to demonstrate that syntactic distance is an important factor. In her experiments, she used complex subject
DPs as preambles which contained two nouns following the nominal head of the
subject DP (see the examples in (22)). The error patterns revealed that in such
constructions, if an error occurred, the verb was more likely to agree with the
semi-local plural noun, that is, the noun closer to the nominal head of the subject
DP but linearly more distant from the verb. Clearly, the syntactic path from the
local noun to AgrS is always longer than the path from the nominal head of the
subject DP. Therefore, agreement of the verb with features of a local noun is an
unlikely event. In cases like (26), linear proximity may favour the occurrence of
an SVA-error. Nicols data, however, show that linear proximity is not everything.
If the syntactic distance between erronoeous agreement controller and agreement
target gets too large then proximity (even adjacency) loses its effect (see Franck
et al. (2002) for similar results with French and English preambles). Note that in
a spreading activation architecture (for example, Dell (1986)), this phenomenon
could be accounted for by assuming that the number information loses some of
Grammar as Processor
its activation with every step along the way. Thus, the more steps it has to take, the
less activation remains forinterference.26
In my corpus, there are five SVA-errors in which the complex subject DP contains two nouns following the nominal head. Hence, the structure of the subject
DP in these errors resembles the structure of the preambles used in the experiments by Nicol (1995) and Franck et al. (2002). Although the small number of
errors does not allow for safe conclusions, it is interesting to note that in four
out of these five errors, the error source is semi-local to the verb, that is, a singular noun intervenes between the error source and the verb. In all four cases, two
prepositional phrases follow the nominal head of the complex subject DP; see,
for instance, the example in (28a), in which the verb agrees with the semi-local
plural noun Freunde (friends). The one exception is given in (28b). In this case,
the nominal head of the subject DP is followed by a PP which contains a genitive
modifier. It is only the last of the three nouns within the subject DP that is specified for plural. We must therefore assume that the local noun Zustnde (situation)
is responsible for theSVA-error.
(28) a. [unsere Reise mit Freund-en nach Sdamerika] war-en,
h,
our
trip with friend-pl to
SouthAmerica be.past-3.pl, er,
war
leider
ein ziemlicher Reinfall
be.past.3.sg unfortunately a quite
disappointment
Unfortunately,ourtripwithfriendstoSouthAmericawasquite
adisappointment.
b. [das Problem bei der Einschtzung der
dortigen Zustnd-e]
the problem with the evaluation
the.gen.pl there
situation-pl
Theproblemwithevaluatingthesituationthereisthathardlyany
reliable sourcesexist.
. The same line of reasoning can explain why SVA-errors are more likely to occur after
non-clausal complements (for instance, a prepositional phrase as in (21a)) than after clausal
complements (for instance, a relative clause as in (21c)) following the nominal head. Bock &
Cutting (1992) argue that this difference is due to a clause packaging organization of production units. Franck et al. (2002), however, argue that the different error rates are better
explained in terms of syntactic structure. Again, the syntactic path is shorter for the plural
feature of a local noun contained in a PP modifier than for the plural feature of a local noun
contained in a relative clause.
c. weil
[die Debatte ber die
Kondition-en] die Entscheidung
because the debate about the.pl condition-pl the decision
verzger-n weil
die Debatte die Enscheidung verzger-t
delay-3.pl because the debate the decision
delay-3.sg
becausethedebateabouttheconditionsisdelayingthedecision.
Moreover, in three SVA-errors from my corpus, in which the verb agrees with a
semi-local plural noun, the local singular noun is not part of the complex subject DP. Rather, in these cases, the local singular noun is contained in a direct
object or a prepositional adjunct intervening between the complex subject DP and
the verb. In (28c), for instance, the singular noun Entscheidung (decision) precedes the verb. Still, the verb erroneously agrees with the semi-local plural noun
Konditionen(conditions).
What all the German and English slips discussed so far have in common is
that a verb agrees with a plural noun that is part of a complex subject DP which
precedes the verb. Obviously, this is not the only conceivable option regarding
defective feature copy. On the one hand, a verb might also agree with a plural noun
that follows it, that is, a verb might anticipate the number feature. On the other
hand, the verb might be preceded by a constituent which is not (part of) the subject of the clause. In the remainder of this section, I will consider the first option;
the second option will be addressed in the nextsection.
In my corpus, there is only one slip that seems to contradict the generalization that the erroneous agreement controller precedes the verb. In example (29a),
a prepositional phrase precedes the finite verb, while the complex subject DP
appears in post-verbal position. As in (26b), the subject DP contains a plural
noun within a PP modifier, but in contrast to (26b), the noun preceding the verb
(Wort (word)) is singular. Therefore, in this case, we are not dealing with local
agreement at least not at surface structure (see Section 5.2.3 for discussion).27
Note, however, that Mehrzahl (multitude), the nominal head of the subject DP,
is semantically plural. Hence, the utterance resulting from the error may be
ungrammatical, but it is certainly not as deviant as (26a) and (26b). For comparison, consider the English equivalent: a multitude of words has/have to be activated
(see Section5.2.2.8).
. Note that (29a) might also be analyzed as a phrasal blend in which the two DPs [eine
Mehrzahl an Wrtern] (a multitude of words) and [mehrere Wrter] (several words) are in
competition. The latter DP requires a plural verb and this is the form that surfaces in the
error. Actually, in my corpus, this slip is classified as a blend. See Section 5.2.2.4 for further
discussion.
Grammar as Processor
(29) a.
aktiviert werd-en
muss
eine Mehrzahl aktiviert
activated be.pass-inf must.3.sg a
multitude activated
werd-en
be.pass-inf
Foreveryutteredword,amultitudeofwordshastobeactivated.
b. Arethehelicopterfortheflightssafe?
Similarly, in the spontaneous English data, there is not a single case of plural feature
anticipation (that is, a case like the hypothetical error My grandmother like the pictures). However, in an experiment, Vigliocco & Nicol (1998) were able to elicit SVAerrors in yes/no-questions, that is, in a configuration in which the verb precedes the
subject. In the experiment, participants were first presented an adjective (for example,
safe) and then a complex subject DP (for example, the DP the helicopter for the flights).
Their task was to form a question from that material, for instance, Is/was the helicopter for the flights safe?. Interestingly, the frequency and distribution of agreement
errors was similar to the pattern previously described for declarative sentences with
complex subject DPs. In particular, participants produced SVA-errors like the one in
(29b) although in this string due to subject-auxiliary inversion the plural noun is
not local to the verb. Again, this implies that linear proximity of plural noun and verb
is not the only decisive factor (see Section 5.2.3.4 for furtherdiscussion).
Table (30) gives an overview of the SVA-errors from my corpus in which the
verb erroneously agrees with a (semi-)local noun from within the subject DP. In all
cases, the error source (the agreement controller) intervenes between the nominal
head of the subject DP and the verb; it is either part of a genitive complement or a
prepositionalphrase.
(30) SVA-errors: agreement of verb with local noun in subject DP (n =76)
Error source
plural
40
2
within prepositionalphrase
and local toverb
and semi-local toverb
23
5
Total
singular
42
1
1
-
28
5
5
-
70
DP but which still precedes the verb. Obviously, in English, the possibilities for a
non-subject to directly precede the verb are highly restricted in matrix as well as
in embedded clauses. There is, for instance, no construction in English like The
colour of the cars likes Peter, in which the plural object DP the cars preceding the
verb would be a potential error source.28 In fact, English objects can only directly
precede the verb in wh-questions. The spontaneous slip in (31) shows that in this
particular configuration, a verb may indeed agree in number with an object whphrase (Levelt & Cutler1983:206).
(31) Whatthingsarethiskid,isthiskidgoingtosaycorrectly?
Things are different in German. Two aspects of German word order are relevant
in the present context. First, due to the underlying SOV word order, object DPs
always precede the verb in embedded clauses. Moreover, non-subjects (arguments
and adjuncts) can directly precede the verb in main clauses due to verb second.
That is, in contrast to English, it is common for an object DP or a DP within an
adjunct to be more local to the verb than the subject DP. We therefore expect to
find SVA-errors in which the verb erroneously agrees with a DP from within an
object or an adjunctphrase.
And indeed: in 63 out of 149 local SVA-errors from my corpus, the verb agrees
with a DP that is contained in an object phrase or a prepositional adjunct. Interestingly, 53 out of these 63 errors are observed in embedded clauses, as exemplified
by the examples in (32). In (32a), the direct object reife Damen (mature ladies)
precedes the verb to which it passes on its plural feature. (32b) is different in that
the verb is preceded by an adjunct, a prepositional phrase, containing the plural
noun Nebenwirkungen (side effects). Again, the verb agrees in number with the
local pluralnoun.29
. It is, of course, possible to topicalize objects in English, as, for instance, in (i) and in the
cleft construction in (ii). Still, in contrast to German, in these constructions, the object presumably occupies an extra-clausal position and the subject DP always intervenes between the
topicalized phrase and the verb.
(i)
(ii)
. Due to the homophony of the German plural allomorph -(e)n with the 3rd person plural
affix on verbs, (32a) as well as some of the slips in the previous subsection might also be
analyzed as phonological errors (perseverations). Such an alternative analysis is not available
for (32b), because the stem vowel of the verb mssen (must) changes in the error because of
its new number specification.
Grammar as Processor
(32) a.
weil
er offensichtlich [reif-e
Dame-n] bevorzug-en
because he obviously
mature-pl lady-pl prefer-3.pl
weil
er bevorzug-t
because he prefer-3.sg
becauseheobviouslyprefersmatureladies.
b. weil
der Arzt mich
[ber mgliche Nebenwirkung-en]
because the doctor 1.sg.acc about possible side.effect-pl
becausethedoctorhastoinformmeaboutpossiblesideeffects.
A tree structure representation for (32a) is given in (33). In this structure, I neglect
the adverbial offensichtlich (obviously). Note, however, that the adverbial, which
presumably adjoins to VP, would make the syntactic path for the plural feature of
the direct object evenlonger.
Again, it is only from a strictly linear point of view that the error source is
closer to the agreement target. From a structural point of view, the erroneous
agreement controller is not local to the verb because the syntactic path the plural
feature has to take to reach AgrS is obviously longer than the path the relevant
features from the (nominal head of the) subject DP would have to take. As before,
we may assume that the linear proximity of two constituents is sometimes capable of cancelling the structural prerequisites in particular, Spec-head relation
between controller and target for the establishment ofagreement.
(33)
TnsP
Tns
DP
VP
[3rd] []
Tns
[]
DP
D
tL
LP
LP
[] [+]
Tns
Tns
AgrS
[]
[+]
feature copy
Similar to the data presented in the previous section (see (28)), we find cases in
which two nouns one plural, one singular intervene between the nominal head
of the subject DP and the verb. There are 12 such cases in my corpus. In four of
them, the verb agrees with a semi-local plural noun, while in the other eight, the
plural noun is local to the verb.30 I will not go into all possible configurations, but
only present one semi-local and one local case both of which involve a complex
object XP. The object DP in (34a) is structurally similar to the complex subject DPs
in (22a) and (28a): a singular head noun is followed by two prepositional phrases
where the first one contains a plural noun and the second one a singular noun.
Given that the semi-local noun Autos (cars) is the only plural noun in the clause,
it seems likely that the verb agrees with thatnoun.
(34) a. weil
er sowieso [das Buch mit den Auto-s auf dem Cover]
because he anyway the book with the car-pl on the cover]
nehm-en weil
er sowieso das Buch nimm-t
take-3.pl because he anyway the book take-3.sg
becausehewilltakethebookwiththecarsonthecoveranyway.
mehrerer Verarbeitungs-komponente-n]
more
processing-component-pl
eine hirnorganische
a
brainorganic
fhr-en wenn
lead-3.pl when
Strung
zur Strung
fhr-t
malfunction to.the breakdown lead-3.sg
whenabrainorganicmalfunctionleadstothebreakdownofoneormore
processingcomponents
In contrast, (34b) involves a complex object PP. The head of this PP is singular but
is followed by a genitive complement containing the plural noun Komponenten
. In three slips, both intervening nouns are specified for plural and it can therefore not be
decided whether the verb agrees with the local or the semi-local plural noun. One of these
slips is given in (i). In this example, the nominal head of the subject DP is followed by a
genitive complement containing the plural noun Haare (hair). Moreover, the indirect object
preceding the verb contains the plural noun Vorgaben (requirements).
(i)
weil
die Lnge seiner Haar-e nicht den Vorgabe-n
entsprech-en
because the length of.his hair-pl not the requirement-pl comply.with-3.pl
weil
die Lnge nicht entsprich-t
because the length not comply.with-3.sg
because the length of his hair does not comply with the requirements.
Grammar as Processor
(components). In this example, the plural noun is local (adjacent) to the verb
which surfaces in its pluralform.
In all examples discussed so far, defective feature copy is observed in embedded clauses. SVA-errors in main clauses in which the verb is preceded by a constituent that is not the subject of the clause are less frequent, but they do occur.
Ten slips in my corpus follow this pattern. Two examples with direct objects in
clause-initial position are given in (35). In both examples, the verb surfaces in its
plural form because of a preceding pluralnoun.
(35) a. [dein-e Kumpel-s] knn-en, h, kann-st du
doch
auch
your-pl buddy-pl can-3.pl, er, can-2.sg you(sg) mod.part also
morgen
noch treff-en
tomorrow still meet-inf
Youcanstillmeetyourbuddiestomorrow,cantyou?
b. [die
unschn-en
Ding-e] vergess-en ich, vergess-e
ich
the.pl unpleasant-pl thing-pl forget-3.pl I
forget-1.sg I
meist ziemlich schnell
mostly rather
quickly
Mostly,Iforgettheunpleasantthingsratherquickly.
A simplified tree structure for example (35a) is given in (36). In this example, the
V-Tns complex has moved to C and the direct object occupies SpecCP. Hence,
the subject pronoun (marked by the broken circle) is lower in the structure than
the verb. In Section 5.2.3, I will discuss the significance of these surface structure
properties for the implementation ofagreement.
(36)
CP
C
DP
D
LP
[+]
[2nd]
[] [+]
Tns
feature copy
Tns
TnsP
Tns
DP
Tns
AgrS
[]
[+]
[2nd]
VP
tTns
tDP...tL
distribution of errors in which the verb agrees with either an (in)direct object or
with a noun contained in a prepositional phrase is given in Table (37). In both
groups, I make a distinction based on whether the error occurs in an embedded
or a mainclause.
(37) SVA-errors: agreement of verb with (noun from within) object XP (n =63)
Error source
plural
singular
39
31
8
3
3
-
withinPP-adjunct
preceding verb in embeddedclause
preceding verb in mainclause
18
16
2
3
3
-
Total
57
As pointed out at the outset of this section, comparable SVA-errors have not
been reported for English with the notable exception of (31). However, the
possibility of object attraction in the construction of subject-verb agreement
has been experimentally investigated for Dutch by Hartsuiker, Antn-Mndez
& van Zee (2001).31 In Dutch, just as in German, the object precedes the verb
in embedded clauses. In the experiment, participants first saw a main clause
(like Karen says that ), then an embedded clause and a verb stem. Their task
was to complete the embedded clause using a perfect tense auxiliary (which is
specified for number) and the past participle of the verb. In one condition, the
embedded clause contained a complex subject DP with a prepositional modifier; in the other condition, the embedded clause contained a direct object. In
both conditions, participants produced SVA-errors. In 47 out of 48 errors, the
verb erroneously agreed with a local plural noun. More errors occurred in the
subject-modifier condition (34 errors) than in the object condition (14 errors).
An example for the former type of error is given in (38a), while the latter type of
error is illustrated in(38b).
. Hartsuiker et al. (2001: 550) also cite two spontanous Dutch slips in which the verb
agrees with a preceding local plural object; one of these slips is given in (i).
(i)
Grammar as Processor
(38) a. Karin zegt dat [het meisje met de krans-en] hebb-en
Karin says that the girl
with the garland-pl have-3.pl
ge-wonn-en
part-win-part
Karinsaysthatthegirlwiththegarlandshaswon.
Just like the spontaneous data discussed above, these experimental results clearly
show that number information from outside the subject DP can interfere with
the establishment of subject-verb agreement albeit less frequently than activated
features from within the subjectDP.32
5.2.2.3 Agreement with local nominative DP
At present, my corpus contains ten slips, in which the verb agrees with a local
noun which is neither part of a complex subject DP nor contained in an object DP
or prepositional phrase. In all these cases, the error source is part of a comparative
construction in which it receives nominative case. For illustration, consider the
two examples in (39). In (39a), the comparative phrase is embedded under the
adjective schneller (faster); in this environment, the second person pronoun
appears in the nominative. In the error, the future tense auxiliary agrees with the
features of the local pronoun instead of with the sentential subject, which is 3rd
personsingular.
(39) a. ich denke, dass der M. [schneller als du]
reagier-en wir-st
I think that the M. faster
than you(sg) react-inf will-2.sg
IthinkthatM.willreactfasterthanyou.
. In addition, Hartsuiker et al. (2001) demonstrate that in Dutch, plural direct object
pronouns are also capable of triggering erroneous agreement (see Section 5.2.2.6 for further
discussion). In my corpus, there is only one error in which the verb agrees (in person and
number) with a plural object pronoun. Note, however, that the object pronoun euch in (i) is
assigned dative case.
(i)
die
Bar-s, die
the.pl bar-pl, rel.pl
er euch
empfohlen
he 2.pl.dat recommend.part
h, empfohlen
hat
er, recommend.part has.3.sg
habt,
have.2.pl
b. weil
du,
[genau wie die
meisten andere-n Leute],
because you(sg), just like the.pl most
other-pl people.pl
weil
du denk-st
because you(sg) think-2.sg
becauseyou,justlikemostotherpeople,thinkthatsignlanguage
isinternational.
Structurewise, (39b) is different because in this utterance, the comparative construction is part of an apposition which separates the subject (the 2nd person singular pronoun) and the verb. But just as in (39a), the comparative phrase contains
a nominative DP; this DP includes the inherently plural noun Leute (people). In
the error, the verb denken (think) agrees with this local pluralnoun.
All errors of this type occur in embedded clauses because it is only in embedded clauses that the comparative phrase containing the local (pro)noun precedes
the finite verb. In five of these errors, the question of whether the error source is
singular or plural is irrelevant because it is only the person feature of the local
(pro)noun that is responsible for the error; see, for instance, (39a). In four of the
errors in this group, the verb agrees with a local DP that is specified for plural
one of these cases is (39b). There is only one instance in which the subject is
specified for plural, while the intervening nominative pronoun issingular.
What we have to be aware of with respect to these errors is the fact that these
constructions are commonly assumed to involve PF-deletion of material. Underlyingly, the comparative phrase in (39a) contains the verb reagieren (react) and the
2nd person future auxiliary wirst, while the apposition in (39b) contains the plural
form of the verb denken (think). In (40), the material that should be deleted at PF
is crossed out (see Lechner (2001) for an elaborate analysis which involves extraposition and deletion ofmaterial).
(40) a. dassderM.[schnelleralsdureagieren wirst]reagierenwird
b. weildu,[genauwiediemeistenanderenLeutedenken],denkst,dass
It could therefore be suggested that the feature mismatch observed in (39) is not
due to erroneous feature copy from the local (pro)noun but rather that it is the
result of a deletion error. That is, at PF, the deletion mechanism targets the wrong
verb or verb-auxiliary combination, thereby giving rise to theSVA-error.
5.2.2.4 SVA-errors in blends
Before turning to factors that possibly have an influence on the occurrence of local
SVA-errors, let me introduce one final type of error: SVA-errors in phrasal blends.
In all of the errors discussed so far (with the notable exception of (29a)), the error
Grammar as Processor
becauseone//wecannotaskthechild.
b. das ist
mein-e Wort-e
that be.3.sg my-pl word-pl
das ist
meine Rede // das sind
mein-e Wort-e
that be.3.sg my
speech // that be.3.pl my-pl word-pl
ThatswhatIvealwayssaid//Thesearemywords.
My collection of spontaneous slips contains 14 SVA-errors in blends. What distinguishes these errors from the ones discussed in the previous sections is the
. The error in (i) is quite similar to the one in (41a). Under a blend analysis, the same
two subject DPs as in (41a) are in competition. Note, however, that in (i), the VP contains the
plural noun Methoden (methods). Thus, this error could also be analyzed as a local SVA-error
in which the verb agrees with the preverbal direct object (in line with the examples in (32)).
find-en mss-en
find-inf must-3.pl
fact that no defective feature copy (from a local noun) is involved in the blend
errors. Rather, subject-verb agreement is established successfully in both competing frames, as shown for (41a) in the structures in (42). Subsequently, that is, after
MS, the two frames blend into one and the singular subject34 (frame ) combines
with the plural feature under the agreement node (frame).
(42)
1
DP
[]
TnsP
Tns
LP
TnsP
DP
AgrS
[]
[1st]
[+]
Tns
LP
AgrS
[+]
Given the structural parallelism between both frames, we can state that the competing subject DPs the one that surfaces in the erroneous utterance and the one
that loses the competition are equally close to AgrS. Despite the fact that the DP
which is responsible for the error is not local to the verb (in the sense of being
closer), I include SVA-errors in blends in the group of local SVA-errors for the
simple reason that they are not non-local. In six of the 14 cases, a singular DP combines with a plural verb, which means that the error-inducing DP (that is, the one
that does not make it into the utterance) is plural (41a). In the other eight cases, a
plural DP combines with a singular verb(41b).
5.2.2.5 The prominence of [+plural]
One aspect of the SVA-errors we still need to consider is the question why, for the
most part, it is the plural feature of a local noun which triggers erroneous agreement. This pattern has not only been found in all of the experimental studies on
proximity concord, it also clearly holds for the spontaneous data. Going back to
the 26 spontaneous English slips mentioned above (see (24)), we find that there are
only four cases in which the local noun is singular; two of these are given in (43).
In (43a), the agreement source problem is part of a reduced relative clause within
the complex subject DP (Bock & Cutting 1992:102); in (43b), the local singular
noun coverage is part of a PP modifier following the plural noun disputes (Bock &
Eberhard1993:59).
. In structure , the subject position is occupied by man. Note, however, that the
Vocabulary item man might also be taken to be the realization of the features [3rd] and [pl]
in the absence of a gender feature.
Grammar as Processor
In my corpus, too, spontaneous slips in which the verb agrees with a local singular
DP be it part of a complex subject DP or part of an object XP are the exception.
As shown in Tables (30) and (37), the verb erroneously agrees with a (semi-)local
plural noun from within the subject DP in 70 out of 76 cases, and with a (semi-)
local plural noun from within an object XP in 57 out of 63 cases. Taken together,
this means that the erroneous agreement controller is specified for plural in 91.4%
of theseerrors.
Two of the few exceptions to this pattern are given in (44). In (44a), the verb
agrees with the singular noun Familie (family) which is part of a prepositional
modifier within the subject DP. In contrast, in (44b), the local singular noun
Freund (boyfriend) which triggers erroneous agreement is contained within the
direct objectDP.
(44) a. [die Angabe-n
ber seine Familie] entsprich-t,
the statement-pl about his family
be.in.accordance.with-3.sg,
entsprech-en
nicht der Wahrheit
be.in.accordance.with-3.pl not the truth
Thestatementsabouthisfamilyarenotinaccordancewiththetruth.
b. weil
ihre Eltern
dann endlich [ihren neuen Freund]
because her parents.pl then finally her
new
boyfriend
kennenlern-en wird,
h, werd-en
get.to.know-inf will.3.sg, er, will-3.pl
becauseherparentswillfinallymeethernewboyfriendthen.
Bock & Eberhard (1993), Nicol (1995), and Eberhard (1997) claim that the
observed bias receives a straightworward explanation by assuming that there is
an asymmetry in the grammatical representation underlying singular and plural count nouns. The authors propose that the asymmetry is due to plural nouns
possessing a grammatical feature for number that singular nouns lack. Thus, verb
agreement is implemented by a mechanism which checks whether the head noun
of the subject phrase possesses a number feature. If there is no such feature (as in
the case of singular nouns), the mechanism interprets the head noun as singular
by default and a singular verb is retrieved. In contrast, if there is an activated number feature (as in the case of plural nouns), this information overrides the default
assignment and a plural verb will be retrieved. On this account, local plural nouns
may occasionally interfere in the agreement process because the mechanism
inadvertently detects the plural feature on the local noun and retrieves a plural
verb. In contrast, singular local nouns are less likely to disrupt the agreement
process, since they do not possess a number feature which might interfere in the
establishment ofagreement.35
In DM terms, this means that singular nouns are not specified for number.
In other words, List 1 does not contain a morphosyntactic feature [pl]. Consequently, there is no such feature to be copied onto AgrS at MS. In case AgrS is void
of features when Vocabulary insertion takes place, the default Vocabulary item /-t/
(representing 3rd person singular) will be inserted. The Vocabulary items competing for insertion under the AgrS node in German are listed in(45).
(45) a.
b. [1st]
c. [2nd]
d. [+pl]
e. [2nd][+pl]
/-t/
/-/
/-st/
/-n/
/-t/
For the SVA-errors, the line of reasoning is the same as sketched above: an intervening plural DP makes available a number feature (which may be copied onto
AgrS by mistake), while an intervening singular DP has no such feature to offer.
In the rare instances in which a verb happens to erroneously agree with a local
singular noun, we must assume that we are dealing with the unlikely case of a
copy failure, that is, no number feature at all is transmitted to the AgrS node and
consequently, the default item will be selected for insertion. Based on this line of
reasoning, in the remainder of this study, I will not specify the feature [pl] anymore in the tree structures and the Vocabulary items. Absence of a number feature
in a terminal node implies default singularspecification.36
. In her investigation, Eberhard (1997) somewhat refined Bock and Millers original experimental setting. For instance, she compared sentence preambles like the ones in (i) and (ii):
Consistent with Bock and Millers investigation, her results showed that plural local nouns
elicited more agreement errors. In addition, however, she was able to show that plural nouns
are less likely to elicit erroneous agreement when the head noun is explicitly marked as singular by a quantifier (as in (ii)) in comparison to when it remains unmarked by the determiner
the (as in (i)). Eberhard claims that this finding supports the hypothesis that the absence of
number marking for singular count nouns is a contributing factor in the asymmetry found
in SVA-errors.
. Further evidence for the prominence of the plural feature comes from defective agreement patterns in Belfast English reported by Henry (1995). In Belfast English, plural subjects
Grammar as Processor
Let me finally remind you that in nine of the local SVA-errors from my corpus, defective agreement is not due to the number feature but only to the person
feature. Five of these belong to the group of errors in which the verb agrees with
a local nominative DP (see Section 5.2.2.3); an example has been given in (39a).
In two other cases, a prepositional modifier within the subject DP contains a personal pronoun. In (46a), for instance, the second person feature of the local singular dative pronoun dir has been copied onto the adjacent verb. In other words, the
local pronoun is singular but the head noun Nachricht (message) is singular,too.
(46) a. [die
the
erstaun-t
surprise-part
Themessagefromyouhassurprisedme.
b. weil
ich seit Tag-en [diese Schei-firma] anzuruf-en versuch-t
because I for day-pl this shit-company call-inf
try-3.sg
becauseIvebeentryingtocallthisdamnedcompanyfordays.
weil
ich anzuruf-en versuch-e
because I
call-inf
try-1.sg
Finally, there are two slips in which the verb agrees in person with a local (pro)
noun from within a direct object. In (46b), the subject pronoun is specified for the
feature [1st]. The verb, however, agrees with the noun Firma (company). This case
is not only structurally different from (46a) in that the error source is contained
within a direct object DP; it also differs from (46a) with respect to what causes the
error. In (46a), the local pronoun is equipped with a feature that can be copied
onto AgrS, the feature [2nd]. In contrast, the local noun in (46b) is 3rd person
may optionally occur with singular verbs. The opposite pattern, that is, singular subjects occurring with plural verbs, however, is ungrammatical. Therefore, the sentences in (i) and (ii)
are grammatical, while (iii) is not (Henry 1995:16f).
If the insertion of singular verbs required the presence of a feature [pl], sentences (i) and (ii)
could not be explained. Obviously, in Belfast English, it is possible not to copy a [+pl] feature
onto AgrS. Whenever this happens, the verb will be spelled out in its default singular form.
In contrast, example (iii) is ungrammatical because in this case, the subject DP is unspecified
for number. Insertion of a plural verb, however, requires the presence of a plural feature (see
Mohammad (2000) for similar patterns of partial agreement in Arabic).
singular, which implies that the terminal node only contains firma which brings
along its gender feature [f]. Hence, in this particular error, as in (44) above, we
must assume that defective agreement is due to copy failure and that the default
Vocabulary item /-t/ is inserted into an empty AgrSnode.
5.2.2.6 Morphosyntactic factors: Case and gender
In this and the following section, I want to briefly discuss further factors that might
have an influence on the occurrence of subject-verb agreement errors. I will start
this discussion by looking at two morphosyntactic factors: the effect of overt case
marking on the local noun and the effect of gender of the head noun. With respect
to the former, the relevant question is whether the combination of the error source
(the local noun) and the inflected verb when looked at in isolation would make
for a grammatically well-formedstring.
Obviously, in English with its rather poor case system, such a situation is
much more likely to occur. Generally, the case feature of a full noun phrase is not
morphophonologically visible, that is, the phonological surface form of a constituent which is not the sentential subject does not indicate its non-subject status. For
illustration consider the examples in (47). (47a) is a spontaneous SVA-error (Francis 1986:314). In this error, the verb make agrees with the local plural noun figures.
Note that in isolation, the bracketed string [all these figures make them harder to
understand] could be interpreted as a grammatical sentence, since the DP these figures has the same phonological form in the dative and in the nominative. The same
is true for basically all of the sentences that were elicited in the various experiments reported in Section 5.2.1. In (47b) one of the sentences elicited by Bock &
Miller (1991) the bracketed string [the babies were small] is grammatical when
considered inisolation.
(47) a. thesheerweightof[allthesefiguresmakethemhardertounderstand]
thesheerweightofallthesefiguresmakesthem
b. theblanketon[thebabiesweresmall]
Things are quite different in German with its comparably rich case system. Within
German DPs, case is marked on determiners, adjectives and quantifiers, and
sometimes on the noun. Consequently, in local SVA-errors, the combination of
the error source and the verb would not usually pass for a grammatical string. In
(48a), the reflexive verb richten (depend) agrees with the plural noun Zahlungen
(payments) contained in a genitive modifier. Due to the fact that the possessive
pronoun within this modifier is overtly marked for genitive case, the DP cannot
be interpreted as being nominative and consequently, the string within brackets is
ungrammatical in contrast to those in (47). The same holds for (48b) where the
verb entsprechen (comply with) agrees with the preverbal plural object Vorgaben
Grammar as Processor
(requirements). In this error, the object noun is accompanied by a definite determiner that is overtly marked for dative case. Hence, the bracketed part cannot be
interpreted as a subject-verbcombination.
(48) a. die Hhe [mein-er Unterhaltszahlung-en
richt-en
sich nach
the size my-gen.pl maintenance.payment-pl depend-3.pl refl on
Thesizeofmymaintenancepaymentsdependsonthesizeofmyincome.
b. weil
die Lnge sein-er
Haar-e nicht [den
because the length his-gen.pl hair-pl not the.dat.pl
Vorgabe-n
entsprech-en]
weil
die Lnge nicht
requirement-pl comply.with-pl because the length not
entspricht
comply.with.3.sg
becausethelengthofhishairdoesnotcomplywiththerequirements.
In a number of experiments, it has been shown that overt case marking on the
local noun does indeed have an influence on the occurrence of SVA-errors
(Hartsuiker et al. 2001; Hartsuiker, Schriefers, Bock & Kikstra 2003; Nicol &
Antn-Mndez, in press). Using Dutch sentence preambles, Hartsuiker et al.
(2001) found that erroneous agreement of the verb with a preceding object pronoun was more likely when the pronominal object was ambiguously case-marked
(as, for instance, the plural pronoun ze (they/them) which can be nominative
or accusative). In fact, these ambiguously case-marked pronouns caused as many
SVA-errors as the corresponding full NPs (for instance, de autos (thecars)).
Even more interesting in the present context is an experiment conducted
by Hartsuiker et al. (2003), because they used German preambles consisting
of a complex subject DP. This complex DP consisted of a definite article and a
noun (the head noun), followed by a prepositional phrase that included a preposition and a noun phrase (the local noun phrase). Two different prepositions
were used, one assigning dative case, the other one assigning accusative case.
Crucially, the plural accusative determiner (die) is homophonous with the plural nominative determiner. In other words, in the plural, dative local nouns are
not case-ambiguous but local accusative nouns are. This is illustrated in (49). In
(49b), the local plural DP [die Demonstrationen] is ambiguous between accusative and nominative, while the local plural DP in (49a) is unambiguously dative
(Hartsuiker et al. 2003: 1318). This implies that the combination of the local
die
Stellungnahme zu den
Demonstration-en
the.f.nom.sg position(f)
on the.dat.pl demonstration-pl
thepositiononthedemonstrations
b. die
Stellungnahme gegen die
Demonstration-en
the.f.nom.sg position(f)
against the.acc.pl demonstration-pl
thepositionagainstthedemonstrations
c.
(gegen) [die
Demonstration-en
(against) the.nom/acc.pl demonstration-pl
(against)thedemonstrationsareuseless
sind sinnlos]
be.pl useless
In the experiment, erroneous local agreement was observed reliably only in the condition with singular head nouns and case-ambiguous local plural nouns. This result
suggests that case marking can have an effect on the occurrence ofSVA-errors.37
While the results from these studies, that is, from controlled settings, allow
for safe conclusions about the role of case marking in SVA-errors, the same is not
true for the spontaneous data analyzed in the present study. Browsing through the
corpus, it appears that there are only few errors with case-ambiguous local nouns.
Let us first look at those slips in which the verb agrees with a local noun that is
part of a complex subject DP (see Section 5.2.2.1). These slips come in two types.
First, there are 43 errors in my corpus in which the local noun is contained within
a genitive modifier. In all of these cases, the local noun phrase is overtly marked
for genitive case (see (48a) above). Secondly, in 33 errors, the local noun is part of a
prepositional phrase (see (26b) above). In only three of these errors, the local plural NP can be considered case-ambiguous. It is noteworthy, that in all three cases,
the PP contains a bare noun; that is, none of these errors parallels the structure of
(49b). One example is given in (50a). In this error, the copula agrees with the local
. Nicol and Antn-Mndez (in press) were able to demonstrate that the presence of overt
case-marking of a local NP also affects the incidence of SVA-errors in English. In their experiment, they also used complex subject DPs in which the head noun is followed by a prepositional phrase. In one condition, the PP contained a full NP (i), in the other condition, it
contained a case-marked pronoun (ii).
(i)
(ii)
Subjects were asked to fill in the copula verb (note that Nicol and Antn-Mndez tested singular
and plural local (pro)nouns). The results indicate that plural pronouns induced substantially
fewer errors than full NPs; that is, subjects were significantly more likely to produce an SVAerror like The bill from the accountants were reasonable than The bill from them were reasonable.
Grammar as Processor
plural noun Soldaten (soldiers), which is assigned dative case by the preposition
(also in contrast to (49b)). It is only due to the lack of a determiner that the bracketed string could be misinterpreted as a grammatical sequence which has Soldaten
as itssubject.
(50) a. eine Gruppe von [Soldat-en
sind
in dem
Film
a
group of
soldier-pl.dat be.3.pl in the.dat movie
draufgegangen]
bite.the.dust.part
Agroupofsoldiershasbittenthedustinthemovie.
thatthetheoryallowsfordifferentexplanations.
Things are somewhat different for the cases in which the verb erroneously agrees
with a noun from within an object XP, be it an (in)direct object or a PP adjunct (see
Section 5.2.2.2). My corpus contains 63 errors of this type. In 13 of these errors,
the local NP, which is assigned accusative case by the verb or the preposition,
is potentially case-ambiguous. For illustration, consider the example in (50b), in
which the verb agrees with the plural direct object verschiedene Erklrungen (different explanations). However, in this error (as well as in another eight errors),
an important reservation has to be made: while the bracketed sequence in (50b)
might pass for a grammatical sentence beginning (the NP being misinterpreted as
nominative), it does not pass for a grammatical sentence in contrast to (49c) and
(50a) because one argument ismissing.38
. All of these nine errors occur in embedded clauses. The remaining four errors are structurally different in that they involve agreement of the verb with a preverbal object in a matrix
clause (see (35) for examples). Notably, in all four slips, we observe self-correction immediately after production of the verb.
Another factor that becomes relevant in SVA-errors in embedded clauses is word order.
Actually, there is a fair number of errors in my corpus in which the local object NP is caseambiguous but in which a potential misinterpretation is prevented by the presence of a
sentence-final auxiliary in combination with a participial. This phenomenon is illustrated in
(i). The NP seine Kumpels (his buddies) is case-ambiguous (nom/acc) but the following participial excludes the nominative interpretation.
(i)
de
straat
bij de
kerk-en
the.c street(c) near the.pl church-pl
the street near the churches
(ii)
het plein
bij de
kerk-en
the.n square(n) near the.pl church-pl
the square near the churches
Grammar as Processor
The local SVA-errors from my corpus show a somewhat less dramatic but still
similar distribution. For the sake of comparison, I only considered SVA-errors in
which the verb agrees with a plural noun from within a complex subject DP. There
are 70 errors of this type in my corpus. All singular head nouns are accompanied
by a determiner; the distribution of determiners is given in(51).
(51) Determiners accompanying singular head noun in localSVA-errors
Type of determiner
Number of SVA-errors
38
10
12
7
3
Total
70
Apparently, local SVA-errors are most likely to occur when the head noun is
accompanied by a number-ambiguous determiner (note that the others category
includes a demonstrative pronoun, a possessive pronoun, and aquantifier).
In conclusion of this section, I want to briefly consider the results from a
Distributed Morphology point of view. As far as case marking of the local noun
is concerned, we only find few cases in the spontaneous data in which the local
DP is ambiguously case-marked. Remember that case-ambiguity is based on
the phonological form of the local DP. This implies that the errors occur at a
point in the derivation at which the phonological surface form of the string has
not yet been specified. Hence, we must conclude that these errors occur at MS,
that is, before Vocabulary insertion takes place. Obviously, the fact that at this
point, case has already been assigned does not prevent the selection of the local
non-nominative DP as source for feature copy. Things get more complicated when
we take into account the results from the experimental studies which indicate that
case-ambiguity does facilitate the occurrence of local SVA-errors. Given that caseambiguity is a PF phenomenon, this tendency cannot be explained when we stick
to the DM assumption that the utterance is not yet phonologically specified when
feature copy takesplace.
The same holds for the second pattern described in this section. In both the
experimental and the spontaneous data, a local SVA-error is more likely to occur
when the determiner accompanying the head noun is number-ambiguous. As
before, this ambiguity is phonological in nature and is therefore only manifest at
PF. At MS, the determiner position is only specified for the features [f] and [+def].
I therefore have to conclude that this pattern does not receive a straightforward
explanation inDM.
Everyarticleinthesecataloguesistotallyoverpriced.
ThemajorityofAmericansisforthedeathpenalty.
Grammar as Processor
As before, the low number of errors in which this factor might play a role does
not allow for any safe conclusions since we are dealing with spontaneous data. It
is interesting to note, however, that there are more than three times as many errors
in my corpus (n = 13) in which the local plural noun is zero-marked. In (52b),
for instance, the local plural noun Amerikaner (Americans) contained in a genitive modifier has the same form in the singular. Clearly, and as predicted by DM,
overt plural marking on the noun is not a prerequisite for SVA-errors tooccur.41
5.2.2.8 Semantic factors: Animacy and collectivity
In Section 5.2.1, I already briefly discussed the role of the semantic feature animacy in local SVA-errors. Experimental results from Bock & Miller (1991) and
Barker et al. (2001) indicate that only animacy of the head noun but not animacy
of the local noun has an impact on SVA-errors. In my corpus, there are 27 slips in
which head and local noun differ in animacy (note that I excluded from the count
errors in which one of the two was a pronoun). Some representative examples have
already been given above. Examples in which the head noun is inanimate but the
local noun animate are (28a), (50a), and (52b); a case in which the head noun is
animate but the local noun inanimate is (32b). For the readers convenience, in
(53), I list the nouns participating in these fourerrors.
(53) example headnoun
(28a) Reise(trip)
(50a) Gruppe(group)
(52b) Mehrheit(majority)
(32b) Arzt(doctor)
localnoun
Freunde(friends)
Soldaten(soldiers)
Amerikaner(Americans)
Nebenwirkungen(sideeffects)
In my corpus, errors in which only the local noun is animate (first three rows in
(53)) are twice as common as errors in which only the head noun is animate (last
row); there are 18 SVA-errors of the former and 9 of the latter type. Hence, the
distribution is different from the one found in the experiments mentioned above.
Moreover, the constellation animate head noun inanimate local noun is only
observed in errors in which the verb agrees with a local DP from within an object/
adjunct XP. In contrast, 16 out of 18 errors showing the opposite pattern are found
. Bock & Eberhard (1993) also investigate whether the morphology of regular plural
marking on nouns somehow contributes to agreement marking on the verb. If this was the
case, then agreement errors should be more unlikely after irregularly inflected local nouns
(like children). The results, however, indicate that local nouns with irregular plural marking
cause as many SVA-errors as regularly inflected nouns. Again, this shows that abstract features
trigger the agreement operation. In my corpus, there are no SVA-errors involving irregularly
inflected plural nouns.
in SVA-errors in which the verb agrees with a local DP from within a complex
subject DP. Based on this distribution, one may therefore conclude albeit with
some caution that in the spontaneous data that have a structure parallel to that
of the elicited errors, animacy does play a role: if both nouns within the complex
subject DP have different specifications for animacy, then it is always the local
noun that is animate. Note that within DM, a compositional semantic feature like
[+animate] is assumed to be present in the syntax (Marantz 1997). Consequently,
the fact that it has an influence on SVA-errors does not come as a surprise. What
remains unclear is why, at least in my corpus, this influence is only observed in one
type of SVA-error. I leave this question for furtherresearch.
Finally, Bock & Eberhard (1993) and Bock et al. (2001) considered yet another
possible semantic influence on SVA-errors. They note that some nouns are syntactically singular but represent a group of individuals, that is, they bear a plural
meaning (for instance, collective nouns like army or fleet). Bock and Eberhard
point out that collective nouns are polysemous in the sense that they refer to either
the individual entities they contain (distributive sense) or to an undifferentiated
whole (collective sense). If abstract features are indeed capable of influencing the
establishment of agreement, then it may be worthwhile to investigate whether the
semantic plural of a local noun, when interpreted in its distributive sense, facilitates agreement errors. Following this line of reasoning, local SVA-errors should
not only be observed after a sentence preamble with a local plural noun (54a) but
also following a preamble like (54b) with a local noun that is morphosyntactically
singular but semantically plural (Bock et al.2001:107).
(54) a. therecordoftheplayers
b. therecordoftheteam
The experimental results suggest that grammatically singular collective nouns have
no greater potential to cause a local SVA-error (in which the verb would appear in
its plural form) than grammatically singular individual nouns.42 Similarly, in my
. In addition, grammatically plural collectives (e.g., teams) elicited no more local SVAerrors than grammatically plural individual nouns (e.g., players) did. The influence of the conceptual number of a noun on SVA-errors was further investigated in a series of experimental
studies; see Bock et al. (1999), Eberhard (1999), and Haskell & MacDonald (2003) for English;
Berg (1998) for comparison of English and German; Bock et al. (2001) for comparison of
English and Dutch; Vigliocco et al. (1995) for Italian; Vigliocco, Butterworth & Garrett (1996)
for Spanish; and Vigliocco, Hartsuiker, Jarema & Kolk (1996) for French and Dutch.
The results from these studies show that the notional number of the local noun (individual versus collective) had no significant effect on verb number (see Bock, Butterfield, Cutler,
Cutting, Eberhard & Humphreys (2006) for a comparison of British and American English).
Grammar as Processor
corpus, there is not a single SVA-error in which a verb would show up in its plural
form in the context of a local singular collectivenoun.
However, singular collective nouns might play another role in local SVAerrors. Note that two of the head nouns listed in (53), namely Gruppe (group)
and Mehrheit (majority), are syntactically singular but semantically plural. Possibly, the combination of a singular collective head noun with a plural local noun
increases the likelihood of an error, due to the creation of a stronger plural atmosphere. In my corpus, there are seven errors of this type, three of which have
already been cited in (29a), (50a), and (52b). Just like (52b), the error in (55) is
particularly interesting because it combines two of the properties discussed in this
section: the head noun Menge (amount) is a singular collective noun and the local
plural noun Fehler (mistakes) iszero-marked.43
(55) [die Menge der
Fehler]
sind
echt
unglaublich
the amount the.gen.pl mistake.pl be.3.pl really unbelievable
die Menge ist
echt unglaublich
the amount be.3.sg really unbelievable
Theamountoferrorsisreallyunbelievable.
In a series of experiments, Bock & Cutting (1992) and Haskell & MacDonald
(2003) were able to demonstrate that collectivity of the head noun does indeed
increase the likelihood of an SVA-error. In other words, production of a pluralmarked verb is more likely following a sentence preamble like The jury at the trials
than following The judge at the trials (Bock & Cutting1992).
In a sense, the spontaneous data pattern with the experimental findings. We
do not observe slips in which erroneous subject-verb agreement is triggered by a
local singular collective noun, but there is a small number of SVA-errors in which
a singular collective head noun is followed by a local plural noun. However, while
the experimental analyses reveal a main effect of collectivity, such that collective
. In addition, there are seven slips in my corpus in which the local plural noun displays
another type of semantic plurality in that it does not have a singular form. The relevant local
nouns are Eltern (parents; four instances), Leute (people; two instances), and Gebrder
(brothers). An example in which the verb agrees with the local noun Eltern contained in a
genitive modifier is given in (i). Note that in two SVA-errors, Eltern is not contained within
the complex subject DP but is a direct object.
(i) [die Kche meiner
Eltern] knn-en, h, kann
einen neuen Anstrich
the kitchen my.gen.pl parents can-3.pl, er, can.3.sg a
new painting
gebrauch-en
use-inf
head nouns elicited more plural verbs than non-collective heads, no such conclusion can be drawn on basis of the spontaneous data. First, in the controlled
experimental settings, the number of collective and non-collective head nouns
was balanced. In contrast, in spontaneous speech, non-collective nouns are probably much more frequent than collective ones. Second, as already mentioned in
the discussion of (29a), the combination of a singular collective noun with a plural
verb may be more likely to go unnoticed because the resulting utterance tends to
sound less deviant than one in which a singular non-collective noun combines
with a plural verb (consider, for instance, the English example a number of paintings were notsold).44
5.2.2.9 Summary
With that, all types of local SVA-errors from my corpus have been described.
Table (56) serves as a synopsis of the different error types and their distribution.
The comparison of the spontaneous data to those elicited in various experiments
reveals two interesting findings. First, in most of the experiments, use was made
of complex subjects DPs in which head noun and local noun mismatch in number. Those errors from my corpus that have a comparable structure show properties similar to the elicited ones in that the verb agrees with a local plural noun
in most of the cases (92.1%). Second, we also find spontaneous errors that are
structurally different from those elicited in the English experiments. These are the
slips in which the verb erroneously agrees with a local noun contained within an
object DP or PP. Again, in most of these errors (90.5%), the local noun is specified
for plural. The existence of such errors can be attributed to the different syntactic
structure of German (as also evidenced by the fact that similar errors were elicited
in experiments using Dutch elicitation materials). In particular, in German and
Dutch, but not in English, objects frequently precede theverb.
. Just like Barker et al. (2001) (see Footnote 22), Pittman (2004) and Pittman & Smyth
(2005) investigated the effect of the predicate on SVA-errors. In their experiments, they did
not only manipulate the number of the head and the local noun but also properties of the
predicate. In particular, they compared sentence preambles in which the adjectival predicate is plausible with both nouns in the complex subject phrase (i) with preambles in which
the predicate is plausible solely with the head noun (ii).
In contrast to Barker et al. (2001), they found that more agreement errors occurred in the
condition in which the local noun would also be a plausible subject for the predicate (also see
Thornton & MacDonald 2003).
Grammar as Processor
singular
plural
6
6
6
8
70
57
4
6
26
137
The striking prominence of the plural feature in the elicited and the spontaneous SVA-errors has been accounted for by assuming that only plural nouns come
with a number feature that can interfere in the agreement process. Other factors that might facilitate the occurrence of a spontaneous SVA-errors (and that
were also tested in various experiments) are gender of the singular head noun,
animacy of the local plural noun, and collectivity of the head noun. In contrast,
case-ambiguity of the local noun and zero-marking of plurality on the head noun
do not appear to increase the likelihood of an SVA-error. Remember, however,
that, given the nature of spontaneous data, all of these potential facilitating factors have to be taken with a pinch ofsalt.
In addition, Table (56) lists two less frequent types of SVA-errors. First, a
verb may agree with an intervening nominative DP which is part of a comparative construction. From the numbers, it seems as if in these errors, the error
source was more likely to be singular. This distribution, however, is due to the
fact that for the most part, it is only the person feature which causes these errors.
Secondly, SVA-errors may be observed in blends. Remember, however, that
strictly speaking, these errors are not local SVA-errors. Rather, the verb agrees
with the subject DP from a competing planning frame. Hence, out of a total of
163 SVA-errors, only 149 can be considered localerrors.
5.2.3 Transformations and feature copy
In Section 2.2.3, I have already discussed psycholinguistic studies which are
concerned with the psychological reality of syntactic transformations. Fay
(1980a,b), for instance, reports a number of errors which, according to him,
are best analyzed as the result of either a wrong or a non-application of some
transformational rule. Consider, for example, the spontaneous slip in (57) (Fay
1980b:114).
(57) Lookatthosecloudsaremoving[howfast]
Lookathowfastthosecloudsaremoving
Fay (1980b) takes this error to be the result of a failure in applying a transformational rule, namely the wh-fronting rule. As a consequence, the wh-phrase
how fast appears in its deep structure position. Fay claims that the position of
the wh-phrase in the error follows naturally from a transformational account,
since an element not moved by a transformation must remain in its deep
structureposition.
The impact of syntactic transformations on speech errors that I will be
concerned with in this section is of a different nature. In the following, I will
investigate further the syntactic properties of the local SVA-errors introduced
in the previous section by looking at the interaction of movement operations
and agreement feature copy. In particular, I will try to evaluate whether the DM
assumption that the implementation of agreement nodes and subsequent agreement feature copy takes place after syntax (Halle & Marantz 1993) is supported
by the errordata.
In the model of language production sketched in Section 3.1, agreement is
assumed to be computed at the level of grammatical encoding (that is, at the
functional level). At this level, lemmas are retrieved from the mental lexicon
and are assigned to slots in a hierarchical structure. That processing stage is followed by the stage of phonological encoding at which the phonological form of
a sentence is spelled out (Garrett 1975, 1980a; Levelt 1989). As we have seen, the
course of processing as assumed in the production model is compatible with DM
ideas, since in DM, too, the assignment of phonological forms (Vocabulary items)
follows all processes of grammatical encoding. Pictures diverge, however, with
regard to the precise locus of agreement computation. Within many syntactic and
psycholinguistic theories (Chomsky 1981; Kempen & Hoenkamp 1987; Levelt
1989), agreement relations are taken to be established during the construction of
the hierarchical structure. In contrast, DM endorses late insertion of agreement
nodes. Agreement nodes are adjoined to functional nodes at the level of Morphological Structure, that is, after syntactic operations have taken place but before
Vocabulary insertion isexecuted.
This assumption has important consequences for the interpretation of
speech error data. In particular, a DP which is local to a verb (or to AgrS) at
deep structure may be separated from that verb by a syntactic movement operation. The DP would then no longer be local to the verb when agreement feature copy takes place at Morphological Structure. Alternatively, a DP may be
more local to the verb at surface structure, that is, after movement operations
have applied. Linear proximity of a verb and a DP is, of course, not a prerequisite
for agreement processes to take place. The only prerequisite is a certain structural relation (Spec-head agreement) between agreement controller and agreement target at some point in the derivation. Above, I already pointed out that
Grammar as Processor
occasionally, linear proximity may supersede syntactic distance between controller and target. As far as SVA-errors are concerned, my hypothesis is that linear proximity between erroneous agreement controller and verb must be given
at surface structure, since agreement features are copied after syntactic movement has takenplace.
In order to test this hypothesis, I will only consider SVA-errors in which the
verb erroneously agrees with a plural DP that is either part of a complex subject DP or part of an object DP/PP. In my corpus there are 70 slips of the former
and 57 slips of the latter type (see Table (56)). The following discussion is divided
into two parts. I will first look at local SVA-errors in embedded clauses (Section
5.2.3.1) before discussing the locality patterns observed in local SVA-errors in
matrix clauses (Section 5.2.3.2). The summary of the findings for these two data
sets (Section 5.2.3.3) is followed by a brief discussion of SVA-errors in polar questions (Section5.2.3.4).
5.2.3.1 Local SVA in embedded clauses
Out of the 127 local SVA-errors under investigation, 64 (50.4%) are observed in
embedded clauses. As is well-known, word order in German embedded clauses
(SOV) is commonly assumed to be the underlying word order (Thiersch 1978;
Grewendorf 1988). Consequently, in most of the embedded clause errors from my
corpus, no XP-movement has applied except for movement of the subject DP
from SpecVP (or SpecvP) to SpecTnsP, which, however, does not alter the locality
patterns within the embedded clause. This is illustrated by the examples in (58). In
(58a), the (zero-marked) plural DP Hrer (listeners), a genitive modifier within
the complex subject DP, is local to the verb at deep and at surface structure. That is,
at neither of the two levels, another DP would be (linearly) closer to the verb. Similarly, in (58b), the verb agrees with the preceding plural direct object Erklrungen
(explanations) within the embedded clause. Again, the locality patterns are the
same at deep and surfacestructure.
(58) a. dass [der Geschmack der
Hrer]
extrem schlecht sind
that the taste
the.gen.pl listener.pl extremely bad
be.3.pl
thatthetasteofthelistenersisextremelybad.
thatthetheoryallowsfordifferentexplanations.
Actually, out of the 64 SVA-errors in embedded clauses, only three are informative in the present context because in only three errors, the locality patterns are
different at deep and surface structure. In two of these, the verb agrees with a
local plural DP within the subject DP, but an object phrase has been scrambled
to pre-subject position. Crucially, before scrambling, the object phrase which,
in both cases, contains a singular DP intervenes between the subject and the
verb. For illustration, consider the example in (59a). Presumably, in this example, the object PP beim Lernen (for learning) containing a nominalized verb
follows the complex subject DP at deep structure, as indicated in the simplified
bracketed deep structure in (59a). Above, I have pointed out that occasionally,
a verb agrees with a semi-local plural DP (see (28) and (34a) for examples).
This implies that in principle, an SVA-error might have occurred in (59a) even
without scrambling of the PP. What is crucial about this error, however, is that
the plural noun Netze (networks) is semi-local to the verb at deep structure but
local at surfacestructure.
(59) a. weil
beim Lernen [die Beteiligung komplexer
because for.the learning the involvement complex.gen.pl
weil
beim Lernen die Beteiligung notwendig ist
because for.the learning the involvement necessary be.3.sg
becausetheinvolvementofcomplexneuralnetworksisnecessary
forlearning.
b. weil
da
oft
ber Sache-n ge-rede-t
because expl often about thing-pl part-talk-part
wurd-en,
die
fr seine Arbeit wichtig
sind
be.pass.past-3.pl which for his work important be.3.pl
weil
da geredet
wurd-e
because expl part-talk-part be.pass.past-3.sg
becauseoften,itwastalkedaboutthingsthatareimportantforhiswork.
The slip in (59b) has different characteristics. In this slip, the passive auxiliary
agrees with the plural DP Sachen (things) contained in a PP-adjunct. Due to the
extraposed relative clause, this plural DP is proximal to the verb only at surface
structure. As for extraposition, I follow Bring & Hartmann (1995) who argue that
the relative clause is right-adjoined to IP (or TnsP). Hence, at deep structure, the
Grammar as Processor
relative clause intervenes between the plural DP and the verb and consequently,
the singular DP Arbeit (work) is closest to the verb at that level. This is illustrated
in (59b) where the intervening relative clause isunderlined.
5.2.3.2 Local SVA in matrix clauses
Obviously, errors in matrix clauses are more informative when it comes to evaluating the locality patterns at deep versus surface structure because all German
matrix clauses are assumed to involve Tns-to-C movement of the verb followed
by movement of some XP to preverbal position. In my corpus, 63 out of 127 local
SVA-errors (49.6%) are attested in matrix clauses. For these errors, three different
locality patterns have to be distinguished. First, the error source may be local to
the verb at deep and surface structure. Second, the error source may only be local
to the verb at surface structure. Finally, there are also a few cases in which different
plural DPs are local to the verb at deep and surfacestructure.
In my error collection, there are 27 instances of matrix clause errors in which
the error source is local to the verb before and after XP-movement has applied. 22
of these are cases in which the verb agrees with a plural DP inside a complex subject DP, while only five are slips in which the verb agrees with a plural DP within
an object phrase. The error in (60a) exemplifies the former type of error. In this
error, the passive auxiliary agrees with the (zero-marked) plural noun Gemlde
(paintings). (60b) is an instance of the less frequent latter type. Here, the modal
verb agrees with the pre-verbal plural object Resultate(results).
(60) a. [der Wert der
Gemlde] werd-en
nicht,
the value the.gen.pl painting.pl be.pass-3.pl not,
wird
nicht bekannt gegeben
be.pass.3.sg not known part-give-part
Thevalueofthepaintingswillnotbedisclosed.
b. die
Resultat-e knn-en, h, kann
ich dir
noch nicht
the.pl result-pl can-3.pl, er, can.1.sg I 2.sg.dat yet not
sag-en
tell-inf
Icannottellyoutheresultsyet.
An interesting property of the errors in (60) is that in both cases, the DP which
passes on its number feature to the verb is more local, that is, adjacent to the
verb at surface structure. Although no other DP intervenes at deep structure,
additional material separates the error source and the verb. This is illustrated in
the simplified bracketed deep structures in (60) in which the material that would
intervene at deep structure is underlined. In the syntax, both verbs raise to Tns
and then to C.In (60a), the subject subsequently raises from SpecTnsP to SpecCP,
while in (60b), it is the direct object which moves from within VP toSpecCP.
(60) a. [CPC0[TnsPderWertderGemlde[VPnichtbekanntgegebenwerden]]]
b. [CPC0[TnsPich[VPdirdieResultatenochnichtsagenknnen]]]
More illuminating are, of course, cases in which the erroneous agreement source
is local to the verb only at surface structure because a singular DP would intervene between the error source and the verb at deep structure. In my corpus, there
are 30 such cases. In 25 of these, we observe erroneous agreement of the verb with
a plural noun within a complex subject DP (61a), while in only five, the verb agrees
with a plural object after topicalization(61b).
(61) a. [die Lnge der
Stck-e] spiel-en eine wichtige Rolle
the length the.gen.pl stick-pl play-3.pl an important role
Thelengthofthesticksplaysanimportantrole.
b. den
Student-en woll-en,
h, will
er die Lsung
the.pl student-pl want-3.pl, er, want.3.sg he the solution
Hewantstoshowthesolutiontothestudentsonlylater.
. The spontaneous English slip in (31), repeated below as (i), has similar properties. In this
error, wh-movement of the direct object has applied. Due to that movement operation, the
auxiliary to be appears adjacent to the plural noun things at surface structure (Levelt & Cutler
1983:206). At deep structure, however, adjacency of the error source things and the auxiliary
is not given, as is shown in (ii).
(i)
(ii)
What things are this kid, is this kid going to say correctly?
[CP C0 [TnsP this kid are going [VP to say what things correctly]]]
Grammar as Processor
b. [CPC0[TnsPer[VPdenStudentendieLsungerstspterzeigenwollen]]]
Finally, there are six more slips in my collection, in which different DPs are
local to the verb at deep and surface structure. The crucial difference to the
errors given in (61), however, is that both these DPs are specified for plural. In
all errors of this type, the verb erroneously agrees with a plural noun within a
complex subject DP. Consider, for example, the slip in (62a), in which the verb
agrees with the plural DP Studenten (students). As can be seen in the deep
structure representation in (62b), the material intervening between Studenten
and the verb contains another plural DP, namely Nerven(nerves).
(62) a. [das Genrgel der
Student-en] geh-en mir
manchmal
the nagging the.gen.pl student-pl go-3.pl 1.sg.dat sometimes
Sometimes,thenaggingofthestudentsannoysme.
b. [CP C0 [TnsP das Genrgel der Studenten [VP mir manchmal auf die Nerven
gehen]]]
Hence, in this slip, just as in the other five errors of this type, different plural DPs
are adjacent to the verb at deep and surface structure. In other words: at both
levels, a plural DP is more local to the verb than the singular head noun of the
complex subject DP, a characteristic which in a sense makes these errors comparable to the ones cited in (58) and(60).
5.2.3.3 Summary
Table (63) summarizes the locality patterns described in the previous sections
for the 127 slips under investigation, that is, slips in which the verb agrees with
a plural noun from within a complex subject DP or an object DP/PP. Obviously,
the label local SVA suggests that in all these errors, the error source is linearly closer to the verb than the head of the subject DP. In fact, in 88 out of
127 local SVA-errors (69.3%), the error source is local to the verb at deep and
surface structure. In addition, however, the distribution in Table (63) also shows
that in the 33 cases (25.9%) in which locality patterns differ at deep and surface
structure, the error source is local to the verb only at surface structure. In all
these cases, a singular DP would be linearly closer to the verb at deep structure. This pattern suggests that agreement is established after syntactic operations have been executed. As argued above, this is exactly the view taken by DM,
where agreement nodes are taken to be inserted at the postsyntactic level of
MorphologicalStructure.
(63) Locality of error source at deep (DS) and surface structure (SS) (n =127)
Agreement of verb with noun
within
in matrix clause
in embedded clause
22*
15
25
2
6
-
5*
46
5
1
Total
88
33
subject dp
= error source is local at deep and surface structure, but more local at surfacestructure.
The distribution in (63) raises the question what an SVA-error could look like in
which a plural DP is local to the verb only at deep structure. Various scenarios
are possible. One of these would involve, at surface structure, a singular subject
DP preceding the verb, a plural object following the verb, and a verb that shows
up in its plural form. Clearly, in such a case, the plural noun would only be local
to the verb at deep structure. For the sake of illustration, in (64a), I give a hypothetical error that satisfies these requirements. At surface structure, the verb follows the singular subject Umsatz (turnover), while at deep structure, the plural
object Erwartungen (expectations) directly precedes the verb. My corpus does not
contain an error with comparable structural properties (see Section 5.2.3.4 for an
alternativescenario).46
(64) a. der Umsatz bertreff-en unser-e khnste-n Erwartungen
the turnover exceed-3.pl our-pl boldest-pl expectation-pl
Theturnoverexceedsourboldestexpectations.
b. weil
sie fast
wortgetreu dasselbe ti ge-sag-t
hast
because she almost literally
the.same part-say-part have.2.sg
[wie du]i
weil
sie dasselbe ge-sag-t
hat
as you(sg) because she the.same part-say-part have.3.sg
becauseshehassaidalmostliterallythesameyoudid.
. The only error that comes close is the one cited in (29a) above. Remember, however,
that I classified this error as a blend. Moreover, it is different from the one in (64a) in that it
occurred in a passive construction and involves a topicalized PP.
Grammar as Processor
Still, there is a single error in my collection which points in the opposite direction,
that is, an error in which the DP transmitting its plural feature appears to be local
to the verb only at deep structure. This error is not included in Table (63) because
it is not one of the 127 slips that were analyzed in the preceding sections. Rather,
it is one of the slips in which the verb agrees with a nominative DP from within
a comparative phrase (see Section 5.2.2.3). This error is given in (64b). In this
utterance, an embedded clause, the verb appears in its 2nd person singular form
although the subject pronoun is specified for 3rd person singular. The comparative
phrase wie du (as you) containing the responsible second person pronoun has
been extraposed out of an object DP, as indicated by the coindexed trace in (64b).
What makes this error peculiar with respect to linear proximity is that at both
deep and surface structure, one element intervenes between error source and verb:
at deep structure the participial, at surface structure the comparative particle wie.
Because of extraposition, however, the syntactic distance between the pronoun
and the verb is bigger at surfacestructure.
I conclude that the general picture that emerges is that the local SVA errors
from my corpus are compatible with DM assumptions concerning feature
copy. The data analysis reveals that in erroneous subject-verb agreement, the
verb either agrees with a DP which is local to it at deep and surface structure
or with a DP that is local at surface structure only. The data therefore support the idea that the implementation of agreement nodes is executed only
after syntactic movement operations have taken place, that is, at the level of
MorphologicalStructure.
5.2.3.4 A note on local SVA in polar questions
These remarks on local SVA-errors in polar questions follow the summary of locality patterns in spontaneous speech errors because actually, to date, my error corpus
does not contain a single SVA-error in a polar question. However, given experimental results by Vigliocco & Nicol (1998), it is worth to briefly discuss thisoption.
Remember from Section 5.2.2 that Vigliocco and Nicol were able to elicit
SVA-errors in polar questions, that is, in a configuration in which the verb precedes the subject. In the experiment, participants first saw an adjective like safe
and then a phrase like the helicopter for the flights. Their task was to form the
question Is/was the helicopter for the flights safe?. It turned out that the frequency
and distribution of SVA-errors was similar to the patterns found in declarative
sentences with complex subject DPs. In particular, participants produced agreement errors like Are the helicopter for the flights safe?. This is a striking finding
given that in this condition due to subject-auxiliary inversion the verb is
not local to the plural noun. Therefore, Vigliocco and Nicol hypothesize that
these errors arise as a consequence of syntactic proximity, that is, deep structure
CP
TnsP
C
C
DP
Tns
[+] Tns
[]
AgrS
[+]
Tns
LP
D
[+] L
LP
PP
P
[for]
feature copy
tTns
DP
D
LP
[+]
[+]
Therefore, the agreement errors elicited by Vigliocco & Nicol (1998) appear not to be
consistent with the spontaneous speech error data from my corpus. I think, however,
that the validity of the experimental paradigm can be questioned. Given that, to the
best of my knowledge, there is no evidence for comparable spontaneous SVA-errors,
it does not seem unlikely that the participants use an agreement strategy that is different from the on-line strategy used in spontaneous speech. Remember that, in the
experiment, the participants knew that they were required to produce a question.
One may therefore hypothesize that they were already computing an inflected auxiliary for the sentence preamble while the preamble appeared on the computer screen
(duration: one second). After the preamble disappeared, they would then use that
auxiliary when forming the polar question. In other words: the participants were in
fact computing the auxiliary for the declarative sentence (in which the plural noun is
local to the verb) and not for the syntactic question structure given in(65).
Grammar as Processor
Following feature copy, the verb will move to AgrS in order to receive its morphological specification for agreement. In SV-languages, the subject will subsequently
leave its base position within VP and move to SpecAgrS, thereby creating a local
Spec-head relationship between the moved subject and AgrS. Franck et al. (2006,
2007) suggest that in SV-structures, agreement features are checked twice: first,
through AGREE, and secondly, under Spec-head agreement.47 The trees in (67),
adopted from Franck et al. (2006), illustrate the relevant structural configuration
before (67a) and after (67b) movement of the subject toSpecAgrS.
(67) a.
AgrSP
AgrS
Spec
AgrS
VP
Subj
V
AGREE
b.
V
Obj
AgrSP
AgrS
Subj
VP
V+AgrS
tSubj
Spec-head
tv
Obj
In order to test the role of precedence, dominance, and Spec-head relation in SVAerrors, Franck et al. (2006) tried to experimentally elicit SVA-errors in six different
structural configurations. In Franck et al. (2007), another six configurations were
added. In their experiments, Franck et al. used elicitation techniques similar to
those described in Section 5.2.1. Here, I will only report the results for some of
their experiments, based mainly on the discussion in Franck et al.(2007).
First, the authors consider the hypothesis that interference (attraction) in SVA
arises as a consequence of intervention. In a configuration like A > B > C, with >
being some structural relation, B intervenes between A and C.Applied to SVA, A
would be the subject, C the inflected verb, and B an intervening plural DP. Under
the intervention hypothesis, attraction is not expected when B is not intervening. In
Grammar as Processor
order to test the hypothesis that intervention is a necessary condition for interference, Franck et al. tested French structures like the ones given in (68a) and (68b).
In the former, a plural object follows the verb (configuration A > C > B); in the latter, a plural object is part of a matrix clause, while the complement clause contains
a singular subject (configuration B > A > C). In both configurations, the plural DP
does not intervene between singular subject and verb. As expected, no significant
effect of attraction was found for these two configurations, that is, subjects would
not produce plural verbforms.
(68) a. Lenseignant dcrit
les
roman-s
theteacher describe.3.sg the.pl novel-pl
Theteacherdescribesthenovels.
b. Jean dit
aux
patient-es que le mdicament gurit
Jean say.3.sg to.the.pl patient-pl that the medicine
cure.3.sg
Jeantellsthepatientsthatthemedicinecures.
The accusative clitic pronoun was found to trigger significantly more SVA-errors
than the intervening plural DP that is part of the complex subject DP. Hence, the
error distribution supports the hypothesis that an intervening element which
c-commands AgrS is more likely to cause an SVA-error than an element that only
precedes AgrS. Remember, however, that in all of the English data (spontaneous
and elicited), the erroneous agreement controller does not c-commandAgrS.
Finally, Franck et al. (2006, 2007) also investigated whether traces generated
by movement of a plural DP can disrupt subject-verb agreement. In order to test
this possibility, they used object cleft constructions with surface OSV order like
the one given in (68e). Presumably, this construction involves stepwise movement
of the object to the left periphery: first, it moves to SpecAgrO, which is located
between the VP and SpecAgrSP, then it moves on to the complementizer system.
Crucially, neither the objects base position within VP nor its final position intervene between AgrS and the subject. The only point in the derivation at which the
plural object does intervene is when it transits through SpecAgrOP. At this point,
it disrupts the AGREE relation between the VP-internal subject and AgrS. Franck
et al. found significant attraction rates in structures like (68e). Actually, the error
rate was higher than that observed with intervening object clitics (68d). They take
this as strong evidence for the assumption that intermediate positions can interfere inSVA.
(68) e. Cest les
ngociation-s que le ministre suspend
thatis the.pl negotiation-pl that the minister stop.3.sg
Itsthenegotiationsthattheministerstops.
f.
Cest les
ngociation-s que suspend le ministre
thatis the.pl negotiation-pl that stop.3.sg the minister
Itsthenegotiationsthattheministerstops.
Note that in (68e), the subject is still in SpecAgrSP at surface structure. That is,
while the first feature checking operation (AGREE) may be disrupted, the second (Spec-head agreement) is not. Things are different in (68f), an object cleft
construction in which stylistic inversion (SI) has applied, giving rise to an OVSorder. While the object undergoes the same movement operations as in (68e),
the subject does not leave the VP and therefore, SVA can only be established by
way of AGREE. Again, the error pattern shows that intermediate representations
are capable of triggering SVA-errors. In addition, however, Franck et al. found
that sentences with SI yielded a much stronger attraction effect than sentences
without SI. They relate this result to their double check hypothesis: agreement
based on AGREE only is more fragile than agreement based on AGREE and Spechead agreement. In (68e), the plural object may disrupt the AGREE relation (in
its intermediate position in SpecAgrOP) but SVA can be saved when the subject
moves to SpecAgrSP, thereby entering into a Spec-head relation with AgrS. The
latter operation is not available in (68f). In other words: once the AGREE relation
is disrupted, SVA cannot besaved.
The studies by Franck et al. (2006, 2007) illustrate in an impressive way how
psycholinguistic investigations can profit from theoretical syntax and vice versa.
Let us take a moment to reconsider the local SVA-errors from my corpus in the
light of their findings. The first thing to note is that there are no errors in my corpus that have the structure in (68a), that is, errors in which the plural DP would
Grammar as Processor
follow the verb. This is in accordance with Franck et al.s experimental results. For
French, this is expected because the plural DP does not intervene between singular subject and verb at any point in the derivation. For German with its underlying SOV-structure, however, it may be less expected because at deep structure,
the plural object intervenes between subject and verb, that is, it precedes and
c-commands the verb. Still, it has to be noted that even at this point, the plural
object does not interfere in the AGREE relation, which is established between
AgrS and the VP-internal subject. Hence, it seems that even in German, the position of the object vis--vis the verb does not disrupt the agreementrelation.
Why, then, are there numerous German errors in which the verb agrees with a
preceding plural object in an embedded clause (see Table (63))? Crucially, if AgrSP
is left-headed (as indicated in (67)), then the plural object neither precedes nor
c-commands AgrS at any point in the derivation since it remains in situ, while the
subject moves to SpecAgrSP (the resulting configuration being subject > AgrS >
object). Only when we assume right-headedness of AgrSP, we can account for this
rather common German error pattern. However, even under this assumption, the
plural object does not c-command AgrS; it only precedes it. But again, we would
have to ask the question why agreement of the verb with a post-verbal plural object
in a matrix clause (see the hypothetical error in (64a)) is not observed given that
in a right-headed configuration, the plural object should be able to disrupt the
AGREE relation in the same way as the intermediate trace in (68e) does. Surely,
SVA could be saved by Spec-head agreement at surface structure, but still, some
errors of this type should beobserved.
I argued above that the absence of such errors can be accounted for when we
assume that agreement is a surface structure phenomenon and that AgrS is adjoined
to Tns at MS (remember that I assume that TnsP is right-headed). According to
this analysis, the only relevant structural property is precedence because neither
the plural object in an embedded clause nor a plural DP contained within the subject DP c-command AgrS. Remember that there is only one error in my corpus in
which the plural DP (a post-verbal subject) does not precede the verb (see (29a)).
Actually, the only SVA-errors in my corpus in which c-command can be argued to
play a role are the few cases in which a plural object precedes the inflected verb in
a matrix clause; see, for instance, (60b) and (61b). In these errors, the object DP
occupies SpecCP and AgrS is adjoined to Tns after Tns-to-Cmovement.
In principle, precedence can also explain the occurrence of errors in the
experimental configurations in (68c) to (68f). After all, in all these configurations, the plural DP precedes AgrS. Yet, it has to be noted that precedence alone
cannot account for the differences in error rates between (68c) and (68d) on
the one hand and (68e) and (68f) on the other hand. Franck et al. (2007) argue
that lack of c-command in (68c) can be held responsible for the first difference,
while the second one can be accounted for by the lack of a Spec-head relation
between the subject and AgrS in (68f). I want to point out, however, that other
factors might also help us in accounting for the attested differences. On the one
hand, shorter syntactic distance between erroneous agreement controller and
AgrS could be argued to be responsible for the higher error rate in configuration
(68d). On the other hand, the increased error rate in configuration (68f) might
result from the fact that at surfacestructure, a singular DP intervenes between
the plural DP and the verb in (68e).
Without doubt, the analysis of elicited French SVA-errors within the Minimalist Program framework presented in Franck et al. (2006, 2007) is as innovative as it
is thought-provoking. The authors show that the attested errors patterns, in particular, differences in error probability, can be accounted for with the help of theoretical constructs such as c-command, AGREE, and Spec-head agreement. For some
of the configurations that were used in the experimental studies, there are no comparable examples in my corpus of spontaneous German speech errors. For others,
the same reasoning that Franck et al. suggest could be applied to the German data,
most importantly, to the errors in which the verb agrees with a plural DP within a
complex subject DP. These errors, however, are the least interesting when it comes
to their theoretical explanation because the factor responsible for their occurrence
is precedence and not one of the before-mentioned MP-specific mechanisms.
Crucially, as I have shown, a common German error pattern agreement of the
verb with a preceding plural object does not receive a straightforward explanation in the MP-model. While this area certainly awaits further investigation, the
German data suggest that precedence may play a more important role in SVA than
assumed in the Franck et al. studies. In addition, syntactic distance and intervention effects may come to fruition in the speecherrors.
5.2.5 Local agreement involving pronouns
Obviously, in German and in English, feature copy does not only take place
between a subject and a verb but also between a noun and a co-referent pronoun.
While the relevant features in subject-verb agreement are person and number, in
pronominal agreement, the gender feature does also play a role albeit to varying degrees in German and English (see Footnote 20 for studies looking at gender
agreement on predicativeadjectives).
Interference in the establishment of pronominal agreement has been investigated in a number of experiments which used elicitation techniques similar
to those previously described. Bock et al. (1999) and Bock, Eberhard & Cutting
(2004), for instance, focused on number agreement and investigated in how
far errors in verbal agreement (as discussed above) equal errors in pronominal
Grammar as Processor
agreement (also see Eberhard, Cutting & Bock 2005). In one of the sentence completion tasks, subjects were required to produce tag questions containing personal
pronouns. The results indicate that pronouns exhibit the same patterns of attraction as verbs. That is, subjects would occasionally produce tag questions in which
the pronoun agrees in number with a local plural noun, as in the representative
error in (69a). In addition, Bock et al. (1999) found that tag and reflexive pronouns
were the same in susceptibility to plural attraction(69b).
(69) a. Theactorinthesoapoperasrehearsed,didntthey?
b. Theactorinthesoapoperaswatchedthemselves.
In contrast, Meyer & Bock (1999) investigated the establishment of gender agreement, in particular, the production of gender-marked pronouns in Dutch. In
Footnote 39, I have already pointed out that Dutch has two grammatical genders:
neuter and common. Singular neuter nouns take the definite determiner het, while
singular common nouns require the definitive determiner de (the indefinite determiner een as well as the plural determiner de are not marked for gender). The corresponding gender-marked singular demonstrative pronouns are dat (neuter) and
die (common). In the experiments, participants heard a preamble sentence like the
one given in (70). Note that the nouns contained in this sentence are of different
gender but are bothindefinite.
(70) Kijk, daar ligt een aardappel bij
een badpak
Look there lies a
potato(c) next.to a swimsuit(n)
Look,theresapotatolyingnexttoaswimsuit.
der,
the.one.m
die
rel.f.acc
ich suche
I look.for
ThecolouroftheskirtisexactlytheoneImlookingfor.
b. [die Freundin
mein-es
Bruder-s] hat sein-en
Zug
the.f girlfriend(f) my-m.gen brother-gen has poss.3.sg.m-m train(m)
verpass-t
hat ihr-en
Zug
verpass-t
miss-part has poss.3.sg.f-m train(m) miss-part
Mybrothersgirlfriendhasmissedhertrain.
als mein-er
ist anders als mein-e
from poss.1.sg-m is different from poss.1.sg-f
Youropinionabouttheconflictisdifferentfrommine.
Moreover, six of the gender errors involve possessive pronouns. In (71), I give
two of these errors because they have different properties. In a sense, possessive
pronouns agree twice. First, the pronoun itself agrees in person and number (and
gender in the 3rd person) with an antecedent. Second, the pronoun takes a suffix to agree in gender with the noun it modifies. In (71b), it is the first type of
agreement that goes wrong. Instead of agreeing with the feminine noun Freundin (girlfriend), the possessive pronoun picks the local masculine noun Bruder
(brother) as antecedent. Within DP, however, gender agreement (with the masculine noun Zug (train)) is the same in the erroneous and the correct utterance (see
Slevc, Wardlow Lane & Ferreira (2007) for similar experimental data). (71b)
Grammar as Processor
exemplifies an error of the second type of agreement. Here, the features of the
possessive pronoun itself (1st person singular) remain the same, but the pronoun
agrees with the local noun Konflikt (conflict) ingender.48
The remaining 34 pronoun errors all involve relative pronouns. In 23 of these, it
is the gender feature that can be held responsible for the error. In all cases, the relative
pronoun agrees in gender with a local noun that follows the head noun in a complex
subject DP or an object DP. The former holds in (72a), where the relative pronoun
agrees with the gender feature of the noun Holz (wood) contained in a prepositional
modifier following the head noun. Given that in the plural, there is no gender distinction on relative prounouns, in the remaining 11 relative pronoun errors, it is only the
number feature that can be held responsible for the error. In the accusative and in
the nominative, the plural relative pronoun is die. In (72b), the selection of the plural
pronoun is triggered by the presence of the local plural nounZombies.
(72) a. [der
Vogel aus Palmen-holz], das
er mir,
the.m bird(m) of palm-wood(n) rel.n.acc he 1.sg.dat,
den
er mir
aus Ecuador mit-ge-brach-t
hat
rel.m.acc he 1.sg.dat from Ecuador with-part-bring-part have.3.sg
thebirdmadeofpalmwoodthathebroughtformefromEcuador
b. [der
Film
mit den
Zombie-s],
die
er mir
the.m movie(m) with the.pl.acc zombie(m)-pl rel.pl he 1.sg.dat
ge-lieh-en,
den
er mir
geliehen
hat
part-lend-part, rel.m.acc he 1.sg.dat part-lend-part have.3.sg
themoviewiththezombiesthathehaslentme
c.
die
Anbieter
von kologisch-er Kleidung, die
auch
the.pl supplier(m).pl of ecological-f clothing(f) rel.pl also
soll
shall.3.sg
thesuppliersofecologicalclothingthatshouldalsobeattractive
. Just as in error (2c) at the beginning of this chapter, the possessive pronoun is part of an
elliptic comparative construction. In such an elliptic construction, the form of the masculine
pronoun is different from when it co-occurs with an overt noun. Compare the sentence in (i)
with that in (71c).
(i)
In the intended utterance in (72b), the relative clause is supposed to modify the
head noun of the complex subject DP. It is worth noting that in two of the relative
pronoun errors, the relative clause modifies the head noun instead of the local
noun, that is, the relative pronoun erroneously agrees with a non-local noun. In
both these cases, the local noun is feminine. Given that the feminine and the plural relative pronoun are homophonous (die), it is only the form of the verb within
the relative clause that indicates that the relative pronoun is indeed specified for
plural. In (72c), plural agreement on the verb within the relative clause tells us that
the relative pronoun agrees in number with the non-local plural noun Anbieter
(suppliers) instead of the local feminine noun Kleidung (clothing). The resulting utterance is not ungrammatical; it is not even particularly awkward. What is
being said is that the suppliers should be attractive but not necessarily the clothing.
Clearly, however, this was not the intention of the speaker. Rather, the preceding
conversation dealt with the regrettable fact that usually, ecological clothing is not
particularlyglamorous.
In sum, the pronoun data show interesting parallels to the SVA-errors.
Although in most cases, it is the gender feature (rather than the number feature)
that is responsible for the error, the structural conditions that facilitate the error
appear to be the same. Except for two out 45 errors, that is, in 95.5% of all cases,
the error source is more local to the agreement target than the correct agreement
controller just as in the SVA-errors. Moreover, in all these cases, the local noun is
part of a genitive modifier or a prepositional complement within a complexDP.
5.2.6 Summary
In this section, I have considered different types of local agreement errors in German. Within this error group, two types of errors have to be distinguished: local
SVA-errors and local errors involving pronouns. My corpus contains 149 slips of the
former and 43 slips of the latter type (remember that two of the pronoun errors discussed in the previous section are non-local), that is, a total of 192 local agreement
errors, the distribution of which is given in Table (73). In this table, I distinguish different features that due to defective feature copy may be responsible for the error.
Clearly, the number feature is the most important one for the SVA-errors, while the
gender feature plays the most prominent role in the pronoun errors. In contrast to
previous tables, I do not distinguish between singular and plural error sources. The
reader will remember that, just as in the spontaneous and elicited English errors,
there are very few cases in which a local SVA-error is caused by a singular local noun.
I have argued that within DM, this tendency can be accounted for by assuming that
singular nouns are not specified for number. As for the SVA-errors, there are also a
few spontaneous slips that are due to erroneous copy of the personfeature.
Grammar as Processor
number
74
61
5
person
gender
140
2
2
5
9
9
149
34
23
6
5
34
The analysis of the spontaneous German errors also brought to light an interesting difference to the elicited English data. In German, it is not uncommon for a
verb to erroneously agree with a noun from within an object DP or an adjunct PP,
while, due to structural differences, similar agreement patterns are not attested
inEnglish.
I want to conclude this section with a brief typological note on SVA-phenomena.
To the best of my knowledge, there are no languages in which agreement of the
verb with a noun from within a complex subject DP would be the rule. That is,
local agreement of the type discussed in Section 5.2.2.1 appears not to be attested
in natural languages.49 Things are quite different with respect to object agreement,
of course. At least agreement of the verb with a direct object is a very common
phenomenon cross-linguistically. In my corpus, there are 32 slips in which the verb
agrees in number with a direct object (out of a total of 63 object/PP-adjunct agreement errors; see Table (37)). For these cases, it could be argued that the speaker
applies an agreement strategy which, in principle, is provided by UG but is actually
not available in her/his language. It has to be noted, however, that in these errors,
object agreement does not combine with subject agreement. Rather, it substitutes
subject agreement. Cross-linguistically, this is a marked pattern because usually,
object agreement exists in addition to subject agreement (occasionally realized
by a portmanteau morpheme). Finally, agreement with an indirect object or a
DP from within a prepositional phrase is much less common cross-linguistically,
although it does exist (Bobaljik & Wurmbrand 2002; Corbett2006).
5.3. Defective feature copy II: Long-distance agreement
It could be argued that the local SVA-errors discussed in the previous section,
as intriguing as they are, are still the expected case. If, for whatever reason, the
agreement target picks an erroneous agreement controller, then it seems plausible that it picks one that is closer to home (remember that this implies linear
distance, not syntactic distance). Interestingly, however, my corpus also contains
a fair number of slips which contradict this prediction, that is, slips in which
the erroneous agreement controller is clearly more distant to the verb than the
correct controller (the head of the subject DP). I refer to errors of this type as
long-distance agreement (LDA) errors. For the most part, in LDA-errors, an
agreement relation is established between a matrix clause and an embedded
clause. For the sake of illustration, consider the example in (74a). In this error,
the matrix verb wissen (to know) agrees in person and number with the 2nd person plural pronoun ihr which is the subject of the finite embedded clause. While
for some errors of this type, it might be argued that they do not exemplify LDA
but are the result of a suffix anticipation or perseveration, such an account is not
available for (74a) because the matrix verb is subject to a stem-internal change in
its 2nd person pluralform.
(74) a. ich wiss-t,
[dass ihr
nicht Recht hab-t]
I
know-2.pl that you(pl) not right have-2.pl
ich wei,
dass ihr
nicht Recht hab-t
I know.1.sg, that you(pl) not right have-2.pl
Iknowthatyourenotright.
b. damit du,
[wenn jemand komm-st, komm-t], nicht
so.that you(sg) when someone come-2.sg come-3.sg not
ge-hinder-t
wir-st
part-hinder-part be.fut-2.sg
Sothatyouwontbehinderedwhensomeonecomes.
With the notable exception of Meringer & Mayer (1895), to date only few studies
on speech errors have acknowledged the existence of spontaneous LDA-errors.
Grammar as Processor
Amongst the examples that Meringer & Mayer (1895:42f) cite is the one given in
(74b). In this error, in contrast to the one in (74a), the embedded verb kommen
(come) agrees with the second person singular subject of the matrix clause (also
see MacKay (1979:485) and Pfau(2003)).
Different types of LDA-errors from my corpus will be presented in Section 5.3.2.
Before turning to the error data, however, I will point out in Section 5.3.1 that
LDA, as unusual as it may seem, is also attested as a regular phenomenon in some
languages. In Section 5.3.3, I summarize the main findings and compare the patterns attested in the speech errors to the regular LDA-patterns to be described in
the nextsection.
5.3.1 LDA in natural languages
Long-distance agreement refers to the phenomenon of a verb agreeing with an
argument that is not its own. LDA-patterns have been described for a number of unrelated languages, such as, for instance, Hindi-Urdu (Mahajan 1989;
Bhatt 2005), the Daghestanian languages Godoberi and Tsez (Haspelmath 1999;
Polinsky & Comrie 1999; Polinsky & Potsdam 2001), the Chukotko-Kamchatkan
language Itelmen (Bobaljik & Wurmbrand 2002, 2005), and the Algonquian languages Passamaquoddy and Innu-aimn (Bruening 2001; Branigan & Mackenzie
2002). In the following, I will focus on the main characteristics of LDA in Tsez
andHindi.
Polinsky & Comrie (1999) provide a description of LDA in Tsez, a Daghestanian language spoken in the Northeastern Caucasus (see Polinsky & Potsdam
(2001) and Bobaljik & Wumbrand (2005) for syntactic analyses). Tsez has four
noun classes. Noun class (just like gender in German) is a covert category
which, however, surfaces in agreement prefixes on verbs and modifiers. Crucially, verbal agreement can only be triggered by an absolutive DP. In cases
in which the absolutive argument of a verb is a sentential complement, Tsez
offers two possibilities for agreement: either the matrix verb agrees with the
sentential complement as a single complex phrase, thus assigning it class 4, or
the matrix verb agrees with the absolutive DP inside the complement clause,
as illustrated in (75a). In this example, class 3 agreement on the matrix verb
iy (know) can only be triggered by the DP magalu (bread) within the complement clause because this is the only absolutive DP in the sentence (note
that in Tsez, most proposition-attitude verbs, such as to know, take the experiencer argument in the dative). As a consequence, both the embedded and
the matrix verb appear with the same agreement prefix (Polinsky & Comrie
1999:117). Example (75b) shows that Tsez also permits LDA out of infinitival
complements. In this example, class 2 agreement on the predicative adjective
hai]
be.pres.3.sg
FirozthoughtthatMonasingsghazals.
. Similar patterns have been described for Godoberi, another Daghestanian language,
which, however, is not closely related to Tsez (Haspelmath 1999). In Godoberi, LDA is only
observed with two types of sentential complements, namely infinitival complements and converbal complements (used with the complement-taking verbs begin and finish). As in Tsez,
matrix verbs may agree in gender and number with an embedded absolutive (direct object)
argument. In (i), for instance, the matrix verb eu (forget) agrees with the absolutive neuter
plural argument gyazeta (newspaper) contained in the infinitival complement (Haspelmath
1999:131; cont = contessive case).
(i) ali-u
[gyazeta-be
r-ax-i]
r-eu-a
Ali-cont paper-pl.abs pl.n-take-inf pl.n-forget-aor
Ali forgot to buy newspapers.
Grammar as Processor
Before turning to the German speech errors, I wish to point out that by citing
the examples in (75) and (76), I do not mean to suggest that regular LDA in
Tsez and Hindi and erroneous LDA in German speech errors exemplify the same
phenomenon in fact, they probably dont. Still, I take the above examples to be
illuminating in that they illustrate that verbal agreement in natural languages is
not necessarily confined to a single clause. Rather, agreement features may be
transferred across clause boundaries, and it is exactly this kind of transfer which
also manifests in the speech error data. In Section 5.3.3, I will provide a comparison of the two data sets in terms of agreementdomains.
5.3.2 LDA in speech errors
Let me now turn to the patterns of long-distance agreement as attested in my
corpus which, at present, contains 56 spontaneous LDA-errors. As far as the
relation between matrix and embedded clause is concerned, we are going to see
that in the speech errors, LDA can go both ways. On the one hand, a matrix
verb can agree with an embedded argument, as in (74a); errors of this type will
be discussed in Section 5.3.2.1. On the other hand, agreement of an embedded
verb with an argument of the matrix clause is also attested (see (74b)). Slips that
follow this pattern are introduced in Section 5.3.2.2. In addition, there are also a
few cases in which LDA is observed between conjoined clauses. This special type
of LDA, which I refer to as anticipatory agreement, will be subject to discussion
in Section5.3.2.3.
5.3.2.1 Matrix verb agrees with embedded argument
Based on the available data, we can draw two conclusions concerning regular LDA. First, regular LDA always involves agreement of a matrix verb with
an embedded argument. I will therefore discuss spontaneous errors exhibiting
a comparable pattern first. There are 23 such errors in my corpus. Secondly,
agreement of a matrix verb with an argument from a finite embedded clause
is more marked. The data discussed in the literature suggest that languages
that allow for this type of LDA (for example, Tsez and Passamaquoddy), also
allow for LDA with an argument from a non-finite clause, while the opposite
is not the case. In Hindi and Itelmen, for instance, LDA is restricted to infinitival clauses. Hence, the first error data I will present are those that exhibit the
unmarked pattern: agreement of a matrix verb with an argument from a nonfinitecomplement.
Given that the subject position is phonologically empty in German infinitival
clauses, in these cases, the agreement controller can only be a direct or indirect
object. Actually, all seven cases from my corpus, two of which are given in (77),
involve LDA with a direct object. In (77a), the embedded clause containing the
plural noun Zitronen (lemons) intervenes between the 3rd person singular subject and the matrix verb vergessen (forget). Instead of with its own subject, the
matrix verb agrees with the embedded object. The error in (77b) is structurally
similar. In this case, however, the error source is a 2nd person singular pronoun
contained in the non-finite clause. That is, the feature responsible for the error is
the person feature, not the numberfeature.51
(77) a.
dass er [die
Zitrone-n zu kauf-en] vergess-en
that he the.acc.pl lemon-pl to buy-inf forget-3.pl
thatheforgetstobuythelemons.
b. weil
sie [dich
schon ewig
zu treff-en] versuch-st,
because she 2.sg.acc already for.ages to meet-inf] try-2.sg,
h, zu treff-en versuch-t
er, to meet-inf try-3.sg
becauseshesbeentryingtomeetyouforages.
In (78), I give a syntactic structure for the slip (77a). In this structure, the matrix
verb selects an IP-complement the specifier position of which is occupied by PRO.
As in previous structures, the arrows along the branches of the tree illustrate the
syntactic path the relevant feature of the local noun (number in this case) has to
take to reach the agreement node. Clearly, this path, which crosses four maximal
projections, is longer than that of the features of the subject DP in SpecTnsP. Linearly, the error source may be as as close to the verb as the error sources in many
of the local agreement errors discussed in Section 5.2. The crucial difference, however, is that in (78) a clausal projectionintervenes.
Grammar as Processor
TnsP
(78)
Tns
DPi
[3rd]
[]
LP
Tns
IP
Spec
PROi
tL1
I
L1
LP
DP
[+]
tL2
Tns
Tns
AgrS
[]
[+]
feature copy
Let us now turn to the cases in which a matrix verb agrees with an argument of a
finite embedded clause. Interestingly, this type of LDA is more common in my corpus, despite the fact that it is the more marked type in natural LDA. My corpus contains 16 slips that follow this pattern. 13 of these slips contain a complement clause,
the other three an adverbial (adjunct) clause. Here, I will only present two errors of
the former type. In (79a), for instance, we are dealing with a main clause containing
a complement clause. In this case, the matrix verb wollen (want) agrees in number
with the the plural subject of the that-clause, which follows the verb. (79b) is different in that the embedded CP intervenes between the subject pronoun and the verb
in an embedded clause actually, a marked structure in German because usually,
complement clauses are extraposed to sentence-final position in embedded contexts
(Bring & Hartmann 1995). In the error, the matrix verb bedauern (regret) erroneously agrees with the intervening direct object pronoun in the complementclause.
(79) a. ich woll-en,
h, ich will
eigentlich, [dass wir uns
I want-1.pl, er, I
want.1.sg actually
that we rec
ft-er
seh-en]
often-comp see-pl
Iactuallywantforustoseeeachothermoreoften.
b. weil
sie manchmal, [dass
because she sometimes that
sie dich
kenn-t],
bedauer-st,
she 2.sg.acc know-3.sg regret-2.sg
h, bedauer-t
er, regret-3.sg
becausesometimessheregretsthatsheknowsyou.
When finite complements are involved, LDA-errors in which the matrix verb
agrees with the subject of the embedded clause (as in (74a) and (79a)) appear to
be more common than errors in which the matrix verb agrees with an embedded object (79b). Actually, only three of the errors from my corpus are of the
lattertype.
In (80), I provide a syntactic structure for the error (79a). Note that the
adverbials are neglected in this structure. In this error, it is the root woll that
takes a CP-complement. This root moves to matrix Tns and then on to C where
AgrS is adjoined to Tns. Since the embedded as well as the matrix verb agree with
the embedded plural subject, we must assume that the agreement features of the
embedded subject are copied twice, as indicated in the structure. The same is, of
course, also true for all the other errors in which a matrix verb agrees with the
subject of a finite embedded clause. In all of these cases, two verbs agree with one
subjectDP.
(80)
CP
C
DP
[1st]
Tns
L1
TnsP
Tns
tDP
Tns
AgrS
[]
[+]
LP
CP
C
[dass]
long-distance
feature copy
Tns
tTns
tL1
TnsP
DP
Tns
[1st]
[+]
Tns
LP
...tL2
Tns
L2
Tns
[]
regular feature copy
AgrS
[+]
Grammar as Processor
bist]
be.2.sg
Weunderstandwellthatyouareannoyed.
b. dir
ist
doch
vollkommen wurscht, [ob
er
2.sg.dat be.3.sg mod.part completely sausage, whether he
Recht hast,
ob
er Recht hat
oder nicht]
right have.2.sg, whether he right have.3.sg or
not
Youcouldntcarelesswhetherhesrightornot.
In (82), I give a structure for the error in (81a). While this structure is almost
identical to the one given in (80), the agreement copy process is completely different. In this example, there is no agreement feature copy within the lower CP.
Rather, as indicated by the arrows, both agreement nodes receive the feature
[+pl] from the matrix subject position. While regular feature copy and LDA
take different routes in (80) one going down, the other up they both proceed
downwards in (82). In both cases, however, LDA crosses one CPboundary.
(80)
CP
C
DP
[1st]
[+]
Tns
L1
TnsP
tDP
Tns
Tns
[]
LP
AgrS
[+]
CP
C
regular feature copy
[dass]
Tns
tTns
tL1
TnsP
Tns
DP
[2nd]
LP
Tns
tL2
L2
Tns
Tns
long-distance
feature copy
[]
AgrS
[+]
In six slips, the embedded verb which erroneously agrees with the matrix subject
is not contained in a complement clause but rather in an adjunct clause. This is
true, for instance, for example (83a), which contains a temporal adverbial clause in
sentence-initial position. The verb of the adverbial clause agrees with the second person singular pronoun of the matrix clause. Finally, in example (83b), I present one of
four slips in which the verb in a relative clause agrees with an argument of the main
clause instead of with the subject of the relative clause. In this example, the plural head
noun Dinge (things) is the subject of the matrix clause but the object of the relative
clause; the subject of the relative clause is the singular DP kein Mensch(nobody).
(83) a.
h, hat],
war-st
du wahrscheinlich
er, have.3.sg be.past-2.sg you probably
Thelasttimeshevisitedus,youwereprobablyonholiday.
in Urlaub
on holiday
Grammar as Processor
b. hier passier-en
Ding-e, [die kein Mensch fr mglich
here happen-3.pl thing-pl rel.pl no person for possible
ge-halt-en
hab-en]
part-hold-part have-pl
Therearethingsgoingonthatnobodyconsideredpossible.
It should be noted that some of the errors which I classify as LDA-errors might
also be instances of phonological errors. The slip in (83a), for instance, could also
be analyzed as a phonological anticipation of -st, which is part of the coda of the
following verb. As pointed out earlier, only few of the speech errors are perfect
in the sense that they allow for an unambiguous analysis. However, slips like (81a),
which involve suppletive verb forms, are clear evidence that LDA is attested in the
spontaneousdata.
5.3.2.3 A special case: Anticipatory agreement
So far, all the LDA-phenomena I have discussed involve the interaction between a
matrix and an embedded clause. Interestingly, however, my error collection contains yet another type of LDA-error. In this type, two clauses are conjoined in a
coordination structure and the verb of the first conjunct erroneously agrees with
the subject of the second conjunct. I am referring to this special case of longdistance agreement as anticipatory agreement (Anderson 1992:112). There are
13 such errors in my collection, two of which are given in (84). In (84a), two
embedded clauses are coordinated and the verb of the first conjunct agrees with
the first person singular pronoun of the second conjunct. Given the suppletive
nature of the inflected form of the verb sein (to be), I consider this error a clear
case of anticipatory agreement. Things are somewhat different in (84b). In this
error, two matrix CPs are combined. The verb kommen (to come) in the first conjunct surfaces with the second person singular agreement suffix, thereby agreeing
with the subject of the second conjunct. Note, however, that the verb is the same
in both conjuncts and is gapped in the second clause. Presumably, before deletion
of the second verb, the structure of the second conjunct was du kommst hoffentlich auch. It might therefore be the case that we are dealing with a phonological
or word anticipation rather than an LDA-error. That is, some material was anticipated before it got deleted in the secondconjunct.
(84) a. weil
er wtend bin
und ich keine Lust
hab
because he angry be.1.sg and I no
inclination have.1.sg
weil
er wtend ist
because he angry be.3.sg
becauseheisangryandIdontfeellikedoingit.
Mybrotherwillcomeandhopefullyyou,too.
TnsP
TnsP
DP
[3rd]
[]
AP
BP
B
Tns
VP
[]
Tns
tL1 L1
[]
AgrS
LP
LP
[]
[+]
Tns
tL2 L2
DP
[1st] D
long-distance
feature copy
Tns
DP
[1st]
Tns
Tns
TnsP
Tns
Tns
[]
AgrS
[1st]
While in the errors discussed in the previous subsections, the embedded clause
is dependent on the matrix clause and cannot appear on its own, in anticipatory
agreement, the two clauses are completely independent of each other, a fact which
makes the erroneous feature copy appear even more exotic. After all, in case of a
dependency relation, it may seem more likely for an agreement feature to cross a
clausalboundary.
. In (85), BP stands for Boolean Phrase, the head of which may be occupied by Boolean
operators such as und (and), oder (or), and sowie (as well as). For further aspects of the
syntax of coordination see Hartmann (2000).
Grammar as Processor
The Papuan examples resemble the speech error data in that the first conjunct
verb always agrees with features of a second conjunct subject. To date my corpus does not contain an error in which agreement features would be perseverated in a coordination structure. Still, there is an important difference between
the spontaneous errors and regular anticipatory agreement. In the speech errors,
anticipatory agreement overwrites regular subject agreement, while in the
Papuan examples, regular subject agreement and anticipatory agreement cooccur, that is, the first conjunct verbs in (86) agree with their own subject and
with the subject of the second conjunct. In contrast, such double marking is not
attested in the speecherrors.
5.3.3 Summary: Agreement domains
The spontaneous speech errors presented in the preceding sections illustrate
that in the German errors, verbs may not only erroneously agree with a local
DP be it part of a complex subject DP or part of an object phrase but also
with the subject/object of another clause. In long-distance agreement, two
. Similar clause chaining strategies have been described for Fore (Scott 1978), Awa
(Loving & McKaughan 1973), and Hua (Haiman 1980). For a comprehensive survey of clause
chaining in Papuan languages see Foley (1986:175ff).
subject
10
3
9
6
4
object
13
10
7
3
-
19
1
1
-
anticipatoryagreement
13
Total
45
11
I have also shown that some of the LDA-patterns are attested as a regular phenomenon in various languages. For instance, LDA of a matrix verb with an argument of
an embedded clause is attested in Hindi as well as in some Daghestanian languages,
while anticipatory agreement of a first conjunct verb with a second conjunct subject has been described for some languages spoken in Papua NewGuinea.
Still, there are some crucial differences between the regular and the error
data. First of all, regular LDA always involves agreement of a matrix verb with a
constituent inside the verbs clausal complement. Clearly, this is not true for the
German speech errors, where an embedded verb may also agree with a matrix
argument, as has been shown in Section 5.3.2.2. That is, 20 out 56 LDA-errors
exhibit a pattern which, to the best of my knowledge, is not attested as a regular
agreement pattern. Hence, we are left with 23 slips in which the matrix verb
agrees with an embedded argument. Remember, however, that in Tsez and Hindi
a matrix verb can only agree with an embedded object (according to Bobaljik &
Wurmbrand (2005), the same is true for Itelmen). In Tsez, this is the absolutive
argument within a finite (75a) or non-finite (75b) complement; in Hindi, it is
the accusative object within a non-finite complement (but only when not overtly
case marked, (76a)). Generally, agreement of the matrix verb with an embedded
Grammar as Processor
subject appears to be highly uncommon.54 Therefore, we are left with only ten
spontaneous errors which resemble the regular LDA-patterns as described in
Section 5.3.1. Out of these ten errors, seven involve infinitival complements,
which could be taken as evidence that LDA out of infinitival complements is
lessmarked.
Obviously, these data pose a challenge to theories that assume that a strictly
local relationship such as the specifier-head relationship is necessary for agreement. Based on their discussion of Itelmen and Tsez data, Bobaljik & Wurmbrand
(2005) suggest two scenarios for the establishement of LDA. For the Itelmen case,
they argue that the infinitival clause constitutes an agreement domain and that
agreement with the matrix predicate (with the matrix vP, in their terms) is not
possible without movement. Therefore, the object has to undergo LF-movement to
the matrix clause (to SpecvP). In other words: what looks like LDA is really covert
movement into a higher agreementdomain.
Things are different in Tsez. In order to account for the Tsez data, Bobaljik &
Wurmbrand (2005) follow a suggestion made by Polinsky & Potsdam (2001).
First, they borrow an idea from phase theory (Chomsky 2000), in particular,
the edge condition, according to which elements at the high periphery of one
locality domain are accessible to the next higher domain. Applied to the Tsez
data this means that the specifier of the embedded CP would be accessible for
agreement with the matrix predicate. In the Tsez example (75a), however, the
absolutive argument is not at the periphery of the embedded clause. Therefore,
they also adopt Polinsky and Potsdams proposal that an agreeing DP carries
an obligatory topic interpretation. Hence, at LF this DP moves to the specifier
of a topic phrase in the left periphery of the embedded clause. According to
Polinsky & Potsdam (2001: 585), [i]n this position, the topic is in a sufficiently local configuration with the embedding verb with which it agrees, as is
illustrated in(88).
. Data presented by Branigan & MacKenzie (2002:388) indicate that Innu-aimn is an exception to this generalization. In this language, a matrix verb which takes a complement clause
optionally agrees with the embedded subject or the embedded direct object. In (i) the matrix
verb takes the suffix -at, thereby showing agreement with the (coordinated) plural subject of
the finite embedded clause.
(i)
(88)
TnsP
Tns
Spec
Tns
Tns
VP
AgrS V
agreement
domain
TopP
Spec
Top
Top
IP
While these proposals may be convincing for the languages under consideration,
it is not clear how they can help us in accounting for the spontaneous errors. Note
that according to these analyses, agreement is determined at LF, not at MS. In
other words: agreement (and presumably other morphosyntactic processes) follows all syntactic operations, not just overt syntax. I assume that this view is still
compatible with the course of the derivation as sketched in Sections 3.2.2 and
5.2.3 provided that feature copy at LF feeds Vocabulary insertion at PF. For the 23
errors in which the matrix verb agrees with an embedded argument we might then
assume that at LF, the responsible argument (a subject or an object) is erroneously
moved to a position in which it is sufficiently local to the matrix verb, or rather to
the matrix agreement node. Remember that in the German errors, just as in Tsez,
the matrix verb only agrees with the embeddedargument.
This scenario, however, is highly stipulative. Also, it does not account for the
cases in which the embedded verb agrees with a matrix argument. For these configurations, it is hard to imagine how the agreement source and the agreement
target could end up in a sufficiently local configuration at LF. The same, of course,
holds for the cases of anticipatory agreement. Certainly, in this configuration, the
subject of the second conjunct is neither interpreted in the first conjunct (which
might motivate LF-movement) nor can it be moved to a peripheral position where
it would be accessible for agreement with the first conjunct predicate. I therefore
conclude that the analyses proposed for regular LDA-phenomena do not offer
plausible explanations for the error data. What happens in these errors is exceptional in the sense that an agreement feature targets an agreement node which
should be inaccessible because it lies outside of the agreement domain of that feature. Hence, it appears that for unknown reasons, the agreement domain becomes
permeable, thereby allowing for feature transmission across the domain boundary.
Grammar as Processor
properties (Fromkin 1971:43). In contrast to (89a), however, this slip could also be
analyzed as the exchange of complete DPs because both nouns are accompanied
by a definite determiner.55 Clearly, such an analysis is not available for(89a).
(89) a. ich habe der
Hilfe fr ihr-e
Kind-er ge-dank-t
I
have the.f.dat help(f) for their-pl child-pl part-thank-part
den
Kind-er-n
fr ihr-e
Hilfe
the.pl.dat child-pl-dat for their-f help(f)
Ihavethankedthechildrenfortheirhelp.
b. examinethehorseoftheeye-s theeye-softhehorse
My corpus contains 23 errors which involve number stranding. All of these errors
are either noun exchanges or incompletes, that is, errors that are self-corrected by
the speaker after the first error element. In (90a), for instance, wort (word) and
buchstabe (letter) are exchanged and wort combines with the stranded plural feature. This slip also illustrates that we are in fact dealing with feature stranding and not with suffix stranding since, after the exchange has taken place, the
appropriate plural allomorph is chosen for wort and umlaut formation is triggered. Things are not as clear in the error in (90b). Given that the exchanged roots
stufe (step) and treppe (stair) take the same plural affix, this error could also
be analyzed as an instance of suffix stranding. In other words: this error is ambiguous between a root exchange at MS and a (phonological) stem exchange at PF. The
same holds for the English exchange in (90c) (Fromkin 1973b:258).
(90) a. ein Buchstabe ist vier Wrt-er lang, h,
a letter
is four word-pl long, er,
Awordisfourletterslong.
Howmanystepsdoesthisstairactuallyhave?
c. aholefulloffloor-s afloorfullofhole-s
. That the exchange of whole DPs is in fact an option, is illustrated by the slip in (i) in
which the respective determiner positions contain different material (Garrett 1980a: 192).
Hence, this error can only be analyzed as an exchange of [DP a discussion] and [DP this guy].
(i)
I got into [this guy] with [a discussion] into a discussion with this guy
Grammar as Processor
Stemberger (1985) reports that in his corpus, stranding of the plural, as in (90), is
about four times as frequent as non-stranding cases, such as (89). When checking the Frankfurt corpus for noun exchanges, it contained only 18 informative
errors, that is, errors in which a plural and a singular noun interact. In 14 of these
exchanges, however, the plural feature strands a ratio very similar to that reported
byStemberger.
Generally, the feature [+pl] is selected from List 1 and enters the computation whenever a node multiple(x) is active at the conceptual level. In tree
structures presented in the previous sections, it was assumed that roots share
a terminal L-node with the feature [+pl] (see, for instance, (78)). It has to
be noted, however, that this is a simplification. According to recent theorizing, a
plural feature projects its own functional projection which is located between D
and the lexical head L.In (91), I give a more elaborate structure for the plural DP
ihre Kinder (their children) from the error (89a). Following a suggestion made
by Ritter (1991, 1995), I label the relevant functional projection NumP (for functional categories within the noun phrase see also Szabolcsi (1994) and Vangsnes
(2001)). Presumably, in the syntax, the root will raise to SpecNumP. Also, the
plural feature will be copied ontoD.
(91)
DP
D
[]
NumP
Num
LP
[+]
Based on this structure, we can make an interesting prediction. In a sentence containing the DP in (91), three types of exchanges should be possible. I will illustrate
these three types by means of the intended utterance of the slip (89a). First of all,
LP could be accessed in the error. In this case, we would observe number stranding
as well as stranding of the material in D, that is, an error similar to (90a); see (92a)
for the resulting hypothetical error. The second option involves the manipulation
of NumP. In this case, the plural feature is taken along but the material in D is left
behind. This is actually what we observed in (89a). Thirdly, as has been pointed out
above, the whole DP could be the target of the exchange, yielding the hypothetical
error given in (92b). What should be impossible, however, given the structure in
(91), is an error in which the DP is exchanged but the number feature is stranded.
The result of such an operation would look like (92c). Interestingly, this prediction
is borne out: neither the Frankfurt corpus nor my own corpus contains an error
that would exhibit such apattern.
(92) a.
den
Kind-er-n
fr ihr-e
Hilfe
the.pl.dat child-pl-dat for their-f help(f)
Ihavethankedthechildrenfortheirhelp.
Note that despite the additional functional structure argued for in (91), in the
structures to follow, I will stick to the convention of representing the plural feature
under the same node as the root that it combineswith.
Let us now turn to the shift cases. As for the plural feature, my corpus contains
only seven errors which exemplify this type of feature manipulation. The error in
(93a) is the only one in which [+pl] is really shifted in the sense that it is detached
from one root and combines with another root. The plural feature is shifted from
buch (book) to schuber (slipcase) more precisely, from the head of the first
NumP to the head of the second NumP. Subsequently, the plural feature is copied
onto the definite article which is spelled out correctly. The DP vier Buch (four book),
however, is ungrammatical, simply because there is no way to adjust the numeral vier
to a singular noun. What is remarkable about this slip is the fact that the verb does
not accommodate to the new situation, that is, to the singular noun in subject position. This is unexpected given that the error must have taken place before features are
copied; otherwise accommodation of the determiner could not be explained. Possibly, insertion of the plural copula sind is motivated by the presence of the numeral.
Otherwise, we would have to assume that the plural feature is copied onto AgrS
before it is shifted to the following noun. This, in turn, would imply that subject-verb
agreement is established before DP-internal agreement and that an error can occur
between the two copy mechanisms. Given that my corpus contains only one error of
this type, I will not attempt to decide between these twoexplanations.
(93) a.
h, vier Bch-er im
Schuber
er, four book-pl in.the.m.dat slipcase(m)
Inthisedition,therearefourbooksintheslipcase.
b. weil
er in den
letzte-n Woche-n sein-e Shn-e,
because he in the.pl.dat last-pl week-pl his-pl son-pl,
h, sein-en
Sohn nicht ge-seh-en
hat
er, his-m.acc son(m) not part-see-part have.3.sg
becausehehasntseenhissoninthelastweeks.
Grammar as Processor
In the other six shift errors, we are dealing with an anticipation or perseveration of
the plural feature rather than a true shift. That is, the plural feature appears twice
in the erroneous utterance. In (93b), for instance, the plural feature accompanying
woche (week) is perseverated in the DP containing sohn (son). The possessive
pronoun accommodates and the Vocabulary item /zo:n/ appears with the appropriate plural allomorph and undergoes phonological readjustment(umlaut).
5.4.2 Tense
The second feature to be considered in this section is the tense feature. Just like the
plural feature, the tense feature heads a projection of its own, a TnsP. In contrast to
the FP headed by [+pl], however, the TnsP is part of the CP domain. In the syntax, a
root that is licensed by a light verb will raise and adjoin to Tns and consequently, this
root and Tns are sisters under the Tns node. In an error, a root may be exchanged,
anticipated, or perseverated, leaving the adjacent Tns node behind. In other words:
the Tns information may be stranded. In addition, speech errors may involve
infinitival and/of participial verb forms. Leaving technicalities aside, I assume that
infinitivals adjoin to a Tns head that is specified for [tns] (or an empty Tns head)
and that the feature [+part] combines with the verb inside the verbphrase.
For illustration, consider the errors in (94) all of which involve the exchange
of roots. In (94a), schein (seem) and droh (impend) change place and adjoin
to the stranded [+past] and [tns] feature, respectively. In both slots, we observe
accommodation: the Vocabulary item that spells out droh takes the regular past
tense suffix -te and for the Vocabulary item that spells out schein, ablaut is not
triggered in the post-error environment. In (94b), the tensed verbs knnen (can)
and haben (have) are not affected in the error. Rather, in this slip, the roots fahr
(drive) and trink (drink) are exchanged and the stranded features are [tns]
and [+part]. In the English slip in (94c), just as in (94a), an infinitival form interacts with a past tense form (Garrett 1980b: 264). Both error elements undergo
accommodation, that is, stem-internal vowelchange.
(94) a.
es droh-te
zu schein-en, dass es schien
zu droh-en
it threat-past to seem-inf that it seem.past to threat-inf
Itseemedtoimpendthat
Mencanstilldrivewhentheyhavedrunk(alcohol).
c. IdontknowthatIdhearoneifIknewit
thatIdknowoneifIheardit
The syntactic structure for the error in (94a), after root exchange has taken place,
is given in (95). At the point of Vocabulary insertion, the regular past tense suffix /-t/ will spell out [+past] and the infinitival suffix /-n/ will spell out [tns].
Note, however, that strictly speaking, we are not in a position to decide whether
the two roots are actually exchanged after having been adjoined to Tns (as indicated in the below structure) or whether they are exchanged before head movement has applied, that is, while still taking the positions tL1 and tL2, respectively,
where, in fact, they are structurally closer to eachother.
(95)
CP
DP
[3rd]
Tns
TnsP
L1
Tns
[+]
tDP
Tns
LP
tTns
TnsP
DP
PRO
root exchange
tL1
Tns
LP
tL2
Tns
L2
Tns
[]
eigentlich woll-te
er kam,
h, komm-en
actually
want-past he come.past, er, come-inf
Actually,hewantedtocome.
Grammar as Processor
b. Iwinduprewrot-ingtwelvepages Iwounduprewriting
c. theyre just clouds that are been divert-ing, that are being divert-ed from
the North
Fay (1980b) and Stemberger (1985) report some spontaneous English errors
involving tense shift. In the extraordinary slip in (96b), the [+past] feature
is shifted from wind to rewrite, that is, in contrast to (96a), the first verb
loses its [+past] specification (Stemberger 1985: 163). In the intended utterance, however, rewrite is not supposed to appear with any tense feature at
all; rather, it is adjoined to an aspectual head containing a [+continuative]
feature. Interestingly, in the error, the aspectual feature is not substituted for
by the tense feature (which would have given rise to I wind up rewrote twelve
pages). Instead, the tense feature enriches the featural make-up of the structure under the aspectual node. Apparently, [+past] adjoins to the L-node
containing rewrite and this root is spelled out accordingly. Still, the aspectual feature which, of course, is incompatible with the feature [+past] is
also spelled out. (96c) is the only example in which two features (according to
Stemberger two affixes), namely [+past] and [+continuative], fully
exchange (Stemberger1985:163).
It has to be pointed out that, with regard to stranding and shift of the tense
feature, there are also some problematic cases which do not receive a straightforward explanation in the present framework.56 One such tricky case from my
corpus is given in (97a). At first sight, this error looks like a simple anticipation
of kauf (buy) with stranding of the feature [+part]. Note, however, that the
participle form of kaufen is gekauft, while verkauft (as it appears in the error) is
the participle form of the prefixed verb verkaufen (sell). Consequently, anticipation of kauf should have resulted in the sequence was ich ge-kauf-t habe. Obviously, the prefix ver- which is part of the intended verb vergessen (forget) was
left behind. This, however, excludes the possibility that vergess as a whole is
substituted for by kauf at MS. Alternatively, one might argue that it is not a root
that is anticipated but rather the phonological form /kauf/. This analysis would
locate the error at PF. Under this analysis, however, there is no way to account
for the accommodation of the participle suffix. If the error had in fact occurred
after Vocabulary insertion, then the ungrammatical utterance *was ich verkauf-en
. See Thompson (2004) for discussion of English and Spanish speech errors that involve
stranding or shift of tense features. In particular, she compares the involvement of auxiliary
verbs in speech errors in the two languages and interprets the attested differences in Lasniks
(1995) hybrid theory of verbal inflections.
habe should have surfaced. Therefore, this error as simple as it may look at first
sight remainsunexplained.
(97) a.
ich wei,
was ich verkauf-t hab-e,
I know.1.sg what I
sell-part have-1.sg
IknowwhatIforgottobuy.
b. Rosaonlydateshrank-s
Rosaonlydate-dshrink-s
Tns
Tns
L1
LP1
tL1
DP
LP2
L2
L2
Tns
[+]
[+]
Grammar as Processor
Still, due to the [+pl] feature, Vocabulary insertion will also supply a plural suffix. What is probably most surprising about this particular error is the
fact that the Tns node adjoins to an l-node which is licensed by a determiner.
Apparently, the adjunction process in contrast to Vocabulary insertion is blind
with respect to the licensing environment of the node it targets (also see Leuninger
& Keller(1994:102f)).57
5.4.3 Negation
In Section 4.3.3, I have already considered the possible role of semantic negation
in speech errors. In this section, I shall have a closer look at slips that involve
morphological or syntactic negation. As has been pointed out before, both morphological and syntactic negation involve the presence of a morphosyntactic
feature [+neg]. In the former type of negation, [+neg] shares a terminal node
with a root, while in the latter type, the negative feature heads a functional projection of its own, a NegP (Pollock 1989; Haegeman 1995; Zanuttini 1997). Only
syntactic negation changes the polarity of the sentence (see Klima (1964) for a
number of tests that can serve as a diagnostic for negation type in English, for
instance, the tag-question test: syntactically negative sentences (He is not happy)
take positive tags, while morphologically negative sentences (He is unhappy)
take negativetags).
In German, too, speakers often have a choice between between a morphological and a syntactic strategy when it comes to expressing the negation of a
concept. For instance, in order to convey the message that something is not
clear, a speaker may either decide to use the morphologically complex word
unklar (unclear) or the phrase nicht klar (not clear), which involves the
. Besides tense features, mode features, which are taken to head a functional projection
of their own (be it MoodP, FinP, or some other functional projection), are also capable of
stranding and shift. While my corpus does not contain a relevant example, Meringer & Mayer
(1895:42) report errors in which the feature [+conditional] is shifted (also see the discussion in MacKay (1979)). For illustration, see the error in (i), in which the conditional feature
is anticipated and combines with the auxiliary verb haben (have).
(i)
ich erinner-e
mich, wie unser-e Lehrer
immer ge-sag-t
I remember-1.sg refl how our-pl teacher.pl always part-say-part
htte-n,
hab-en, es wre
ja
sehr schn
have.cond-pl, have-pl, it be.cond.3.sg mod.part very nice
negative particle nicht. In fact, the speech error data in (99) suggest that there
are no unitary negative concepts (like, for example, unklar) but rather that
morphologically complex forms result from the combination of a basic concept
and a morphosyntactic feature[+neg].
The slip in (99a) is the only slip from my corpus in which a [+neg] feature is
stranded. Interestingly, the negative prefix accommodates to the anticipated root
relevant. In addition, my corpus contains seven slips in which a negative feature is shifted. In (99b), for instance, the adjectival root mglich (possible) in
the embedded clause is associated with a [+neg] feature. In the error, this feature
is anticipated into the matrix clause, where it does not morphologically combine
with some other root but rather enriches the syntactic structure by projecting a
NegP. Example (99c) is different from (99b) in two respects. First, it involves the
shift of [+neg] from a matrix into an embedded clause. Secondly, the negative
element retains its status as a negative particle, that is, it does not change from
morphological to syntacticnegation.
(99) a.
eine etwas
ir-relevante, h, un-deutliche,
a
somewhat neg-relevant, er, neg-clear,
relevante Aussage
relevant statement
asomewhatunclearbutstillrelevantstatement.
aber dennoch
but still
Hehassaidthatitisimpossible.
c. mir gelingt
es, ihn nicht zu erreich-en
me succeed it, him not to reach-inf
Idontsucceedingettingintouchwithhim.
In (100), I give a syntactic structure for the error in (99b). Note that the embedded CP in (100) has been extraposed, that is, right-adjoined to TnsP. As pointed
out above, I assume that the shifted [+neg] feature projects a NegP in the matrix
clause. Depending on the analysis of German sentential negation that one adopts,
it might, however, also be the case that Neg adjoins to an XP be it TnsP or LP1.
In any case, [+neg], when appearing on its own, will be spelled out by the Vocabularyitem /nit/.
Grammar as Processor
(100)
CP
DP
[3rd][]
C
Tns
NegP
[+] Neg
[+]
TnsP
TnsP
tDP
CP
Tns C
tTns DP
LP1
L1
[+]
shift of Neg
feature
TnsP
tCP
Tns
[3rd][N] LP2
LP3
Tns
tL2 L2
Tns
[]
For comparison, in (101), I cite three English slips which involve the manipulation of the feature [+neg]. In (101a), the negative feature accompanying
precise is anticipated and combines with regard (Fromkin 1973a: 32). As
in (99a), the negative prefix accommodates to the post-error environment. In
principle, the shift might also have resulted in the sequence I dont regard this as
precise, with the negative feature extending the syntactic structure in the same
way as in(99b).
Fromkin does not give an analysis for the error in (101b), but I suppose that
we are dealing with a special type of semantic substitution (Fromkin 1973b: 268).
Semantically, crazy and insane are very close to each other; the former is a root,
while the latter is a combination of sane and [+neg]. In the error, crazy takes
the place of sane. Consequently, the terminal node contains crazy as well as
[+neg]. There is, however, no Vocabulary item to match this feature combination
(that is, there is no Vocabulary item like uncrazy), and therefore Vocabulary insertion must resort to a syntactically complexparaphrase.
(101) a. Idis-regardthisasprecise Iregardthisasim-precise
b. Itellyouhesnotcrazy,Imean,hesin-sane
c. thebonsaididntdiebecauseIwateredit
thebonsaidiedbecauseIdidntwaterit
Finally, the English slip in (101c) is similar to (99c) in that a syntactic [+neg] feature is shifted, in this case, anticipated from the embedded into the matrix clause
(Fromkin 1973b: 268). This error is particularly illuminating because, after the
shift, the presence of the negative feature triggers do-insertion in the matrix clause.
Remember that, according to Halle & Marantz (1993), English main verbs do not
raise to Tns. Rather, Tns merges with V resulting in the structure [V [V] [Tns]]
(see Section 3.2.2.2). Given that the presence of Neg prevents merger of Tns and
the matrix verb, the dummy element do has to be inserted in order to pick up
the Tns feature [+past]. Consequently, this particular error suggests that the Neg
shift has taken place before merger of Tns and V has occurred, since otherwise
do-insertion would not have been triggered and the ungrammatical utterance *the
bonsai not died because I did water it would havesurfaced.
I want to conclude this section with a note on Vocabulary insertion. With
respect to the negative prefixes that figure in the errors in (99a) and (101a), there
are actually two options with respect to Vocabulary insertion. On the one hand,
one may assume that the feature [+neg] and the root share one terminal node. In
this case, there will be one Vocabulary item that spells out the terminal node, as
indicated in (102a) for the error element in(99a).
(102) a. relevant
/irelevant/
[+neg]
b. [+neg]
/ir-/
/
relevant
Alternatively, one may assume that the terminal node containing relevant has
a branching structure with [+neg] being the sister node of the root. In this case,
the root and the negative feature (that is, the negative prefix) will be spelled out by
separate Vocabulary items. Clearly, in this case, the spell-out of [+neg] would be
context-sensitive, so that in the context of relevant it would be spelled out as
/ir-/ (102b), while in the intended utterance, that is, in the context of deutlich,
it would be spelled outas /n-/.
5.4.4 Gender
In Section 5.1, I have argued that gender is an inherent feature of roots and that
roots are drawn from List 1 and inserted into a syntactic structure along with their
gender feature. This implies that roots are more closely related to their gender
feature than they are, for instance, to the plural feature (the choice of which is triggered by the activation of a conceptual node multiple(x) at the conceptual level),
tense features, and the negation feature. Intuitively, it seems therefore unlikely that
a root could be accessed, for instance, exchanged, in an error leaving its gender
featurebehind.
Grammar as Processor
Let us still take a moment to speculate about what such an error might look like.
In contrast to the cases of plural stranding discussed in Section 5.4.1, a displaced
root would, of course, not be spelled out differently because of a stranded gender
feature with which it combines. The only errors which could possibly be analyzed as
instances of gender stranding are those in which a root is displaced without subsequent accommodation of the material in D to the gender feature of that root. There
are eight such cases in my corpus, seven root exchanges and one root anticipation.
For the sake of illustration, I give one of the exchange cases in (103). In this error,
haar (hair) and nase (nose) are exchanged and combine with the gendermarked indefinite and definite determiner, respectively, of the intended roots. Consequently, we observe feature mismatch within both DPs.
(103) ich hab
ein
Nase
auf der
Haar, h,
I have.1.sg a.n.acc nose(f) on the.f.dat hair(n), er,
ein
Haar
auf der
Nase
a.n.acc hair(n) on the.f.dat nose(f)
Ihaveahaironthenose.
A structure for the slip in (103) is given in (104). Following the exchange, the
stranded gender features will be copied onto the respective D-positions and the
Vocabulary items that best match the feature content of D will be inserted: the feature bundle [def][n][acc] will be spelled out by /ain/; the feature bundle [+def]
[f][dat] will be spelled out by /de:/. Note that following this interpretation of the
facts, the error in (103) does not really constitute an instance of feature mismatch
within DP, since the (stranded) gender features associated with the exchanged
roots and the gender features of the determiner domatch.
(104)
DPACC
LP1
D
[][]
PP
L1
[]
DPDAT
[auf]
LP2
[+][]
[]
root exchange
A serious problem, however, concerns the insertion of Vocabulary items for the
exchanged roots. Remember that I have argued in Section 5.1.2 that not only
roots but also the corresponding Vocabulary items are specified for grammatical gender. Following this line of reasoning, the two Vocabulary items /ha:/ and
/na:z/ do not match the respective terminal nodes with respect to the gender
feature. Halle & Marantz (1993) point out that at the point of Vocabulary insertion, the Vocabulary is searched for the entry that best matches the content of a
given terminal node. While Vocabulary items may be underspecified for a given
feature contained in a terminal node, they may not conflict with a morphosyntactic feature present in that node. Consequently, the insertion of the Vocabulary
items /ha:/ and /na:z/ should beblocked.58
Given these problems, it seems much more likely that the root exchange in
(103) has taken place after the respective gender features have been copied onto
the determiners. Only under this assumption, we are in fact dealing with instances
of feature mismatch within DP, since the exchanged roots (which presumably take
along their gender feature) do not agree in gender with the accompanying determiners (see Section 6.7.1.2 for further discussion). I therefore conclude that in
German speech errors, gender stranding is notattested.59
My corpus contains ten spontaneous errors which I classifiy as gender shifts.
Two of these involve pronouns. Gender on pronouns is different, of course, from
the inherent gender specification of roots, because pronouns are nothing but bundles of morphosyntactic features. Hence, individual features can be accessed in an
error. In (105a), for instance, the gender features of a dative relative pronoun (masculine) and an adjacent nominative personal pronoun (feminine) areexchanged.
. The only way to save the gender stranding analysis would be to assume a flaw at the point
of Vocabulary insertion. That is, the Vocabulary items which best match the roots contained
in the terminal nodes are selected and inserted in spite of the feature conflict that arises.
However, an explanation along these lines, that is, the assumption of an additional error, seems
to be very unlikely.
. Note that this does not imply that gender stranding is ruled out in general. In languages
with morphological gender or noun class systems, one would expect the morphological exponents to be able to strand in a root exchange. For instance, Kihm (2005), who investigates noun
class in Manjaku (a Niger-Congo language spoken in Guinea-Bissau and South Senegal) and
gender in Spanish, suggests that the respective morphological markers head a functional projection which c-commands the root they attach to, similar to the plural feature in (91). The relevant functional projection is nP (see Section 6.4.2). Hence, a Spanish root like gat (cat) can
combine with the masculine suffix -o (class I) or the feminine suffix -a (class II). The root will
be attracted by the functional item and left-adjoin to it. Given that the gender (or noun class)
markers in these languages are not inherent features of roots, stranding should be possible.
Grammar as Processor
(105) a.
leb-t
mit dem
sie
leb-t
live-3.sg with who.sg.m.dat 3.sg.f.nom live-3.sg
Shehassomeonewithwhomshelivestogether.
b. mir
steht
die Schwei auf der
Stirn
1.sg.dat stand-3.sg the.f sweat(m) on the.f.dat forehead(f)
Myforeheadisallsweaty.
ein
Blatt vor
den
Mund
a.n.acc leaf(n) in.front.of the.m.acc mouth(m)
Nooneminceshiswords.
The other eight errors in the gender shift category are more equivocal with respect to
their classification. Consider, for instance the slip in (105b), in which we observe gender mismatch within the first DP. The fact that the masculine noun Schwei (sweat)
is accompanied by a feminine definite determiner is probably due to the presence
of the feminine noun Stirn (forehead). Hence, one might assume that the gender
feature of the second root was anticipated to the first root. In this case, the second
root does not loose its gender feature. Rather, just as in the [+pl] shift in (93b) and
the [+past] shift in (96a), the manipulated feature surfaces twice in the utterance.
However, given that the gender feature has to be copied onto the D-position, the slip
in (105b) might as well be analyzed as an instance of erroneous feature copy, that is,
long-distance gender feature copy from stirn onto the first determiner (note that
an analysis along similar lines is not available for (93b) and (96a)). In other words:
the feature that is responsible for the error is not shifted butcopied.
The same is true for the other seven slips in the gender shift category. For the
sake of illustration, I give one more example in (105c). Here, the gender feature of
blatt (leaf or sheet) appears to be perseverated. But again, the gender feature
might have been copied from blatt onto the determiner of the following DP
(which subsequently cliticizes to thepreposition).
Due to the scarcity of data and to the dubious status of the examples discussed
in this section, it is in contrast to the features [+pl], [+past], and [+neg] questionable whether gender features can in fact be manipulated, that is, whether roots
can be exchanged leaving their gender feature behind and whether a gender feature
alone can be shifted in an error. It is only for pronouns that the separation and
manipulation of a gender feature seems possible. Still, for the time being, I shall
stick to the classification of the eight dubious cases as gender shift errors (see Table
(109)). I acknowledge, however, that they might better be analyzed as a special case
of long-distanceagreement.
5.4.5 Case
Last but not least, I shall discuss the manipulation of case features in language production. Once again, we are dealing with different structural facts. In contrast to
the features discussed in the previous sections, case features are neither associated
with an l-node nor are they base-generated under an f-node. Rather, case features
are assigned to a DP as a whole according to the case-assignment properties of
a lexical item. From the DP node, the case feature will percolate down to all elements dominated by this DP. In German, case features may influence the spell-out
of pronouns, roots, adjectives, anddeterminers.60
Actually, an example of case stranding has already been given in (105a) above.
This example has been discussed in the context of gender shift because the gender
features of the two pronouns involved are exchanged in the error. In addition,
however, the case features dative and nominative are stranded. The same is true
for the German error in (106a) and the English error in (106b) (Stemberger 1982a:
345), in both of which all features except for the case features are exchanged. In
(106b), case stranding is only obvious in the second position. There are also some
errors in which case stranding is observed when a root is manipulated. In (106c),
for instance, kind (child) and familie (family) are exchanged. The two DPs
containing these roots are assigned different case by the respective prepositions,
the first one accusative case and the latter one dative case. Case stranding is obvious in the second position because, in the plural, Kind is overtly marked for dative
case by means of anaffix.
(106) a.
ich
soll-te
doch
ihn,
h,
1.sg.nom shall-past mod.part 3.sg.m.acc, er,
er
soll-te
doch
mich
anruf-en
3.sg.m.nom shall-past mod.part 1.sg.acc call-inf
Hewassupposedtocallme.
b. youmustbetootightforthem theymustbetootightforyou
. In the following discussion, I will neglect the suggestion made in the literature that there
is a functional category related to case above the DP-level. Lamontagne & Travis (1986), for
instance, label the relevant functional projection K(ase)P and view it as the nominal parallel to
the clausal functional projection CP (see Lbel (1994) for further elaboration of that idea).
Grammar as Processor
c.
besonders
problematisch fr Familie-n aus auslndisch-en
particularly problematic
for family-pl from foreign-pl
Kind-er-n
Kind-er aus auslndisch-en Familie-n
child-pl-dat child-pl from foreign-pl
family-pl
particularlyproblematicforchildrenfromforeignfamilies.
My corpus contains 27 slips that are classified as case stranding errors. Note, however, that the term stranding may not be fully accurate in this context. In contrast
to other morphosyntactic features, case features are not drawn from List 1 for
insertion into a terminal node. That is, they do not necessarily share a terminal
node with other features or a root when the error occurs. Hence, it is quite possible
that the exchanges in (106) took place before the respective case features have percolated down to the terminal nodes. Use of the term stranding is therefore only
justified insofar as in the erroneous utterance, the same case features are assigned
to the same DPs as in the intendedutterance.
The structure in (107) illustrates the process that can be held responsible
for the error in (106a), namely exchange of features. As pointed out before, case
assignment is not affected in the error. When contained in a DP that is assigned
nominative case, the feature [1st] will be spelled out as /i/, while the combination
of [3rd] and [m] with accusative case will be spelled out as /i:n/.
(107)
CP
DPNOM
[1st]
C
Tns
TnsP
Tns
[+]
tDP
Tns
LP
DPACC
[3rd][]
tTns
L
tL
feature exchange
While in the above examples, case features are assigned to the same positions
within a syntactic structure before and after the exchange of features or roots has
taken place, in three errors in my corpus case features are shifted in the sense that
they are assigned to a different DP projection in the error. The verb ausspannen
(to pinch) in (108a) requires three case-marked arguments, one nominative, one
accusative, and one dative. In the error, however, accusative case is assigned to both
object pronouns, thereby violating the subcategorization properties of theverb.
(108) a. ich bin mir fast
sicher, dass er sie
mich,
h,
I am refl almost sure
that he 3.sg.f.acc 1.sg.acc, er,
dass er sie
mir
ausspann-en
that he 3.sg.f.acc 1.sg.dat pinch-inf
woll-te
want-past
Iamalmostsurethathewantedtopinchherfromme.
h, dich
ihm
er, 2.sg.acc 3.sg.m.dat
Ireallywanttointroduceyoutohim.
. Table (2) in Chapter 1 lists 133 errors in the category stranding or shift of abstract
feature because, in addition to the 116 errors in which morphosyntactic features are stranded/
shifted, this group also contains the 17 cases discussed in Section 4.3, which involve stranding/
shift of a compositional semantic feature.
Grammar as Processor
stranding
shift
number: [+pl]
tense: [+past], [tns], [+part]
person: [1st], [2nd], [3rd]
negation: [+neg]
gender: [m], [f], [n]
case: [nom], [acc], [dat]
23
20
8
1
27
7
2
8
7
10
3
Total
79
37
Clearly, the way in which a particular feature enters the computational system, its
participation in structure building, and its manipulation in the syntax differ from
feature to feature. Generally, features that project their own functional projection
can be stranded in an error. This is frequently observed with the feature [+pl] and
with tense features. In contrast, shift of these features is rare. Negation is different in that it may either head a functional projection or combine with a root. The
numbers in the table make it look like as if negation is more frequently shifted
than stranded. These numbers may be misguiding, however, since in the stranding
category, only cases were considered in which morphological negation is stranded
in a root exchange, as, for instance, (99a). Cases in which two (verbal) roots are
exchanged while a negative particle stays in place are attested but have not been
included in this category although, strictly speaking, they also involve stranding
of[+neg].62
I have not discussed cases in which a person feature is stranded or shifted.
Still, I wish to point out that for these features, the stranding and shift cases constitute different phenomena. In all the stranding cases, verbal agreement is stranded
in a root exchange or blend (see, for example, the slip in Footnote 62). In contrast,
shift of a person feature is only observed with pronouns. The slip in (106a), for
. In the example in (i), for instance, behaupt (claim) and kenn (know) have been
exchanged. The negative feature [+neg] within the infinitival complement, however, remains
in place and projects a NegP. In this sense, negation is stranded in the error. Note also that the
person feature [3rd] in the matrix clause as well as the tense feature [tns] in the embedded
clause are stranded.
(i)
er kenn-t
sie nicht
he know-3.sg her not
er behaupt-et
he claim-3.sg
zu behaupt-en
to claim-inf
instance, which has been presented as an example of case stranding, also exemplifies the exchange of person features. Actually, in this example, the exchange of the
features [1st] and [3rd] (in combination with [m]) constitutes the real error, while
case stranding could be argued to be a by-product (also see (13) in Chapter6).
Gender features are the only features that are not capable of stranding. This
is not surprising, given that gender features, in contrast to all the other features,
are inherently specified for roots. Due to this tight link, gender features cannot be
detached from a root in a root exchange, anticipation, or perseveration. Gender
shift is attested but we have seen that the only true cases of gender shift are those
in which a gender feature is part of a feature bundle that will be spelled out as a
pronoun. Crucially, in this case, the gender feature is not an inherent feature but
is drawn from List 1 on its own. I have also argued that other cases that are subsumed in the gender shift category are probably better analyzed as instances of
long-distance gender copy. Finally, case features are commonly stranded in root
and feature exchanges, while shift, that is, wrong assignment of a case feature,
appears to be an uncommon phenomenon. We have to keep in mind, however,
that the manipulation of case features differs from that of other morphosyntactic
features because case features are not drawn from List 1. Rather, they are assigned
to DPs in the course of the derivation. Hence, case stranding does not involve case
features that are left behind in some terminal node. Rather, it implies assignment
of case to a DP that contains a displacedroot.
5.5 Conclusion
The wealth of speech error data discussed in this section leaves little doubt about
the crucial role the manipulation of morphosyntactic features plays in the course
of language production. In order to account for the error patterns, three types
of feature manipulation have to be distinguished: feature copy, feature stranding,
and feature shift. The above discussion has made clear that feature copy plays two
distinct roles. On the one hand, copy processes may ensure that a grammatical
utterance surfaces after an error has taken place. In this case, the context of a substituted or displaced root accommodates to the gender feature of that root. We
have seen that the attested accommodation patterns are in line with the predictions made on the basis of the DM architecture: accommodation is only observed
for errors that occur before or at the level of MS, the level at which agreement relations are established (note that gender feature copy will make another appearance
in Section 6.2.1). On the other hand, feature copy can also be the cause of an error
in cases where an erroneous target is chosen for agreement feature copy. In most
of the errors involving feature mismatch, the erroneous target be it a verb or a
Grammar as Processor
pronoun is linearly closer to the agreement controller than the correct target.
Interestingly, however, we also find a fair number of errors in which the mismatch
is due to long-distanceagreement.
In feature shift, just as in feature copy, it is the morphosyntactic feature itself
that is manipulated in the error. In a shift error, the relevant feature is not copied,
that is, no agreement relation is involved. Rather, some feature present in the hierarchical structure appears under a wrong functional node. Feature shift is either
replacing (for instance, when a tense feature takes the place of another tense feature) or structure-building (as in some cases of [+neg] shift). Moreover, a shifted
feature may disappear from its original position (true shift) or may appear twice
in the structure as a result of feature anticipation or perseveration. Feature stranding is different from both feature copy and feature shift because it is a defining
characteristic of stranding that some element an affix or a feature is not affected
in the error. Still, just like the feature shift cases, stranding errors prove that morphosyntactic features can be separated from a root or from other morphosyntactic
features they accompany. Some consequences of feature stranding for spell-out
and phonological readjustment will be discussed in more detail in Section6.3.
chapter 6
Rethinking accommodation
Much of the preceding chapter was devoted to the discussion of spontaneous errors
that result in an ungrammatical utterance, for instance, all types of SVA-errors and
most of the feature shift cases. In addition, however, I have already presented slips
in which a grammatical string surfaces thanks to a post-error repair process commonly referred to as accommodation. In Section 5.1, for instance, we have seen
that following a semantic substitution or a root exchange, a gender accommodation may prevent a gender mismatch withinDP.
In this chapter, I will have a closer look at various types of accommodation
from a DM-perspective. Accommodations have fascinated psycholinguists for
quite some time because of their multifarious and complex nature. As has already
been pointed out in Section 1.3.4, Garrett (1980b: 263) defines accommodations as
errors in which the phonetic shape of elements involved in errors accommodates
to the error-induced environment. He takes accommodations to be a particularly
clear piece of evidence for the existence of several distinct processing levels in language production (see Section 3.1.1). At one level, the actual error occurs, while
at a subsequent level, the structure resulting from the error is brought in line with
some grammaticalconstraint.
The main goal of this chapter is to demonstrate that the concept accommodation, as defined in the speech error literature, is superfluous. In a nutshell,
I will argue that the assumption of repair processes is unnecessary once we make
use of derivational operations as assumed in DM operations like feature copy,
licensing, morpheme insertion, and phonological readjustment. I will start out
by presenting a comprehensive typology of accommodations in Section 6.1. In
Section 6.2, I will consider the first relevant mechanism, feature copy. In speech
errors, feature copy ensures that agreement targets surface in a properly inflected
form. The crucial role of two further mechanisms, stranding and local licensing,
will be discussed in Sections 6.3 and 6.4. Both these mechanisms make sure that
error elements surface in a phonologically and morphologically appropriate form.
In Section 6.5, I present a detailed analysis of two particularly complex slips in
order to further illustrate the application of the DM-mechanisms discussed in
Sections 6.2 to 6.4. Section 6.6 summarizes the arguments against repair strategies
and also presents errors that may contradict my general conclusion that repairs play
Grammar as Processor
no role in the course of language production. Finally, in Section 6.7, I turn to some
remaining issues concerning repairs. I will, for instance, address the question why
despite the mechanisms argued for in the previous sections accommodation is
not always observed in speecherrors.
. See Berg (1987) for a different classification. Berg proposes two parameters. The first one
concerns the character of the accommodation process itself (phonological, morphological,
or lexical), while the second one takes into account what kind of element is displaced in the
error (phoneme, morpheme, or word). The combination of the two parameters results in nine
different categories.
Rethinking accommodation
h, in Kassel [as]?
er, in Kassel
ils
ont convol [k]
they have married
c. pankeren [pak]
(error)
to camp
kamper-en [kamp]
camp-inf
Nasality is also relevant in the Dutch error in (2c) in which the consonants /k/ and
/p/ are exchanged (Cohen 1965:183). Following the exchange, the labial assimilates to the place features of the following velar and surfaces as the velar nasal[].
6.1.2 Morphophonological accommodation
We are to speak of a morphophonological accommodation whenever phonologically triggered allomorphy has an impact on the surface form of a slip. That is, after
the error has taken place, the allomorph that is appropriate for the error element(s)
is selected. In English, for instance, choice of a plural suffix is determined by phonological properties of the stem to which it attaches. In the root exchange in (3a),
the plural suffix accommodates to the error element cow (Fromkin 1973a: 27).
Grammar as Processor
Allomorphy also plays an important role in Turkish. In Turkish, the choice of suffixes is determined by vowel harmony. Turkish shows the characteristic pattern of
a symmetric vowel harmony system, that is, only suffixes alternate and the alternating suffix vowels harmonize with the nearest stem vowel with respect to certain
phonological features. The error in (3b) is due to a vowel exchange. Following the
error, all three suffixes of the resulting stem harmonize; the first one with the stem,
the following ones with the respective preceding suffix.2 Hence, the resulting nonexisting word is phonologicallywell-formed.
b. hukumet kr-l-me-si
hkmet
kur-ul-ma-s
(error) (error)-pass-nmlz-poss government form-pass-nmlz-poss
formation of a government
c. give the [] nipple an [n] infant the [i:] infant a [] nipple
. I am indebted to Bn Ergnal for supplying the Turkish speech error. Note that Turkish
employs two intersecting vowel harmony systems, one involving [back] and the other
[round] (see Clements & Sezer (1982) for details). The letter // represents the [+high]
counterpart to /a/, that is, a vowel which is characterized by the features [+high], [+back],
and [round].
Rethinking accommodation
g e-monat-ete
Arbeit-en ge-arbeit-ete
Monat-e
part-month-part work-pl part-work-part month-pl
months in which one has worked
in der
er wohn-t
in which he live-3.sg
Bank
Bnk-e
bench
benches
Maus
Mus-e
mouse
mice
Bank
Bank-en
bank
banks
Grammar as Processor
accommodations. These are the cases in which a stem involved in the error is
spelled out differently because of a different categorial specification. In (5), I cite
two errors of this type. The German noun Flughafen (airport) contains the stem
flieg (fly) in its nominalized form (Flug). In the anticipation in (5a), Flug takes
the position of the verb fahr (go) in which it does not retain its stem vowel.
In other words, the anticipated stem accommodates to the categorial specification of its landing site. I consider this a morphological accommodation because
noun-verb conversion be it by means of affixation (4b) or stem-internal change
(5a) is a morphological process. Crucially, we are not dealing with a phonological accommodation because the stem-internal phonological change is not
caused by some adjacentsegment.
(5) a. die S8 flieg-t zum Flug-hafen fhr-t zum Flug-hafen
the S8 fly-3.sg to.the flight-port go-3.sg to.the flight-port
The S8 (train) goes to the airport.
b. will-st
du
den Gang, h, die Treppe runter-geh-en
want-2.sg you(sg) the.m aisle(m), er, the.f stairs(f) down-go-inf
Do you want to go down the stairs?
(5b) shows the opposite pattern of (5a). In this error, the error element geh (go) is
anticipated from a verbal slot into a nominal slot. The nominalized form surfacing
in the error is Gang, which can mean various things but is best translated as aisle
(or corridor) in the present context. Errors like those given in (5) will receive a
more detailed analysis in Section6.4.1.
6.1.4 Morphosyntactic accommodation
Let us now turn to the fourth and final type of accommodation, the morphosyntactic ones. We are dealing with a morphosyntactic accommodation whenever the
structure of an utterance is adjusted with respect to some morphosyntactic feature
after the error has taken place. Again, we have to distinguish context from error
accommodations. In the context accommodation in (6a), both determiner positions accommodate to the gender features of the exchanged nouns, the masculine
noun Faden (thread) and the feminine noun Nadel (needle). Similarly, in the
Spanish error in (6b), the indefinite determiner of the first DP accommodates to the
gender feature of duro (five pesetas) (Garcia-Albea, del Viso & Igoa1989:152).4
. In contrast to the examples (6a) and (6b), the exchange in (i) is ambiguous. Since both
DPs contain a definite determiner, it cannot be decided whether the error should be analyzed
as a DP-exchange with stranding of case features (nominative and accusative), or as an NPexchange with subsequent morphosyntactic accommodation of the two determiners.
Rethinking accommodation
(6) a.
ein-en Faden
in die
Nadel
a-m.acc thread(m) in the.f.acc needle(f)
how one gets a thread through (the eye of) the needle
b. un duro
de veinte moneda-s
a.m 5.pesetas(m) of twenty coin-pl
du
sag-st
ja
nicht, h, ich sag-e
ja
nicht,
you(sg) say-2.sg mod.part not, er, I say-1.sg mod.part not
dass du
schuld bist
that you(sg) guilty be.2.sg
b. youre too good for that thats too good for you
c.
ich las
ihm, h, ich empfahl
ihm, den Artikel zu les-en
I read.past him, er, I advise.past him the article to read-inf
I advised him to read the article.
Occasionally, the error element itself changes under the influence of some morphosyntactic feature. Consider, for instance, the anticipation in (7c), in which the
stem les (read) changes its form because it combines with the morphosyntactic
(i)
das
Bild
auf den
Boden
ge-fall-en
the.n.nom picture(n) on the.m.nom ground(m) part-fall-part
Grammar as Processor
feature [+past]. Note that this kind of phonological change is different from the ones
observed in (5). In both errors in (5), the original position of the error element has
a categorial specification different from that of its landing site. In contrast to that, in
(6c), the error element is moved from one verbal slot to another verbal slot. Hence, it
is only the morphosyntactic specification that can be held responsible for the surface
form of the error element. I will return to morphosyntactic context accommodations
in Section 6.2 and to morphosyntactic error accommodations in Section6.3.
6.1.5 Summary
Citing speech errors from various languages, I have argued that four types of
accommodation should be distinguished, namely phonological, morphological,
morphophonological, and morphosyntactic accommodation. In principle, all four
types may come as context and error accommodations. Without going into details,
I want to point out that this classification is quite different from the one proposed
by Berg (1987). For instance, for Berg, the errors involving accommodation of
English determiners (3c) and German plural (4a) or agreement (7a) suffixes are
all examples of morphological accommodation since, following the error, a different morpheme combines with the error element. Such an interpretation, however,
disguises the fact that the selection of the respective morphemes is conditioned by
different factors in these threeerrors.
In an attempt to characterize accommodations, Berg (1987: 277) states that
an accommodation is a process whereby a processing conflict between the actual
error and the context of the original utterance is reconciled. He takes this as evidence for the fact that the processing system is sensitive to the eventual output
and concludes that accommodations can therefore be considered a blind repair
process which brings utterances in line with linguistic constraints. In the remainder of this chapter, I will challenge this characterization. In contrast to Berg, I am
going to claim (a) that no processing conflict is reconciled in an accommodation,
(b) that therefore no repair strategy is involved, and (c) that output-oriented processing need not be assumed. In other words, I will argue that the concept accommodation, while being a convenient descriptive label for various error types, does
not have any theoretical significance at least not when we employ the theoretical
tools made available by DM (also see Pfau(2007)).
6.2 Feature copy
The first mechanism that will help us to account for the so-called accommodations is feature copy. Remember that according to DM, the only elements that
Rethinking accommodation
enter the computation are abstract roots and morphosyntactic features. Agreement relations are established at MS, that is, before Vocabulary insertion takes
place, by copying the relevant morphosyntactic features from an agreement controller onto a functional head. In the German speech errors, two types of feature
copy are of interest: copy of gender and number features within DP (Section
6.2.1) and copy of person and number features from the subject onto the AgrS
node (Section6.2.2).
6.2.1 Gender agreement
Actually, not much has to be said about gender agreement because the phenomenon has already been discussed in the context of gender accommodations in
Section 5.1.2.1. For the sake of completeness, I will repeat the basic facts and
discuss one more example (also see (6a) in the precedingsection).
In German, the roots that are selected from List 1 are specified for gender, that
is, they are linked to a gender feature which the feature copy mechanism at MS
can access. Crucially, correct insertion of a Vocabulary item into a terminal node
that is underspecified for gender (for instance, into D0) could not be guaranteed.
In (8), the roots tasse (cup) and sprung (crack) have been exchanged. The
two roots have different gender features and these features are copied onto the
determiners after the error has takenplace.
(8) a.
oh, der
Sprung hat
ja
ein-e
Tasse
oh, the.m.nom crack(m) has.3.sg mod.part a-f.acc cup(f)
die
Tasse hat
ja
ein-en
Sprung
the.f.nom cup(f) has.3.sg mod.part a-m.acc crack(m)
b. er kann
kein-e
Fliege trb-en kein
Wsser-chen
he can.3.sg no-f.acc fly(f) cloud-inf no.n.acc water-dim(n)
trb-en
// kein-er
Fliege was
zuleide tun
cloud-inf // no-f.dat fly(f) something harm do.inf
Grammar as Processor
into the competing planning frame and subsequently, its gender feature is copied
ontoD0.5
Assuming that both errors occur before features are copied and Vocabulary
items are inserted, we must not resort to any accommodatory process. These
mechanisms are automatic and they apply in the errors in exactly the same way
as they would have applied in the intended utterances. Or, to put it differently, the
determiners need not be accommodated because they are not yet present when
the error occurs. Rather, at this point, the determiner positions only contain the
features [def] (8a) or [def] and [+neg] (8b). After feature copy and case assignment has taken place, the Vocabulary items given in (9) will be inserted into the
respective D0-positions at PF (remember that I argued in Section 5.2.2.5 that List
1 does not contain a feature[pl]).
(9) a. [+def] [m] [nom]
b. [def] [f] [acc]
c. [def] [+neg] [f] [acc]
/de:/
/ain/
/kain/
Note that the Vocabulary items given in (9b) and (9c) are probably too simplistic.
It is quite likely that the feature [def] and the feature combination [+neg][def]
are spelled out by one Vocabulary item, while a separate item (the suffix /-/) will
be inserted for the feature combination[f][acc].
6.2.2 Subject-verb agreement
The argument for subject-verb agreement is basically the same as for gender
agreement. As before, I argue that feature copy at MS ensures that a functional
element here: an agreement suffix is correctly spelled out and that therefore, no
repair process is involved in theerror.
In principle, there are two scenarios in which the adaptation of agreement
may be required in a speech error. In the first case, a root is dislocated together
. Generally, a phrasal blend can be equated with a substitution process (Wiegand 1996). In
(8b), it is clear that Fliege is the intruding element. In the blend in (i), however, the two competing frames have the same syntactic structure, namely [DP D Adj NP]. It is therefore impossible to decide whether the noun or the adjective is the intruder. Still, under both analyses, the
gender feature will be copied from the noun onto the adjective after the error has taken place.
(i)
ein dick-er
Hund //
a.m thick-m.nom dog(m) //
ein stark-es
Stck
a.n heavy-n.nom piece(n)
Rethinking accommodation
with the feature [+pl] and takes the place of a singular root (in subject position).
Remember, however, from the discussion in Section 5.4.1 that this error type
is infrequent because in most cases, the plural feature is stranded. One of the
few examples is given in (10a). In this slip, student (student) is anticipated
together with its plural feature. In other words, the NumP is anticipated. The
plural feature is not only copied onto D0 but also onto AgrS. The second scenario
involves pronouns, that is, bundles of features. In this case, the verb may adapt to
the pronoun in subject position with respect to person and/or number features.
In (10b), the feature [2nd] replaces the features [3rd] and [m], the feature [2nd]
is copied onto AgrS, and the agreement suffix is spelled outaccordingly.
(10) a.
die
Student-en hab-en, h, der
D. hat
the.pl student-pl have-pl, er, the.m D. have.3.sg
einige seiner
Student-en durchfall-en lass-en
some of.his.pl student-pl fail-inf
let-inf
b. du
denk-st
wohl,
h, er denk-t
wohl
you(sg) think-2.sg probably, er, he think-3.sg probably
dass du
fr den
Job nicht
that you(sg) for the.acc job not
geeignet
suitable
He probably thinks that you are not suitable for the job.
bist
be.2.sg
Again, I want to stress that it is not the case that the agreement suffixes accommodate to the new situation. Rather, they simply spell out the features that have been
copied onto AgrS after the error has taken place. The relevant Vocabulary items for
the error elements in (10) are listed in(11).
(11) a. hab
[+pl]
/ha:b/
/-n/
b. denk
[2nd]
/d7k/
/-st/
6.2.3 Summary
Traditionally, all of the errors I discussed in this section would be considered context accommodations. Moreover, according to the typology I proposed in Section
6.1, they would be classified as morphosyntactic. I have claimed, however, that all
morphosyntactic context accommodations are in fact the result of feature copy.
Clearly, the relevant process is morphosyntactic in nature, but accommodation
is not involved in these errors. Rather, the grammatical outcome results from the
application of a mechanism which according to DM is an essential part of the
derivation anyway. Table (12) shows the distribution of errors from my corpus in
Grammar as Processor
which a grammatical outcome is due to feature copy (remember from the discussion in Sections 5.2 and 5.3 that feature copy may also be erroneous, thereby giving
rise to an ungrammaticalutterance).
(12) Slips involving feature copy following the actual error (n =103)
feature copy within DP
genderfeature
numberfeature
feature copy onto AgrS
caseassignment
Total
88
83
5
13
2
103
Table (12) also includes two special cases which I did not address in this section.
These are errors in which, following a root exchange, a different case is assigned to
a DP. Note, however, that one of these errors will be subject to detailed discussion
in Section6.5.2.
Rethinking accommodation
Film,
dass
movie(m), that
ihr
Film
so gut
3.sg.f.poss.m movie(m) so well
mir
1.sg.dat
gefall-en
hat
please-part have.3.sg
b. ich
soll am Montag zu mir
komm-en
1.sg.nom shall on Monday to 1.sg.dat come-inf
zu ihm
komm-en
to 3.sg.m.dat come-inf
Berg (1987) analyzes all errors of this type as error accommodations. In particular, he argues that the exchanged or perseverated pronouns accommodate to the
syntactic environment in which they appear after the error has taken place. For the
slip in (13b), for instance, this would imply that ich is changed to mir under the
influence of a dative feature. Clearly, within a model that assumes the manipulation
of abstract features, such an analysis is unnecessary. The errors receive a straightforward explanation if we assume that a person feature (possibly in combination
with a gender feature) is shifted, while at the same time, a case feature is stranded.
In their post-error position, the shifted features combine with the case features
(which are assigned to DPs at MS). Moreover, in (13a), the gender feature of Film
(movie) will be copied onto the possessive pronoun. The resulting feature bundles
are spelled out at PF by the Vocabulary items listed in(14).
(14) a. [3rd] [f] [dat]
[1st] [poss] [n]
/i:/
/main/
b. [1st] [dat]
/mi:/
Grammar as Processor
In other words: given that no phonological forms participate in the above errors,
one need not assume that the phonological surface forms of the pronouns
involved undergo repair due to the different case specification of their respective
landingsites.
6.3.2 Phonological readjustment
The stranding errors discussed in the previous section are featural in nature;
they do not involve the manipulation of roots. I shall now turn to errors in
which the phonological form of a root changes under the influence of a stranded
feature. Despite their different nature, the slips to be presented in this section
would also be classified as morphosyntactic error accommodations. That is, the
error element itself undergoes a change. I will claim that the mechanism which
can be held responsible for this change is phonological readjustment atPF.
Amongst the morphosyntactic features that may trigger phonological readjustment in German are [+past], [+part], [+pl], and [3rd]. Here, I will only discuss examples involving the first two features (but see (17b) in the next section for
an example involving [+pl]). In (15a), les (read) is anticipated into a terminal
node in which it combines with [+past], while in (15b), lg (lie) is anticipated
into a slot in which it combines with [+part]. Hence, both errors are instances of
tense stranding (of the type discussed in Section5.4.2).
(15) a.
ich las
I
read.past
ihr frs,
h, ich dank-te
her for.the, er, I
thank-past
frs
Korrektur les-en
meines Handout-s
for.the correction read-inf of.my handout-gen
b. du
hast
doch
ge-log-en,
nicht
you(sg) have.2.sg mod.part part-lie-part not
versprech-en,
promise-inf
h, versproch-en, nicht
er, promise-part not
ihr
her
mehr
zu
anymore to
mehr
zu lg-en
anymore to lie-inf
As far as (15a) is concerned, the Vocabulary item (VI) /le:z/ will spell out les
at PF. Subsequently, a phonological readjustment rule (PRR) will trigger a steminternal change in the context of [+past]; see (16a) for the relevant operations.
Things are quite similar in (15b), only that in this error, the feature [+part] is
responsible for the application of a rule which changes the stem vowel of /ly:g/
(16b). Note that the phonological readjustment rules specify the phonological
change and the context in which it applies. Frequently, a readjustment rule applies
to more than one Vocabulary item; see, for instance,(16a).
Rethinking accommodation
/ X + [+part]
Note that the exchange in (15b) also illustrates the opposite phenomenon. In this
error, the Vocabulary item that spells out versprech (promise) is subject to phonological readjustment in the intended utterance. In the error, however, readjustment
is not called for because the root combines with [tns]. In all three errors, the change
in the surface form of the respective error elements comes for free since only abstract
roots are manipulated in the course of the derivation. Consequently, we are not dealing with an error accommodation in the sense of a post-error repairprocess.
6.3.3 Context-sensitive spell-out of features
The phenomenon I wish to introduce in this section is different from the ones
discussed above in that a stranded feature does not have an impact on the error
element be it a feature bundle or a root but rather influences the choice of
some inflectional morpheme. In other words: a stranded feature is spelled out differently because of its combination with a different root in the error. Given that it
is the morphological context that changes, errors of this type can be classified as
morphological context accommodations. As before, however, I will argue that no
accommodation is at work in theseerrors.
The two morphosyntactic features which are relevant in this context are [+pl]
and [+part] since spell-out of both these features is subject to allomorphic variation. In both errors in (17), a root which ends up in a nominal slot is accompanied
by a plural allomorph different from that present in the intended utterance. The
relevant roots are akzent (accent) in the exchange (17a) and bad (bath) in
the perseveration (17b). Note that in the latter error, the stranded plural feature
also triggers umlaut in the Vocabulary item that spells out bad, that is, a change
of the type discussed in the previoussection.
(17) a. die silben-tragenden Akzent-e die akzent-tragenden Silbe-n
the syllable-bearing accent-pl the accent-bearing syllable-pl
the syllables that bear accent
b. im
Schwimm-bad knn-en sich die
Bd-er, h,
in.the swim-bath
can-pl refl the.pl bath-pl, er,
die
Jung-s richtig austob-en
the.pl boy-pl really romp.about-inf
Grammar as Processor
While there are five different plural allomorphs in German (see Footnote 3), there
are only two choices for the participial affix. In German, the feature [+part] is
usually spelled out by a circumfix. The prefix-part is the same for both allomorphs
(ge-) but the suffix-part may be realized by either -t or -en.6 Both the errors in (18)
are self-corrected anticipations (incompletes), and in both of them, the anticipated root takes a participial suffix different from that of the intended utterance.
Also, in both of the slips, the error element originates from a position where it
combines with[tns].
(18) a.
e r hat
mich ge-drng-t,
he has.3.sg me part-push-part
ge-bet-en,
ihn nicht
part-ask-part him not
zu drng-en
to push-inf
b. er hat
ge-seh-en,
h, er hat
ge-sag-t,
he has.3.sg part-see-part, er, he has.3.sg part-say-part
In (19), I list the Vocabulary items which spell out the relevant morphosyntactic
features in the above errors (also see the discussion in Section 3.2.3.1). The two
plural allomorphs that surface in the slips in (17) are given in (19a), the participial
allomorphs that combine with the anticipated roots in (18) are given in (19b). For
all allomorphs, the context in which they are inserted has to be specified. I leave
open whether this context is a root (for example, X = akzent) or a Vocabulary
item (for example, X=/akts7nt/).
(19) a. [+pl]
/-/ / X
(where X = Akzent, Tag (day), Maus (mouse), )
[+pl]
/-r/ / X
(where X = Bad, Mann (man), Haus (house), )
b. [+part] /g--t/
/ X
(where X = drng, lieb (love), lach (laugh), )
[+part] /g--n/ / X
(where X = seh, helf (help), fahr (drive), )
. Note that on prefixed verbs such as vergeben (to forgive) and loan verbs such as
amsieren (to amuse), the participial form is marked by a suffix only.
Rethinking accommodation
Again, given that the plural and participial allomorphs are not yet phonologically
specified when the errors occur, it is unnecessary to assume the application of accommodatory processes that would change the phonological form of thesuffixes.
6.3.4 Summary
In Section 2.2.2, I have presented speech errors that are commonly taken to
prove that the language processor has access to the morphological structure of
words during language production, that is, errors in which derivational and/
or inflectional affixes are stranded in their original position. The discussion in
the present section has been a first step towards a revision of this assumption.
With respect to inflectional elements, the above analyses make clear that it is
not really affixes that are stranded in these errors but rather morphosyntactic
features. Obviously, this holds for errors in which the inflectional affix appears
in a different form in the error (see (17) and (18)) as well as for errors in which
an affix is the same in the intended and the erroneous utterance. In both cases,
the affix is the context-sensitive spell-out of a stranded feature. Sometimes, a
post-error context simply does not require a different affix. In addition, we have
seen that a stranded feature can also influence the spell-out of a feature bundle
(in the form of a pronoun) or a root. The distribution of errors from my corpus
in which a stranded feature is responsible for different spell-out of a feature
bundle, a root, or an affix is given in Table (20). Note that there is one error in
which a stranded [+neg] feature is spelled out by a different allomorph. This
error has not been discussed in this section but it has already been cited as
example (99a) in Chapter5.
(20) Slips in which a stranded feature has an impact on spell-out (n =53)
case stranding spell-out of feature bundle
14
4
16
9
1
Total
18
26
53
I have argued that in none of these cases, the different surface form be it of an
error element or an accompanying morpheme should be considered the result of
an accommodation, that is, a repair process which applies after the error has taken
Grammar as Processor
place. Rather, I suggest that the elements manipulated in the syntax are sufficiently
abstract to make the assumption of such a process superfluous. I conclude that, as
far as inflectional morphology is concerned, the processer need not have access to
morphological structure. In fact, it cannot have access to morphological structure
because no such structure is available at the point of the derivation at which the
errors occur. In the following section (in particular, Section 6.4.2), I will extend
this argument to derivationalmorphology.
Rethinking accommodation
. Corver (1997) also argues that quantifier-like degree items behave differently from other
degree words in various respects. He therefore claims that besides DegP, a functional quantifier phrase (QP) projection should be distinguished within the functional domain of the
extended adjectival projection.
Grammar as Processor
Rethinking accommodation
anticipated into a different licensing environment and therefore surfaces in a different phonologicalform.
(21) a.
ich hab-e
ein-en Wurf
ge-blick-t
I have-1.sg a-m.acc throw.nmlz(m) part-glance-part
ein-en Blick
ge-worf-en
a-m.acc glance(m) part-throw-part
b. d
er
Sprung,
h, der
Funke
spring-t
ber
the.m jump.nmlz(m), er, the.m spark(m) jump-3.sg over
It clicks (between them).
In (22), I give the relevant Vocabulary items and readjustment rules for the slips in
(21). Note that the Vocabulary items that spell out the error elements are specified
for the different licensing environments in which they may be inserted (remember that the symbol should be read as is licensed by). The Vocabulary item
/v7rf/ in (22a), for instance, can occur in various environments. Firstly, it may be
inserted in a [+v] environment (remember that [+v] is [d] by implication) but
only when the light verb head is filled by a cause morpheme, as is the case in the
intended utterance in (21a). Secondly, it may also appear in a [v][+d] environment (which, of course, is unspecified for [cause]), as is true in the error (21a). The
Vocabulary item in (22b) is specified for the same licensingenvironments.8
In both errors, a phonological readjustment rule is responsible for the observed
stem alternations (ablaut). The readjustment rules see to it that the stem vowel of a
certain Vocabulary item X changes in a certain licensingenvironment.
(22) a. VI: werf /v7rf/ / [v][d][+cause]
PRR: /7/
//
/ X [v][+d]
(where X = werf, brech (break), sprech (speak), )
b. VI
spring
PRR: /i/
/pri/ / [v][d][+cause]
// / X [v][+d]
Given that the changes in (21) only affect the respective stem vowel of the Vocabulary items, it is safe to assume that these errors involve the application of phonological readjustment rules. Things are different in the errors in (23), where the
phonological form of the error element in its post-error position differs more
. I acknowledge that things are probably more complex for (21b) since the involved verb
is actually not springen (jump) but rather the particle verb berspringen (jump to). In the
analysis in (22b), I neglect the particle. Also note that I leave open the question whether the
two Vocabulary items may be inserted in a [v][d] environment (as their Enlish equivalents
in the constructions the thrown ball or the jumping boy).
Grammar as Processor
dramatically from that in its intended position. In (23a), steh (stand) originates
from a position in which it is licensed by a light verb but is anticipated into a slot
in which it is licensed by a determiner. The same licensing conditions hold for the
error in (23b), in which zieh (drift) isanticipated.9
(23) a.
auf ein-em
Stand,
auf ein-em
Bein
on one-m.dat stand.nmlz(m), on one-n.dat leg(n)
b. Rauch-zg-e,
h, Rauch-wolke-n zieh-en Richtung Westen
smoke-drift.nmlz-pl, er, smoke-cloud-pl drift-3.pl direction West
Clouds of smoke are drifting westwards.
When it comes to Vocabulary insertion, two scenarios are possible. On the one
hand, one might assume that the regular Vocabulary items /te:/ and /tsi:/ spell
out the roots steh and zieh, respectively, and that subsequently, just as in (22),
phonological readjustment rules change the form of these items in an environment where they are locally licensed by a determiner. On the other hand, we might
be dealing with instances of suppletion. According to this scenario, different (that
is, more specified) Vocabulary items are inserted in a D-environment. Given the
phonological differences between the intended forms and the forms that surface
in the errors in (23), I adopt the latter scenario (see Harley & Noyer (1998a) for
discussion of the distinction between readjustment and suppletion). Hence, both
Vocabulary items in (24) are only specified for one specificcontext.10
. Note that the verb ziehen as well as the noun Zug have various meanings: ziehen can
mean to drift, as in the error, but also to pull and to move, march, wander, roam, amongst
other things; Zug can mean train but also tension, procession, and draft, amongst other
things. In (23b), the verb has the drift/move semantics, while the anticipated root (glossed as
drift.nmlz) would probably best be translated as draft.
. In principle, suppletion could also be observed in errors involving feature stranding (see
Section 6.3.2). However, my corpus does not contain an error in which the combination of a
root and a stranded feature would trigger the insertion of a different Vocabulary item. For the
sake of illustration consider the hypothetical English error in (i).
(ii) go
/wend/ / [+past]
(iii) [+past] /-t/
/ X
(where X = wend, send, buy, loose, )
According to Halle & Marantz (1993) go is spelled out by the Vocabulary item /wend/ in the
Rethinking accommodation
steh
/tant/ / [v][+d]
b. VI:
PRR:
As also indicated in (24b), two further changes are of interest in the slip (23b). First,
the [+pl] feature is spelled out by the appropriate allomorph. Secondly, the stem vowel
of the Vocabulary item /tsu:k/ undergoes umlaut in the context of the pluralfeature.
The English slip in (25a) is also of interest in this context. In this error, the
two roots pull.up and die are exchanged (Fromkin 1973a: 31). The first root
originates from a position in which it is licensed by a light verb but ends up in a
position in which it is licensed by a degree head. In this [v][d] environment,
spell-out of this root requires participial morphology. At its landing site, die is
spelled out by the Vocabulary item in (25b). This Vocabulary item may appear in a
[+v] environment, however, only when v is filled by a become morpheme, that is,
a combination of [cause] and [be] (as in The victim died). It is this very restriction which makes the slip in (25a) awkward, since in the error, the element die
appears in a [+cause] environment.11 I assume that spell-out of die is subject to
suppletion in other licensing environments. When licensed by a degree element,
for instance, it will be spelled out by the Vocabulary item in(25c).
(25) a. the gardener has to die the pulled up flowers
to pull up the dead flowers
b. die /dai/ / [+v][cause][be]
c. die /ded/ / [v][d]
I conclude from the above discussion that the (extended) functional head analysis
in combination with the notion of local licensing of l-nodes allows for a straightforward explanation of speech errors such as those given in (21) and (23). In many
cases, there is only one Vocabulary item for a given root, irrespective of licensing environment, and the difference in surface form is due to the application of a
phonological readjustment rule at PF. In other cases, it appears to be more likely
that we are dealing with suppletion. That is, a given root may be spelled out by
context of a [+past] feature (ii). Moreover, the regular past tense suffix /-t/ will be inserted under
the Tns node (iii) (and the final /-d/ of /wend/ will delete before /-t/ just as in send/sent).
. Adopting ideas of Generative Semantics, the combination of the atomic predicates cause
and die that is, cause and die should have given rise to the insertion of the Vocabulary item kill, since it is exactly the combination of these two predicates that paraphrases the
meaning of kill (Katz 1970; Lakoff 1971).
Grammar as Processor
welch-er
what-m
Schreib-er,
Quatsch, welch-er
write-nmlz(m), nonsense, what-m
schreib-t denn
so
was
write-3.sg mod.part such a.thing
Idiot
idiot(m)
b. er hat
ein-e
Erzhl-ung, h, ein-en
Schwank
he have.3.sg a-f.acc tell-nmlz(f), er, a-m.acc tale(m)
aus sein-er
Jugend
erzhl-t
from his-f.dat youth(f) tell-part
Rethinking accommodation
The exchange in (26c), despite the fact that it is English, is also from my corpus.
This error differs from the previous ones in two respects. First, it involves the
insertion of an adjectival suffix. Secondly, there is also an adjectival suffix in the
intended utterance, yet a different one. This illustrates that morpheme insertion
does not only depend on the licensing environment but also on the specific root
that is being licensed. This is taken into account in the morpheme insertion rules
in (27). Suffixes like the ones given in (27a) and (27b) are not only in competition
with each other but also in competition with all other suffixes that are specified
for insertion in a [v][+d] environment. It is only the additional specification of
the context X which guides the selection of one suffix over the other. Frequently,
in German, a given root can combine with various suffixes in one and the same
licensing environment (there may, for instance, be various nominalized forms for
one root); I will come back to this additional complexity in the next section. Note
that the morphemes specified in the below rules are abstract (as indicated by the
sumbol ); just like roots and features, they will be spelled out by Vocabulary
items atPF.
(27) a. Insert [-er(m)] / x [v][+d]
(where x = schreib, tanz (dance), lehr (teach), )
b. Insert [-ung(f)] / x [v][+d]
(where x = erzhl, wohn (live), hoff (hope), )
c. Insert [-al]
/ x [v][d]
(where x = nation, person, category, )
. Wiese (1996) points out, for instance, that -e, just like any other derivational suffix, determines its plural marker, the corresponding plural suffix being -n.
Grammar as Processor
Finally, the phonological readjustment rule also given in (28b) will trigger ablaut
in aD-environment.
(28) a.
sie hat
mir ihr-e Gab-e,
h, ihr-e Nummer
she have.3.sg me her-f give-nmlz(f), er, her-f number(f)
nicht
not
b. MIR:
VI:
PRR:
ge-geb-en
part-give-part
Insert [-e(f)] /
x [v][+d]
(where x = geb, folg (follow), lang (long), )
geb /ge:b/ / [v][d]
/e:/
/a:/
/ X [v][+d]
(where X = geb, )
DP
nP
D
n
LP
/ion/
/distrfi/
Hence, according to both proposals, morpheme insertion takes place at PF. What
is inserted is not an abstract morpheme but a phonological form. In light of the
German error data, this assumption is problematic because, as we have seen,
Rethinking accommodation
morpheme insertion must precede feature copy since nominalizing suffixes are
endowed with genderfeatures.13
Kihm (2005), discussing Spanish nominalizations, acknowledges the fact
that derivational morphemes can be gender-relevant. Note, however, that for
him, gender-relevant does not necessarily imply that the derivational suffix
comes with a gender feature. Rather, the distinction between gender-relevant
and non-gender-relevant suffixes it motivated by the observation that certain
Spanish suffixes may combine with different gender markers, while for others, there is no such choice available. He further considers derivational morphemes to be functional roots that, because of their functional character, must
combine with a nonfunctional root when they are taken from the lexicon (Kihm
2005:484). In other words, derivational morphemes, just like roots, are drawn
from List 1. For the English nominalization destruction, Kihm (2005:489) proposes the structure in(30).
(30) [nP tion [vP [P destroy ]]]
At first sight, this structure looks quite similar to the one in (29) (P being the
equivalent of LP).14 Crucially, however, tion is an abstract morpheme that is
present in thesyntax.
Extending Kihms proposal to German, one might suggest that derivational affixes
are present in List 1 and that German nominalizing suffixes are gender-relevant in the
sense that, just like roots, they are inherently specified for gender. While this analysis would allow us to account for the observed gender agreement in errors such as
(26b) in a straightforward way, it brings with it a serious conceptual problem. In particular, it makes it difficult to account for errors in which a derivational affix replaces
another affix present in the intended utterance. If the intended affix enters the computation together with the root that it is supposed to accompany, then we would have
to assume that it is substituted for by another affix after the error has taken place. In
(26c), for instance, the intended adjectival suffix -ous does not meet the subcategorization requirements of nation. Hence, ous has to be deleted and List 1 has to be
. Note, however, that Marantz (1998) does away with a separate level of morphology (MS).
As far as agreement morphology is concerned, he assumes that it is purely morphophonological and that therefore, agreement nodes are only added at PF. I will not further pursue this
proposal here. Still, I want to point out that following this proposal, one could still account
for the speech errors if one assumes that spell-out is cyclic and that morpheme insertion
precedes feature copy at PF (but see Harbour (2003) for arguments in favour of a separate
morphological level).
. According to Kihm (2005:490), the inclusion of v in the representation captures the fact
that these forms, despite being nouns, have the argument structure of the verb they nominalize.
Grammar as Processor
accessed again in order to select the appropriate suffix. This procedure constitutes a
real repair process (that is, a morphological accommodation) in that, following the
error, an abstract element that has been merged must be replaced. In contrast to that,
I have shown above that accommodation is not required when we assume that derivational morphemes are inserted at MS in specific licensing environments. I therefore
conclude that morpheme insertion at MS should be maintained, given that it allows
for a more elegant and more economical explanation of the speech errordata.
6.4.3 Competing nominalizations and DP-internal structure
As already implied in the previous section, more has to be said about errors that
involve nominalizations. Closer inspection of some of these errors reveals that
things are probably more complex than assumed above. Consider, for instance,
again the slip in (26a). Interestingly, in this case, there are (at least) three different
conceivable nominalizations of the abstract root schreib. We must therefore ask
why, after the error has taken place, the root is spelled out as Schreib-er (writer)
and not as Schreib-ung (spelling) or Schrift((hand)writing)?
Following Abney (1987), Szabolcsi (1994), Marantz (1997), and Harley &
Noyer (1998b), I assume that the functional structure within DP is much richer
than has been assumed in the above discussion, actually paralleling the functional structure of the clause and involving additional functional projections (see,
for instance, Marvin (2002) for Slovenian nominalizations). Without going into
details of the syntactic representation of nominalizations, I want to argue that the
nominalization which is spelled out in the error is usually the one that best fits the
internal semantics, that is, the DP-internal functional structure, of the intended
noun. For this reason, schreib in (26a) is spelled out as Schreiber, which, just
like the intended noun Idiot (idiot), has agentive semantics, and not as Schreibung
(which could be argued to be eventive) or Schrift (which isstative).
Interestingly, my corpus also contains an error in which schreib is spelled
out as Schrift, see (31a). In this error, too, schreib is perseverated into a position
where it is licensed by a determiner. For this case, I suggest that Schrift, just like
the replaced element Strich (line), has stative semantics and therefore best fits
the functional make-up of the slot it is perseverated into.15 Note that Schrift, in
contrast to Schreiber, does not involve the insertion of a morpheme at MS. Rather,
just like the cases in (24), it is probably an instance of suppletion in a [v][+d]
environment (and not the result of phonologicalreadjustment).
. In addition, it could be argued that the phonological similarity of Strich and Schrift
contributes to the selection of Schrift over the other two forms (see Section 6.6.3 for further
discussion).
Rethinking accommodation
The errors in (31b) and (31c) can be explained along similar lines. In (31b),
terror is anticipated into a position where it is licensed by D but is neither
spelled out as Terror (terror) nor as Terrorismus (terrorism) in that position. Crucially, both the intended noun Direktor (director) and Terrorist (terrorist), the noun surfacing in the error, have agentive semantics and refer to
individuals. That is, the nominalization Terrorist best fits the slot into which it
isanticipated.
(31) a. schreib-t man das mit Binde-schrift
write-3.sg one that with connect-write.nmlz(f)
mit Binde-strich
with connect-line(m)
die
gesamte Belegschaft terror-isier-t
the.f.acc whole staff(f) terror-ize-3.sg
c.
der
Tour-ismus,
die Ignoranz
der
Tour-ist-en
the.m tour-nmlz(m), the.f ignorance(f) of.the tour-nmlz(m)-pl
nimm-t
von Jahr zu Jahr zu
increase-3.sg from year to year particle
Finally, the slip in (31c) is particularly interesting because here, tour originates
from a position where it is licensed by a determiner and is anticipated into another
position where it is also licensed by D.Still, the root combines with a different derivational morpheme at its landing site. In its original slot, tour combines with
the abstract morpheme [-ist(m)] and receives an agentive interpretation (tourist). In contrast, both the intended noun Ignoranz (ignorance) and the resulting
nominalization Tourismus (tourism) can be argued to be stative. This line of reasoning also explains why the anticipated root tour is not simply spelled-out as
Tour (tour) which has eventivesemantics.
The errors in (31), as well as those in (26a) and (26b), are representative for
the nominalization patterns in my corpus. Whenever a root is anticipated or perseverated into a slot where it is licensed by a determiner (that is by [v][+d]) and
there is a choice with respect to morpheme insertion, the morpheme which best
fits the semantics of the intended noun will be inserted. In Table (32), I give three
more examples which involve the insertion of a semantically appropriate morpheme at MS. I specify the nominalization surfacing in the error, the intended
noun, as well as conceivable alternative nominalizations of the errorelements.
Grammar as Processor
Intendednoun
Alternative
nominalization
Sicher-heit
(safety)
bel-keit
(nausea)
Sicher-ung
(protection)
Herrsch-aft
(reign,power)
Diktatur
(dictatorship)
Herrsch-er
(ruler)
Verkuf-er
(salesperson)
Freund
(friend)
Verkauf
(sale)
I did not go into the internal syntax of these nominalizations but I take it to be
likely that DP-internal light verb heads (see (30)) that are equipped with the features [be][cause] can be held responsible for the insertion of one morpheme
over another. Alternatively, if we adopt Marantz (2001) proposal of a little n head,
it might be argued that this functional head hosts (some of) the relevant features.
In addition, a DP-internal aspectual head may be involved in some of the nominalizations, as has been argued, for instance, for destroying (as in Johns destroying
the city) by Marantz (1998). Actually, I consider the speech error patterns discussed in this section intriguing psycholinguistic evidence for the assumption of
such additional functionalstructure.
6.4.4 Accounting for categorial identity
In the preceding section, I have argued that a number of complex speech errors
receive a straightforward explanation when we adopt the DM assumption that
only abstract, acategorial roots and features are manipulated within the computational system. Most importantly, we do not need to call upon the service of
costly accommodatory processes in order to rectify possible morphological or
morphosyntacticmismatches.
With respect to speech error patterns, however, abandoning category labels
also gives rise to a serious conceptual problem. It is a well-known and wellsubstantiated fact that the grammatical category of the involved elements plays an
important role in word exchanges. In particular, there is a strong tendency for the
elements participating in the exchange to be of the same grammatical category.
Things are different for stranding errors and sound exchanges, however, which
typically involve words from different grammatical categories. It is that very property of exchange errors which lead Garrett (1975, 1980a) to the assumption that
different types of exchanges occur at different processing levels. He argues that
word exchanges take place at the functional level, at which phrasal membership and
Rethinking accommodation
grammatical category of lemmas are taken into account, while sound exchanges
and stranding errors occur at the positional level, at which the serial order of
words as well as aspects of their form are specified (see Section 3.1.1). Three representative exchange errors from Garrett (1980a: 179,188) are given in (33), a word
exchange (33a), a stranding error (33b), and a sound exchange(33c).
(33) a. I left the briefcase in my cigar
the cigar in my briefcase
b. I thought the park was truck-ed the truck was parked
c. on a sot holdering iron
a hot soldering iron
Garrett (1980a: 189) states that 85% of the word exchanges from his corpus obey
the same category constraint, while the same is true for only 43% of the stranding
errors and 39% of the soundexchanges.
To date, the Frankfurt corpus contains 163 clear instances of word exchanges.
Of these, only twenty involve words of different grammatical categories. That is,
87.7% of the word exchanges obey the same-category constraint a number very
similar to that reported by Garrett. With respect to sound exchanges, however,
percentages differ. There are 394 sound exchanges in the Frankfurt corpus. 190
of these sound exchanges occur within a word, be it morphologically complex
(for instance, Wesserbisser Besserwisser (know-all)) or monomorphemic (for
instance, Kvalier Klavier (piano)), and were therefore not considered. Of the
remaining 204 sound exchanges, 107 (52.5%) involve words of different categories.
Interestingly, for the 85 stranding errors, the bias towards mixed-category errors is
much stronger: in 76 of them (86.4%), words of different categories are involved.
For the readers convenience, these numbers are summarized in Table(34).
(34) Grammatical category constraint in exchanges (n =452)
Type of exchange
Same category
Different category
143 (87.7%)
9 (10.6%)
97 (47.5%)
20 (12.3%)
76 (89.4%)
107(52.5%)
Total
249
203
Leaving sound exchanges aside for the moment, we still need to account for the
fact that the same-category constraint obviously holds for word exchanges but not
for stranding errors. Clearly, following DM assumptions, we cannot assume that
categorially specified words or morphemes are exchanged before Vocabulary insertion takes place, since the only elements available for exchange prior to spell-out are
abstract features and acategorial roots. Above, I argued that stranding errors occur
before spell-out. Consequently, they resemble word exchanges in that in both roots
are exchanged (but see Section 6.7.1.1 for exceptions). The difference between word
Grammar as Processor
exchanges and stranding errors then reduces to the fact that in the latter, a feature
which has an impact on spell-out is stranded (for instance, [+past] in(33b)).
How, then, can the same-category constraint be accounted for? In the following,
I will reconsider word exchanges and stranding errors in light of the DM framework.
As far as root exchanges of the word-type are concerned, I will suggest that the same
category constraint can be reformulated in terms of licensing environments (Section
6.4.4.1). Things are different for root exchanges of the stranding-type. For these,
I will argue that they are facilitated by the fact that, for the most part, the interacting
roots are adjacent to each other (Section 6.4.4.2). In addition to root exchanges, I will
discuss data that show that exchanges can also occur after Vocabulary insertion. In
this case, however, we are not dealing with root exchanges but rather with word or
morpheme exchanges after spell-out. Consequently, in such errors, a grammatical
outcome cannot be guaranteed (also see Ferreira & Humphreys (2001) for the role
of syntactic category in experimentally elicited stemexchanges).
6.4.4.1 The role of licensing in root exchanges
Generally, roots are not randomly exchanged in speech errors. Rather, there is
a strong tendency for a manipulated root to take a position in which it is locally
licensed by the same kind of functional head as in its original position. This holds
for the English slip in (33a) and also for the two errors in (35). Traditionally, these
errors would be classified as noun exchanges. Within DM, however, they have
to be treated as root exchanges. This is particularly evident for the error in (35b)
because here, the two interacting roots schwalbe (swallow) and sommer
(summer) are specified for different gender features and both determiners surface in their appropriately gender-marked form. In other words, the error must
have taken place before gender copy at MS. I assume that the same is true for (35a),
although in this error, it cannot be decided whether the gender features have been
copied before or after the exchange. For both errors, it is true that the interacting
roots are locally licensed by adeterminer.
(35) a. eine Theorie
ist
eine Grammatik des
Wissens
a.f theory(f) be.3.sg a.f grammar(f) of.the.n knowledge(n)
b. ein Sommer
mach-t
noch kein-e
Schwalbe
a.m summer(m) make-3.sg yet
no-f.acc swallow(f)
Rethinking accommodation
However, the constraint on interacting elements only holds for errors that occur
prior to Vocabulary insertion. After spell-out of Vocabulary items possibly followed by phonological readjustment in certain licensing environments the job of
the licensing elements is done. All errors occurring after that point can no longer
be constrained by the licensing environment of the involved elements; that is, they
are purelyphonological.
The fact that words may be exchanged after spell-out is illustrated by the
errors in (36) and (37). Note that both these errors are from the Frankfurt corpus. They are not included in my corpus because they do not fall into one of
the four error classes that my corpus contains (see Table (2) in Chapter 1). The
exchange in (36a) affects elements from different licensing environments. Before
spell-out, blass (pale) is licensed by a degree element, while neid (envy)
is licensed by a determiner. If we were dealing with a root exchange, then we
would expect both roots to be properly spelled out at their landing sites. For
both roots, this would involve morpheme insertion. In addition, for blass,
it would involve phonological readjustment (umlaut) in a D-environment. For
the sake of illustration, I give the hypothetical outcome of a root exchange in
(36b). The English equivalent of the properly spelled out string would be envious
withpaleness.
(36) a.
da
wird
mancher Neid
there will.3.sg some
envy
da
wird
mancher neid-isch vor Blss-e
werd-en
there will.3.sg some
envy-adj with pale-nmlz become-inf
In (37a), the two interacting roots are licensed by a determiner and a light verb,
respectively, before spell-out. The phonological form glnzt (glitters) is actually
the spell-out of glanz (glitter, shine) in combination with third person singular
agreement. Umlaut formation is triggered in this Vocabulary item in the environment of a light verb but not in a D-environment. The fact that the error element is shifted in its phonologically readjusted and inflected form clearly indicates
that the error must have taken place after spell-out. If the exchange had occurred
before spell-out, then the expected (nominalized) form would have been Glanz,
as is indicated in the hypothetical error in (37b). For gold (gold), it is actually
not clear how it could be spelled out when licensed by a light verb. The only possible verbalization that comes to mind is vergolden (to gild, gold-plate) but this
Grammar as Processor
verb has a causative meaning and probably doesnt fit the content of the light verb
head in the intended utterance. Hence, in the hypothetical error in (37b), I leave
open how gold would have been spelled out had the error taken place before
Vocabularyinsertion.
(37) a.
With respect to the so-called word exchanges, I therefore conclude that they come
in two different types. The first one is actually a root exchange, which happens
before Vocabulary insertion is carried out. This kind of exchange is constrained
by the licensing environments of the roots participating in the error. Following
the error, the usual mechanisms that is, morpheme insertion, feature copy, spellout, and phonological readjustment apply and therefore a grammatical, albeit
possibly awkward, outcome is guaranteed. The second type of exchange occurs
after Vocabulary insertion and involves phonological words. This type is not
constrained by licensing environments. Since all of the before-mentioned DMmechanisms have applied at this point, there is no way for the exchanged elements
to adapt to their post-error environment. It should be noted, however, that this
type of error appears to be quiterare.
6.4.4.2 The role of adjacency in root exchanges
As pointed out above, the licensing conditions appear to be different for stranding errors, for which it has been argued that they are not subject to the samecategory constraint. As shown in Table (34), in almost 90% of the cases, the
exchanged morphemes constitute part of elements of different grammatical
category. At first sight, this observation implies that stranding errors (just like
exchanges of phonological words) occur at a stage at which licensing elements
are no longer relevant, that is, after spell-out. For these cases, we therefore predict that neither adaptation of the exchanged morphemes to their new environment nor adaptation of stranded material to the exchanged morphemes should
beobserved.
This is in fact true for the error in (38a). In this error, the stems pfeif
(whistle) and Tanz (dance) are exchanged, while a past tense and a nominalizing suffix are stranded. The outcome of the error is not grammatically wellformed. If the error had occurred before spell-out, that is, if roots had been
exchanged, then the expected outcome would be the one given in (38b). In this
Rethinking accommodation
hypothetical error, the Vocabulary item which spells out pfeif will undergo
phonological readjustment in the context of [+past]. Moreover, tanz will
combine with a zero-affix which is specified for masculine gender when licensed
by a determiner.16 The gender feature will be copied onto the determiner
(the possessivepronoun).
(38) a. er pfeif-te
nach ihr-er
Tanz-e,
h,
he whistle-past to
her-f.dat dance-nmlz(f), er,
tanz-te
nach ihr-er
Pfeif-e
dance-past to
her-f.dat whistle-nmlz(f)
er pfiff
he whistle.past
nach ihr-em
Tanz
to
her-m.dat dance.nmlz(m)
The fact that in this and other stranding errors, the exchanged elements are
not spelled out correctly follows automatically from the assumption that these
errors occur after Vocabulary insertion. Remember, however, that I have already
shown above that there are also stranding errors in which either the form
of the exchanged elements or the form of an inflectional or derivational affix
changes after the error (see, for instance, the slips in (21a) and (26c)). I have
argued that the grammatical outcome can be explained when we assume that in
these errors, roots are exchanged before spell-out. This implies that, in contrast
to (38), we are not dealing with stranding in the sense of morpheme stranding,
for the simple reason that there a no morphemes present when the error occurs.
Rather, as also argued above, the surface form of these errors results from the
stranding of morphosyntactic features and/or local licensing. For further illustration consider the error in (39a). In this exchange, two things are of interest.
First, the Vocabulary item which spells out brech (break) is phonologically
readjusted to Bruch when licensed by a determiner. Secondly, the correct participial allomorph ge- -t is chosen for bann (spell) in its post-error environment. Hence, the error must have occurred before Vocabulary insertion. The
hypothetical ungrammatical outcome of a morpheme exchange after spell-out
is illustrated in(39b).
. Another possible nominalization of the Vocabulary item that spells out tanz would be
Tnz-er (dancer). Tanz, however, is the nominalization which best fits the semantics (that is,
the DP-internal functional structure) of Pfeife (whistleN); see Section 6.4.3.
Grammar as Processor
der
Bann
ge-broch-en
the.m spell(m) part-break-part
da
war der
Broch
ge-bann-en
there was the.m break.part part-spell-part
From the above data, I conclude that the so-called stranding errors also come
in two types: as root exchanges before spell-out and as morpheme exchanges
after spell-out. If one wanted to stick to the term stranding for the former type,
then one could argue that these errors involve stranding of features and of a
licensingelement.
Now, this being said, we have to return to the licensing environment constraint argued for in the previous section. How can errors like the one in (39a) be
accounted for if we stick to the assumption that identical licensing environments
are a precondition for the exchange of roots? One way might be to somewhat
loosen the same-licenser constraint.17 It is a well-known fact that the so-called
word exchanges typically involve elements from different phrases, while stranding
errors (as well as sound exchanges) typically involve elements which appear under
the same maximal projection (Garrett 1980a: 189). We may therefore hypothesize
that the closer the exchanged elements are to each other in a syntactic tree structure, the less influence the same-licenser constraint has. This is exemplified by the
bracketed structures for the two representative root exchanges in (35b) and (39a),
respectively, given in(40).
(40) a. [CP [DP ein [LP Sommer]] [TnsP macht [LP [DP keine [LP Schwalbe ]]]]]
b. [LP2 [DP der [LP1 Bruch]] [L ge-bann-t]]
These are simplified structures, of course. In fact, Garretts statement that stranding errors typically involve members of a single phrase is not completely adequate.
In (40b), for instance, a lexical phrase (LP1) and a DP separate the two error
. Alternatively, one might argue that the roots that interact in (39a) are actually licensed
by the same type of element because the nominalization Bruch contains a light verb head
which c-commands brech (see the discussion in Section 6.4.3). I will not further pursue this
possibility here. Let me just point out that an analysis along these lines would require further
stipulations in order to guarantee that the appropriate phonological readjustment rules apply.
After all, phonological readjustment of the Vocabulary item brech is not required in a light
verb environment.
Rethinking accommodation
elements, which, however, appear under the same maximal projection LP2 (which
corresponds to VP). Still, it is true that in almost all of the exchanges which affect
roots from similar licensing environments, these roots are separated from each
other by a larger number of maximal projections. This is true, for instance, for the
slip in (40a), in which the exchanged roots both appear in a D-environment. The
first root is contained in a DP that occupies SpecCP, the second root is part of a DP
which is embedded under a LP (again, corresponding toVP).
Interestingly, a closer look at the root exchanges reveals that in almost all of
them, the roots which interact in the error are adjacent to each other in the sense
that no other root that is, no other possible candidate for exchange intervenes
between the exchanged elements. Actually, my corpus contains only two stranding
errors in which the exchanged roots appear in considerable distance from each
other. In both cases, however, the involved roots come from the same licensing
environment. Consider, for instance, the error in (41a) in which lach (laugh)
and sprech (speak) are exchanged, while agreement features are stranded.
Both roots are licensed by a light verb. Clearly, the error must have taken place
before spell-out because the Vocabulary item that spells out sprech undergoes
the required phonological readjustment in its post-error position (ablaut in the
context of the feature [3rd]). You will notice that another root, namely span
(Spanish), appears between the exchanged elements; this root, however, is licensed
by adeterminer.
(41) a.
er sprich-t
immer, wenn ich Spanisch lach-e
he speak-3.sg always when I
Spanish laugh-1.sg
er lach-t
immer wenn ich Spanisch sprech-e
he speak-3.sg always when I Spanish laugh-1.sg
Moreover, there are twelve word exchanges in my corpus, that is, root exchanges
without stranded morphemes, in which another root intervenes between the
exchanged elements. In all twelve cases, however, the intervening root is licensed
by a different functional head. Two examples have already been given in (35)
where the intervening roots are sein (be) and mach (make), respectively.
One more example is provided in (41b). As before, the exchanged roots mensch
((hu)man) and affe (ape) both come from a D-environment, while the intervening root stamm (descend) is licensed by a lightverb.
Grammar as Processor
The picture that emerges from the above discussion is that there are two
scenarios that allow for the exchange of roots in speech errors. In the first scenario, the two roots are adjacent to each other. In this case, the two roots need
not be licensed by the same type of functional element. In the second scenario,
another root, that is, another possible candidate for exchange, intervenes between
the exchanged elements. However, whenever this is the case, the exchanged roots
are locally licensed by the same licensing element, while the intervening element
appears in a different licensing environment. In other words: these errors obey the
licensing environmentconstraint.
As usual, a few problematic cases remain. In conclusion of this section, I want
to briefly discuss two root exchanges which are not readily accounted for following the above analysis. One such tricky case is the English stranding error cited in
(4c) above, repeated here as (42a). In that error, the appropriate adjectival suffix
-ful is inserted following the exchange and we must therefore assume that the error
occurred before Vocabulary insertion. Since the exchanged roots appear in considerable distance from each other, we would expect them to be licensed by the same
kind of element (as in (41a)). This, however, is not the case: in the intended utterance reason is licensed by a degree element and care by a determiner. Moreover, there is another root, namely measure, intervening between the exchanged
elements which is licensed by yet another element (a light verb). Consequently,
in this sequence, one would either expect reason to interact with measure
(giving rise to the error its measurable to reason with care) or measure to interact with care (yielding its reasonable to care withmeasure).
(42) a. I think its care-ful to measure with reason
its reasonable to measure with care
b. das ist Marc-s
Bruder Anke, Anke-s
Bruder Marc
that is Marc-gen brother Anke Anke-gen brother Marc
That is Ankes brother Marc.
Rethinking accommodation
elements of the same grammatical category tend to interact. I have argued that in
DM, this tendency can be explained without bothering category labels, when we
assume that prior to Vocabulary insertion, the interaction of roots in an error is
constrained by the licensing environment in which they occur. This constraint,
however, may become ineffective whenever the affected roots are sufficiently close
to each other and no other root intervenes.
Moreover, I have argued that the common distinction between word exchanges
and stranding errors needs to be reconsidered. In psycholinguistic models of language production, it is generally assumed that word exchanges take place early (at
the functional level), while stranding errors take place at a later point in the derivation (at the positional level). The discussion of German speech errors suggests
that this characterization may be too simplistic. In fact, I have shown that both
types of errors may occur before and after the insertion of Vocabulary items into
terminal nodes. In the former case, we are actually dealing with instances of root
exchanges. Whenever an error occurs prior to spell-out, DM-mechanisms such as
morpheme insertion, feature copy, and phonological readjustment ensure a grammatical outcome. These mechanisms may influence the surface form of an error
element and/or the context in which it appears. In addition, exchanges may also
take place after Vocabulary insertion. In these cases, however, we are actually dealing with an exchange of phonological material, be it a word or a morpheme. At this
point, the licensing environment does no longer constrain the interaction of error
elements. Generally, word or morpheme errors that occur after spell-out tend to
result in ungrammatical utterances because all the relevant DM-mechanisms have
already applied at thispoint.18
. In this respect, these two types of exchanges resemble sound exchanges. There is, for instance, no sound exchange in the Frankfurt corpus in which the exchange accidentally results
in an existing noun which in turn triggers accommodation on an adjacent element, for instance, a determiner. For the sake of illustration, consider the consonant exchange in (i).
(i)
ihr
drf-t
die Kraut
bss-en,
you(pl) may-2.pl the.f cabbage(n) (error)-inf
die
Braut
kss-en
the(f) bride(f) kiss-inf
In this error, the first word resulting from the exchange of /k/ and /b/ happens to be the existing German word Kraut (cabbage). However, in contrast to Braut (bride), which is feminine, Kraut is of neuter gender. Still, the definite article does not surface in its neuter form
(das). Such an accommodation would be quite surprising, of course, since all feature copy
processes have already been executed when the error occurs.
Grammar as Processor
6.4.5 Summary
It turns out that local licensing is a powerful tool when it comes to explaining
speech errors. The above discussion shows that many errors that have traditionally
been considered the result of a (morphological or morphosyntactic) accommodation receive a straightforward explanation under a licensing analysis. Crucially,
under such an analysis, the fact that an error element or a derivational morpheme
surfaces in a form different from the intended one is not seen as the result of a
post-error repair process. Rather, the observed changes result from local licensing
of an abstract root by a c-commanding functional head. I have shown that local
licensing can have an effect on the surface form in three different ways. First, it may
trigger the insertion of abstract morphemes at MS in certain licensing environments. In German, some of these morphemes are equipped with a gender feature.
Secondly, it may give rise to suppletion when a more specified Vocabulary item
is selected for insertion at PF due to the licensing environment in which the root
appears. Finally, after Vocabulary insertion has taken place, local licensing may
also be responsible for the application of a phonological readjustment rule. Table
(43) shows the distribution of errors from my corpus across the different licensing effects and it also specifies how many errors are attested in a specific licensing
environment. Remember that two effects may combine in one error (for instance,
morpheme insertion and phonological readjustment, as in (28a)). The table also
shows that the effect of licensing is most pronounced in D-environments, probably
because the choice of competing suffixes is most extensive fornominalizations.
(43) Slips in which local licensing has an impact on spell-out (n =70)
local licensing triggers morpheme insertion
in D-environment (nominalizingsuffix)
in Deg-environment (adjectivalsuffix)
local licensing triggerssuppletion
inD-environment
inDeg-environment
inv-environment
local licensing triggers phonol.readjustment
inD-environment
inDeg-environment
inv-environment
Total
34
25
9
11
7
3
1
25
15
3
7
70
Rethinking accommodation
morphology. I have suggested that numerous errors which seem to involve stranding of a derivational affix actually involve stranding of a licensing element which
triggers morpheme insertion. Again, this implies that prior to Vocabulary insertion, the language processor does not have access to morphological structure
because there is no such structure. It is only after spell-out that morphemes can be
the target of an error and only these errors are real strandingerrors.
On basis of the speech error data presented in this section, I conclude that language production, when assisted by DM-mechanisms, can do without reference to
category labels. In fact, many of the errors receive a more elegant explanation once
we assume that only acategorial roots and morphosyntactic features are manipulated
in the syntax. In particular, under this assumption, the postulation of repair processes that change the form of a stem or a derivational affix becomessuperfluous.19
6.4.6 An alternative account: Minimize Exponence
Before concluding this section, I want to discuss a recent theoretical account
brought forward by Siddiqi (2006) which, despite the fact that it is also couched
within DM, takes a somewhat different perspective on MS-operations and spellout. I will first sketch the crucial characteristics of his account and then show what
implications it has for the analysis of speecherrors.
Let me first point out that all the central ideas of DM that figured prominently
in the previous discussion are also adopted by Siddiqi. In particular, he assumes
that the computational system only manipulates abstract morphosyntactic features
and acategorial roots, that well-defined operations at MS may change the number
and organization of terminal nodes, and that Vocabulary insertion follows syntax
(late insertion). As for the roots that are manipulated in the syntax, Siddiqi follows
Pfaus (2000) proposal that roots are specific to the particular concept they are
linked to (for instance, dog; see Section 4.1). What he adds to the general picture
is an economy constraint on the grammar which he labels Minimize Exponence.
This constraint is defined asfollows.
Minimize Exponence (ME)
The most economical derivation will be the one that maximally realizes all the
formal features of the derivation with the fewest morphemes. (Siddiqi 2006:82)
. See Barner & Bale (2002) for further psycholinguistc arguments in favor of categorial
underspecification. The authors claim that a theory without lexical categories (like DM) offers
a natural solution to the boostrapping problem in language acquisition. In addition, they
discuss category-specific neurological deficits and show how these can be accounted for in
a model that assumes the manipulation of acategorial roots (see Panagiotidis (2005) for a
critique and Barner & Bale (2005) for a rejoinder).
Grammar as Processor
In other words: the most economical derivation is the one that is spelled out by the
fewest Vocabulary items. In a nutshell, Siddiqi argues for a proliferation of merger
and fusion operations at MS, for a larger inventory of Vocabulary items, and for a
reduction (or even abolition) of phonological readjustmentrules.
Let me illustrate his proposal with the irregular plural form mice (Siddiqi
2006:48f). In traditional DM, the syntactic structure for the pluralization would
look like (44). mouse is licensed by little n (which will be realized by a null morpheme). At PF, mouse will be spelled out by the Vocabulary item /mas/. In this
specific context, the plural feature in Num0 will not be spelled out by the regular
suffix -s but by a zero affix. In addition, in the context of [+pl], a phonological
readjustment rule will change the formof /mas/ to /mais/.
(44) a.
NumP
nP
Num0
[+]
LP
Siddiqi points out that there is strange interdependence in the derivation. On the
one hand, the null plural morpheme is licensed by the specific root mouse; on
the other hand, the readjustment rule is triggered by the presence of the plural
feature. He suggests that this awkward situation can be avoided when we assume
merger and fusion of terminal nodes at MS. In a first step, the root undergoes
head movement (that is, morphological merger) to adjoin to the functional heads
above it. This operation yields the complex adjunction structure below Num0 in
(44b). Subsequently, the resulting complex head (marked by the broken circle) will
undergo fusion to incorporate the features and the root into one simplex head,
see(44c).
(44) b.
NumP
Num0
Num0
[+]
nP
tn
LP
tL
Rethinking accommodation
c.
NumP
nP
Num0
[+]
[n]
tn
LP
tL
The resulting node, containing the root and several grammatical features, will then
be the target for Vocabulary insertion. Siddiqi assumes that the specific Vocabulary item given in (45a) spells out the root-feature bundle in (44c). Crucially, this
item differs from the one that would spell out the node in the absence of [+pl]
(45b). That is, there are two separate Vocabulary items for the singular and the
plural form and the derivation does neither involve the insertion of zero affixes
nor the application of a phonological readjustmentrule.
(45) a. [+pl] [n]
mouse
/mais/
b. [n] mouse
/mas/
Exactly the same argument can be made for roots that are licensed by other functional elements. run, for instance, when licensed by little v and appearing in a
[+past] context, will merge with the light verb head and Tns, and after fusion, the
resulting node will be spelled out by the more specified Vocabulary item /rn/
(Siddiqi2006:54f).
As a result of the suggested merger and fusion operations before spell-out,
more than one terminal element is realized by just a single Vocabulary item because
Vocabulary items can be specified for both a root and formal features. Actually,
fusion is driven by the need to make the utterance contain as few morphemes as
possible (Siddiqi 2006:83). In this sense, the derivation obeys the ME-constraint.
Siddiqis proposal is elegant in that it renders unnecessary the application of phonological readjustment rules. Since in his model, Vocabulary items which are
linked to the same root compete with each other for insertion, readjustment rules
are no longer needed to alter the phonological form of Vocabulary items in certain contexts. In addition, the proposal is attractive because it drastically reduces
the number of null morphemes that DM is forced to propose. It has to be noted,
however, that the suggested ubiquitous application of merger and fusion makes
the computational load heavier: more computation is needed for every derivation. Also, it clearly increases the size of the Vocabulary. Siddiqi argues that these
Grammar as Processor
negative side effects have to be put up with in order to satisfy the larger economy
constraint MinimizeExponence.20
Following this exposition, I shall now have have another look at two of the
speech errors that have been analyzed in Sections 6.3.2 and 6.4.1, namely (15a)
and (21b), repeated here as (46a) and (46b). Above, I argued that both errors
involve the application of a phonological readjustment rule. In the first error,
the readjustment rule is triggered by a stranded [+past] feature, in the second
error, readjustment is due to the licensing environment in which the anticipated
rootappears.
(46) a.
ich las
I
read.past
ihr frs,
h, ich dank-te
her for.the, er, I
thank-past
frs
Korrektur les-en
meines Handout-s
for.the correction read-inf of.my handout-gen
ihr
her
b. d
er Sprung,
h, der Funke
spring-t
ber
the.m jump.nmlz(m), er, the.m spark(m) jump-3.sg over
It clicks (between them).
. With respect to Vocabulary size, Siddiqi (2006:82) also argues that a possible compromise might be to have fused forms for the most frequently used roots while leaving less
frequent forms to regular morphological processes.
Let me also point out that Siddiqi shows that the ME-constraint he proposes has additional applications. In particular, it offers a new way to account for inflection within English
nominal compounds and for subcategorization (argument selection).
. Given that agreement inflection is regular in the past tense even for strong verbs, one
might also assume that the root only fuses with [+past] and that a separate Vocabulary item
spells out the agreement features. Note, however, that the agreement suffix is zero for the first
and third person singular.
Rethinking accommodation
additional merging operations may be required to ensure selection of the appropriate Vocabulary item, for instance, merger of the root with a feature contained
in a DP-internal light verbhead.
(47) a. [+past] [1st]
les
b. [n] spring
/la:z/
/pr~]/
I conclude from this brief discussion that Siddiqis Minimize Exponence model is
well compatible with the speech error data. In principle, all the error data which
I claimed to involve the application of a phonological readjustment rule can also be
accounted for when we assume the insertion of more highly specified Vocabulary
items after merger and fusion of roots and grammatical features. What remains to
be explored though is how ME handles morpheme insertion. Just like phonological readjustment rules, morpheme insertion rules might become superfluous once
we adopt Siddiqis proposal. Fusion of a root (for example, dance) with little n,
for instance, might result in a terminal node in which a morphologically complex
Vocabulary item (dancer) is inserted. Intuitively, however, the suggested fusion
processes seem much more plausible for cases which involve zero affixes and steminternalchanges.
Grammar as Processor
determiner and the adjective are spelled out in their neuter form. This change
in surface form can only be explained when we assume that we are actually
dealing with an exchange of bel (bad) and klag because only bel
possesses a neuter feature in its nominalized form. We may therefore assume
that the complete erroneous DP that was planned was ein ganz klgliches bel
(a very miserableevil).
(48) das ist
wirklich ein
ganz klg-lich-es,
h,
this be.3.sg really
a.n.nom very charge-adj-n, er,
ein-e
ganz bl-e Klage
a-f.nom very bad-f charge(f)
This is really a very bad charge.
A syntactic structure for the DP after root exchange is given in (49). In this structure,
the degree element ganz (very) is taken to occupy the head of DegP (Corver1997).
(49)
DP
LP
D
[]
DegP
Deg
LP
root exchange
Following the root exchange, various things happen at MS. First of all, the DP
will be assigned nominative case. Second, bel will combine with a zero suffix
when licensed by D (or by little n). This suffix is equipped with a neuter gender
feature; see the morpheme insertion rule in (50a). In an account that assumes the
presence of a nP between the DP-and the LP-layer (for instance, Kihm (2005)),
this null suffix may be inserted in little n, followed by merger of the root with little
n.22 In any case, the gender feature will be copied onto other positions within DP
before spell-out. Thirdly, another morpheme insertion rule will combine klag
. Alternatively, on might argue that bel is selected from List 1 along with an inherent
gender feature, no matter what licensing environment this root appears in. Since this feature
only plays a role when the root takes a position in which it is licensed by a determiner, it
may be deleted in other licensing environments. This deletion could be accounted for by a
Rethinking accommodation
with the abstract morpheme given in (50b) when licensed by a degree element.
After MS-operations have applied, the relevant part of the structure looks as is
illustrated in(51).
(50) Morpheme insertion at MS
a. Insert [-(n)]
x [v][+d]
b. Insert [-lich]
x [v][d]
(51)
DPNOM
LP1
D
DegP
[]
[]
L1
Deg
LP2
+ [-lich]
[]
+ [-]
[]
feature copy
This structure is the basis for Vocabulary insertion at PF. The relevant Vocabulary
items are listed in (52). The first one is the item that spells out the feature bundle
in D, that is, the feature [def] drawn from List 1, the copied gender feature, and
the case feature that percolated down from DP. The Vocabulary item that spells out
klag is given in (52b). As you can see, I assume that this Vocabulary item can be
inserted in all licensing environments. In a [+v] environment, however, it is only
licenced when the light verb head is specified for [+cause] because klag requires
an agentive argument in SpecvP. The abstract morpheme accompanying klag is
spelled out by the phonological form in (52c) (remember that capital X represents
an underspecified fricative; see Footnote 19 in Chapter 3). Finally, the gender feature under LP2 is realized by the suffix in (52d). It is important to point out that
this suffix is only inserted in the context of a [def] feature in D (in context of a
[+def] feature, the suffix /-/ will beinserted).
Grammar as Processor
/ain/
b. klag
c. [-lich]
d. [n] [nom]
/kla:g/
/-liX/
/-s/
[v][d][+cause]
[def]
Last but not least, the phonological readjustment rule in (53) will trigger umlaut
formation in the Vocabulary item /kla:g/ when licensed by a degree element (that
is, by[v][d]).
(53) Phonological readjustment at PF
[+back] [back] / X [v][d]
(where X = klag, tag (day), gefahr (danger), )
In other words: the error in (48) can be accounted for by a series of regular processes which apply blindly to the sequence that results from the root exchange.
No repair is involved because none of the elements that surface in a form different
from that of the intended utterance are present when the erroroccurs.
6.5.2 Error #2: Case assignment, morpheme insertion & feature copy
The second slip that I want to discuss in some detail is the one cited in (54). Again, we
are dealing with a root exchange, this time, however, without self-correction. In this
exchange, the roots folg (follow) and versuch (tempt, try) are affected and the
resulting utterance is fully grammatical. By means of exception, for this error, I give
a translation both for the intended and the erroneous utterance. This is important to
capture meaning differences that are not evident from the interlinear translation. In
the intended utterance, folg means follow but in its nominalized form in the error,
it will be interpreted as order or sequence. Also, in the intended utterance, versuch
combines with a nominalizing suffix which yields the meaning temptation, while in
the error, the verbal surface form will be interpreted as try. This difference will turn
out to be crucial because the corresponding German verbs assign differentcase.
(54) ich versuch-e die
Folg-e
I tempt-1.sg the.f.acc follow-nmlz(f)
folg-e
der
Versuch-ung
follow-1.sg the.f.dat tempt-nmlz(f)
intended: I follow the temptation.
error: I try the order/sequence.
The post-error structure in (55) indicates that I assume that folg and versuch
change place after movement of folg to Tns and C (and movement of the subject
Rethinking accommodation
to SpecCP). This assumption is not crucial for the analysis. Obviously, it might as
well be the case that the error occurs at deep structure when both roots are still
adjacent to each other under LP1. Subsequently, versuch would undergo head
movement to Tns andC.
(55)
CP
DP
[1st]
Tns
TnsP
L1
Tns
[]
tDP
Tns
LP1
DP
root exchange
tTns
tL1
LP2
[+]
Grammar as Processor
(57)
CP
DPNOM
[1st]
C
Tns
TnsP
L1
Tns
tDP
Tns
AgrS
[]
[1st]
feature copy
Tns
LP1
DPACC
tTns
tL1
LP1
[+]
[]
+ [-e]
[]
The Vocabulary items that spell out the terminal nodes in this structure are listed
in (58). The first three items are those that spell out features or feature bundles: the
subject pronoun, the agreement suffix, and the definite determiner. The Vocabulary items for the two roots that participate in the error are given in (58d) and
(58e). Both roots are licensed in all three possible environments. A difference,
however, concerns their insertion in a [+v] environment. While versuch is only
permissible in a [+cause] context, folg can also be inserted in a [cause] context (as, for instance, in wichtige Informationen folgen (important information follows)). Finally, the abstract morpheme will be spelled out by the Vocabulary item
in (58f). Phonological readjustment is not required in thiserror.
(58) Vocabulary items inserted at PF
a. [1st] [nom]
b. [1st]
c. [+def] [f]
[acc]
d. folg
e. versuch
f. [-e(f)]
/iX/
/-/
/di:/
/flg/
/f7zu:X/
/-/
/
/
[v][d][cause]
[v][d][+cause]
Rethinking accommodation
that exactly the same operations would apply in the intended utterance, which also
requires case assignment, feature copy, and morphemeinsertion.
6.6 Against repair strategies
The errors discussed so far in this chapter make an important contribution to our
understanding of the processing and manipulation of morphosyntactic features
in language production. In particular, the above analysis of two complex speech
errors illustrates that the interplay of various DM-mechanisms provides for an
elegant explanation of these and many other spontaneous errors. Table (59) gives
an overview of all the errors in my corpus that would traditionally be analyzed as
accommodations, or, to be more precise, as morphological and morphosyntactic
accommodations (see Section 6.1). In the table, the errors are grouped together
according to the respective mechanism which makes the assumption of a repair
process superfluous; these mechanisms are feature copy, feature stranding, and
local licensing (note that Table (2) in Chapter 1 gives a total of 241 errors involving accommodation because it also includes 15 instances of lexical construal; these
will only be discussed in Section6.6.2).
(59) Errors in which DM-mechanisms help us to accountfor
so-called accommodations (n =226)
feature copy following the error
feature copy withinDP
feature copy ontoAgrS
caseassignment
88
13
2
featurestranding
has impact on spell-out of featurebundle
has impact on spell-out ofroot
has impact on spell-out ofaffix
9
18
26
locallicensing
triggers morphemeinsertion
triggerssuppletion
triggers phonologicalreadjustment
34
11
25
Total
103
53
70
226
On the one hand, I have claimed that all context accommodations are either due
to feature copy (types , , and ), to context-sensitive spell-out of a morphosyntactic feature (type ), or to morpheme insertion due to licensing (type ).
The first of these mechanisms has an impact on the syntactic context of the error
Grammar as Processor
element, while the latter two impose a change on the morphological context (an
inflectional or derivational morpheme). On the other hand, I argued that error
accommodations can be seen as the result of either feature stranding (types and
), suppletion (type ), or phonological readjustment (type ), where the latter
two mechanisms are triggered by the licensing environment in which the error
elementappears.
I conclude from the discussion in the previous sections that the concept
accommodation is unnecessary and should therefore be abandoned. Once we
adopt DM-mechanisms, the assumption of costly repair processes that would
bring the erroneous utterance in line with some grammatical constraint becomes
superfluous. By claiming this, I go one step further than Leuninger & Keller
(1994:89) who argue that accommodations are cost-free adaptations of linguistic errors to grammatical [] well-formedness restrictions. My point is: even the
postulation of cost-free adaptations is unnecessary because no adaptation whatsoever is involved. None of the elements that surface in a form different from that of
the intended utterance for instance, agreement suffixes, derivational morphemes,
and determiners are present when the error occurs. After the error has taken
place, the processor will spell out blindly the abstract roots and features contained
in terminal nodes. The insertion of a different morpheme in a certain licensing
environment or the application of a phonological readjustment rule, to name just
two mechanisms, do no constitute an extra burden for the processor, that is, they do
not constitute repairs. Rather, these mechanisms simply reflect the normal workings of the morphological system when confronted with a slightly incorrectinput.
This being said, I want to consider two further aspects. First, I will reconsider
phonological and morphophonological accommodations (Section 6.6.1). I will
argue that these accommodation types do not involve the application of a repair
operation either. Second, I will present two error types that are problematic in
light of my claim that repair processes play no role during language production.
The first error type, lexical construal, I take to be a real challenge for that claim
(Section 6.6.2). In contrast, the second error type to be discussed only poses a
potential challenge; these are self-corrected errors that seem to involve the application of a surface filter (Section6.6.3).
6.6.1 Reconsidering (morpho)phonological accommodation
The discussion in Sections 6.2 to 6.5 centered around the analysis of errors in
which we observe morphological or morphosyntactic changes. But what about the
other two types of accommodation that were distinguished in Section 6.1, namely
phonological and morphophonological accommodations? As for the first type,
consider the three errors in(60).
Rethinking accommodation
The Dutch within-word sound exchange (Cohen 1965:183) that has already
been presented in (2c) is repeated here as (60a). In this slip, the phonemes /k/
and /p/ change place. Subsequently, the nasal /m/ assimilates to the place features of the adjacent /k/ and surfaces as the velar nasal []. If the phoneme /m/
was in fact fully specified for all phonological features, then we would have to
assume that one feature value is changed, that is, that we are dealing with a true
context accommodation. This, however, is most probably not the case. Within
underspecification theory (see, for instance, Archangeli 1988; Yu 1992; Steriade
1995), it is assumed that certain feature values may be underlyingly unspecified. The missing values will either be supplied by complement or default rules
or they will be filled in by processes of assimilation (feature spreading). For
Dutch and German, for instance, it has been argued that nasal consonants are
not specified for place of articulation; rather, they assimilate to a neighboring
obstruent with respect to the place feature. Consequently, in (60a), the underspecified nasal receives its place feature [velar] from the neighboring /k/ and
surfaces as [], while in the intended utterance, it would have received the feature [labial] from theadjacent /p/.23
In speech errors, feature spreading is not only observed from an error element onto the context but also from the context onto an (underspecified) error
element. In (60b), for instance, a nasal is perseverated. In its post-error position, the underspecified nasal assimilates to the place features of /p/ and surfaces
as[m].
(60) a.
b. Bauern-tmpel Bauern-tlpel
farmer-(error) farmer-bumpkin
country bumpkin
c. ich
I
hab-e
den Buch [bx], Butt
have-1.sg the (error),
Butt
ge-schrieb-en
part-write-part
. See Yu (1992:187ff) for an account of nasal assimilation in German. Note that a similar
process of place assimilation is observed in English, as is exemplified by the two prefixed
forms in-definite vs. im-perfect.
Grammar as Processor
. Obviously, it is the German author Gnter Grass who uttered this slip. The English translations of the two books he refers to are The flounder and The tin drum, respectively.
. Meara & Ellis (1981) discuss intriguing speech errors from Welsh which show that phonological errors may actually occur before and after the application of phonological rules. Just
like other Celtic languages, Welsh has a phonological rule known as initial consonant mutation, which changes word-initial consonants according to the words syntactic environment.
Three different types of mutation have to be distinguished. The authors show that phoneme
reversals are observed with mutated and unmutated consonants. In (i), the two word-initial
consonants /gw/ and /m/ have undergone mutation before they were reversed and no accommodation to their post-error position is observed (for this error, Meara & Ellis do not specify
the underlying (unmutated) form of the consonants).
(i)
(ii)
y mn wahaniethau
In contrast, unmutated consonants are exchanged in (ii). The underlying forms of the two
words involved in the error are mn (little) and gwahaniethau (differences). Nouns that
follow adjectives are subject to the rule of soft mutation. Therefore, the noun would surface as
wahaniethau in the intended utterance (loss of initial /g/). In the error, however,/gw/surfaces
in its unmutated form. At the same time, soft mutation correctly applies to the other error
element and changes it from /m/ to /v/ (Meara & Ellis 1981:800). That is, in (ii), we observe
phonological accommodation. Meara & Ellis take these errors as evidence for the psychological reality of deep and surface phonological representations.
Rethinking accommodation
suffix and the past tense suffix. In example (61a), the number feature is stranded
(Stemberger 1985: 162), while in (61b), the same is true for the tense feature
(Garrett 1976: 238). In both examples, the respective suffixes are spelled out in
their appropriate form atPF.
(61) a. you just count wheel-s [-z] on a light light-s [-s] on a wheel
b. he roast-ed [-id] a cook
he cook-ed [-t] a roast
c. [+past]
/-t/ / Z +___
(where Z = cook, dwell, buy )
There are two possibilities to account for the correct spell-out of these suffixes. On
the one hand, one might argue that there is only one underlying (tense or plural)
morpheme which is underspecified for a certain feature (for example, [voice]
for the plural morpheme), the specification of this feature depending on the context of insertion. Moreover, a default vowel insertion rule will apply in certain
contexts (for instance, in (61b)).26 On the other hand, it could be the case that
the Vocabulary contains several suffixes that compete for insertion under a given
node, each Vocabulary item specifiying the context in which it may be inserted.
At least for the feature [+past], the latter option is adopted by Halle & Marantz
(1993:125f), who postulate the Vocabulary item in (61c) for one of the English
past tense allomorphs. Independent of the position one adopts, morphophonological accommodation of the suffix need not be assumed. Rather, we are either
dealing with a phonological spreading rule that fills in a missing feature value
(option 1) or with competition among more highly specified Vocabulary items
(option 2). In any case, exactly the same operations would have been effective in
the intendedutterances.27
. I assume that phonological underspecification of suffixes also allows for a straightforward explanation of vowel harmony in Turkish speech errors, as observed, for instance, in
example (3b) in Section 6.1.2 (see Mester & Ito (1989) for an underspecification account of
Turkish vowel harmony).
. It is not entirely clear whether a similar strategy can help us in explaining the accommodation of definite and indefinite articles in English errors such as (i) (Garrett 1976:238).
(i)
That is, it is not clear whether there are two allomorphs for each article, the insertion of which
depends on the first segment of the following word, or whether we are dealing with a phonetic
phenomenon (comparable to, for instance, liaison in French).
Note that a similar phenomenon is observed in Italian, where the selection of definite
determiners is not only influenced by a nouns gender feature (as in German and French) but
for masculine determiners also by phonological characteristics of the word that follows the
determiner. That is, the gender feature determines only the allomorphic set of determiners
Grammar as Processor
(il and lo in the singular, i and gli in the plural), while the choice of one of the allomorphs over
the other depends on phonological factors (see Miozzo & Caramazza 1999).
Rethinking accommodation
im
Netz
in.the.n net(n)
Schinken-semmel
ham-roll
kein-e
Krmm-ung, kein-en
Finger
krumm
no-f.acc bent-nmlz(f) no-m.acc finger(m) bent
Grammar as Processor
pointed out by Leuninger & Keller (1994), this second step is of the type of a formal
substitution. In contrast to the different types of adaptations discussed in Sections 6.1
to 6.5, this second step is definitely not cost-free, that is, it does not involve the application of a mechanism that would apply in the course of the derivation anyway. I therefore conclude that lexical construal is in fact the result of a repair. Lexical construal
may not be a very frequent phenomenon (my corpus contains 15 errors of this type),
but it is frequent enough to consider it a challenge to my previous claim that language
production, when assisted by DM-mechanisms, can do without repairprocesses.
6.6.3 A possible surface filter
Albright (2007), in his insightful commentary on Pfau (2007), discusses further
examples that possibly contradict the assumption of an entirely repair-free derivation. In addition, he reconsiders the processing of grammatical gender in DM and
the strict division of semantics and phonology as assumed in DM. I will briefly
summarize his treatment of the latter two topics before turning my attention to
errors that suggest that the computational system does not operate entirely blindly
and that some surface filter might be at work afterall.
Albright argues that according to a conceptually appealing null hypothesis,
the syntax would only care about a limited set of universal syntactic and semantic features, while lexically arbitrary and language-particular morphological and
phonological features would be available only after Vocabulary insertion (Albright
2007: 38). Clearly, gender agreement is a challenge to this strong hypothesis. He
sketches an alternative scenario according to which the processor generates parallel structures, including masculine, feminine, and neuter versions of DP-internal
material. At PF, only the structure that turns out to be compatible with the (gendermarked) Vocabulary item will be spelled out. Hence, for a German sentence which
includes two roots that are licensed by D, nine (32) structures would have to be
generated clearly a somewhat awkward solution. Mechanically, this solution might
work but the speech error data (in particular, the identical gender effect and accommodation patterns) show that it cannot be on the right track. Rather, the error data
support the claim that gender features are available early in the derivation. With
respect to gender processing, Albright (2007:41) therefore concludesthat
Pfaus findings concerning the asymmetry in gender accommodation go beyond
simply confirming a prediction of Distributed Morphology; they actually
disconfirm a strong version of the theory, and inform our understanding of
genderagreement.
Rethinking accommodation
a matter of on-going debate (also see Section 3.1.3) and I will not go into this issue
here. However, I want to have a brief look at errors that appear to speak for the
early availability of phonological information. In this context, Albright (2007:43f)
discusses the error which has been cited as (31a) in Section 6.4.3, repeated here as
(63a) for convenience. Remember that I argued that in this error, the surface form
of the perseverated root schreib (write) is motivated by the fact that it is licensed
by a determiner in its post-error position. In this environment, the suppletive
form /rift/ will be retrieved from the Vocabulary. Furthermore, I argued that the
nominalization Schrift is the best syntactic/semantic match for the intended stative
DP. In Footnote 15, however, I already admitted that the phonological similarity of
intended and produced noun might play a facilitating role in this error: both are
monosyllabic, share the onset and the vowel, and have similar syllablestructure.
But how could the spell-out of schreib be influenced by phonological
similarity with the Vocabulary item that spells out strich (line)? Albright
rightly points out that the DM architecture does not provide any obvious way
for the phonological form of Strich to play a role in the error. According to DM,
phonological forms are only available at the time of Vocabulary insertion, but
at this point, schreib has already taken the slot of strich and there is absolutely no reason for the Vocabulary item /tri/ to be considered when Vocabulary insertion takes place. After all, we dont want to assume that the processor
remembers that strich was once part of the derived structure. DM therefore predicts that substitutions that occur early in the derivation should not be
encouraged additionally by phonological similarity (Albright 2007:44). Hence,
in this theory, the phonological similarity between intruder and intended item
can only be anaccident.
(63) a.
mit Binde-strich
with connect-line
b. [die Motivation
der
Athlet-en] knn-en durchaus variier-en
the.f motivation(f) of.the.pl athlete-pl can-3.pl certainly vary-inf
die
Motivation kann variier-en
the.f motivation(f) can.3.sg vary-inf
Another relevant observation concerns the fact that the phonological form of agreeing determiners appears to have an impact on speech errors that presumably occur
early in the derivation. In an experimental study, Vigliocco et al. (2004) showed that
in elicited German noun substitution errors, gender preservation is observed only
Grammar as Processor
when a change in gender would also require a change in determiner.29 This implies
that the likelihood of a semantic substitution is influenced by the eventual surface
form of the determiner. Given that semantic substitutions take place early in the derivation (selection of roots from List 1), while the phonological form of the determiner
should only be available late (at PF), this is clearly a problematicobservation.30
Remember from the discussion in Section 5.2.2.6 that a similar property has
been found to facilitate the occurrence of local SVA-errors. In my corpus, SVAerrors in which the verb agrees with a local noun from within a complex subject
DP are most likely to occur when the singular head noun is accompanied by the
feminine definite determiner die, which is homophonous to the plural determiner
(see Table (51)); a representative example is given in (63b). Again, we are confronted with the paradoxical situation that the phonological form of an error element should have an influence on an error that takes place before spell-out. Clearly,
these data are a challenge for the theory and I agree with Albright (2007:46) who
states that further inquiry is needed to determine to what extent the data is compatible with a modular, unidirectional model such as DistributedMorphology.
Most important in the present context is the third observation that Albright
(2007) makes in his commentary because it touches on the issue of repairs in language
production. Under the header Is free too cheap? he discusses my claim that additional repair operations are not required once we are equipped with the tools made
available by DM. He suggests that this hypothesis may actually make repairs too easy,
and that there may exist additional constraints that limit errors from making use of
the full computational power of the morphological system (Albright2007:46).
. In the relevant test condition, subjects had to produce nouns accompanied by indefinite
determiners in a speeded picture naming task. In German, the indefinite determiner is homophonous for masculine and neuter nouns (ein), while feminine nouns require a different
determiner (eine). The results indicate that masculine nouns are freely substituted for neuter
nouns (and vice versa). In contrast, the substitution of feminine nouns for either masculine
or neuter nouns was avoided.
. Albright (2007:45) argues that this finding raises the possibility that the identical gender
effect is actually an identical determiner effect. He points out that a similar effect may possibly be lurking in Pfaus own data. When discussing gender accommodation in Section 5.1.3,
I showed that the Frankfurt corpus contains a total of 91 singular noun substitutions in which
target and intruding noun do not share the same gender feature. In these 91 cases, accommodation could in principle be observed. However, 57 of the noun substitutions that violate
the identical gender effect (62.6%) actually satisfy determiner-matching, either because the
accompanying determiner is gender-ambiguous or because there is no determiner (see the
examples in (12) in Section 5.1.3). This distribution implies at least the possibility that even
semantic substitutions are sensitive to the surface morphological/phonological structure of
the agreement context (Albright 2007:45).
Rethinking accommodation
Albright reconsiders the English slip in (42a): I think its care-ful to measure
with reason (Fromkin 1973a: 31). I argued above that in this error, reason and
care change place and that following the error, a morpheme insertion rule will
insert the abstract morpheme [-ful] in a context where care is licensed by a
degree element. Albright wonders what would have happened if the exchange
had involved a root that does not happen to have an existing degree adjective,
for instance, gusto. What would the processor make of the resulting structure?
Would gusto be spelled out as gustoful given that -ful appears to be the productive affix for creating degree adjectives? If, as I claim, so-called repairs simply reflect the ordinary workings of the computational system, then this leads to
the prediction that errors, too, should be free to productively create neologisms
(Albright2007:47).
However, at least in my corpus, this prediction is not borne out. In all errors in
which a root appears in a different licensing environment (giving rise to morpheme
insertion and/or phonological readjustment), an existing word surfaces. Moreover,
as pointed out by Albright (2007), there is evidence from self-corrections which
suggests that errors generally tend not to result in productively formed neologisms. For the sake of illustration, consider the error in (64a). Given that the error
is self-corrected after Tnzer (dancer), we are not in a position to decide whether
we are dealing with an anticipation or an incomplete exchange. Note, however,
that in the latter case, linguist would target a position in which it is licensed by
a light verb. While tanz (dance) is readily spelled out in a D-environment, there
is no Vocabulary item that could spell out linguist in a v-environment. Only a
neologism could do the job (Albright suggests the forms linguistizieren and linguistieren). Albright (2007:47) conjectures that it is precisely the lack of a suitable
existing verb that lead the speaker to an immediateself-correction.
(64) a. dass ein Tnz-er,
h, ein Linguist
so wild tanz-t
that a.m dance-nmlz(m), er, a.m linguist(m) so wildly dance-3.sg
that a linguist dances so wildly.
b. die
Sitz-e, h, die
Kind-er hab-en im
Kreis
the.pl sit-pl, er, the.pl child-pl have-pl in.the.m circle(m)
ge-sess-en
part-sit-part
c.
er hat
eine
he have.3.sg a.f
Konstrukt-ion-en
diskutier-t
construct-nmlz-pl discuss-part
Grammar as Processor
It is interesting to note that all the errors from my corpus in which a full exchange
of roots would have required the spell-out of a non-existing (nonce) form are selfcorrected immediately after the first error element. A second example is provided
in (64b). For the sake of the argument, lets assume that kind (child) and sitz
(sit) were exchanged. The former root is spelled out according to its licensing
environment (note that this involves the choice of a different plural allomorph).
For kind, however, it is not clear how it could be spelled out when being combined with [+part] and licensed by a light verb (gekindert?). Again, following
Albrights suggestion, we may hypothesize that the lack of a suitable Vocabulary
item is responsible for the self-correction (see (4b), (5b), (23a), and (26a) for further examples).31 It has to be pointed out, however, that self-correction in root
exchanges is not necessarily an indicator of the lack of an appropriate Vocabulary item. The slip in (64c), for instance, is self-corrected after the first error element (note the stem-internal change in a D-environment) despite the fact that
konstruier (construct) is licensed in a light verb environment; the resulting
utterance would have been Diskussionen konstruiert (see (21b), (31b), and (48) for
addititionalexamples).
Still, if Albrights argumentation is on the right track, the self-correction data
suggest the existence of a strong lexical filter which blocks morphological structures that do not correspond to pre-existing Vocabulary items or, to put it differently, structures that require the insertion of Vocabulary items in an environment
in which they are not licensed (remember that Vocabulary items are specified for
legitimate licensing environments). This in turn implies that the system does not
operate entirely blindly. Rather, it anticipates a licensing conflict and stops the
derivation. Albright (2007:47) tentatively concludes that a surface filter on preexisting combinations seems to lead to a very common and powerful repair, of
stopping and correcting theutterance.
Albrights proposal is highly thought-provoking and certainly deserves further in-depth study. Still, it seems to me that it is not absolutely necessary to postulate a surface filter in order to account for the observed self-corrections. Why not
simply assume that the derivation crashes at the point of Vocabulary insertion?
. See Ferreira & Humphreys (2001) and Wardlow Lane, Slevc, and Ferreira (2007) for experimental evidence that points in a similar direction. In particular, experimental findings
reported in Ferreira & Humphreys (2001) indicate that words that belong to more than one
syntactic category (verb and noun in their case; e.g., tape and record) are more likely to participate in an experimentally elicited exchange. It has to be noted, however, that the authors
do not consider morphologically complex nominalizations. That is, a word like decorate is said
to belong to only one syntactic category because decorate cannot appear in a nominal slot and
hence, an exchange like caked the decorate is ill-formed.
Rethinking accommodation
In (64a), for instance, after having spelled out tanz in a D-environment, the
processor would proceed to spell out linguist in a light verb environment. The
processor will find a Vocabulary item that matches the root; however, this item
does not match the roots licensing environment (presumably, the Vocabulary item
is specified for insertion in a [v][d] context). Hence, the derivation crashes and
the speaker will start anew. Obviously, an explanation along these lines raises the
question of how spell-out proceeds. Above, I argued that, within an XP, spell-out
proceeds from the root outward (Bobaljik 2000). In order to make my suggestion
work, however, one would also have to assume a certain amount of left-to-right
spell-out. Otherwise, it is difficult to explain why tanz is spelled out in the first
place given that Vocabulary insertion for linguist already failed (note that the
same caveat holds for Albrights account). I leave this fascinating and complex
issue for futureresearch.
6.6.4 Summary
In this section, I have summarized my findings concerning the speech error
data presented in Sections 6.1 to 6.5. The discussion has culminated in the
claim that language production can do without repair strategies. On the one
hand, I have argued that apparent morphosyntactic accommodations are due
to feature stranding and feature copy. On the other hand, I have shown that
morphological accommodations receive a straightforward explanation when
we assume local licensing of acategorial roots and context-sensitive spell-out
of features. I have added to the picture phonological and morphophonological
accommodations which I take to be the result of underspecification, feature
spreading, and possibly a competition amongst alternative Vocabulary items.
Crucially, all of the mechanisms needed to account for the errors are mechanisms that apply in the course of the derivation anyway. For the processor, it
does not make any difference (in terms of a processing effort) what feature is
copied or what morpheme is inserted. Hence, it is the application of regular
mechanisms that yields a fully grammatical surface form, not the application of
exceptional repairstrategies.
I have also discussed error data that are potentially challenging to my general
claim. Admittedly, for cases of lexical construal, it is difficult to come up with an
explanation that does not involve the application of a repair strategy. Indeed, these
data suggest that the processor accesses the Vocabulary again after spell-out in
order to retrieve an existing word. Further examples that proof problematic for
a DM account are those in which phonological information (that is, information
that is only available at PF) seems to have an influence on errors that take place
before spell-out (for instance, semantic substitutions and SVA-errors). Finally,
Grammar as Processor
Rethinking accommodation
b. im
Wolken nord-ig
im
Norden wolk-ig
in.the.m.dat cloud(f) north-adj in.the.m.dat north(m) cloud-adj
cloudy in the North
In (65b), the stranded element -en happens to be the plural suffix of Wolke (cloud),
but this is a mere coincidence, since no plural feature is present in the intended
utterance.32 The stranded element -ig, on the other hand, is an adjectival suffix;
still, it is certainly not the appropriate one for nord (North). Possible adjectival forms for nord are nrdlich (Northern) and nordisch (Nordic). Moreover,
there is a feature mismatch between the cliticized article and the noun. All these
characteristics clearly indicate that the error must have occurred after Vocabulary
insertion, that is, that phonological material has been exchanged at PF. If two roots
had been exchanged before Vocabulary insertion, then the interplay of morpheme
insertion and feature copy would have yielded another surfaceform.
6.7.1.2 Feature mismatch within DP
While a mismatch between morphological form and sentence position (that is,
licensing environment) is the hallmark of stranding errors, the other two types
. Note that the suffix -en which accompanies the Vocabulary item that spells out nord
(North) is possibly a nominalizing suffix.
Grammar as Processor
ast
h
du
Karte-n frs
Filiale,
h,
have.2.sg you(sg) ticket-pl for.the.n branch(f), er,
frs
Finale
for.the.n final(n)
leicht-es
Kopfweh
// leicht-e
Kopfschmerz-en
slight-n.acc headache(n) // slight-pl.acc headpain(m)-pl
Blends have by far the highest share in DP-internal agreement errors. There are
48 blends in my corpus that give rise to a feature conflict. The error in (66b) is
interesting because in this particular case, the mismatch is due to two features:
gender and number. In German, the two compounds Kopfweh (headache) and
Kopfschmerzen (literally headpains) are synonymous; the former is always singular, while the latter is usually used in its plural form. Moreover, the heads of these
two compounds are of different gender: -weh is neuter, -schmerz is masculine. In
Rethinking accommodation
the erroneous utterance, the adjective leicht (slight) is marked for singular and
neuter, while the noun it combines with is masculine and marked forplural.
Except for one case, in all blends that result in a DP-internal feature mismatch, there are two nouns in competition, while other material within DP is
identical for both alternative structures (for instance, the adjective in (66b)).33
We may therefore assume that two roots within List 1 receive activation from
the conceptual level and that both roots enter the computation competing for
the same l-node within the syntactic DP-frame. At MS, the gender feature of one
of the competitors will be copied but at PF, the other root will be spelled out.
That is, the error proceeds in two steps. Crucially, just as in phonological substitutions, the second error takes place at spell-out, that is, after feature copy has
beenexecuted.34
It is important to point out that things may be different in blends in which
two different phrases compete (and not just two synonymous roots with different
gender features). For these blends, we must assume that two alternative frames are
constructed in parallel and that a root from one frame intrudes into the competing frame. Consider, for instance, the blend in (67), which involves two idiomatic
expressions. Clearly, in this error, fahne (flag) takes the position of mund
(mouth) in the competing frame. This substitution process (Wiegand 1996)
takes place within the computational system. Subsequently, the gender feature of
fahne is copied onto the determiner and a grammatically well-formed utterance surfaces (also see the blend in (8b), which involves two idiomatic expressions, and the blend in Footnote 5, in which the two competing DPs contain
differentadjectives).
. The one exception is given in (i). In this error, two prepositional phrases are in competition. The two PPs contain different prepositions both of which, however, assign dative case.
In the error, the masculine noun Mund (mouth) combines with the feminine (dative) determiner of the competing frame.
(i)
das muss
man sich auf der
Mund
zergeh-en lass-en
that must.3.sg one refl on the.f.dat mouth(m) melt-inf let-inf
auf der
Zunge
// im
Mund
on the.f.dat tongue(f) // in.the.m.dat mouth(m)
. Alternatively, one might argue that two structurally identical frames are activated in parallel. At MS, gender agreement is established successfully within both frames but at spell-out,
the two frames blend into one thereby combining the noun of one frame with the gender
marked adjective of the competing frame (see Section 5.2.2.4).
Grammar as Processor
No matter whether one adopts a competing frame analysis for both types of blends
or not, it is clear that the intended element in (67) is substituted before MS, while
in (66b) a decision between the two competitors is made only after MS. Therefore,
only in the first case, a grammatical outcome isguaranteed.
The third type of error that may give rise to a feature mismatch within DP
are noun displacements. My corpus contains twelve errors of this type, two noun
anticipations and ten noun exchanges. Above, we have already seen that usually,
following a root displacement, features of the affected root are properly copied
onto other material within DP; see, for instance, the examples in (6) and (8a).
In the exchange in (68a), however, gender copy is not observed. Note that a feature clash is only visible for the first error element Mund (mouth) because in the
dative, the definite determiner (which cliticizes to the preposition) happens to be
the same for masculine and neuter gender. The lack of correct feature copy in this
error indicates that it takes place after spell-out. That is, we are dealing with a
word exchange and not with a root exchange (compare examples (36) and (37) in
Section6.4.4.1).35
. Given that the two error elements are neither affected by morpheme insertion nor phonological readjustment, another scenario could be suggested. According to this alternative
scenario, the error takes place after feature copy but before Vocabulary insertion. That is, first
gender features are copied, then the roots are exchanged (still at MS), and then Vocabulary
items are inserted. I dont see a way to decide between the two possible explanations. The same
holds for the French noun anticipation in (i). Here, the feminine noun ligne (line) takes the
place of the masculine noun livre (book) but still, the determiner surfaces in its masculine
form (Rossi & Defare 1995:28).
(i)
cest
le
mme ligne, le
mme livre
that.is the.m same line(f), the.m same book(m)
la
ligne
prs
to the.f line(f) exact
Rethinking accommodation
b. Nagel-studio fr den
Dame, h, fr die
Dame
nail-studio for the.m.acc lady(f), er, for the.f.acc lady(f)
und den
Herrn
and the.m.acc gentleman(m)
Finally, a gender mismatch within DP can also result from the displacement
of a determiner, as is illustrated in (68b), where the feminine noun Dame
(lady) combines with the masculine accusative determiner which appears to
be anticipated from the second DP. The self-correction suggests that we are
in fact dealing with a determiner anticipation and not with a noun anticipation. Again, the error must have taken place after feature copy. I assume
that it occurs after spell-out but in principle, nothing excludes the possibility
that a feature bundle has been anticipated at MS, following feature copy and
caseassignment.
6.7.1.3 Subcategorization errors
A different type of feature conflict is observed in subcategorization errors, the last
error type that I wish to discuss in this section. Agreement conflicts are not at
issue here; rather, subcategorization errors result from a conflict between a caseassigning element (mostly, a verb) and a case-marked argument. Verbs, for
instance, have specific lexically determined subcategorization frames, that is, they
require certain case-marked arguments. The selection of a wrongly case-marked
argument or the lack of an argument leads to ungrammaticality and this is what
we find in the subcategorization errors. 46 errors in my corpus belong to this
group; 43 of these are the result of a blend. In most of the blends, two verbs which
require differently case-marked arguments are in competition. Two examples are
given in (69). In (69a), the competing verbs are widersprechen (contradict) and
verletzen (violate); the first one assigns dative case, the second one accusative
case. In the blend, the accusative-assigning verb appears with the dative argument
(as is evident from the suffix on the adjective). Things are somewhat different in
(69b). In this error, the verb ahnen (suspect) combines with the reflexive pronoun
that is required by the competing verb denken (think). Hence, the error is not due
to the presence of a wrongly case-marked argument but rather to the presence of
a superflouosargument.
Grammar as Processor
(69) a.
eine Sprache,
die
bestimmt-en
UG-Prinzip-ien verletz-t
a.f language(f) rel.f certain-pl.dat UG-principle-pl violate-3.sg
die
bestimmt-en
UG-Prinzip-ien widersprich-t
//
rel.f certain-pl.dat UG-principle-pl contradict-3.sg //
die
bestimmt-e
UG-Prinzip-ien verletz-t
rel.f certain-pl.acc UG-principle-pl violate-3.sg
b. ich habs
mir schon ge-ahn-t
ich habs
I have.1.sgit refl already part-suspect-part I have.1.sgit
mir ge-dach-t
// ich habs
ge-ahn-t
refl part-think-part // I
have.1.sgit part-suspect-part
I already thought so//I already suspected it.
The explanation for these two errors is quite similar to the one given above for
the DP-internal mismatches resulting from blends (see (66b)). That is, a decision
between two competing roots that entered the computation is only made after case
has been assigned at MS. For the cases with a wrongly case-marked argument,
we may assume that the roots compete for one and the same position within one
syntactic frame (a position that is licensed by a light verb), while in the errors that
contain a superfluous argument, it seems more likely that two frames are constructed which blend into one beforespell-out.
6.7.1.4 Summary
I conclude that all the errors discussed in this section have in common that they
occur too late in the derivation for the resulting morphological or morphosyntactic conflict to be repaired. They either occur at MS after features have been copied
and assigned, at PF when Vocabulary items are inserted, or after spell-out. Table
(70) gives an overview of the errors from my corpus that result in a morphosyntactic (gender or case) mismatch. Remember that my corpus does not contain morpheme stranding errors (the errors in (65) are taken from the Frankfurt corpus).
Therefore, stranding errors are not included in Table(70).
(70) Errors that result in a morphosyntactic mismatch (n =131)
feature mismatch within DP
due to ablend
due to a nounsubstitution
due to a nounexchange/anticipation
due to determinerexchange/anticip./persev.
48
13
12
12
subcategorizationerror
due to ablend
due to noun/determineranticip./perseveration
43
3
Total
85
46
131
Rethinking accommodation
Before concluding this section, I want to say a few more words about DP-internal
agreement errors that result from a noun displacement. Berg (1987) states that in
his corpus of German speech errors, non-accommodation of determiners is the
rule in such errors. In 1987, his corpus contained 36 noun errors in which adjustment of the determiner to the new noun would be required; these comprise 31
(incomplete) noun anticipations, four perseverations, and one exchange. Interestingly, in only five of these cases, accommodation of the determiner is observed.
In contrast, in my corpus, non-accommodation of a determiner after a noun (or
root) shift is the exception. As can be seen in Table (70), there are twelve noun
exchanges/anticipations that result in a feature mismatch. However, my corpus
also contains 43 noun displacements (exchange, anticipation, or perseveration) in
which the determiner matches the gender/number feature of the noun it combines
with in the error (these 43 cases are included in group in Table (59)). Note that
I exclude blends and noun substitutions from the count in order to allow for a
comparison with the distribution in Bergs corpus. In other words: in Bergs corpus, the determiner agrees with the new noun in 5 out of 36 noun errors (13.9%),
while in my corpus, correct DP-internal agreement is established in 43 out of 55
noun errors(78.2%).36
Given the analysis advocated above, we have to assume that the majority of
noun displacements in Bergs corpus takes place after spell-out, that is, that phonological words are anticipated, perseverated, or exchanged. In contrast, most
of the comparable errors from my corpus are root exchanges that occur before
feature copy at MS. I have no explanation for this striking difference in distribution. It is noteworthy, however, that all of the 36 errors that Berg (1987) cites are
self-corrected. Of these, eight are corrected within the word and 24 immediately
after the error element. Especially for some of the cases that involve word-internal self-correction, it seems quite likely that they are indeed phonological errors.
Example (71a), for instance, is clearly a phonological error because umlaut in
the anticipated element /m/ is triggered by a phonological readjustment rule
(Berg 1987: 297). This rule affects the Vocabulary item that spells out mann
(man) in the context of [+pl]. Given the self-correction, we cannot exclude the
possibility that the speaker was about to articulate in der Mtung. Consequently,
the error takes place too late to allow for gender accommodation (in the sense of
featurecopy).
. The distribution is different for blends that involve competing nouns of different gender
and/or number. In a total of 60 blends, we observe DP-internal feature mismatch in 48 cases
(see, for instance, (66b)). Thus, in only twelve blends (20%), the determiner (or adjective)
agrees with the noun (see (67) for an example).
Grammar as Processor
(71) a.
hast-e
ge-les-en,
was in der
M-,
have.2.sg-you part-read-part what in the.m.dat (error),
in der
Zeitung
ber die
Mnn-er steh-t
in the.m.dat newspaper(m) about the.pl man-pl stand-3.sg
Did you read what was written in the newspaper about men?
b. dass die
Text-, die
Schler den
Text
that the.pl text(m), the.pl pupil.pl the.m.acc text(m)
ge-kann-t
hab-en
part-know-part have-pl
In other errors, it is not clear whether they involve feature mismatch at all. Consider, for instance, (71b). Actually, here we have two options to account for the
apparent gender mismatch. According to the first scenario, text (text) is anticipated, while the plural feature is stranded (see Section 5.4.1). In combination with
[+pl], text should be spelled out as Texte but self-correction stops the articulation during the noun. In this case, there is no feature mismatch because die
Texte would have been a grammatically well-formed DP. The alternative scenario
involves morpheme insertion. Note that Schler is morphologically complex; it is
a combination of schule (school) with the agentive suffix -er. The derived form
takes a zero plural suffix. Consequently, given agentive semantics, the anticipated
text will combine with the abstract morpheme [-er(m)] at MS. Due to the presence of a plural feature, the DP would be spelled out as die Texter (the text/story
writers) but self-correction stops the articulation just before the derivational suffix. As in the first scenario, no feature mismatch is involved in theerror.
Still, even given a possible reanalysis of some of Bergs data, it has to be
acknowledged that in his corpus, feature mismatch (failure to accommodate in
his terms) appears to be significantly more common in errors involving the displacement of nouns than in mycorpus.
6.7.2 Partial repair
Basically, I have argued that there are two options for speech errors. Either they
occur before DM-operations such as feature copy and morpheme insertion apply,
thereby resulting in a grammatical string, or they take place at or after spell-out
and give rise to an ungrammatical utterance. I shall now turn to a phenomenon
that suggests that grammaticality is not all or nothing: partial repair. For the
sake of illustration, consider the error in (72a). In this error, two roots that are
licensed by a determiner, bel (bad) and wurzel (root), are exchanged.
What makes this error problematic is the fact that following the error, gender
copy is observed in only one position. While the first DP resulting from the error
Rethinking accommodation
das
bel an der
Wurzel pack-en
the.n.acc evil(n) at the.f.dat root(f) grab-inf
One has to tackle the root of the problem.
b. da
bring-t
er dem
Hasen
den
Kind
there bring-3.sg he the.m/n.dat rabbit(m) the.m.acc child(n)
dem
Kind
den
Hasen
the.n.dat child(n) the.m.acc rabbit(m)
The explanation of (72a) is a challenge for every theory. Fortunately, this error is
one of a kind and I therefore consider it exceptional. At first sight, the slip in (72b)
appears to have similar properties. Here, too, two roots are exchanged kind
(child) and hase (rabbit) and a feature mismatch is only attested in the second DP (the fully grammatical sequence would be dem Hasen das Kind). In this
error, however, the apparent partial repair is most probably due to the fact that
in the dative, the definite determiner is homophonous for masculine and neuter
nouns. Thus, the fact that the first DP sounds grammatical is a coincidence; featurewise there may still be amismatch.
The second instantiation of partial repair I want to briefly discuss is hypothetical in nature.37 The error in (73a) is an instance of morpheme stranding of the type
. I owe thanks to Heidi Harley for drawing my attention to this possibility and for pointing
out potential theoretical implications.
Grammar as Processor
discussed in Sections 6.4.4.2 and 6.7.7.1. The exchange of the morphemes /tra:f/
and /ly:g/ happens after spell out and consequently, the utterance resulting from
the error is not well-formed. The stranded plural suffix happens to be the appropriate one for Strafe (punishment) but the participial form that surfaces is incorrect.
If two roots had been exchanged before spell-out than the expected grammatical
outcome would be the one given in (73b). Note that lg (lie) requires a different participial allomorph and moreover, the Vocabulary item that spells out lg
will be subject to phonological readjustment in a [+part] context. That is, two
mechanisms are required to yield a grammatical (fully repaired) output. Now
consider the error in (73c). In this hypothetical case, lg does not combine with
the appropriate participial allomorph but the corresponding Vocabulary item does
undergo readjustment, that is, we observe partial repair. The question we need to
ask is whether the system predicts a case like (73c) to bepossible.38
(73) a. der
Mann hat
mich
the.m man(m) have.3.sg me
Lg-en ge-straf-t
lie-pl part-punish-part
Straf-en ge-lg-t
punish-pl part-lie-part
. Here I only consider partial repair for an example involving a participial form. Exactly
the same argument could be made for an error that includes a noun which requires phonological readjustment (umlaut) in the context of [+pl]. In this case, partial repair would imply
that this noun correctly undergoes readjustment in its post-error position but does not show
up with the appropriate plural allomorph.
Also note that there is yet another conceivable partial repair scenario for the slip in (73a),
namely one in which the root shows up with the appropriate participial allomorph but phonological readjustment is not observed; see (i).
(i)
Lg-en ge-straf-t
I assume that there is no way for such a structure to be derived (unless we assume that the
required phonological readjustment simply doesnt take place) but I leave it to the reader to
do the detectives work necessary to verify this claim.
Rethinking accommodation
point of Vocabulary insertion, the roots will be spelled out first and subsequently,
the appropriate plural and participial allomorph will be inserted (the result being
Lg-en ge-straf-t). Then the phonological forms of the stems are exchanged,
and finally, the Vocabulary item /ly:g/ undergoes phonological readjustment, as
required in a [+part] context. This sequence of operations is illustrated in(74).
(74) Output of syntax:
Spell-out of roots:
Spell-out of features:
Stem exchange:
Readjustment:
{lg; [+pl]};
{/ly:g/; [+pl]};
/ly:g-n/;
/tra:f-n/;
/tra:f-n/;
{straf; [+part]}
{/tra:f/; [+part]}
/g-tra:f-t/
/g-ly:g-t/
/g-lo:g-t/
The fact that my corpus does not contain a single error that shows this type of
partial repair requires an explanation. One possibility would be to claim that phonological readjustment is tightly linked to the spell-out process and that errors
cannot occur between the spell-out of roots and readjustment. In other words:
displacement errors can only take place before the spell-out process sets off or after
phonological readjustment has applied (phonological substitutions, of course, take
place at the point of spell-out when Vocabulary items are selected from List1).
Alternatively, the absence of partial repair might be taken as evidence for a stemallomorphy account (as advocated, for instance, in Siddiqi (2006); see Section 6.4.6).
According to this line of reasoning, the surface form of lg in the context of
[+part] is not the result of a phonological readjustment rule but rather results from
the insertion of a different (more specified) Vocabulary item. Under such an analysis, the partial repair sequence in (73c) could never surface. In order to retrieve the
stem allomorph /lo:g/ from the Vocabulary, the exchange would have to take place
prior to Vocabulary insertion (so that lg appears in a [+part] context). However,
in this case, the processor would automatically supply the contextually appropriate
participial allomorph. As a consequence, the repair would be full, notpartial.
More research is required to further evaluate these two options (which, by the
way, are not mutually exclusive). But even this brief discussion highlights ways in
which the investigation of spontaneous errors can add to our understanding of
regular derivational processes in a formal model like DM. Once again, the study of
certain error patterns, and the absence thereof, may help us in uncovering details
of the workings of the system that are difficult to infer from flawlessproductions.
6.7.3 Summary
The errors discussed in this section contrast sharply with those presented in Sections 6.1 to 6.6. Obviously, despite the powerful DM-tools argued for previously, a
grammatical outcome is not always guaranteed. I have argued that all errors that
violate some grammatical well-formedness constraint must take place after MS,
Grammar as Processor
either during or after spell-out. That is, these errors occur too late for adaptation
to take place. I have shown that such adaptation failures may concern proper spellout of stems and affixes, feature copy within DP, and case assignment. Conversely,
whenever an error results in an utterance that is morphologically and morphosyntactically well-formed, this error must have taken place beforespell-out.
Furthermore, I have investigated the possibility of partial repair, that is, errors
in which one element be it a determiner or a stem changes its form as required
in its post-error environment, while another element does not. Generally, partial
repair seems to be highly exceptional. In particular, the fact that my corpus does
not include cases in which only the displaced stem adapts but not the affix may
shed light on the workings of certain components of the grammarmodel.
6.8 Conclusion
The existence of accommodations, that is, of errors that appear to involve the application of some post-error repair strategy, has intrigued psycholinguists for more
than thirty years. Still, in this chapter, I have advocated the claim that the concept
accommodation, while being a convenient descriptive label, does not have any
theoretical significance. Once we are equipped with the theoretical tools made
available by Distributed Morphology, apparent repair processes can be attributed
to the blind application of various DM-mechanisms. Hence, the postulation of
repair strategies in language production becomessuperfluous.
Let me briefly summarize the account I have given for the different types of
accommodations. First, in the framework I adopt, morphological error accommodations are the result of suppletion and phonological readjustment in certain
licensing environments, while morphological context accommodations follow
from morpheme insertion (due to local licensing) and context-sensitive spell out
of features. Second, morphosyntactic error accommodations are either due to
feature stranding or phonological readjustment, while morphosyntactic context
accommodations always result from feature copy. Finally, phonological and morphophonological error and context accommodations receive a straightforward
explanation when we assume phonological underspecification (of segments and
affixes) in combination with assimilation and feature spreading processes. The distribution of errors across the different processes I argued for is repeated in the top
part of Table(75).39
. For the sake of completeness, let me point out that, according to a traditional accommodation analysis, the 226 apparent accommodations in Table (75) would be classified as follows:
Rethinking accommodation
226
103
53
70
15
241
In Table (75), I add to the picture the 15 cases that involve lexical construal. I have
argued that only these cases can be considered true accommodations because
they do indeed involve a repair following the actual error. Following a phonological error that results in a non-word, the Vocabulary is accessed again to
retrieve a phonologically similar Vocabulary item. Further research will have to
show whether an additional surface filter can be held responsible for certain selfcorrectionpatterns.
The discussion has demonstrated that a formal framework like DM can serve
as a valuable tool in making predictions about possible and impossible error types.
Conversely, however, a systematic study of spontaneous speech errors can also
usefully inform our understanding of how the grammatical systemworks.
chapter 7
Conclusion
After discussing various types of spontaneous speech errors in the context of the
Distributed Morphology framework, it is now time to come back to the multilevel models of language production introduced in Section 3.1. In the previous chapters, we have seen how speech errors which involve the manipulation
of semantic and morphosyntactic features can be accounted for within DM. In
some errors, the erroneous manipulation of a feature gives rise to a speech error
(for instance, an SVA-error). In other slips, the actual error concerns a root and
regular feature manipulation following the error, often in combination with other
mechanisms such as local licensing and phonological readjustment, can be held
responsible for the resulting surface form. Clearly, morphosyntactic (and semantic) features play a prominent role in flawless as well as in erroneous production. In contrast to that, I have argued that reference to lexical categories is not
required in order to account for the error patterns. Quite to the contrary, a number of intricate speech errors, which appear to involve the application of a repair
strategy, receive a straightforward account once we assume the manipulation of
acategorial roots, as is also assumed in DM. Roughly speaking, whenever an error
occurs within the computational system, that is, before or at the level of MS, MSand PF-operations guarantee that a grammatical utterance surfaces. Errors that
occur after MS-operations have applied, however, may result in an ungrammatical utterance since at this stage of the derivation, it is simply too late for adaptation to takeplace.
What remains to be done is to investigate how the morphosyntactic model
and the psycholinguistic model can be mapped onto each other. In Section 3.3,
I already pointed out that both models advocate a similar allocation of tasks. In
the production model, grammatical encoding (functional level) precedes phonological encoding (positional level). In the grammar model, the mechanisms
that derive a syntactically complex expression are taken to be strictly separate
from the mechanisms that supply the corresponding phonological expressions.
In Section 7.1, I will sketch how the DM-model can be mapped onto the production model. Basically, I claim that the DM-model takes the place of the formulator in Levelts (1989) model of language production. Secondly, in Section 7.2,
Grammar as processor
I will reconsider at which stages of the grammatical derivation the various types
of spontaneous speech errors occur. In addition, I will comment on some of
the benefits provided by a DM-inspired account of spontaneous speech errors.
Finally, in Section 7.3, I will re-examine error patterns that are a potential challenge to some of the claims made in thisbook.
Obviously, in any model, a prerequisite for the generation of an utterance is a communicative intention of a speaker, the wish to convey some message. On the basis
of the message intention, nodes are activated at the conceptual level (or conceptual stratum). In the present example, the concept nodes katze (cat), schnurr
(purr), multiple(x) (which must be linked to katze), and possibly definite
(indicating that the speaker refers to some particular cats) are activated. The conceptual level is a highly interconnected network and therefore some of these concept nodes feed activation to neighboring nodes as, for instance, tier (animal),
hund (dog), miau (miaow), and bellen (tobark).
In the DM-model, activated concept nodes spread activation to corresponding roots and features in List 1. Note that the conceptual level must also supply
information about the temporal setting of the message to be conveyed, activating
the feature [past] in the present example, and about the type of action, activating
the feature [+cause]. Moreover, katze is inherently linked to a gender feature, to
compositional semantic features (for example, [animate]), and, due to conceptual
activation, to the number feature[+pl].
The most highly activated roots and features enter the computational system.
In the present context, I shall not go into the details of how a deep structure is generated from these elements. We must assume that roots and features trigger specialized syntactic procedures in some orderly fashion. Firstly, all roots are inserted
into l-nodes and automatically project LPs. Secondly, on the basis of the communicative intention, these LPs appear in different functional licensing environments. In the above example, schnurr, for instance, is licensed by a light verb
head which contains the feature [+cause]. Remember that this feature requires
Conclusion
. The example discussed here is, of course, a very simple one. Whenever more roots are
selected from List 1, the conceptual level must supply further information which guarantees
that roots are inserted into appropriate slots (for example, one concept node being marked as
the actor, the other as theme or goal). See Kempen & Hoenkamp (1987) for various categorial
and functional procedures that are responsible for syntactic tree formation (also see Levelt
(1989:236ff) for an illuminating illustration of the basic operations). Following DM assumptions, however, categorial procedures (that is, building instructions for categorially specified
phrases) do not play any role.
Grammar as processor
structure is converted into an instruction to the articulatory apparatus and the phonetic sequence in (2a) is uttered. Of course, phonetic reduction and assimilation
rules such as schwa deletion and r-vocalisation may apply, giving rise to the alternative phonetic form in(2b).
(2) a. [di:]
b. [di:]
[katsn] [nrn]
[katsn] [n:n]
Let us now consider how these processes can be mapped onto the production
model. As has been explained in Section 3.1.2, in Levelts (1989) model of language production, three processing components are distinguished: the conceptualizer, the formulator, and the articulator (see Figure (4) in Chapter 3). According
to Levelt, the formulator accepts conceptual information as characteristic input
from the conceptualizer. On the basis of this information, lemmas are activated in
the speakers mental lexicon and syntactic building procedures are set off, both of
these operations happening in parallel. Remember, however, from the discussion
in Section 4.2.1 that I do not assume that conceptual (non-compositional semantic) information enters the computational system. Rather, I take the conceptualizer and the formulator to be linked via List 1, as is illustrated in the scheme
in (3). In line with DM assumptions, lexical selection, that is, selection of the
most highly activated roots and features from List 1, takes place before entering
the computational system and consequently, the building of syntactic structure is
only indirectly controlled by conceptual information. In other words: while in the
production model, the lemma lexicon is seen as a data pool that is accessed from
within the formulator, I suggest that List 1 (which equals the lemma lexicon) is a
sort of gate that has to be passed in order to enter theformulator.
(3) Conceptualizer
List 1
Formulator
This, of course, implies that those elements within List 1 that receive activation
from the conceptual level and that are finally retrieved must somehow be marked
for their syntactic role in order to guarantee their insertion into the appropriate
node at deep structure.2 Obviously, the same holds within all types of production models. In these models, a predicative frame is assumed to be built independently of the lemmas that are retrieved from the lexicon. Still, some mechanism
. Note that this is true even for an intransitive sentence like the one in (1). Since according
to DM, the activated roots in List 1 are not equipped with categorial information, the insertion of schnurr into SpecvP would otherwise be as likely as the insertion of katze.
Conclusion
must ensure that selected lemmas appear in the appropriate slots in that frame (for
example, the boy chased the dog vs. the dog chased theboy).
Garrett (1975, 1980a) as well as Levelt (1989) further assume that within the
formulator, the translation of a conceptual structure into a linguistic structure proceeds in two steps. The first step is called grammatical encoding (functional level
in Garretts terminology), while the second step is called phonological encoding
(positional level). This crucial distinction grammatical encoding taking place
before phonological encoding is also advocated in the DM-model. I therefore suggest that the DM-model of grammar is to be equated with the formulator module,
that is, both structure-building and spell-out happen within that processing module. First, grammatical encoding is carried out within the computational system,
where surface structure is generated from deep structure, and at the post-syntactic
level of MS, where further structure-changing operations may apply. Secondly, phonological encoding takes place at the level of PF by means of Vocabulary insertion
and phonological readjustment. This mapping is illustrated in the scheme in(4).
(4)
List 1
GRAMMATICAL ENCODING
PHONOLOGICAL
Computational System
ENCODING: PF
MS
FORMULATOR
List 2
In this model, the grammar is the processor. It receives input from the conceptualizer and hands its output over to the articulatory system. Given that DM
not only allows for the generation of correct utterances but also provides us
with the necessary tools to account for the attested speech error patterns (see
next section), I conclude that it makes for a psychologically plausible model
ofgrammar.3
. See Phillips (1996) for a similar view of the comprehension side of language processing.
His work is an attempt to abandon the parser-grammar distinction. The PIG (Parser Is
Grammar) model he proposes has two components: a grammar (containing language universals, a lexicon, structure building procedures, etc.) and a finite set of resources (like working
memory and world knowledge). Phillips, drawing on earlier analysis-by-synthesis models,
claims that we perceive sentences by generating them for ourselves. That is, parsing is an
active process, in which the grammar tries to generate a sentence the phonetic form of which
matches the incoming sentence, using the normal structures and operations of the grammar
(see Garrett (1982, 2000) on the relation of production models to comprehension models).
Grammar as processor
Conclusion
Inherent gender features may only be involved in copy errors. All slips
involving erroneous feature copy occur at MS, the level at which subject-verb
agreement and DP-internal agreement is established. For the most part, SVAerrors involve local agreement but we have seen that long-distance agreement
errors are also attested. In addition, erroneous case assignment is to be located
atMS.
Since the phonological form of elements is neither available within the
computational system nor at MS, it is clear that all errors that involve phonological information must occur at PF, that is during or after spell-out. First of
all, phonologically specified Vocabulary items are retrieved from List 2. At this
point, form-based substitutions may occur when a phonological neighbor of an
intended item receives too much activation and is selected for insertion. As in
meaning-based substitutions, the substituted and substituting item tend to share
the same gender feature. This suggests that Vocabulary items, too, are specified
for gender. Following the insertion of Vocabulary items into terminal nodes, all
errors that involve the exchange, anticipation, perseveration, or shift of phonological material take place, no matter whether a phonological feature, a segment
or segment cluster, a morpheme, or a word is affected by the slip. Remember that
I argued in Section 6.4.4 that the traditional distinction between word exchanges
and stranding errors obscures the fact that both these error types may occur
before or after spell-out. In the former case, we are actually dealing with root
exchanges, while in the latter case, the error is phonological in nature. Crucially,
only following a root exchange, adaptation of the error elements to their new
licensing environment isobserved.
As is well-known, blends are a particularly diverse and intricate matter. They
shall therefore receive a somewhat more detailed treatment.4 Blends come in two
types: as word blends and as phrasal blends. In a word blend, two related concepts
are activated at the conceptual level. In contrast to meaning-based substitutions,
however, no decision between the two corresponding roots is achieved in List 1
and consequently, both roots enter the computational system where they compete for a single terminal node. In case the competing roots occupy a position
that is licensed by a determiner, the gender feature of one of them will be copied
onto other elements within DP at MS. At PF, two Vocabulary items are selected
from List 2 and the two phonological forms are fused in the terminal node, as is
illustrated in(5a).
. For extensive discussion of German blends see Leuninger (1987), Gies (1993), and
Wiegand (1996); for English blends see Fay (1982) and Coppock (2007).
Grammar as processor
(5) a.
Hatte-st du
ein-en
gut-en
Plitz
had-2.sg you(sg) a-m.acc good-m.acc (error)
ein-en gut-en
Platz
// ein-en gut-en
Sitz
a-m.acc good-m.acc place(m) // a-m.acc good-m.acc seat(m)
b. er kann
kein-e
Fliege trb-en kein
Wsser-chen
he can.3.sg no-f.acc fly(f) cloud-inf no.n.acc water-dim(n)
trb-en // kein-er
Fliege was
zuleide
cloud-inf // no-f.dat fly(f) something harm
tun
do.inf
c.
er wohn-t in der
vierten Stock,
h, Etage
he live-3.sg in the.f.dat fourth story(m), er, floor(f)
He lives on the fourth floor//story.
As already pointed out in Section 6.7.1.2, phrasal blends may have different etiologies. All phrasal blends have in common that within List 1, roots and features that
belong to two different, yet semantically equivalent, speech plans receive activation from the conceptual level. Whenever a phrasal blend results in a grammatical
utterance, the substitution of one root for another must have taken place before
feature copy at MS, that is, either when roots are selected from List 1 or within the
computational system. In (5b), for instance, fliege (fly) takes the position of
wasser (water) in the competing frame and subsequently, the gender feature of
fliege is copied onto thedeterminer.
The phrasal blend in (5c) has other characteristics. Most importantly, we
observe a gender mismatch in the resulting utterance. For this error, we must
assume that within List 1, no decision is made between the semantically related
roots stock (floor, story) and etage (floor, story). Therefore, both roots
enter the computational system and share one terminal node in the syntactic
structure just like the competing roots in the word blend in (5a). At MS, the
gender feature of etage is copied onto D but at PF, the Vocabulary item that
corresponds to the competing root stock is retrieved from the Vocabulary.
That is, in contrast to word blends, only one phonological form is selected
forinsertion.
Figure (6) presents a synopsis of what error types occur at what processing
level. Note that blends are mentioned twice because, as argued above, for most of
the blends, we have to assume that they involve two error sites: List 1, where roots
receive activation from the conceptual level, and List 2, where Vocabulary items
are selected forinsertion.
Conclusion
(6)
CONCEPTUALIZER
conceptual network
List 1
semantic substitutions,
semantic anticipations &
perseverations, blends I
FORMULATOR
Computational System
exchanges, anticipations
& perseverations of roots,
feature shift & stranding
Morphological Structure
Phonological Form
phonological errors
List 2
ARTICULATOR
phonological
substitutions, blends II
I conclude from the above discussion that within such a combined model, the
attested spontaneous speech errors can be accounted for without further stipulation. Moreover, the model makes a number of interesting predictions about the
type of information that is available to the processor and about factors that may
facilitate the occurrence of speech errors. Also, it allows us to explain why certain
error types are not attested. Let me remind you of the following five findings that
emerged from the analysis of speech errors within the DMframework.
First, it has become clear that the manipulation of morphosyntactic features
is of major importance during the generation of an utterance. The features which
are drawn from List 1 and which are present within the computational system
(with the exception of inherent gender features) may be accessed separately in an
error, that is, they may be subject to shift or stranding. As far as inherent gender
features are concerned, we have seen that an identical gender effect is observed in
Grammar as processor
meaning- and form-based substitutions. This indicates that the grammatical gender feature is specified throughout thederivation.
Secondly, and also related to the manipulation of features, the operations that
take place at MS most importantly, feature copy and case assignment may
be defective. This allows for a straightforward explanation of feature copy errors
some of which have only received little attention in the literature on language production so far. On the one hand, characteristics of local SVA-errors support the
idea that agreement nodes are only implemented after movement operations have
taken place. On the other hand, the existence of long-distance agreement errors
indicates that sometimes, agreement may be established in the absence of a sufficiently local relation between agreement controller and agreementtarget.
Thirdly, I have proposed that the formulator can operate without reference to
syntactic categories. That is, categorial features are not amongst the features drawn
from List 1. This claim, while being in the spirit of DM, departs from assumptions
made in psycholinguistic production models, where lemmas are taken to bear category labels. Still, acategorial roots do not randomly interact in an error. Rather,
their interaction is constrained by the licening environment in which theyappear.
A fourth important point argued for in this book is that, based on the DM
architecture, only errors that occur before the level of MS can be followed by morphosyntactic or morphological adaptation processes such as, for instance, the
insertion of a derivational affix or the choice of an appropriate determiner. This
property of the model allows for interesting predictions concerning the grammatical well-formedness of erroneous utterances. First, only meaning-based but not
form-based substitutions are expected to be followed by morphosyntactic adaptation due to gender feature copy. Moreover, morphological or morphosyntactic
adaptations should never be observed in a phonological error. In my corpus, both
these predictions are borneout.
Finally, once we are equipped with the tools made available by DM, the notion
of accommodation in the sense of a repair strategy or control mechanism
becomes superfluous and can therefore be abandoned. In fact, the adaptation processes mentioned in the previous paragraph are the result of the application of a
number of operations (morpheme insertion, feature copy, phonological readjustment) that apply in the derivation of an utterance anyway. I take this to be a welcome result since it reduces the processing load considerably. In contrast, it is hard
to see how a model in which words are taken to enter the derivation fully inflected
(for instance, Chomsky (1995, 2000)) could account for the error data without
postulating the application of post-error repairprocesses.5
Conclusion
morphemes. Again, this implies that accommodation must have taken place in speech errors
in which a word surfaces with a different (contextually appropriate) derivational affix.
Grammar as processor
processor would have to step back one level in order to repair the utterance.6 A
hypothetical example that I mentioned would be a phonological error which happens to result in an existing noun the gender feature of which is then copied onto
the determiner (see Footnote 18 in Section 6.4.4). For this type of repair to occur,
the processor would have to go back from PF (where the error takes place) to MS
(where gender agreement is established). If the ban on repair is weakened such
that repair is allowed as long as it takes place within one and the same processing
level, then it would be interesting to investigate whether similar phenomena are
attested at MS. I leave this issue for future research. In any case, the fact remains
that, for the most part, DM-mechanisms render unnecessary the application of
costly repairprocesses.
Besides the issue of repair, I also commented on the possible application of a
surface filter that stops the derivation when it runs the risk of generating an unattested morphologically complex form (see Section 6.6.3). Thanks to that surface
filter, self-correction takes place whenever a root appears in an environment in
which it is not licensed. Strictly speaking, in this case, we are not dealing with
a repair, since the erroneous utterance is not brought in line with some grammatical well-formedness constraint. Rather, as Albright (2007) points out, the selfcorrection suggests that the processor detects the upcoming conflict and avoids
it by stopping the derivation and starting anew. On the face of it, the data in my
corpus seem to confirm the existence of such a surface filter. As a matter of fact, all
errors in which a root would have to be spelled out in an environment in which it is
not licensed (for example, kind (child) in a [+v]-environment) are self-corrected
by the speaker before the conflict site is reached. There are two caveats, however.
First, my corpus contains only twelve relevant errors. Self-correction, however, is
very common and is frequently attested in cases in which it is not required by the
surface filter. Secondly, given the self-correction, it cannot be decided whether
these twelve errors are really root exchanges and not root anticipations. Obviously,
the surface filter argument relies on the assumption that we are dealing with an
incomplete exchange because in the case of an anticipation, no licensing conflict
would have arised. Despite these qualifications, it is certainly intriguing that, at
least in my corpus, unattested morphologically complex forms never surface. This
issue, too, awaits further in-depth investigation because the existence of such a
surface filter might run counter to my claim that output-oriented processing need
not beassumed.
Conclusion
A final problematic issue that came up at various points in the discussion concerns the availability of phonological information. DM endorses late insertion of
phonological forms (Vocabulary items). That is, phonological information should
not have any influence on errors that occur earlier in the derivation. As is wellknown, this issue is a matter of ongoing debate. Some scholars have claimed, for
instance, that semantic substitutions are facilitated by phonological similarity (see
Section 3.1.3). They take this as evidence for an interactive model (contra Garrett),
in which phonological information is available already at the point at which roots
are selected from List 1. As argued by Albright (2007:44), a baseline that takes
into account not only the expected rate of chance phonological similarity, but also
chance semantic similarity is required in order to evaluate whether semantic and
phonological similarity truly interact in substitution errors. If further refined measures support the idea of interaction, then the strict separation of the information
contained in List 1 (the lemma lexicon) and List 2 (the phonological lexicon) may
have to be reconsidered. A full discussion of the interaction issue is beyond the
scope of this study. Note, however, that interaction does not necessarily require the
existence of a mixed list. Rather, one might, for instance, postulate a link between
the two lists which due to experience (that is, repeated activation) strengthens
the connections between semantic neighbors that are also phonologically related,
and viceversa.
But even if we assume a link between the two lists, we are still left with the
problem of the apparent influence of phonological information on local SVAerrors discussed in Section 5.2.2.6. Remember that in my corpus, local SVA-errors
are most likely to occur when the singular head noun of the subject DP is accompanied by the definite feminine determiner die, which is homophonous to the plural determiner. That is, the error distribution suggests that the processor knows
that the feature bundles [+def, +fem] and [+def, +pl] are spelled out by phonologically identical Vocabulary items. As pointed out before, such a facilitating
effect is unexpected given that phonological information is not available when feature copy takes place. I cannot offer a resolution to this problem here. A possible
hypothesis to pursue would involve underspecification of the grammatical gender
feature [f]. One might argue then that agreement with a local plural noun is more
likely when the determiner accompanying the head noun is unspecified for gender. Admittedly, however, a solution along these lines is ad hoc and therefore not
very satisfactory. Further inquiry is needed to determine whether a slight adaptation or refinement of DM would allow us to account for these problematicdata.
Despite the problematic cases discussed in this section, the overall picture
that emerges from the discussion of spontaneous speech errors is that the DM
architecture is fit to account for the observed error patterns. On the one hand, it
makes with few exceptions correct predictions about possible influences on
Grammar as processor
speech errors; on the other hand, it helps us in explaining why certain error types
are notattested.
The endeavour undertaken in the present study has been motivated by the
assumption that the grammar directly determines and constrains the generation of possible utterances. It is therefore desirable to contemplate if and how the
grammar influences the translation of a preverbal communicative intention into
an articulated string of sounds. Merrill Garrett puts this requirement into the
followingwords.
When we contemplate language processing questions, there are multiple sources
that we draw on to constrain theoretical claims. These begin with formal
grammars, and should in a quite real sense, end with formal grammars. When
we ask how words are learned, recognized, pronounced, integrated in utterance,
or interpreted in their discourse ensembles, we pose those questions in terms
of structural features that grammars associate with words as a function of their
embedding in grammatical rule systems. (Garrett2000:31f)
In the model sketched above, I have taken this view seriously. Above all, I have
argued that grammatical operations in a real sense constrain language processing. In the modified model I propose, the grammar interacts with the processing
component as directly as possible in that it takes over the job of the processor
(or formulator, respectively). That is, the grammar and the processor are the
samesystem.
appendix
b. [die Farbe
der
Karte-n]
sind
nicht wichtig
the.f colour(f) of.the.pl card(f)-pl be.3.pl not important
die Farbe
ist
nicht wichtig
the.f colour(f) be.3.pl not important
Grammar as Processor
The grammatical information given in the interlinear translation (for example, gender and case features) is usually restricted to the information relevant to a
particular error. For instance, in the local subject-verb aggrement error (1b), case
information is left unspecified. As can also be seen in (1b), the inherent gender
specification of a noun is given between brackets in the interlinear translation. In
blends, the two competing words or phrases are separated by //. Finally, only the
intended utterance is translated, except for blends where whenever possible a
translation for both competitors isgiven.
49
semanticanticipations
semanticperseverations
18
31
406
219
(163)
(56)
45
96
46
133
87
46
241
103
53
(9)
(18)
(26)
70
(34)
(11)
(25)
15
Total
829
III.1Semantic anticipation
a.
ich htte
meine Bohne mit Karotte-n und Erbse-n
I have.cond my
bean with carrot-pl and pea-pl
mitbring-en knn-en
bring-inf
can-inf
meine Dose
my
can
b. schreib-st du
mir bitte, h, gib-st
du
mir bitte
write-2.sg you(sg) me please, er, give-2.sg you(sg) me please
ein-en Stift
pen(m) a-m.acc
beim Laufen
at.the running
Grammar as Processor
+ b. i ch habe die Wsche, h, die Woche so viel zu bgel-n
I have this laundry, er, this week so much to iron-inf
I have so much to iron this week.
III.2Semantic perseveration
h, am
Tisch angestossen
er, at.the table knock.on.part
b. i ch woll-te
den Hund anbell-en anbind-en
I
want-past the dog bark.at-inf tie.up-inf
I wanted to tie up the dog.
a.
in die Srg-e
into the coffin-pl
+ b. an mein-em
Weiher ist
kein See mehr
at my-m.dat pond(m) be.3.sg no lake anymore
or with an intervening nominative DP (9). Within a complex subject DP, the local
noun may be contained in a genitive modifier (e.g., (7a)) or in a PP (e.g., (7n)).
Also, the plural noun may be local to the verb (e.g., (7a)) or semi-local (e.g., (7r)).
Erroneous agreement of the verb with a local singular noun is the exception (e.g.,
(7w)). Moreover, in a few errors, it is only the person feature that triggers local
agreement (e.g.,(7x)).
wird
nicht bekannt gegeben
be.pass.3.sg not known part-give-part
Grammar as Processor
h. [die Motivation
der
Athlet-en] knn-en durchaus variier-en
the.f motivation(f) of.the.pl athlete-pl can-3.pl certainly vary-inf
kann
einen neuen Anstrich gebrauch-en
can.3.sg a
new painting use-inf
+ j. dass [ein Teil der
Vergebrdler]
dadurch
that a
part the.gen.pl slip.of.the.hand.pl from.that
entstand-en sind,
h, ist
result-part be.3.pl, er, be.3.sg
m. weil
beim Lernen [die Beteiligung komplexer
because for.the learning the involvement complex.gen.pl
weil
beim Lernen die Beteiligung notwendig ist
because for.the learning the involvement necessary be.3.sg
n. weil
morgen [die Abstimmung ber die
neu-en Gesetz-e]
because tomorrow the voting
about the.pl new-pl law-pl
stattfind-en
weil
morgen
die Abstimmung
take.place-3.pl because tomorrow the voting
stattfind-et
take.place-3.sg
because the voting about the new laws will take place tomorrow.
+ o. [eine betrchtliche Anzahl von Gebrde-n] sind
lexikalisch
a
considerable number of sign-pl
be.3.pl lexically
markier-t
mark-part
p. [eine
a
draufgegangen
eine Gruppe ist
in dem
Film
bite.the.dust.part a
group be.3.sg in the.dat movie
draufgegangen
bite.the.dust.part
war
leider
ein ziemlicher Reinfall
be.past.3.sg unfortunately a quite
disappointment
sind,
dass es kaum zuverlssige Quelle-n gib-t
be.3.pl that it hardly reliable
source-pl give-3.sg
The problem with evaluating the situation there is that hardly any reliable
sources exist.
Grammar as Processor
+ t. was passier-t,
wenn [die Aktivierung aller Knoten]
what happen-3.sg if
the activation of.all node.pl
What happens if the activation of all nodes goes below the threshold?
u. weil
[die Debatte ber die
Kondition-en] die Entscheidung
because the debate about the.pl condition-pl the decision
verzger-n weil
die Debatte die Enscheidung verzger-t
delay-3.pl because the debate the decision
delay-3.sg
aktiviert werd-en
muss
eine Mehrzahl aktiviert
activated be.pass-inf must.3.sg a
multitude activated
werd-en
be.pass-inf
w. [die Angabe-n
ber seine Familie] entsprich-t,
the statement-pl about his family
be.in.accordance.with-3.sg,
entsprech-en
nicht der Wahrheit
be.in.accordance.with-3.pl not the truth
The statements about his family are not in accordance with the truth.
erstaun-t
surprise-part
The message from you has surprised me.
In local agreement with an object/adjunct XP, the verb may agree with an
(in)direct object (e.g., (8a)) or with a DP within a PP (e.g., (8e)). The plural DP
may be local (e.g., (8a)) or semi-local (e.g., (8i)) to the verb. Most of these errors
occur in embedded clauses but occasionally, the object DP precedes the verb in
a matrix clause (e.g., (8j)). Again, agreement with a singular DP is the exception
(e.g.,(8p)).
Erklrung-en] ermglich-en
explanation-pl make.possible-3.pl
b. weil
er offensichtlich [reif-e
Dame-n] bevorzug-en
because he obviously
mature-pl lady-pl prefer-3.pl
+ c.
weil
er bevorzug-t
because he prefer-3.sg
hab-en
have-3.pl
e.
weil
der Arzt
mich
[ber mgliche Nebenwirkung-en]
because the doctor 1.sg.acc about possible side.effect-pl
f.
weil
da oft [ber Sache-n] ge-rede-t
wurd-en,
because expl often about thing-pl part-talk-part be.pass.past-3.pl
die
fr seine Arbeit wichtig
sind
which for his
work important be.3.pl
weil
da geredet
wurd-e
because expl part-talk-part be.pass.past-3.sg
because often, it was talked about things that are important for his work.
hat
have.3.sg
+ g. dass das [mit Linguistikstz-en]
ab und zu
so
sind
that that with linguistic.sentence-pl from time to time like.that be.3.pl
dass das
that that
that from time to time, its like that with linguistic sentences.
ab und zu
from time to time
so
ist
like.that be.3.sg
Grammar as Processor
mehrerer Verarbeitungs-komponente-n]
more
processing-component-pl
fhr-t
lead-3.sg
fhr-en
lead-3.pl
nehm-en
take-3.pl
weil
er sowieso das Buch nimm-t
because he anyway the book take-3.sg
because he will take the book with the cars on the cover anyway.
morgen
noch treff-en
tomorrow still meet-inf
k. [die
unschn-en
Ding-e] vergess-en ich, vergess-e
ich
the.pl unpleasant-pl thing-pl forget-3.pl I
forget-1.sg I
l. [die
Resultat-e] knn-en, h, kann
ich dir
noch nicht
the.pl result-pl can-3.pl, er, can.1.sg I 2.sg.dat yet not
sag-en
tell-inf
I cannot tell you the results yet.
m. [den
Student-en] woll-en,
h, will
er die Lsung
the.pl student-pl want-3.pl, er, want.3.sg he the solution
n. weil
die Lnge seiner Haar-e nicht [den
because the length of.his hair-pl not
the
Vorgabe-n]
requirement-pl
entsprech-en
comply.with-3.pl
weil
die Lnge nicht entsprich-t
because the length not comply.with-3.sg
because the length of his hair does not comply with the requirements.
( alternative analysis: agreement of verb with noun within subject DP)
o. die
Bar-s, die
er euch
empfohlen
habt,
the.pl bar-pl, rel.pl he 2.pl.dat recommend.part have.2.pl
h, empfohlen
hat
er, recommend.part has.3.sg
p. weil
ihre Eltern
dann endlich [ihren neuen Freund]
because her parents.pl then finally her new boyfriend
kennenlern-en wird,
h, werd-en
get.to.know-inf will.3.sg, er, will-3.pl
because her parents will finally meet her new boyfriend then.
q. weil
ich seit Tag-en [diese Schei-firma] anzuruf-en versuch-t
because I for day-pl this shit-company call-inf
try-3.sg
weil
ich anzuruf-en versuch-e
because I
call-inf
try-1.sg
because Ive been trying to call this damned company for days.
As argued in Section 5.2.2.3, cases in which the verb agrees with a local nominative DP might also be analyzed as deletionerrors.
a.
b. weil
du,
[genau wie die
meisten andere-n Leute],
because you(sg), just
like the.pl most
other-pl people.pl
du denk-st
you(sg) think-2.sg
because you, just like most other people, think that sign language is
international.
Grammar as Processor
c.
weil
sie fast
wortgetreu dasselbe ge-sag-t
hast
because she almost literally
the.same part-say-part have.2.sg
[wie
as
du]
weil
sie dasselbe ge-sag-t
hat
you(sg) because she the.same part-say-part have.3.sg
because she has said almost literally the same you did.
+ d. das gabs noch nie, dass ich [spter als du]
that gaveit yet
never that I
later than you(sg)
auf-ge-stand-en
bist
dass ich spter als du
up-part-get-part be.2.sg that I later than you(sg)
auf-ge-stand-en
bin
up-part-get-part be.1.sg
That never happened before, that I got up later than you did.
In Section 5.2.2.4, I pointed out that SVA-errors in blends fall outside the group of
local SVA-errors. In the errors in (10), we are not to speak of a local or non-local
error source because the noun that triggers the erroneous agreement does not
surface in the actualutterance.
(10) SVA-errors in blends
a. weil
man das Kind nicht frag-en knn-en weil
because one the child not ask-inf can-3.pl because
b. das ist
mein-e Wort-e
that be.3.sg my-pl word-pl
das ist
meine Rede // das sind
mein-e Wort-e
that be.3.sg my
speech // that be.3.pl my-pl word-pl
+ d. da
war-en
ein spanisches Prchen da
war-en
expl be.past-3.pl a Spanish couple expl be.past-3.pl
zwei Spanier
// da
war
ein spanisches Prchen
two Spaniard.pl // expl be.past.3.sg a Spanish
couple
ich wiss-t,
h, ich wei,
[dass ihr
Profis
seid]
I know-2.pl, er, I know.1.sg that you(pl) professionals be.2.pl
I know that you are professionals.
b. dass er [die
Zitrone-n zu kauf-en] vergess-en
that he the.acc.pl lemon-pl to buy-inf forget-3.pl
c.
weil
sie [dich
schon ewig
zu treff-en] versuch-st,
because she 2.sg.acc already for.ages to meet-inf] try-2.sg,
h, zu treff-en
versuch-t
er, to meet-inf try-3.sg
d. ich woll-en,
h, ich will
eigentlich, [dass wir uns
I want-1.pl, er, I want.1.sg actually that we rec
ft-er
seh-en]
often-comp see-pl
e. weil
sie manchmal, [dass sie dich
kenn-t],
bedauer-st,
because she sometimes that she 2.sg.acc know-3.sg regret-2.sg
h, bedauer-t
er, regret-3.sg
f. ich wiss-t,
[dass ihr
nicht Recht hab-t]
I know-2.pl that you(pl) not right have-2.pl
ich wei,
dass ihr
nicht
I
know.1.sg, that you(pl) not
Recht
right
hab-t
have-2.pl
Grammar as Processor
+ g. wir bist,
h, wir sind
etwas
berrascht,
we be.2.sg, er, we be.1.pl somewhat surprised
[weil
du
normalerweise schnell-er
reagier-st]
because you(sg) usually
quick-comp react-2.sg
We are somewhat surprised because you usually react more quickly.
h. [dich
zu kritisier-en] fll-st, h, fll-t
mir
schwer
2.sg.acc to criticize-inf fall-2.sg, er, fall-3.sg 1.sg.dat difficult
I find it difficult to criticize you.
In cases in which an embedded verb agrees with a matrix argument, the erroneous agreement controller may be the subject (e.g., (12a)) or the object (e.g.,
(12d)) of the matrix clause. Also, the embedded clause may be a finite complement clause (e.g., (12a)), an adverbial clause (e.g., (12e)), or a relative clause
(e.g.,(12g)).
(12) Embedded verb agrees with matrix argument
a.
wir versteh-en
gut, [dass du
ge-nerv-t
sind],
we understand-1.pl well that you(sg) part-annoy-part be.1.pl,
h, bist
er, be.2.sg
+ b. wenn du
annimm-st, [dass sie berzeugung-en
when you(sg) assume-2.sg that she belief-pl
hast]
dass
have.2.sg that
Meinung-en
opinion-pl
Meinung-en hat
opinion-pl have.3.sg
sie berzeugung-en
she belief-pl
und
and
und
and
+ c.
d. dir
ist
doch
vollkommen wurscht, [ob
er
2.sg.dat be.3.sg mod.part completely sausage, whether he
Recht
right
hast],
ob
er Recht
have.2.sg, whether he right
hat
oder nicht
have.3.sg or
not
e. [als
sie uns das letzte Mal besuch-t
hast],
when she us the last
time visit-part have.2.sg
h, hat,
war-st
du wahrscheinlich
er, have.3.sg be.past-2.sg you probably
in Urlaub
on holiday
The last time she visited us, you were probably on holiday.
+ f. [wenn das Auto nicht angeh-st],
when the car not start-2.sg
g. hier passier-en
Ding-e, [die
kein Mensch fr mglich
here happen-3.pl thing-pl rel.pl no person for possible
ge-halt-en
hab-en]
part-hold-part have-pl
weil
er wtend ist
because he angry be.3.sg
because he is angry and I dont feel like doing it.
+ c. wenn das Hhnchen komm-st
und du
hast
noch
when the chicken
come-2.sg and you(sg) have.2.sg yet
keinen Tisch
no
table
When the (roast) chicken comes and you dont have a table yet.
Grammar as Processor
den
er mir
aus Ecuador mit-ge-brach-t
hat
rel.m.acc he 1.sg.dat from Ecuador with-part-bring-part have.3.sg
the bird made of palm wood that he brought for me from Ecuador
b. [der Film
mit den
Zombie-s],
die
er mir
the.m movie(m) with the.pl.acc zombie(m)-pl rel.pl he 1.sg.dat
ge-lieh-en,
den
er mir
geliehen
hat
part-lend-part, rel.m.acc he 1.sg.dat part-lend-part have.3.sg
c. [die
Anbieter
von kologisch-er Kleidung], die
auch
the.pl supplier(m).pl of ecological-f clothing(f) rel.pl also
sein soll
be shall.3.sg
d. mein-st
du
[die Jacke
mit dem
Pelz-kragen],
mean-2.sg you(sg) the.f jacket(f) with the.m.dat fur-collar(m),
den
rel.m.acc
ge-kauf-t
hab-e
part-buy-part have-1.sg
ich h, die
I
er, rel.f.acc
Do you mean the jacket with the fur collar that I bought last year?
+ e.
e s
gib-t
[Form-en von Kriminalitt], die
expl exist-3.sg form-pl of crime
rel.pl/f
zu sein
schein-t
Form-en
to be.inf seem-3.sg form-pl
die
importiert
rel.pl imported
importiert
imported
von Kriminalitt,
of crime
zu sein
schein-en
to be.inf seem-3.pl
f. [die
Farbe
des
Rock-s]
ist
genau der,
the.f colour(f) the.m.gen skirt(m)-gen be.3.sg exactly the.one.m
den
ich such-e,
h, ist
genau die,
die
rel.m.acc I look.for-1.sg, er, be.3.sg exactly the.one.f rel.f.acc
ich such-e
I
look.for-1.sg
g. [die Freundin
mein-es Bruder-s]
hat sein-en
Zug
the.f girlfriend(f) my-m.gen brother-gen has poss.3.sg.m-m train(m)
verpass-t
hat ihr-en
Zug
verpass-t
miss-part has poss.3.sg.f-m train(m) miss-part
als
mein-er
ist
anders als mein-e
from poss.1.sg-m be.3.sg different from poss.1.sg-f
h, ihm
ein-e Postkarte
schreib-en kann,
kann-st
er, 3.sg.m.dat a-f
postcard(f) write-inf can.3.sg, can-2.sg
He has given you his address so that you can write him a postcard.
Fll-e
case(m)-pl
auf alle
in all.pl.m.acc
+ c. er hat
uns durch
dies-en
Wein-gewlbe ge-jag-t
he have.3.sg us through this-m.acc wine-vault(n) part-chase-part
durch
dies-en
Wein-keller
// durch
dies-es
through this-m.acc wine-cellar(m) // through this-n.acc
Wein-gewlbe
wine-vault(n)
Grammar as Processor
leicht-es
Kopfweh
// leicht-e
Kopfschmerz-en
slight-n.acc headache(n) // slight-pl.acc headpain(m)-pl
As argued in Section 5.1.3, gender mismatch should only be observed after formbased substitutions (e.g., (16a)), while meaning-based substitutions should be
accompanied by gender accommodation (see (23j) to (23m) in Appendix VI.1).
Example (16e) is an exception to thisgeneralization.
(16) Feature mismatch due to noun substitution
a.
b. hast
du
Karte-n frs
Filiale,
h, frs
Finale
have.2.sg you(sg) ticket-pl for.the.n branch(f), er, for.the.n final(n)
Do you have tickets for the final?
+ c.
+ d. er hat
es in best-er
Abschied
ge-tan
in best-er
Absicht
in best-f.dat intention(f)
der
Bruder
von der
N. ist
the.m brother(m) of
the.f.dat N.(f) be.3.sg
Feature mismatch may also be the consequence of a noun exchange (e.g., (17a)), a
noun anticipation (e.g., (17e)), a determiner anticipation (e.g., (17f)), determiner
perseveration (e.g., (17i)), or a determiner exchange (e.g.,(17k)).
(17) Feature mismatch due to noun/determiner displacement
a.
ich hab
ein
Nase
auf der
Haar, h,
I have.1.sg a.n.acc nose(f) on the.f.dat hair(n), er,
ein
Haar
auf
a.n.acc hair(n) on
der
Nase
the.f.dat nose(f)
b. das Mund
luft
mir im
Wasser zusammen
the.n mouth(m) run-3.sg me in.the.m/n.dat water(n) together
c. man muss
die
Wurzel an der
bel pack-en
one must.3.sg the.f.acc root(f) at the.f.dat evil(n) grab-inf
das
bel
an der
Wurzel pack-en
the.n.acc evil(n) at the.f.dat root(f) grab-inf
d. da
bring-t
er dem
Hasen
den
Kind
there bring-3.sg he the.m/n.dat rabbit(m) the.m.acc child(n)
dem
Kind
den
Hasen
the.n.dat child(n) the.m.acc rabbit(m)
+ e. hol-st
du
mir
fetch-2.sg you(sg) me
in der
Handtuch das
in the.f.dat towel(n) the.n.acc
in der
Kche
das
in the.f.dat kitchen(f) the.n.acc
Kchen-handtuch?
kitchen-towel(n)
Kchen-handtuch
kitchen-towel(n)
+ f. dass jemand auf den
Idee
komm-t,
that someone on the.m.acc idea(f) come-3.sg
auf die
Idee
komm-t, den
Herd
zu putz-en
on the.f.acc idea(f) come-3.sg the.m.acc stove(m) to clean-inf
g. Nagel-studio fr den
Dame, h, fr die
Dame
nail-studio for the.m.acc lady(f), er, for the.f.acc lady(f)
und den
Herrn
and the.m.acc gentleman(m)
+ h. der
Helen,
h, die Helen
ist ja
der
the.m Helen(f), er, the.f Helen(f) is mod.part the.m
Versprecher-star
slip.of.the.tongue-star(m)
Grammar as Processor
+ i. das ist
hier nur die Umlautung,
that be.3.sg here only the.f umlaut.formation(f)
die
das
Plural
markier-t
which the.n.acc plural(m) mark-3.sg
markier-t
mark-3.sg
den
Plural
the.m.acc plural(m)
+ j. die Sieben muss
vor
die A,
vor
das A
the.f seven(f) must.3.sg in.front.of the.f A(n), in.front.of the.n A(n)
The seven must come before the A.
k. jetzt hab
ich selber den
Hand vor
die
now have.1.sg I myself the.m.acc hand(f) in.front.of the.f.acc
Mund,
h, die
Hand
vor
den
Mund
mouth(m), er, the.f.acc hand(f) in.front.of the.m.acc mouth(m)
ge-halt-en
part-hold-part
IV.5Subcategorization errors
Most of the subcategorization errors occur in blends. As a result, an argument may
be wrongly case-marked (e.g., (18a)), there may be a superfluous case-marked
argument (e.g., (18e)), or a required case-marked argument may be missing (e.g.,
(18h)). Since prepositions also require case-marked arguments, a subcategorization error may also be observed within a PP (e.g.,(18i)).
(18) Subcategorization errors due to blend
a.
wer hat
dich
am
meisten imponier-t
who have.3.sg 2.sg.acc at.the most
impress-part
b. eine Sprache,
die
bestimmt-en
UG-Prinzip-ien verletz-t
a.f language(f) rel.f certain-pl.dat UG-principle-pl violate-3.sg
die
bestimmt-en
UG-Prinzip-ien widersprich-t
//
rel.f certain-pl.dat UG-principle-pl contradict-3.sg //
die
bestimmt-e
UG-Prinzip-ien verletz-t
rel.f certain-pl.acc UG-principle-pl violate-3.sg
c. mich
mach-t
das nichts
1.sg.acc make-3.sg that nothing
mich
str-t
das nicht // mir
mach-t
das nichts
1.sg.acc bother-3.sg that not // 1.sg.dat make-3.sg that nothing
+ d. dies-en
Ausdruck
gefll-t
mir
gut
this-m.acc expression(m) please-3.sg 1.sg.dat well
dies-en
Ausdruck
find-e
ich gut //
this-m.acc expression(m) find-1.sg I
good //
dies-er
Ausdruck
gefll-t
mir
gut
this-m.nom expression(m) please-3.sg 1.sg.dat well
e.
das lohnt
sich den
Aufwand nicht das lohnt
that is.worth refl the.m.acc effort(m) not that is.worth
f.
ich habs
mir schon ge-ahn-t
ich habs
I
have.1.sgit refl already part-suspect-part I have.1.sgit
mir ge-dach-t
// ich habs
ge-ahn-t
refl part-think-part // I
have.1.sgit part-suspect-part
+ g. es konn-te nicht bewies-en
werd-en,
dass es Gott nicht
it can-past not proof-part be.pass-inf that it god not
nicht existier-t
not exist-3.sg
man sag-en
one say-inf
Thats possible//You can say that.
+ i. das hat echt Stil, Leberwurst mit s-er
Senf
that has real style liver.sausage with sweet-m.nom mustard(m)
mit s-em
Senf
// und s-er
Senf
with sweet-m.dat mustard(m) // and sweet-m.nom mustard(m)
Grammar as Processor
V.1Feature stranding
Morphosyntactic features that may be stranded are number (e.g., (19a)), person on
verbs (e.g., (19f)) and pronouns (e.g., (19g)), tense (e.g., (19h)), case (e.g., (19l)),
and negation (e.g.,(19q)).
(19) Stranding of morphosyntactic feature
a. bist
du
so nett und leg-st
die
Bnk-e auf das
be.2.sg you(sg) so kind and put-2.sg the.pl bench-pl on the.n
Kissen
und leg-st
die
Kissen auf die Bank
pillow(n) and put-2.sg the.pl pillow.pl on the.f bench(f)
b. ge-monat-ete
Arbeit-en ge-arbeit-ete
Monat-e
part-month-part work-pl
part-work-part month-pl
months in which one has worked
c.
+ f.
du
komm-st zu versprech-en versprich-st zu komm-en
you(sg) come-2.sg to promise-inf promise-2.sg to come-inf
You promise to come.
g. ich mcht-e
dir
ihn
wirklich vorstell-en,
I
want-1.sg 2.sg.dat 3.sg.m.acc really
introduce-inf
h, dich
ihm
er, 2.sg.acc 3.sg.m.dat
es droh-te
zu schein-en, dass es schien
zu droh-en
it threat-past to seem-inf that it seem.past to threat-inf
It seemed to impend that
j.
hab-en
have-pl
k. ich las
ihm, h, ich empfahl
ihm, den Artikel
I
read.past him, er, I
advise.past him the article
zu les-en
to read-inf
l.
sie hat
jemanden, mit der
er
zusammen
she have.3.sg someone with who.sg.f.dat 3.sg.m.nom together
leb-t
mit dem
sie
leb-t
live-3.sg with who.sg.m.dat 3.sg.f.nom live-3.sg
m. ich
soll-te
doch
ihn,
h,
1.sg.nom shall-past mod.part 3.sg.m.acc, er,
er
soll-te
doch
mich
anruf-en
3.sg.m.nom shall-past mod.part 1.sg.acc call-inf
Kind-er-n
Kind-er aus auslndisch-en Familie-n
child-pl-dat child-pl from foreign-pl
family-pl
Grammar as Processor
o. dass ihr
mein
Film,
dass mir
that 3.sg.f.dat 1.sg.poss.m movie(m), that 1.sg.dat
ihr
Film
so gut gefall-en
hat
3.sg.f.poss.m movie(m) so well pleas-part have.3.sg
p. ich
soll am Montag zu mir
komm-en
1.sg.nom shall on Monday to 1.sg.dat come-inf
zu ihm
komm-en
to 3.sg.m.dat come-inf
q. eine etwas
ir-relevante, h, un-deutliche, aber dennoch
a
somewhat neg-relevant, er, neg-clear,
but still
relevante Aussage
relevant statement
Semantic features that appear to be stranded in speech errors are [count] (e.g.,
(20a)) and semantic (natural) gender (e.g., first error step in(20e)).
(20) Stranding of (compositional) semantic feature
a.
b. du
kann-st schon mal Tpf-e in
den Reis tun
you(sg) can-2.sg already
pot-pl into the rice do.inf
Reis
rice
c.
fll-st
du
bitte die
Flasche-n in
den Sand,
fill-2.sg you(sg) please the.pl bottle-pl into the.m sand(m),
h, den Sand
in die Flasche
er, the.m sand(m) into the.f bottle(f)
in
den Topf
into the pot
+ d. es
war-en
total viel-e
Briefksten in mein-er Post,
expl be.past-pl really a.lot.of-pl mailbox.pl in my-f.dat mail(f),
e.
ich mein-e
die Mutter
mein-es
Bruder-s,
I
mean-1.sg the.f mother(f) my-m.gen brother(m)-gen,
h, den Vater
mein-er Schwester, nee, noch mal,
er, the.m father(m) my-f.gen sister(f), no, once again,
ich mein-e
die
Schwester
I mean-1.sg the.f sister(f)
mein-es
Vater-s
my-m.gen father(m)-gen
V.2Feature shift
Morphosyntactic features that may be shifted are number (e.g., (21a)), person on
pronouns (e.g., (21d)), tense (e.g., (21f)), case (e.g., (21g)), negation (e.g., (21i)),
and gender on pronouns (e.g., (21k)). Under shift, I subsume true shift (e.g.,
(21b) and (21i)) but also anticipation, perseveration (e.g., (21a)), and exchange
(e.g., (21d)) of features. As pointed out in Section 5.4.4, cases in which a determiner appears with an inappropriate gender feature due to the presence of another
noun with that feature (e.g., (21l)) should probably not be analyzed as gender shifts
but rather as instances of long-distance featurecopy.
(21) Shift of morphosyntactic feature
a.
e r hat-te
he have-past
Vorsprung
lead
fnf Punkt-e
five point-pl
h, vier Bch-er im
Schuber
er, four book-pl in.the.m.dat slipcase(m)
c.
weil
er in den
letzte-n Woche-n
because he in the.pl.dat last-pl week-pl
h, sein-en
Sohn
nicht ge-seh-en
hat
er, his-m.acc son(m) not part-see-part have.3.sg
sein-e Shn-e,
his-pl son-pl,
d. ich
soll-te
doch
ihn,
h,
1.sg.nom shall-past mod.part 3.sg.m.acc, er,
er
soll-te
doch
mich
anruf-en
3.sg.m.nom shall-past mod.part 1.sg.acc call-inf
Grammar as Processor
e.
ich
soll am Montag zu mir
komm-en
1.sg.nom shall on Monday to 1.sg.dat come-inf
zu ihm
komm-en
to 3.sg.m.dat come-inf
f.
eigentlich woll-te
er kam,
actually
want-past he come.past,
Actually, he wanted to come.
h, komm-en
er, come-inf
g. ich mcht-e
dir
ihn
wirklich vorstell-en,
I want-1.sg 2.sg.dat 3.sg.m.acc really
introduce-inf
h, dich
ihm
er, 2.sg.acc 3.sg.m.dat
dass er sie
mir
ausspann-en woll-te
that he 3.sg.f.acc 1.sg.dat pinch-inf
want-past
i.
er hat ge-sag-t,
dass es un-mglich ist
he has part-say-part that it neg-possible is
j.
mir geling-t
es, ihn nicht zu erreich-en
me succeed-3.sg it, him not to reach-inf
mir geling-t
es nicht, ihn zu erreich-en
me succeed-3.sg it not, him to reach-inf
l.
mir
steht
die Schwei auf der
Stirn
1.sg.dat stand-3.sg the.f sweat(m) on the.f.dat forehead(f)
mit
with
dem
sie
leb-t
who.sg.m.dat 3.sg.f.nom live-3.sg
ein
Blatt
vor
den
Mund
a.n.acc leaf(n) in.front.of the.m.acc mouth(m)
+ n. die Zwei Pik
ist neben der
As
Herz
the.f two(f) spade is next.to the.f.dat ace(n) heart
neben dem
As
Herz
next.to the.n.dat ace(n) heart
Semantic features that are subject to shift in speech errors in my corpus are semantic negation (e.g., (22a)) and semantic (natural) gender (e.g.,(22c)).
(22) Shift of (compositional) semantic feature
a.
b. die un-vorhergesagte,
the un-predicted,
die vorhergesagte
the predicted
Sturmflut
storm.tide
blieb
glcklicherweise aus
fail.to.appear.past fortunately
particle
+ c.
die Frau
hat ihr-en
ganz-en
Kram,
the.f woman(f) has 3.sg.f.poss-m.acc whole-m.acc stuff(m)
hat sein-en
ganz-en
Kram auf die Strae
has 3.sg.m.poss-m.acc whole-m.acc stuff(m) on the.f street(f)
ge-stell-t
part-put-part
h, mein-es
Bruder-s
er, my-m.gen brother(m)-gen
Grammar as Processor
e.
es war
das Gleiche wie bei
it be.past the same
as with
bei sein-er
Mutter
with 3.sg.m.poss-f.dat mother(f)
ihr-er
Mutter
3.sg.f.poss-f.dat mother(f)
VI.1Feature copy
Two types of feature copy have to be distinguished: feature copy within DP (23)
and feature copy from the subject DP onto AgrS (24). Within the former type,
feature copy may be observed following a root exchange (e.g., (23a)), a root perseveration (e.g., (23i)), a noun substitution (e.g., (23j)), or a blend (e.g., (23o)). As
argued in Section 5.1.3, gender copy following a phonological substitution, as in
(23n), isunexpected.
(23) Feature copy within DP
a. sein-e
his-f
Seite
zum
Sprung
ist schief-ge-gang-en
side(f) to.the.m.dat jump(m) is wrong-part-go-part
sein Sprung
zur
Seite
his.m jump(m) to.the.f.dat side(f)
b. oh, der
Sprung hat
ja
ein-e
Tasse
oh, the.m.nom crack(m) has.3.sg mod.part a-f.acc cup(f)
die
Tasse hat
ja
ein-en Sprung
the.f.nom cup(f) has.3.sg mod.part a-m.acc crack(m)
c.
irgendwie
somehow
Knoten
knot(m)
hab-e
ich heute ein-e
Zunge
im
have-1.sg I
today a-f.acc tongue(f) in.the.m.dat
ein-en
Knoten in der
Zunge
a-m.acc knot(m) in the.f.dat tongue(f)
d. ein Sommer
mach-t
noch kein-e
Schwalbe
a.m summer(m) make-3.sg yet no-f.acc swallow(f)
den
Kind-er-n
the.pl.dat child-pl-dat
fr ihr-e Hilfe
for their-f help(f)
ein-en Faden
in die
Nadel
a-m.acc thread(m) in the.f.acc needle(f)
how one gets a thread through (the eye of) the needle
g. ich mein-e
die Mutter
mein-es
Bruder-s,
I mean-1.sg the.f mother(f) my-m.gen brother(m)-gen,
h, den Vater
mein-er Schwester, nee, noch mal,
er, the.m father(m) my-f.gen sister(f), no, once again,
ich mein-e
die
Schwester mein-es
Vater-s
I
mean-1.sg the.f sister(f) my-m.gen father(m)-gen
+ h.
e r hat
das
Geld
voller Tasche-n
he have.3.sg the.n.acc money(n) full.of pocket-pl
die
Tasche-n
voller Geld
the.pl.acc pocket(f)-pl full.of money(n)
i.
von Frage-satz
und normal-em
Satz
of
question-clause and normal-m.dat sentence(m)
j.
ob
dein
Irrtum
genauso ausfllt wie mein-er
whether your.m error(m) exactly turn.out like mine-m
ob
dein
Urteil
genauso ausfllt wie mein-es
whether your.n judgement(n) exactly turn.out like mine-n
Grammar as Processor
k. aus
dem
Magen,
aus
out.of the.m.dat stomach(m), out.of
dem
Sinn
the.m.dat mind(m)
aus
den
Auge-n
out.of the.pl.n.dat eye(n)-pl
l.
du
muss-t
die Tr
dann festhalt-en, Quatsch, das
you(sg) must-2.sg the.f door(f) then hold-inf, rubbish, the.n
Fenster
window(n)
h, mein-es
Bruder-s
er, my-m.gen brother(m)-gen
n. wo
sie ber den Kalender guck-t
ber das Gelnder
where she over the.m calendar(m) look-3.sg over the.n railing(n)
where she looks over the railing
o. das ist wirklich ein dick-es
Stck
that is really
a.n thick-n.nom piece(n)
ein dick-er
Hund // ein stark-es
Stck
a.m thick-m.nom dog(m) // a.n heavy-n.nom piece(n)
p. er kann
kein-e
Fliege trb-en
kein
Wsser-chen
he can.3.sg no-f.acc fly(f) cloud-inf no.n.acc water-dim(n)
trb-en
// kein-er
Fliege was
zuleide tun
cloud-inf // no-f.dat fly(f) something harm do.inf
q. er hat
eine Menge Leute, die ihm nach der
Fahne
he have.3.sg a
lot.of people who him to
the.f.dat flag(f)
red-en
die ihm nach dem
Mund
red-en
//
speak-inf who him to
the.m.dat mouth(m) speak-inf //
die ihr
Fhn-chen nach dem
Wind
hng-en
who their flag(f)-dim to
the.m.dat wind(m) hang-inf
Feature copy onto AgrS is observed after the exchange of pronouns (e.g., (24a)),
the anticipation of a pronoun (e.g., (24b)), and the anticipation of a root together
with its number feature (e.g.,(24d)).
(24) Feature copy onto AgrS
a.
sie war
21, als
ich ge-storb-en
bin
she be.past 21 when I
part-die-part be.1.sg
ich
I
war
21, als
sie ge-storb-en
ist
be.past 21 when she part-die-part be.3.sg
b. du
sag-st ja
nicht, h, ich sag-e
ja
nicht,
you(sg) say-2.sg mod.part not, er, I say-1.sg mod.part not
dass du
schuld bist
that you(sg) guilty be.2.sg
c. du
denk-st
wohl,
h, er denk-t
wohl
you(sg) think-2.sg probably, er, he think-3.sg probably
dass du
fr den
Job nicht geeignet bist
that you(sg) for the.acc job not suitable be.2.sg
He probably thinks that you are not suitable for the job.
d. die
Student-en hab-en,
h, der
D. hat
einige seiner
Student-en durchfall-en lass-en
some of.his.pl student-pl fail-inf
let-inf
VI.2 Featurestranding
Feature stranding can play three different roles in speech errors. First, a
stranded feature can influence the spell-out of a pronoun, that is, a feature
bundle (25). Second, a stranded tense or person feature can trigger phonological readjustment (26). Finally, a stranded feature ([+part] of [+pl]) can trigger
context-sensitive spell-out of an affix, that is, choice of the contextually appropriate allomorph(27).
(25) Spell-out of pronoun
a. ich
soll-te
doch
ihn,
h,
1.sg.nom shall-past mod.part 3.sg.m.acc, er,
er
soll-te
doch
mich
anruf-en
3.sg.m.nom shall-past mod.part 1.sg.acc call-inf
Grammar as Processor
b. dass ihr
mein
Film,
dass mir
that 3.sg.f.dat 1.sg.poss.m movie(m), that 1.sg.dat
ihr
Film
so gut gefall-en
hat
3.sg.f.poss.m movie(m) so well please-part have.3.sg
c.
ich
soll am Montag zu mir
komm-en
1.sg.nom shall on Monday to 1.sg.dat come-inf
zu ihm
komm-en
to 3.sg.m.dat come-inf
+ d. was mich
zu mir
fhr-t
zu ihnen
what 1.sg.acc to 1.sg.dat lead-3.sg to 3.pl.dat
What brings me to you
(26) Phonological readjustment
a.
b. i ch las
ihm, h, ich empfahl
ihm, den Artikel zu les-en
I read.past him, er, I advise.past him the article to read-inf
I advised him to read the article.
c.
du
hast
doch
ge-log-en,
nicht mehr
zu
you(sg) have.2.sg mod.part part-lie-part not anymore to
d. ich las
ihr frs,
h, ich dank-te
ihr
I read.past her for.the, er, I thank-past her
frs
Korrektur les-en
meines Handout-s
for.the correction read-inf of.my handout-gen
e.
er sprich-t
immer, wenn ich Spanisch lach-e
he speak-3.sg always when I Spanish laugh-1.sg
er lach-t
immer wenn ich Spanisch sprech-e
he speak-3.sg always when I
Spanish laugh-1.sg
g e-monat-ete
Arbeit-en ge-arbeit-ete
Monat-e
part-month-part work-pl
part-work-part month-pl
months in which one has worked
Silbe-n
syllable-pl
+ c.
er hat-te
schon fnf Monat-e Urlaub, nee, fnf Woche-n
he have-past already five month-pl holiday, no, five week-pl
He already had five weeks of holiday
d. er hat
mich ge-drng-t,
he has.3.sg me part-push-part
ge-bet-en,
ihn nicht zu drng-en
part-ask-part him not to push-inf
e.
e r hat
ge-seh-en,
h, er hat
ge-sag-t,
he has.3.sg part-see-part, er, he has.3.sg part-say-part
+ f.
hast
du
ge-haus-t,
h, ge-seh-en,
have.2.sg you(sg) part-house-part, er, part-see-part
was sie
mit
what they to
VI.3 Locallicensing
The possible influence of local licensing on an error element is two-fold. On the
one hand, morpheme insertion may be triggered in a certain licensing environment (28). On the other hand, a Vocabulary item that spells out an error element
may also be subject to phonological readjustment or suppletion in its post-error
environment(29).
(28) Morpheme insertion
a. ihr-e Schn-heit,
h, ihr-e Frisur
ist total schn
her-f beautiful-nmlz(f), er, her-f hairdo(f) is very beautiful
Her hairdo is very beautiful.
b. i ch fands ein besonders fest-lich-es, h, gelungen-es Fest
I foundit a.n particularly party-adj-n, er, successful-n party(n)
I think it was a particularly successful party.
Grammar as Processor
c.
in der
er wohn-t
in which he live-3.sg
d. welch-er
what-m
Schreib-er,
Quatsch, welch-er Idiot
write-nmlz(m), nonsense, what-m idiot(m)
schreib-t denn
so
was
write-3.sg mod.part such a.thing
dass der
Terror-ist,
h, dass der
Direkt-or
that the.m terror-nmlz(m), er, that the.m direct-nmlz(m)
die
the.f.acc
die
the
S8
S8
Flug-hafen
flight-port
fhr-t
go-3.sg
zum
to.the
b. d
er
Sprung,
h, der
Funke
spring-t
ber
the.m jump.nmlz(m), er, the.m spark(m) jump-3.sg over
It clicks (between them).
c.
auf ein-em
Stand,
on one-m.dat stand.nmlz(m),
auf ein-em
Bein
on one-n.dat leg(n)
mit Binde-strich
with connect-line(m)
VI.4 Combinedeffects
Following an error, two, sometimes three, of the mechanisms mentioned in Appendices VI.1, VI.2, and VI.3 may interact in yielding a grammatical utterance. These
mechanisms are, in the order of their application: morpheme insertion, feature
copy, context-sensitive spell-out, and phonological readjustment/suppletion be
it due to a morphosyntactic feature or the licensingenvironment.
(30) Morpheme insertion and gender copy
a.
er hat
ein-e
Erzhl-ung, h, ein-en Schwank
he have.3.sg a-f.acc tell-nmlz(f), er, a-m.acc tale(m)
aus sein-er
Jugend erzhl-t
from his-f.dat youth(f) tell-part
b. der
Tour-ismus,
die Ignoranz
der
Tour-ist-en
the.m tour-nmlz(m), the.f ignorance(f) of.the tour-nmlz(m)-pl
nimm-t
von Jahr zu Jahr zu
increase-3.sg from year to year particle
+ c.
sein Bedrf-nis
zu befriedig-en
his.n need-nmlz(n) to satisfy-inf
+ d. nerv-e
die
Nahr-ung,
nhr-e
den Nerv
nerve-imp the.f food-nmlz(f) feed-imp the.m nerve(m)
Feed the nerv!
+ e.
ich
I
an den
Glaub-e-n
aufersteh-en
in the.m.acc belief-nmlz(m)-acc resurrect-3.pl
an die
Aufersteh-ung
glaub-en
in the.f.acc resurrect-nmlz(f) believe-3.pl
s ie hat
mir ihr-e Gab-e,
h, ihr-e Nummer
she have.3.sg me her-f give-nmlz(f), er, her-f number(f)
nicht ge-geb-en
not part-give-part
Grammar as Processor
+ b. sein Verkuf-er,
h, sein Freund
hat
his.m sell-nmlz(m), er, his.m friend(m) have.3.sg
sein-en
Mercedes
verkauf-t
his-m.acc Mercedes(m) sell-part
b. will-st
du
den Gang, h, die Treppe runter-geh-en
want-2.sg you(sg) the.m aisle(m), er, the.f stairs(f) down-go-inf
Do you want to go down the stairs?
(33) Context-sensitive spell-out and phonological readjustment/suppletion
a.
b. im
Schwimm-bad knn-en
in.the swim-bath
can-pl
sich die
Bd-er,
refl the.pl bath-pl,
die
Jung-s
the.pl boy-pl
c.
i ch hab-e
I
have-1.sg
ein-en
Blick
ge-worf-en
a-m.acc glance(m) part-throw-part
d. da
war
there was
e.
h,
er,
richtig austob-en
really romp.about-inf
ein-en
Wurf
ge-blick-t
a-m.acc throw.nmlz(m) part-glance-part
der
Bruch
ge-bann-t
the.m break.nmlz(m) part-spell-part
der
Bann
ge-broch-en
the.m spell(m) part-break-part
And so the spell was broken.
Rauch-zg-e,
h, Rauch-wolke-n zieh-en Richtung Westen
smoke-drift.nmlz-pl, er, smoke-cloud-pl drift-3.pl direction West
Clouds of smoke are drifting westwards.
bist
du
so nett und leg-st
die
Bnk-e auf das
be.2.sg you(sg) so kind and put-2.sg the.pl bench-pl on the.n
Kissen
und leg-st
die
Kissen auf die Bank
pillow(n) and put-2.sg the.pl pillow.pl on the.f bench(f)
das ist
wirklich ein
ganz
this be.3.sg really
a.n.nom very
ein-e
ganz bl-e Klage
a-f.nom very bad-f charge(f)
klg-lich-es,
h,
charge-adj-n, er,
+ b. dies-er Sng-er,
h, dies-e Diva
this-m sing-nmlz(m), er, the-f diva(f)
sing-t
echt wie ein-e rostig-e Rassel
sing-3.sg really like a-f rusty-f rattle(f)
ich versuch-e
die
Folg-e
I tempt-1.sg the.f.acc follow-nmlz(f)
folg-e
der
Versuch-ung
follow-1.sg the.f.dat tempt-nmlz(f)
VI.5 Lexicalconstrual
In lexical construal, the Vocabulary is accessed again following a phonological
error, in order to retrieve an existing Vocabulary item that is phonologically similar to the non-word resulting from theerror.
(37) Lexical construal
a.
im
Netz
in.the.n net(n)
Grammar as Processor
b.
+ c. er wohn-t
in Berlin und studier-t
Benzin Medizin
he live-3.sg in Berlin and study-3.sg fuel
medicine
He lives in Berlin and studies medicine.
( expected outcome: Bedizin)
+ d. das kann
man
that can.3.sg one
sich
refl
an fnf
at five
Zimmer-n
room-pl
abzhl-en
count-inf
an fnf Finger-n
at five finger-pl
+ e. nimm dein-e Flgel aus
dem Steigbgel dein-e F-e
take your-pl wing.pl out.of the stirrup
your-pl foot-pl
Take your feet out of the stirrup.
( expected outcome: Fgel)
References
Abd-El-Jawad Hassam & Issam Abu-Salim. 1987. Slips of the tongue in Arabic and their theoretical implications. Language Sciences 9:145171.
Abney, Steven P. 1987. The English Noun Phrase in its Sentential Aspects. Ph.D. dissertation, MIT.
Albright, Adam. 2007. Cheap but not free: Commentary on Pfau Cheap repairs. In C.T.Schtze &
V.S.Ferreira (Eds), 3551.
Anderson, Stephen R. 1992. A-morphous Morphology. Cambridge: Cambridge University
Press.
Anwar, Mohamed S. 1979. Remarks on a collection of speech errors. International Journal of
Psycholinguistics 6:5972.
Antn-Mndez, Ins, Janet L.Nicol & Merrill F.Garrett. 2002. The relation between gender and
number agreement processing. Syntax 5:125.
Archangeli, Diana. 1988. Aspects of underspecification theory. Phonology 5:183207.
Arnaud, Pierre. 1999. Target-error resemblance in French word-substitution speech errors and
the mental lexicon. Applied Psycholinguistics 20:269287.
Aronoff, Mark. 1994. Morphology by Itself: Stems and Inflectional Classes. Cambridge, MA: MIT
Press.
Bahloul, Maher & Wayne Harbert. 1993. Agreement asymmetries in Arabic. In Proceedings
of the 11th West Coast Conference on Formal Linguistics, J.Mead (Ed.), 1531. Stanford,
CA: CSLI.
Baker, Mark C. 1985. The mirror principle and morphosyntactic explanation. Linguistic Inquiry
16:373415.
Bakker, Dik. 2005. Expression and agreement: some more arguments for the dynamic expression model. In Morphosyntactic Expression in Functional Grammar, C. de Groot &
K.Hengeveld (Eds), 140. Berlin: Mouton de Gruyter.
Bakker, Dik & Roland Pfau. 2008. Agreement in the noun phrase: the dynamic expression of
terms and what can go wrong. In The Noun Phrase in Functional Discourse Grammar,
D.Garca Velasco & J.Rijkhoff (Eds), 287320. Berlin: Mouton de Gruyter.
Barker, Jason, Janet Nicol & Merrill F.Garrett. 2001. Semantic factors in the production of number agreement. Journal of Psycholinguistic Research 30:91114.
Barner, David & Alan Bale. 2002. No nouns, no verbs: Psycholinguistic arguments in favor of
lexical underspecification. Lingua 112:771791.
Barner, David & Alan Bale. 2005. No nouns, no verbs? A rejoinder to Panagiotidis. Lingua
115:11691179.
Becker, Donald A. 1979. Speech error evidence for autosegmental levels. Linguistic Inquiry
10:165167.
Berg, Thomas. 1986. The monomorphematic status of diphthongs revisited. Phonetica
43:198205.
Berg, Thomas. 1987. The case against accommodation: Evidence from German speech error
data. Journal of Memory and Language 26:277299.
Berg, Thomas. 1988. Die Abbildung des Sprachproduktionsprozesses in einem Aktivationsflussmodell. Untersuchungen an deutschen und englischen Versprechern. Tbingen: Niemeyer.
Grammar as Processor
Berg, Thomas. 1991. Redundant feature coding in the mental lexicon. Linguistics 2:903925.
Berg, Thomas. 1992. Prelexical and postlexical features in language production. Applied Psycholinguistics 13:199235.
Berg, Thomas. 1993. The phoneme through a psycholinguists looking-glass. Theoretical Linguistics
19:3976.
Berg, Thomas. 1998. The resolution of number conflicts in English and German agreement
patterns. Linguistics 36:4170.
Berg, Thomas. 2007. A typology of suprasegmental structure. In C.T.Schtze & V.S.Ferreira
(Eds), 5373.
Berg, Thomas & Hassam Abd-El-Jawad. 1996. The unfolding of suprasegmental representations:
A cross-linguistic perspective. Journal of Linguistics 32:291324.
Berkum, Jos J.A. van. 1997. Syntactic processes in speech production: The retrieval of grammatical gender. Cognition 64:115152.
Berwick, Robert C. & Amy S.Weinberg. 1984. The Grammatical Basis of Linguistic Performance.
Cambridge, MA: MIT Press.
Bever, Thomas G. 1988. The psychological reality of grammar: A students-eye view of cognitive
science. In The Making of Cognitive Science: Essays in Honor of George A.Miller, W.Hirst
(Ed.), 112142. Cambridge: Cambridge University Press.
Bever, Thomas G. & Brian McElree. 1988. Empty categories access their antecedents during
comprehension. Linguistic Inquiry 19:3543.
Bever, Thomas G., Katherine Straub, Ken Shenkman, John Kim & Carol Carrithers. 1990. The
psychological reality of NP-trace. Proceedings of NELS 20:4660.
Bhatt, Rajesh. 2005. Long distance agreement in Hindi-Urdu. Natural Language & Linguistic
Theory 23:757807.
Bierwisch, Manfred. 1970. Fehler-Linguistik. Linguistic Inquiry 1:397414.
Bierwisch, Manfred. 1982. Linguistics and language error. In A.Cutler (Ed.), 2972 [English
translation of Bierwisch (1970)].
Bobaljik, Jonathan D. 2000. The ins and outs of contextual allomorphy. In Proceedings of the 1999
Maryland Mayfest on Morphology (University of Maryland Working Papers in Linguistics
10), K.K. Grohmann & C. Struijke (Eds), 3571. University of Maryland: Department of
Linguistics.
Bobaljik, Jonathan D. & Susi Wurmbrand. 2002. Notes on agreement in Itelmen. Linguistic
Discovery 1 (1). [online journal: http://linguistic-discovery.dartmouth.edu].
Bobaljik, Jonathan D. & Susi Wurmbrand. 2005. The domain of agreement. Natural Language &
Linguistic Theory 23:809865.
Bock, Kathryn, Sally Butterfield, Anne Cutler, J.Cooper Cutting, Kathleen M.Eberhard & Karin
R.Humphreys. 2006. Number agreement in British and American English: Disagreeing to
agree collectively. Language 82:64113.
Bock, Kathryn & J.Cooper Cutting. 1992. Regulating mental energy: Performance units in language production. Journal of Memory and Language 31:99127.
Bock, Kathryn & Kathleen M.Eberhard. 1993. Meaning, sound and syntax in English number
agreement. Language and Cognitive Processes 8:5799.
Bock, Kathryn, Kathleen M. Eberhard & J. Cooper Cutting. 2004. Producing number agreement: How pronouns equal verbs. Journal of Memory and Language 51:251278.
Bock, Kathryn, Kathleen M.Eberhard, J.Cooper Cutting, Antje S.Meyer & Herbert Schriefers.
2001. Some attractions of verb agreement. Cognitive Psychology 43:83128.
References
Bock, Kathryn & Carol A.Miller. 1991. Broken agreement. Cognitive Psychology 23:4593.
Bock, Kathryn, Janet Nicol & J. Cooper Cutting. 1999. The ties that bind: Creating number
agreement in speech. Journal of Memory and Language 40:330346.
Bonet, Eulalia. 1991. Morphology after Syntax. Pronominal Clitics in Romance Languages.
Ph.D. dissertation, MIT.
Boomer, Donald S. & John D.M.Laver. 1968. Slips of the tongue. British Journal of Disorders of
Communication 3:112 [reprinted in V.A.Fromkin (Ed.), 1973, 120131].
Branigan, Phil & Marguerite MacKenzie. 2002. Altruism, A-bar movement, and object agreement in Innu-aimn. Linguistic Inquiry 33:385407.
Bresnan, Joan. 1978. A realistic transformational grammar. In Linguistic Theory and Psychological
Reality, M.Halle, J.Bresnan & G.A.Miller (Eds), 159. Cambridge, MA: MIT Press.
Bresnan, Joan & Ronald M. Kaplan. 1984. Grammars as mental representations of language.
In Method and Tactics in Cognitive Science, W.Kintsch, J.R.Miller & P.G.Polson (Eds),
103135. Hillsdale, NJ: Lawrence Erlbaum.
Broecke, Marcel P.R. van den & Louis Goldstein. 1980. Consonant features in speech errors. In
V.A.Fromkin (Ed.), 4765.
Brown, Alan S. 1991. A review of the tip-of-the-tongue experience. Psychological Bulletin
109:204223.
Brown, Jason C. 2002. Active phonological units: Parsimony in sublexical errors. Proceedings of
the Western Conference on Linguistics (WECOL) 2001.
Brown, Jason C. 2004. Eliminating the segmental tier: Evidence from speech errors. Journal of
Psycholinguistic Research 33:97101.
Brown, Roger & David McNeill. 1966. The tip of the tongue phenomenon. Journal of Verbal
Learning and Verbal Behavior 5:325337.
Bruening, Benjamin. 2001. Syntax at the Edge: Cross-clausal Phenomena and the Syntax of
Passamaquoddy. Ph.D. dissertation, MIT.
Bring, Daniel & Katharina Hartmann. 1995. All right!. In Extraction and Extraposition, U.Lutz &
J.Pafel (Eds), 179211. Amsterdam: Benjamins.
Butterworth, Brian. 1982. Speech errors: Old data in search of new theories. In A.Cutler (Ed.),
73108.
Butterworth, Brian. 1989. Lexical access in speech production. In Lexical Representation and
Process, W.Marslen-Wilson (Ed.), 108135. Cambridge, MA: MIT Press.
Caramazza, Alfonso & Michele Miozzo. 1997. The relation between syntactic and phonological
knowledge in lexical access: Evidence from the tip-of-the-tongue phenomenon. Cognition
64:309343.
Caramazza, Alfonso & Michele Miozzo. 1998. More is not always better: A response to Roelofs,
Meyer, and Levelt. Cognition 69:231241.
Carstens, Vicky. 2000. Concord in Minimalist Theory. Linguistic Inquiry 31:319355.
Chanquoy, Lucile & Isabelle Negro. 1996. Subject-verb agreement errors in written productions:
A study of French children and adults. Journal of Psycholinguistic Research 25:553570.
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Chomsky, Noam. 1970. Remarks on nominalization. In Readings in English Transformational
Grammar, R.A.Jacobs & P.S.Rosenbaum (Eds), 184221. Waltham, MA: Ginn & Co.
Chomsky, Noam. 1973. Conditions on transformations. In Festschrift for Morris Halle,
S.R.Anderson & P.Kiparsky (Eds), 232286. New York: Holt, Rinehart & Winston.
Grammar as Processor
Chomsky, Noam. 1980. Rules and Representations. New York: Columbia University Press.
Chomsky, Noam. 1981. Lectures on Government and Binding. The Pisa Lectures. Dordrecht: Foris.
Chomsky, Noam. 1986. Knowledge of Language. Its Nature, Origin, and Use. New York: Praeger.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by Step: Essays in Minimalist
Syntax in Honor of Howard Lasnik, R.Martin, D.Michaels & J Uriagereka (Eds), 89155.
Cambridge, MA: MIT Press.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in Language, M.Kenstowicz
(Ed.), 152. Cambridge, MA; MIT Press.
Chomsky, Noam & Morris Halle. 1968. The Sound Pattern of English. New York: Harper & Row.
Churchland, Paul M. 1989. A Neurocomputational Perspective. The Nature of Mind and the
Structure of Science. Cambridge, MA: MIT Press.
Clark, Herbert H. & Jeffrey S.Begun. 1971. The semantics of sentence subjects. Language and
Speech 14:3446.
Clements, George N. & Samuel J. Keyser. 1983. CV Phonology: A Generative Theory of the
Syllable. Cambridge, MA: MIT Press.
Clements, George N. & Engin Sezer. 1982. Vowel and consonant disharmony in Turkish. In
The Structure of Phonological Representations (Part II), H.van der Hulst & N.Smith (eds.),
213255. Dordrecht: Foris.
Cloitre, Marylene & Thomas G.Bever. 1988. Linguistic anaphors, levels of representation, and
discourse. Language and Cognitive Processes 3:293322.
Cohen, Antonie. 1965. Versprekingen als verklappers van het proces van spreken en verstaan.
Forum der Letteren 6:175186.
Comrie, Bernhard. 1999. Grammatical gender systems: A linguists assessment. Journal of
Psycholinguistic Research 28:457466.
Coppock, Elizabeth. 2007. Alignment in syntactic blending. In Schtze, C.T. & V.S. Ferreira
(Eds), 233251.
Corbett, Greville G. 1991. Gender. Cambridge: Cambridge University Press.
Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press.
Corver, Norbert. 1991. Evidence for DegP. Proceedings of NELS 21:3347.
Corver, Norbert. 1997. The internal syntax of the Dutch extended adjectival projection. Natural
Language & Linguistic Theory 15:289368.
Cowan, Nelson, Martin D.Braine & Lewis A.Leavitt. 1985. The phonological and metaphonological representation of speech: Evidence from fluent backward talkers. Journal of Memory
and Language 24:679698.
Cutler, Anne. 1980a. Errors of stress and intonation. In V.A.Fromkin (Ed.), 6780.
Cutler, Anne. 1980b. La leon des lapsus. La Recherche 11:686692.
Cutler, Anne. 1981. The reliability of speech error data. Linguistics 19:561582.
Cutler, Anne (Ed.), 1982. Slips of the Tongue and Language Production. Amsterdam: Mouton.
Cutler, Anne. 1988. The perfect speech error. In Language, Speech and Mind. Studies in Honour
of Victoria A.Fromkin, L.M.Hyman & C.N.Li (Eds), 209223. London: Routledge.
Cutting, J. Cooper & Victor S. Ferreira. 1999. Semantic and phonological information flow
in the production lexicon. Journal of Experimental Psychology: Learning, Memory, and
Cognition 25:318344.
Dell, Gary S. 1986. A spreading-activation theory of retrieval in sentence production. Psychological Review 93:283321.
References
Dell, Gary S. 1988. The retrieval of phonological forms in production: Tests of predictions from
a connectionist model. Journal of Memory and Language 27:124142.
Dell, Gary S. & Cornell Juliano. 1991. Connectionist Approaches to the Production of Words.
Technical Report CS-9105 (University of Illinois, Institute for Advanced Science and
Technology, Urbana, IL).
Dell, Gary S. & Padraig G.OSeaghdha. 1992. Stages of lexical access in language production.
Cognition 42:287314.
Dell, Gary S. & Peter A.Reich. 1981. Stages in sentence production: An analysis of speech error
data. Journal of Verbal Learning and Verbal Behavior 20:611629.
Eberhard, Kathleen M. 1997. The marked effect of number on subject-verb agreement. Journal
of Memory and Language 36:147164.
Eberhard, Kathleen M. 1999. The accessability of conceptual number to the processes of subject-verb
agreement in English. Journal of Memory and Language 41:560578.
Eberhard, Kathleen M., J.Cooper Cutting & Kathryn Bock. 2005. Making syntax of sense: Number
agreement in sentence production. Psychological Review 112:531559.
Ellis, Andrew W. 1980. On the Freudian theory of speech errors. In V.A. Fromkin (Ed.),
123132.
Embick, David. 2000. Features, syntax, and categories in the Latin perfect. Linguistic Inquiry
31:185230.
Fay, David. 1980a. Performing transformations. In Perception and Production of Fluent Speech,
R.A.Cole (Ed.), 441468. Hillsdale, NJ: Lawrence Erlbaum.
Fay, David. 1980b. Transformational errors. In Fromkin, V.A. (Ed.), 111122.
Fay, David. 1982. Substitutions and splices: a study of sentence blends. In Slips of the Tongue and
Language Production, Cutler, A. (Ed.), 163196. Amsterdam: Mouton.
Fayol, Michel, Pierre Largy & Patrick Lemaire. 1994. When cognitive overload enhances
subject-verb agreement errors: A study in French written language. Quarterly Journal of
Experimental Psychology 47:437464.
Ferber, Rita. 1995. Reliability and validity of slip-of-the-tongue corpora: A methodological note.
Linguistics 33:11691190.
Ferreira, Victor S. & Karin R.Humphreys. 2001. Syntactic influences on lexical and morphological
processing in language production. Journal of Memory and Language 44:5280.
Fodor, Jerry A. 1968. Psychological Explanation. New York: Random House.
Fodor, Jerry A. & Thomas G. Bever. 1965. The psychological reality of linguistic segments.
Journal of Verbal Learning and Verbal Behavior 4:414420.
Fodor, Jerry A., Thomas G.Bever & Merrill F.Garrett. 1974. The Psychology of Language: An
Introduction to Psycholinguistics and Generative Grammar. New York: McGraw Hill.
Fodor, Jerry A. & Merrill F.Garrett. 1967. Some syntactic determinants of sentential complexity.
Perception and Psychophysics 2:289296.
Fodor, Jerry A., Merrill F. Garrett, Edward C.T. Walker & Cornelia H. Parkes. 1980. Against
definitions. Cognition 8:263267.
Fodor, Janet D., Jerry A.Fodor & Merrill F.Garrett. 1975. The psychological unreality of semantic
representations. Linguistic Inquiry 6:515531.
Foldvik, Arne K. 1979. Norwegian speech error data A source of evidence for linguistic
performance models. Nordic Journal of Linguistics 2:113122.
Foley, William A. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge
University Press.
Grammar as Processor
Forster, Kenneth I. & Ilmar Olbrei. 1973. Semantic heuristics and syntactic analysis. Cognition
2:319347.
Francis, Winthrop N. 1986. Proximity concord in English. Journal of English Linguistics
19:309318.
Franck, Julie, Ulrich H.Frauenfelder & Luigi Rizzi. 2007. A syntactic analysis of interference in
subject-verb agreement. In C.T.Schtze & V.S.Ferreira (Eds), 173190.
Franck, Julie, Glenda Lassi, Ulrich H.Frauenfelder & Luigi Rizzi. 2006. Agreement and movement: a syntactic analysis of attraction. Cognition 101:173216.
Franck, Julie, Gabriella Vigliocco & Janet Nicol. 2002. Subject-verb agreement errors in
French and English: The role of syntactic hierarchy. Language and Cognitive Processes
17:371404.
Freud, Sigmund. 1901/2000. Zur Psychopathologie des Alltagslebens. ber Vergessen, Versprechen,
Vergreifen, Aberglaube und Irrtum. Frankfurt/Main: Fischer.
Freud, Sigmund. 2005. The Psychopathology of Everyday Life. Stilwell, KS: Digireads.
Fromkin, Victoria A. 1968. Speculations on performance models. Journal of Linguistics 4:4768.
Fromkin, Victoria A. 1971. The non-anomalous nature of anomalous utterances. Language
47:2752.
Fromkin, Victoria A. 1973a. Introduction. In V.A.Fromkin (Ed.), 1145.
Fromkin, Victoria A. 1973b. Appendix. In V.A.Fromkin (Ed.), 243269.
Fromkin, Victoria A. (Ed.), 1973. Speech Errors as Linguistic Evidence. The Hague: Mouton.
Fromkin, Victoria A. (Ed.), 1980. Errors in Linguistic Performance. Slips of the Tongue, Ear, Pen,
and Hand. New York: Academic Press.
Fromkin, Victoria A. 1988. Grammatical aspects of speech errors. In Linguistics: The Cambridge
Survey. Vol.II: Linguistic Theory: Extensions and Implications, F.J.Newmeyer (Ed.), 117138.
Cambridge: Cambridge University Press.
Gandour, Jack. 1977. Counterfeit tones in the speech of Southern Thai bidialectals. Lingua
41:125143.
Garcia-Albea, Jos E., Susana del Viso & Jos M.Igoa. 1989. Movement errors and levels of
processing in sentence production. Journal of Psycholinguistic Research 18:145161.
Garrett, Merrill F. 1975. The analysis of sentence production. In Psychology of Learning and
Motivation, Vol.9, G.Bower (Ed.), 133177. New York: Academic Press.
Garrett, Merrill F. 1976. Syntactic processes in sentence production. In New Approaches to Language
Mechanisms, R.J.Wales & E.Walker (Eds), 231256. Amsterdam: North-Holland.
Garrett, Merrill F. 1980a. Levels of processing in sentence production. In Language Production,
Vol.1, B.Butterworth (Ed.), 177220. London: Academic Press.
Garrett, Merrill F. 1980b. The limits of accommodation: Arguments for independent processing
levels in sentence production. In V.A.Fromkin (Ed.), 263271.
Garrett, Merrill F. 1982. Remarks on the relation between language production and language
comprehension systems. In Neural Models of Language Processes, M.Arbib, D.Caplan &
J.Marshall (Eds), 209224. New York: Academic Press.
Garrett, Merrill F. 1988. Processes in language production. In Linguistics: The Cambridge Survey. Vol.
III: Language: Psychological and Biological Aspects, F.J. Newmeyer (Ed.), 6996. Cambridge:
Cambridge University Press.
Garrett, Merrill F. 1990. Sentence processing. In An Invitation to Cognitive Science. Vol.1: Language,
D.N.Osherson & H.Lasnik (Eds), 133176. Cambridge, MA: MIT Press.
Garrett, Merrill F. 1993. Errors and their relevance for models of language production. In Linguistic
Disorders and Pathologies: An International Handbook, G.Blanken, J.Dittmann, H.Grimm,
J.Marshall & C.Wallesch (Eds), 7292. Berlin: de Gruyter.
References
Grammar as Processor
Harley, Heidi & Rolf Noyer. 2003. Distributed Morphology. In The Second Glot International
State-of-the-article Book. The Latest in Linguistics, L.Cheng & R.Sybesma (Eds), 463496.
Berlin: Mouton de Gruyter.
Harley, Trevor A. 1984. A critique of top-down independent levels models of speech production:
Evidence from non-plan-internal speech errors. Cognitive Science 8:191219.
Harley, Trevor A. 1990. Environmental contamination of normal speech. Applied Psycholinguistics
11:4572.
Harley, Trevor A. 1993. Phonological activation of semantic competitors during lexical access in
speech production. Language and Cognitive Processes 8:291309.
Harley, Trevor A. 1995. The Psychology of Language. From Data to Theory. Hove: Lawrence
Erlbaum.
Harley, Trevor A. & Siobhan B.G.MacAndrew. 2001. Constraints upon word substitution speech
errors. Journal of Psycholinguistic Research 30:395418.
Hartmann, Katharina. 2000. Right Node Raising and Gapping: Interface Conditions on Prosodic
Deletion. Amsterdam: Benjamins.
Hartsuiker, Robert J., Ins Antn-Mndez & Marije van Zee. 2001. Object attraction in subject-verb
agreement construction. Journal of Memory and Language 45:546572.
Hartsuiker, Robert J., Herbert J. Schriefers, Kathryn Bock & Gerdien M. Kikstra. 2003.
Morphophonological influences on the construction of subject-verb agreement. Memory
& Cognition 31:13161326.
Haskell, Todd R. & Maryellen C.MacDonald. 2003. Conflicting cues and competition in subjectverb agreement. Journal of Memory and Language 48:760778.
Haspelmath, Martin. 1999. Long distance agreement in Godoberi (Daghestanian) complement
clauses. Folia Linguistica XXXIII/2:131151.
Henry, Alison. 1995. Belfast English and Standard English: Dialect Variation and Parameter
Setting. Oxford: Oxford University Press.
Hohenberger, Annette, Daniela Happ & Helen Leuninger. 2002. Modality-dependent aspects of
sign language production: Evidence from slips of the hands and their repairs in German Sign
Language. In Modality and Structure in Signed and Spoken Languages, R.P.Meier, K.A.Cormier & D.G.Quinto-Pozos (Eds), 112142. Cambridge: Cambridge University Press.
Hohenberger, Annette & Eva Waleschkowski. 2005. Language production errors as evidence for
language production processes. The Frankfurt corpora. In Linguistic Evidence. Empirical,
Theoretical and Computational Perspectives, S.Kepser & M.Reis (Eds), 285305. Berlin:
Mouton de Gruyter.
Hotopf, W.H.Norman. 1980. Semantic similarity as a factor in whole-word slips of the tongue.
In: V.A.Fromkin (Ed.), 97109.
Humphreys, Karen. 2002. The Production of Inflectional and Derivational Morphology.
Evidence from Elicited Speech Errors. Ph.D. dissertation, Beckman Institute, University of
Illinois at Urbana-Champaign.
Jackendoff, Ray. 1977. X-syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press.
Janda, Richard D. 1990. Frequency, markedness, and morphological change: On predicting the
spread of noun-plural -s in Modern High German and West Germanic. Proceedings of the
Seventh Eastern States Conference on Linguistics (ESCOL 7): 136153.
Jarvella, Robert J. 1971. Syntactic processing of connected speech. Journal of Verbal Learning
and Verbal Behavior 10:409416.
Jespersen, Otto. 1922. Language: Its Nature, Development, and Origin. London: Allen and Unwin.
Katz, Jerrold J. 1964. Mentalism in linguistics. Language 40:124137.
References
Katz, Jerrold J. 1970. Interpretative semantics vs. generative semantics. Foundations of Language
6:220259.
Katz, Jerrold J. 1972. Semantic Theory. New York: Harper & Row.
Kean, Mary-Louise. 1980. Grammatical representations and the description of language processing. In Biological Studies of Mental Processes, D. Caplan (Ed.), 239268. Cambridge,
MA: MIT Press.
Keller, Jrg, Annette Hohenberger & Helen Leuninger. 2003. Sign language production: Slips
of the hand and their repairs in German Sign Language. In Cross-linguistic Perspectives in
Sign Language Research. Selected Papers from TISLR 2000, A.Baker, B.van den Bogaerde &
O.Crasborn (Eds), 307333. Hamburg: Signum.
Kempen, Gerard & Eduard Hoenkamp 1987. An incremental procedural grammar for sentence
formulation. Cognitive Science 11:201258.
Kihm, Alain. 2005. Noun class, gender, and the lexicon-syntax-morphology interfaces: A
comparative study of Niger-Congo and Romance languages. In The Oxford handbook of
comparative syntax, G.Cinque & R.S.Kayne (Eds), 459512. Oxford: Oxford University
Press.
Klatt, Dennis H. 1981. Lexical representations for speech production and perception. In The
Cognitive Representation of Speech, T.Myers, J.Laver & J.Anderson (Eds), 1131. Amsterdam: North-Holland.
Klein, Monika & Helen Leuninger. 1988. Gestrtes und nicht gestrtes Sprachverhalten: Zur
Struktur lautlicher Fehlleistungen. Frankfurter Linguistische Forschungen 4:120.
Klein, Monika & Helen Leuninger. 1990. Gestrtes und nicht gestrtes Sprachverhalten: Zur
Analyse lautlicher Fehlleistungen im Rahmen nicht-linearer phonologischer Theorien.
Frankfurter Linguistische Forschungen 8:117.
Klima, Edward S. 1964. Negation in English. In The Structure of Language. Readings in the Philosophy of Language, J.Fodor & J.Katz (Eds), 246323. Englewood Cliffs, NJ: Prentice-Hall.
Konishi, Toshi. 1993. The semantics of grammatical gender: A cross-cultural study. Journal of
Psycholinguistic Research 22:519534.
Koopman, Hilda & Dominique Sportiche. 1991. The position of subjects. Lingua 85:211258.
Kpcke, Klaus-Michael. 1993. Schemata bei der Pluralbildung im Deutschen: Versuch einer kognitiven Morphologie. Tbingen: Narr.
Laka, Itziar. 1990. Negation in Syntax: On the Nature of Functional Categories and Projections.
Ph.D. dissertation, MIT.
Lakoff, George. 1971. On generative semantics. In Semantics. An Interdisciplinary Reader in
Philosophy, Linguistics and Psychology, D.D.Steinberg & L.A.Jacobovits (Eds), 232296.
Cambridge: Cambridge University Press.
Lamontagne, Greg & Lisa Travis. 1986. The case filter and the ECP. McGill Working Papers in
Linguistics 3:5175.
Lashley, Karl S. 1951. The problem of serial order in behavior. In Cerebral Mechanisms in Behavior,
L.A.Jeffress (Ed.), 112136. New York: Wiley.
Lasnik, Howard. 1995. Verbal morphology: Syntactic Structures meets the Minimalist Program.
In Evolution and Revolution in Linguistic Theory, H. Campos & P. Kempchinsky (Eds),
251275. Washington, DC: Georgetown University Press.
Laubstein, Ann S. 1987. Syllable structure: The speech error evidence. Canadian Journal of
Linguistics 32:339363.
Laubstein, Ann S. 1999. Word blends as sublexical substitutions. Canadian Journal of Linguistics
44:127148.
Grammar as Processor
Lechner, Winfried. 2001. Reduced and phrasal comparatives. Natural Language & Linguistic
Theory 19:683735.
Leuninger, Helen. 1987. Das ist wirklich ein dickes Stck: berlegungen zu einem Sprachproduktionsmodell. In Grammatik und Kognition (Linguistische Berichte Special Issue 1),
J.Bayer (Ed.), 2440. Opladen: Westdeutscher Verlag.
Leuninger, Helen. 1989. Neurolinguistik. Probleme, Paradigmen, Perspektiven. Opladen:
Westdeutscher Verlag.
Leuninger, Helen. 1993. Reden ist Schweigen, Silber ist Gold. Gesammelte Versprecher. Zrich:
Ammann.
Leuninger, Helen. 1996. Danke und Tsch frs Mitnehmen. Gesammelte Versprecher und eine
kleine Theorie ihrer Korrekturen. Zrich: Ammann.
Leuninger, Helen, Annette Hohenberger, Eva Waleschkowski, Elke Menges & Daniela Happ.
2004. The impact of modality on language production: Evidence from slips of the tongue
and hand. In Multidisciplinary Approaches to Language Production, T.Pechmann & C.Habel
(Eds), 219277. Berlin: Mouton de Gruyter.
Leuninger, Helen & Jrg Keller. 1994. Some remarks on representational aspects of language
production. In Linguistics and Cognitive Neuroscience. Theoretical and Empirical Studies on
Language Disorders, D.Hillert (Ed.), 83110. Opladen: Westdeutscher Verlag.
Levelt, Willem J.M. 1989. Speaking: From Intention to Articulation. Cambridge, MA: MIT Press.
Levelt, Willem J.M. 1992. Accessing words in speech production: Stages, processes and
representations. Cognition 42:122.
Levelt, Willem J.M. & Anne Cutler. 1983. Prosodic marking in speech repair. Journal of Semantics 2:205217.
Levelt, Willem J.M., Ardi Roelofs & Antje S.Meyer. 1999. A theory of lexical access in speech
production. Behavioral and Brain Sciences 22:175.
Liddell, Scott K. 1984. THINK and BELIEVE: Sequentiality in American Sign Language.
Language 60:372399.
Lieber, Rochelle. 1992. Deconstructing Morphology. Word Formation in Syntactic Theory.
Chicago: University of Chicago Press.
Linell, Per. 1979. Psychological Reality in Phonology. A Theoretical Study. Cambridge: Cambridge
University Press.
Lbel, Elisabeth. 1994. KP/DP-syntax: Interaction of case-marking with referential and nominal
features. Theoretical Linguistics 20:3770.
Lopez Cutrin, Belen & Gabriella Vigliocco. 2007. Grammatical gender is everywhere, even on
the tip of the mind: An investigation into retrieval failures. In: C.T.Schtze & V.S.Ferreira
(Eds), 253263.
Loving, Aretta & Howard McKaughan. 1973. Awa verbs II: The internal structure of dependent
verbs. In: McKaughan, H. (Ed.), 5664.
Lumsden, John S. 1992. Underspecification in grammatical and natural gender. Linguistic
Inquiry 23:469486.
MacDonald, Maryellen C. 1989. Priming effects from gaps to antecedents. Language and Cognitive Processes 4:3556.
MacKay, Donald G. 1979. Lexical insertion, inflection, and derivation: Creative processes in
word production. Journal of Psycholinguistic Research 8:477498.
MacWhinney, Brian, Jared Leinbach, Roman Taraban & Janet McDonald. 1989. Language learning:
Cues or rules?. Journal of Memory and Language 28:255277.
References
Mahajan, Anoop Kumar. 1989. Agreement and agreement phrases. In Functional Heads and
Clause Structure (MIT Working Papers in Linguistics 10), I.Laka & A.K.Mahajan (Eds),
217252. Cambridge, MA: MITWPL.
Mann, J.W. 1982. Effects of Number: Experimental Studies of the Grammatical Atmosphere Effect.
Johannesburg: Witwatersrand University Press.
Marantz, Alec. 1982. Re Reduplication. Linguistic Inquiry 13:435482.
Marantz, Alec. 1988. Clitics, morphological merger, and the mapping to phonological structure. In Theoretical Morphology, M.Hammond & M.Noonan (Eds), 253270. San Diego:
Academic Press.
Marantz, Alec. 1991. Case and licensing. Proceedings of the Eights Eastern States Conference on
Linguistics (ESCOL 8): 234253.
Marantz, Alec. 1995. A late note on late insertion. In Explorations in Generative Grammar: A
Festschrift for Dong-Whee Yang, Y.-S.Kim, K.-J.Lee, B.-C.Lee, H.-K.Yang & J.-Y. (Eds),
396413. Seoul: Hankuk Publishing.
Marantz, Alec. 1997. No escape from syntax: Dont try morphological analysis in the privacy of your own lexicon. University of Pennsylvania Working Papers in Linguistics
4.2:201225.
Marantz, Alec. 1998. Cat as a phrasal idiom: Consequences of late insertion in Distributed
Morphology. Unpublished manuscript, MIT.
Marantz, Alec. 2001. Words. Paper presented at the 20th West Coast Conference on Formal
Linguistics (WCCFL 20), University of Southern California, Los Angeles, February 2001.
Marcus, Gary F., Ursula Brinkmann, Harald Clahsen, Steven Pinker & Richard Wiese. 1995.
German inflection: The exception that proves the rule. Cognitive Psychology 29:189256.
Marslen-Wilson, William & Lorraine K.Tyler. 1980. The temporal structure of spoken language
understanding. Cognition 8:171.
Marvin, Tatjana. 2002. Topics in the Stress and Syntax of Words. Ph.D. dissertation, MIT.
Marx, Edeltrud. 1999. Gender processing in speech production: Evidence from German speech
errors. Journal of Psycholinguistic Research 28:601621.
Mathur, Gaurav. 2000. Verb Agreement as Alignment in Signed Languages. Ph.D. dissertation,
MIT.
Matthews, Peter H. 1997. The Concise Oxford Dictionary of Linguistics. Oxford: Oxford University
Press.
Matthews, Robert J. 1991. Psychological reality of grammars. In The Chomskyan Turn, A.Kasher
(Ed.), 182199. Oxford: Blackwell.
McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry
12:373418.
McElree, Brian & Thomas G. Bever. 1989. The psychological reality of linguistically defined
gaps. Journal of Psycholinguistic Research 18:2135.
McKaughan, Howard. 1973. Sequences of clauses in Tairora. In: H. McKaughan (Ed.),
588597.
McKaughan, Howard. (Ed.), 1973. The Languages of the Eastern Family of the East New Guinea
Highland Stock. Seattle: University of Washington Press.
Meara, Paul & Andrew W.Ellis. 1991. The psychological reality of deep and surface phonological
representations: evidence from speech errors in Welsh. Linguistics 19; 797804.
Mehler, Jacques. 1963. Some effects of grammatical transformations on the recall of English
sentences. Journal of Verbal Learning and Verbal Behavior 2:346351.
Grammar as Processor
Melinger, Alissa. 2003. Morphological structure in the lexical representation of prefixed words:
Evidence from speech errors. Language and Cognitive Processes 18:335362.
Meringer, Rudolf & Karl Mayer. 1895. Versprechen und Verlesen. Eine psychologisch-linguistische
Studie. Stuttgart: Gschen [reprinted as: Cutler, Anne & David Fay (Ed.), 1978. Classics in
Psycholinguistics (Vol. 2). Amsterdam: Benjamins].
Mester, R.Armin & Junko Ito. 1989. Feature predictability and underspecification: palatal prosody
in Japanese mimetics. Language: 258293.
Meyer, Antje S. & Kathryn Bock. 1999. Representations and processes in the production of pronouns: Some perspectives from Dutch. Journal of Memory and Language 41:281301.
Miller, George A. & Katherine A. McKean. 1964. A chronometric study of some relations
between sentences. Quarterly Journal of Experimental Psychology 16:297308.
Miozzo, Michele & Alfonso Caramazza. 1997. The retrieval of lexical-syntactic features in
tip-of-the-tongue states. Journal of Experimental Psychology: Learning, Memory, and
Cognition 23:114.
Miozzo, Michele & Alfonso Caramazza. 1999. The selection of determiners in noun phrase production. Journal of Experimental Psychology: Learning, Memory, and Cognition 25:907922.
Mohammad, Mohammad A. 2000. Word Order, Agreement and Pronominalization in Standard
and Palestinian Arabic. Amsterdam: Benjamins.
Mowrey, Richard A. & Ian R.A. MacKay. 1990. Phonological primitives: Electromyographic
speech error evidence. Journal of the Acoustic Society of America 88:12991312.
Mller, Gereon. 2004. Syncretism and iconicity in Icelandic noun declensions: a Distributed
Morphology approach. Yearbook of Morphology 2004:229271.
Newkirk, Don, Edward S. Klima, Carlene C. Pedersen & Ursula Bellugi. 1980. Linguistic
evidence from slips of the hand. In: V.A.Fromkin (Ed.), 165197.
Nicol, Janet L. 1995. Effects of clausal structure on subject-verb agreement errors. Journal of
Psycholinguistic Research 24:507516.
Nicol, Janet L. & Ins Antn-Mndez. In press. The effect of case-marking on subject-verb
agreement errors in English. In Time and Again: Theoretical Perspectives on Formal Linguistics. In Honor of D.Terrence Langendoen, W.Lewis, S.Farrar, S.Karimi & H.Harley (Eds),
Amsterdam: Benjamins.
Nicol, Janet L., Kenneth I.Forster & Csaba Veres. 1997. Subject-verb agreement processes in
comprehension. Journal of Memory and Language 36:569587.
Nooteboom, Sieb G. 1969. The tongue slips into patterns. In Nomen Society: Leiden Studies in
Linguistics and Phonetics, A.G.Sciarone, A.J. van Essen & A.A. van Raad (Eds), 114132.
The Hague: Mouton.
Nooteboom, Sieb G. 1980. Speaking and unspeaking: Detection and correction of phonological
and lexical errors in spontaneous speech. In: V.A.Fromkin (Ed.), 8795.
Noyer, Rolf. 1997. Features, Positions and Affixes in Autonomous Morphological Structure.
New York, NY: Garland Publishing [Revised version of 1992 MIT Ph.D. dissertation].
Noyer, Rolf. 1998. Impoverishment theory and morphosyntactic markedness. In Morphology
and Its Relation to Phonology and Syntax, S.G.Lapointe, D.K.Brentari & P.M.Farrell (Eds),
264285. Stanford: CSLI.
Ouhalla, Jamal. 1993. Subject-extraction, negation, and the anti-agreement effect. Natural
Language & Linguistic Theory 11:477518.
Panagiotidis, Phoevos. 2005. Against category-less roots in syntax and word learning: objections
to Barner & Bale (2002). Lingua 115:11811194.
References
Grammar as Processor
Roelofs, Ardi. 1997. A case for non-decomposition in conceptually driven word retrieval. Journal
of Psycholinguistic Research 26:3367.
Roelofs, Ardi, Antje S.Meyer & Willem J.M.Levelt 1998. A case for the lemma/lexeme distinction
in models of speaking: Comment on Caramazza & Miozzo 1997. Cognition 69:219230.
Rossi, Mario & Evelyne Peter Defare. 1995. Lapsus linguae: Word errors or phonological errors?.
International Journal of Psycholinguistics 11:538.
Rumelhart, David E. & John L. McClelland. (Eds), 1986. Parallel Distributed Processing.
Explorations in the Microstructure of Cognition. Vol.1: Foundations. Cambridge, MA: MIT
Press.
Safi-Stagni, Sabah. 1990. Slips of the tongue in Arabic. In Perspectives on Arabic Linguistics
I.Papers from the First Annual Symposium on Arabic Linguistics, M.Eid (Ed.), 271290.
Amsterdam: Benjamins.
Sandler, Wendy & Diane Lillo-Martin. 2006. Sign Languages and Linguistic Universals.
Cambridge: Cambridge University Press.
Schade, Ulrich. 1992. Konnektionismus. Zur Modellierung der Sprachproduktion. Opladen:
Westdeutscher Verlag.
Scherenberg, Michaele & Karl-Heinz Stier. 1990. Hessen la carte. Frankfurt: Eichborn.
Schriefers, Herbert. 1993. Syntactic processes in the production of noun phrases. Journal of
Experimental Psychology: Learning, Memory, and Cognition 19:841850.
Schriefers, Herbert & Jrg D.Jescheniak. 1999. Representation and processing of grammatical
gender in language production: A review. Journal of Psycholinguistic Research 28:575599.
Schtze, Carson T. & Victor S. Ferreira (Eds), 2007. The State of the Art in Speech Error
Research. Proceedings of the LSA Institute Workshop (MIT Working Papers in Linguistics
53). Cambridge, MA: MITWPL.
Schwichtenberg, Beate & Niels O. Schiller. 2004. Semantic gender assignment regularities in
German. Brain and Language 90:326337.
Scott, Graham K. 1978. The Fore Language of Papua New Guinea. Canberra: Pacific Linguistics
Monographs (Series B, No. 47)
Sevald, Christine A., Gary S.Dell & Jennifer S.Cole. 1995. Syllable structure in speech production:
Are syllables chunks or schemas?. Journal of Memory and Language 34:807820.
Shattuck-Hufnagel, Stefanie. 1979. Speech errors as evidence for a serial-ordering mechanism
in sentence production. In Sentence Processing: Psycholinguistic Studies Presented to Merrill
Garrett, W.E.Cooper & E.C.T.Walker (Eds), 295342. Hillsdale, NJ: Lawrence Erlbaum.
Siddiqi, Daniel. 2006. Minimize Exponence: Economy Effects on a Model of the Morphosyntactic
Component of Grammar. Ph.D. dissertation, University of Arizona.
Slevc, L.Robert, Liane Wardlow Lane & Victor S.Ferreira. 2007. Pronoun production: Word or
world knowledge? In: C.T. Schtze & V.S.Ferreira (Eds), 191203.
Slobin, Dan I. 1966. Grammatical transformations and sentence comprehension in childhood
and adulthood. Journal of Verbal Learning and Verbal Behavior 5:219227.
Sderpalm, Eva. 1979. Speech Errors in Normal and Pathological Speech. Lund: Gleerup.
Spencer, Andrew. 1991. Morphological Theory. An Introduction to Word Structure in Generative
Grammar. Oxford: Blackwell.
Stabler, Edward P. 1983. How are grammars represented?. The Behavioral and Brain Sciences
6:391421.
Stemberger, Joseph P. 1982a. Syntactic errors in speech. Journal of Psycholinguistic Research
11:313345.
References
Stemberger, Joseph P. 1982b. The nature of segments in the lexicon: Evidence from speech
errors. Lingua 56:235259.
Stemberger, Joseph P. 1983a. Inflectional malapropisms: Form-based errors in English morphology.
Linguistics 21:573602.
Stemberger, Joseph P. 1983b. Speech Errors and Theoretical Phonology. A Review. Bloomington,
IN: Indiana University Linguistics Club.
Stemberger, Joseph P. 1984. Length as a suprasegmental: Evidence from speech errors. Language
60:895913.
Stemberger, Joseph P. 1985. An interactive activation model of language production. In Progress
in the Psychology of Language, Vol.1, A.W. Ellis (Ed.), 143186. Hillsdale, NJ: Lawrence
Erlbaum.
Stemberger, Joseph P. 1989. Speech errors in early child language production. Journal of Memory
and Language 28:164188.
Stemberger, Joseph P. 1991. Radical underspecification in language production. Phonology
8:73112.
Stemberger, Joseph P. & Marshall Lewis. 1986. Reduplication in Ewe: Morphological accommodation to phonological errors. Phonology Yearbook 3:151160.
Stemberger, Joseph P. & Brian MacWhinney. 1986. Form-oriented inflectional errors in
language processing. Cognitive Psychology 18:329354.
Steriade, Donca. 1995. Underspecification and markedness. In The Handbook of Phonological
Theory, J.Goldsmith (Ed.), 114174. Oxford: Blackwell.
Stokoe, William C. 1960. Sign language structure: An outline of the communication systems
of the American deaf. Studies in Linguistics, Occasional Papers 8. University of Buffalo
[Re-issued 1978, Linstok Press, Silver Spring, MD].
Strang, Barbara M.H. 1966. Some features of S-V concord in present-day English. In English
Studies Today: Fourth Series, I. Cellini & G. Melchiori (Eds), 7387. Rome: Edizioni di
Storia e Letteratura.
Sturtevant, Edgar H. 1917. Linguistic Change. Chicago: University of Chicago Press.
Szabolcsi, Anna. 1994. The noun phrase. In The Syntactic Structure of Hungarian (Syntax and
Semantics 27), F.Kiefer & K.E.Kiss (Eds), 179274. San Diego: Academic Press.
Tent, Jan & John E.Clark. 1980. An experimental investigation into the perception of slips of the
tongue. Journal of Phonetics 8:317325.
Thiersch, Craig. 1978. Topics in German Syntax. Ph.D. dissertation, MIT.
Thompson, Ellen. 2004. Manipulation of verbal inflection in English and Spanish spontaneous
speech errors. Proceedings of the Western Conference on Linguistics (WECOL 2004)
Thornton, Robert & Maryellen C.MacDonald. 2003. Plausibility and grammatical agreement.
Journal of Memory and Language 48:740759.
Treiman, Rebecca. 1987. The internal structure of the syllable. In Linguistic Structure in
Language Processing, G.N. Carlson & M.K. Tanenhaus (Eds), 2752. Dordrecht:
Reidel.
Twain, Mark. 2000. The Awful German Language/Die schreckliche deutsche Sprache. Waltrop:
Manuscriptum.
Vangsnes, ystein A. 2001. On noun phrase architecture, referentiality, and article systems. Studia Linguistica 55:249299.
Vigliocco, Gabriella, Tiziana Antonini & Merrill F.Garrett. 1997. Grammatical gender is on the
tip of Italian tongues. Psychological Science 8:314317.
Grammar as Processor
Vigliocco, Gabriella, Brian Butterworth & Merrill F. Garrett. 1996. Subject-verb agreement
in Spanish and English: Differences in the role of conceptual constraints. Cognition
61:261298.
Vigliocco, Gabriella, Brian Butterworth & Carlo Semenza. 1995. Constructing subject-verb
agreement in speech: The role of semantic and morphological factors. Journal of Memory
and Language 34:186215.
Vigliocco, Gabriella & Julie Franck. 1999. When sex and syntax go hand in hand: Gender agreement in language production. Journal of Memory and Language 40:455478.
Vigliocco, Gabriella & Julie Franck. 2001. When sex affects syntax: Contextual influences in
sentence production. Journal of Memory and Language 45:368390.
Vigliocco, Gabriella, Robert J. Hartsuiker, Gonia Jarema & Herman H.J. Kolk. 1996. One or
more labels on the bottles? Notational concord in Dutch and French. Language and Cognitive Processes 11:407442.
Vigliocco, Gabriella & Janet Nicol. 1998. Separating hierarchical relations and word order in
language production: Is proximity concord syntactic or linear?. Cognition 68:1329.
Vigliocco, Gabriella, David P. Vinson, Peter Indefrey, Willem J.M. Levelt & Frauke Hellwig.
2004. Role of grammatical gender and semantics in German word production. Journal of
Experimental Psychology: Learning, Memory, and Cognition 30 (2): 483497.
Vigliocco, Gabriella, David P.Vinson, Randi C.Martin & Merrill F.Garrett. 1999. Is count and
mass information available when the noun is not? An investigation of tip of the tongue
states and anomia. Journal of Memory and Language 40:534558.
Vigliocco, Gabriella & Tiziana Zilli. 1999. Syntactic accuracy in sentence production: The case
of gender disagreement in Italian language-impaired and unimpaired speakers. Journal of
Psycholinguistic Research 28:623648.
Vincent, Alexander. 1973. Tairora verb structure. In: H.McKaughan (Ed.), 561587.
Waleschkowski, Eva. 2004. Morphological decomposability of concatenative and nonconcatenative word forms. Evidence from slip experiments. In Proceedings of Console XII:
229243 (available at: http://www.sole.leidenuniv.nl).
Wardlow Lane, Liane, L.Robert Slevc & Victor S.Ferreira. 2007. Exchanging elicits: Stem exchange
errors and syntactic category shifts. In: C.T. Schtze & V.S.Ferreira (Eds), 221232.
Wiegand, Dagmar. 1994. Die Verfgbarkeit lexikalischer Informationen whrend des Sprachproduktionsprozesses. Frankfurter Linguistische Forschungen 15:117.
Wiegand, Dagmar. 1996. Die Sprachplanung als modular organisierter Proze: Zur Berechnung
von Kontaminationen. Ph.D. dissertation. Johann Wolfgang Goethe University, Frankfurt/
Main (Frankfurter Linguistische Forschungen, Special Issue 4)
Wiese, Richard. 1996. The Phonology of German. Oxford: Oxford University Press.
Yu, Si-Taek. 1992. Unterspezifikation in der Phonologie des Deutschen. Tbingen: Niemeyer.
Zanuttini, Raffaella. 1997. Negation and Clausal Structure. A Comparative Study of Romance
Languages. Oxford: Oxford University Press.
Zubin, David A. & Klaus-Michael Kpcke. 1981. Gender: a less than arbitrary grammatical category. In Chicago Linguistic Society 17, R.Hendrick, C.Masek & M.Miller (Eds), 439449.
Chicago: CLS.
Zubin, David A. & Klaus-Michael Kpcke. 1986. Gender and folk taxonomy: The indexical relation between grammatical and lexical categorization. In Noun Classes and Categorization,
C.Craig (Ed.), 139180. Amsterdam: Benjamins.
Subject index
A
ablaut, 19, 39, 70, 78, 79,
206, 207, 209, 243, 248,
259, 287
acategoriality, acategorial
root, 6970, 75, 129, 248,
252, 253, 263, 285, 301,
306, 310
accommodation (also see
gender accommodation)
definition, 9, 18, 223,
230, 274
lexical (see lexical construal)
morphological, 226228,
230, 237238, 246250,
285, 298
morphophonological, 41,
58, 225226, 276278,
285, 298
morphosyntactic, 97, 98,
111113, 121125, 202, 205,
206, 228230, 231233,
236237, 261, 285, 298
of context, 1819, 224,
227, 228229, 231233,
237238, 246250, 273,
275, 298
of error element, 1819, 206,
224, 227228, 229230,
235, 236237, 242246,
274, 275, 298
phonological, 225, 274276,
285, 298
adjacency, 65, 73, 128, 139,
170, 171, 175, 259260,
271, 306
adjunction, 72, 73, 81, 105, 129,
167, 176, 178, 193, 197, 206,
210, 215, 264, 266
AGREE, 176177, 179180
AgrS node (see subject
agreement)
allomorph, allomorphy, 44,
239, 248
case
assignment, 15, 78, 112, 123,
155159, 160, 217219,
234, 268, 270271, 289,
291292, 303, 307, 310
mismatch (see
subcategorization error)
shift/stranding (see feature
shift/stranding)
cause, [cause], 71, 78, 80,
240241, 243, 245, 252,
269270, 272, 302
collectivity, 163165
compound, 13, 14, 39, 111, 115,
123, 266, 279, 288
comprehension, 27, 48, 55,
135, 305
conceptual level/stratum, 56,
59, 8889, 90, 9192,
101102, 117, 120, 204, 213,
289, 302, 304305, 306, 308
connectionism, 61
count/mass noun, 66, 83, 85,
9497, 152153
D
degree element/phrase, 241,
245, 255, 260, 268269,
270, 283
derivational morpheme
(also see morpheme
insertion), 20, 3940,
72, 75, 108109, 115116,
226227, 246250, 251,
257, 262263, 274, 278,
294, 310311
Derivational Theory of
Complexity (DTC), 4748
Dutch, 106, 136, 147148, 159,
165, 182, 225, 275
E
Ecuador, 184
Encyclopedia (List 3), 67, 83,
85, 8990, 91
Subject index
Ewe (Ghana), 4445
exchange
of determiners, 14
of phonological
features, 34, 37
of morphosyntactic
features, 17, 215, 217218,
219, 221, 229, 235
of phrases, 46, 203
of roots/stems/words, 1617,
19, 3941, 4244, 57,
9697, 111112, 202203,
204205, 206208,
214215, 217, 225226, 231,
236237, 242243, 245,
247, 252261, 267268,
270271, 283, 284, 290291,
294295, 296297, 307, 311
of segments, 23, 4, 7, 31, 32,
34, 37, 5758, 61, 224, 225,
226, 227, 228, 252253, 261,
275, 276, 279
of concepts, 9899
extraposition, 149, 169, 174,
192, 211
F
feature copy
gender, 13, 111113,
132133, 181185, 214215,
231232, 235, 254, 257,
268, 271, 273, 279, 288,
289, 294295, 307, 308,
310, 312
subject agreement, 74, 77,
128129, 138, 151, 153, 175,
176, 205, 232233, 271, 303
feature shift, 102, 220, 222,
306, 309
case, 17, 218219
gender, 215217, 221, 235, 288
negation, 211213
number, 17, 205206
person, 221, 235
tense, 207208, 209
feature stranding, 102, 220,
222, 234240, 273274,
298299, 309
case, 217218, 228, 235
gender, 214215
negation, 211
number, 1617, 42, 58,
203205, 237, 277, 294
person, 220
Subject index
List 3 (see Encyclopedia)
local licensing, 6971, 75,
78, 80, 9192, 129, 206,
209210, 240251,
254260, 262, 264266,
268270, 271272, 273274,
279, 280285, 287, 294,
298, 302, 303, 310, 312
Logical Form (LF), 65, 70, 83,
200, 201
M
Manjaku (Guinea-Bissau), 215
mass noun (see count/mass
noun)
mentalism, mentalistic
theory, 2526, 28, 30
merger, 65, 70, 7273, 113,
176, 213, 248, 264265,
266267, 268, 303
Minimalist Program, 64,
176181
Minimize Exponence, 263267
Mirror Principle, 72
morpheme insertion
(derivational), 1920, 66,
75, 227, 246250, 252, 255,
260, 262263, 267, 269,
271, 273, 279, 283, 290,
294, 303, 306, 310
Morphological Structure
(MS), 65, 6876, 79, 112,
120, 126, 153, 160, 167, 175,
201, 221, 246, 247, 249,
263264, 269, 271, 289,
292, 303, 307, 310, 312
N
natural gender (see semantic
gender)
nominalization, 19, 6970,
75, 109, 227, 228, 241,
246249, 250252, 257,
258, 262, 266, 270, 281
Norwegian, 4041
noun class, 188, 215
noun substitution
(see substitution)
numeral phrase
(NumP), 204205, 233,
264265
O
object agreement, 146, 186187
P
participial allomorphy
(see allomorph)
Passamaquoddy
(W Canada), 188, 190
perceptual bias, 8, 159
perseveration
of phonological feature, 3435
of morphosyntactic
feature, 17, 198, 206, 207,
216, 219, 222, 235, 306
of root/stem/word, 111, 237,
250251, 281
of segment(s), 31, 37, 225,
275, 278
semantic, 11, 9193, 98
phonological feature, 34, 36,
225226, 275, 307
Phonological Form (PF), 65,
70, 72, 7680, 8485, 91,
110, 149, 160, 201, 208,
232, 247, 249, 278, 280,
287, 303, 305, 311312
phonological substitution
(see substitution)
plural allomorphy
(see allomorph)
plural shift/stranding
(see feature shift/stranding)
positional level (see processing
level)
processing level
functional level, 5860, 62, 87,
121, 167, 252, 261, 301, 305
positional level, 5860, 62,
121, 253, 261, 301, 305
psychological reality, 5, 2530,
31, 35, 38, 45, 46, 4850, 52,
55, 62, 105, 166, 225, 276, 305
R
reduplication, 4445
repair (see accommodation)
Romanian, 118, 119
S
self-correction, self-repair, 1,
19, 34, 98, 158, 203, 219,
238, 267268, 283285,
291, 293294, 312
semantic anticipation/
perseveration
(see anticipation and
perseveration)
semantic (natural)
gender, 97100, 102, 108
semantic substitution
(see substitution)
shift of feature
(see feature shift)
slip of the hand, 2, 8, 35
Spanish, 108, 113114, 132, 136,
208, 215, 228229, 249
Spec-head relation, 128, 129,
144, 167, 176177, 179181
stem-internal change (also see
ablaut and umlaut), 79,
187, 206, 228, 284
stranding
of feature (see feature
stranding)
of morpheme, 38, 3945,
253, 256261, 279, 287,
296, 307
of suprasegmental
feature, 37
subcategorization error, 15,
219, 291292
subcategorization
requirement, 69, 240,
249, 287
subject agreement
feature copy
(see feature copy)
insertion/adjunction of
AgrS node, 74, 128129,
175, 193, 266, 271, 303
substitution, 113126, 282, 292
phonological (form-based)
substitution, 3, 67, 10,
5861, 115, 117118, 120,
122125, 280, 288, 297, 307,
310, 312
semantic (meaning-based)
substitution, 1011, 5861,
8690, 9596, 98, 112, 114,
117118, 120, 122125, 202,
212, 231, 280, 282, 306,
310, 313
suppletion, 244245, 250, 262,
273, 298
syllable, 3, 31, 32, 35, 36, 39, 45
syntactic (grammatical)
category (also see
acategoriality), 57, 68,
110, 252253, 256, 261,
284, 310
Subject index
syntactic path/distance, 133134,
138139, 140, 144, 168, 174,
181, 191
T
Tairora (Papua New
Guinea), 198
templatic morphology, 43
tense shift/stranding (see
feature shift/stranding)
Thai, 37
tip-of-the-tongue, 60,
9495, 117
trace, 4850, 174, 178179
Tsez (NE Caucasian), 188189,
199, 200
Turkish, 41, 72, 108, 226, 277
U
umlaut, 16, 17, 39, 42, 78, 80,
161, 203, 206, 227, 237, 245,
255, 270, 287, 293
underspecification
of gender, 109119, 120121,
231, 313
of phonological feature, 80,
269, 275277, 285, 298
of Vocabulary item, 6667,
77, 122, 215, 240241
W
well-formedness constraint
(see grammatical
constraint)
Welsh, 276
V
Vocabulary (List 2), 6667,
8990, 99, 110, 118, 121,
122, 215, 240, 277, 278279,
281, 299, 303, 307, 313
Z
zero-marking, zero affix, 16,
7677, 78, 79, 96, 161162,
164, 257, 264265, 267,
268, 294
110 Rothstein, Susan (ed.): Theoretical and Crosslinguistic Approaches to the Semantics of Aspect. 2008.
viii,453pp.
109 Chocano, Gema: Narrow Syntax and Phonological Form. Scrambling in the Germanic languages. 2007.
x,333pp.
108 Reuland, Eric, Tanmoy Bhattacharya and Giorgos Spathas (eds.): Argument Structure. 2007.
xviii,243pp.
107 Corver, Norbert and Jairo Nunes (eds.): The Copy Theory of Movement. 2007. vi,388pp.
106 Deh, Nicole and Yordanka Kavalova (eds.): Parentheticals. 2007. xii,314pp.
105 Haumann, Dagmar: Adverb Licensing and Clause Structure in English. 2007. ix,438pp.
104 Jeong, Youngmi: Applicatives. Structure and interpretation from a minimalist perspective. 2007.
vii,144pp.
103 Wurff, Wim van der (ed.): Imperative Clauses in Generative Grammar. Studies in honour of Frits
Beukema. 2007. viii,352pp.
102 Bayer, Josef, Tanmoy Bhattacharya and M.T. Hany Babu (eds.): Linguistic Theory and South
Asian Languages. Essays in honour of K. A. Jayaseelan. 2007. x,282pp.
101 Karimi, Simin, Vida Samiian and Wendy K. Wilkins (eds.): Phrasal and Clausal Architecture.
Syntactic derivation and interpretation. In honor of Joseph E. Emonds. 2007. vi,424pp.
100 Schwabe, Kerstin and Susanne Winkler (eds.): On Information Structure, Meaning and Form.
Generalizations across languages. 2007. vii,570pp.
99 Martnez-Gil, Fernando and Sonia Colina (eds.): Optimality-Theoretic Studies in Spanish
Phonology. 2007. viii,564pp.
98 Pires, Acrisio: The Minimalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv,188pp.
97 Hartmann, Jutta M. and Lszl Molnrfi (eds.): Comparative Studies in Germanic Syntax. From
Afrikaans to Zurich German. 2006. vi,332pp.
96 Lyngfelt, Benjamin and Torgrim Solstad (eds.): Demoting the Agent. Passive, middle and other
voice phenomena. 2006. x,333pp.
95 Vogeleer, Svetlana and Liliane Tasmowski (eds.): Non-definiteness and Plurality. 2006. vi,358pp.
94 Arche, Mara J.: Individuals in Time. Tense, aspect and the individual/stage distinction. 2006. xiv,281pp.
93 Progovac, Ljiljana, Kate Paesani, Eugenia Casielles and Ellen Barton (eds.): The Syntax of
Nonsententials. Multidisciplinary perspectives. 2006. x,372pp.
92 Boeckx, Cedric (ed.): Agreement Systems. 2006. ix,346pp.
91 Boeckx, Cedric (ed.): Minimalist Essays. 2006. xvi,399pp.
90 Dalmi, Grte: The Role of Agreement in Non-Finite Predication. 2005. xvi,222pp.
89 Velde, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select, Merge,
Copy and Match. 2006. x,385pp.
88 Mohr, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic
languages. 2005. viii,207pp.
87 Julien, Marit: Nominal Phrases from a Scandinavian Perspective. 2005. xvi,348pp.
86 Costa, Joo and Maria Cristina Figueiredo Silva (eds.): Studies on Agreement. 2006. vi,285pp.
85 Mikkelsen, Line: Copular Clauses. Specification, predication and equation. 2005. viii,210pp.
84 Pafel, Jrgen: Quantifier Scope in German. 2006. xvi,312pp.
83 Schweikert, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005.
xii,338pp.
82 Quinn, Heidi: The Distribution of Pronoun Case Forms in English. 2005. xii,409pp.
81 FuSS, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal
inflection. 2005. xii,336pp.
80 Burkhardt, Petra: The SyntaxDiscourse Interface. Representing and interpreting dependency. 2005.
xii,259pp.
79 Schmid, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv,251pp.
78 Dikken, Marcel den and Christina Tortora (eds.): The Function of Function Words and Functional
Categories. 2005. vii,292pp.
77 ztrk, Balkz: Case, Referentiality and Phrase Structure. 2005. x,268pp.
76 Stavrou, Melita and Arhonto Terzi (eds.): Advances in Greek Generative Syntax. In honor of Dimitra
Theophanopoulou-Kontou. 2005. viii,366pp.
75 Di Sciullo, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005.
xviii,398pp.
74 Heggie, Lorie and Francisco Ordez (eds.): Clitic and Affix Combinations. Theoretical perspectives.
2005. viii,390pp.
73 Carnie, Andrew, Heidi Harley and Sheila Ann Dooley (eds.): Verb First. On the syntax of verbinitial languages. 2005. xiv,434pp.
72 FuSS, Eric and Carola Trips (eds.): Diachronic Clues to Synchronic Grammar. 2004. viii,228pp.
71 Gelderen, Elly van: Grammaticalization as Economy. 2004. xvi,320pp.
70 Austin, Jennifer R., Stefan Engelberg and Gisa Rauh (eds.): Adverbials. The interplay between
meaning, context, and syntactic structure. 2004. x,346pp.
69 Kiss, Katalin . and Henk van Riemsdijk (eds.): Verb Clusters. A study of Hungarian, German and
Dutch. 2004. vi,514pp.
68 Breul, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and
intonational approach. 2004. x,432pp.
67 Mieska Tomi, Olga (ed.): Balkan Syntax and Semantics. 2004. xvi,499pp.
66 Grohmann, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003.
xvi,372pp.
65 Manninen, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. xii,275pp.
64 Boeckx, Cedric and Kleanthes K. Grohmann (eds.): Multiple Wh-Fronting. 2003. x,292pp.
63 Boeckx, Cedric: Islands and Chains. Resumption as stranding. 2003. xii,224pp.
62 Carnie, Andrew, Heidi Harley and MaryAnn Willie (eds.): Formal Approaches to Function in
Grammar. In honor of Eloise Jelinek. 2003. xii,378pp.
61 Schwabe, Kerstin and Susanne Winkler (eds.): The Interfaces. Deriving and interpreting omitted
structures. 2003. vi,403pp.
60 Trips, Carola: From OV to VO in Early Middle English. 2002. xiv,359pp.
59 Deh, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii,305pp.
58 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition.
2003. vi,309pp.
57 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. 2003.
vi,405pp.
56 Coene, Martine and Yves Dhulst (eds.): From NP to DP. Volume 2: The expression of possession in
noun phrases. 2003. x,295pp.
55 Coene, Martine and Yves Dhulst (eds.): From NP to DP. Volume 1: The syntax and semantics of noun
phrases. 2003. vi,362pp.
54 Baptista, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003.
xxii,294pp.(incl.CD-rom).
53 Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax.
Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 2627, 2000).
2002. xiv,407pp.
52 Simon, Horst J. and Heike Wiese (eds.): Pronouns Grammar and Representation. 2002. xii,294pp.
51 Gerlach, Birgit: Clitics between Syntax and Lexicon. 2002. xii,282pp.
50 Steinbach, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German.
2002. xii,340pp.
49 Alexiadou, Artemis (ed.): Theoretical Approaches to Universals. 2002. viii,319pp.
48 Alexiadou, Artemis, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Grtner
(eds.): Dimensions of Movement. From features to remnants. 2002. vi,345pp.
47 Barbiers, Sjef, Frits Beukema and Wim van der Wurff (eds.): Modality and its Interaction with the
Verbal System. 2002. x,290pp.
46 Panagiotidis, E. Phoevos: Pronouns, Clitics and Empty Nouns. Pronominality and licensing in syntax.
2002. x,214pp.
45 Abraham, Werner and C. Jan-Wouter Zwart (eds.): Issues in Formal German(ic) Typology. 2002.
xviii,336pp.
44 Taylan, Eser Erguvanl (ed.): The Verb in Turkish. 2002. xviii,267pp.
43 Featherston, Sam: Empty Categories in Sentence Processing. 2001. xvi,279pp.
42 Alexiadou, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x,233pp.
41 Zeller, Jochen: Particle Verbs and Local Domains. 2001. xii,325pp.
40 Hoeksema, Jack, Hotze Rullmann, Vctor Snchez-Valencia and Ton van der Wouden
(eds.): Perspectives on Negation and Polarity Items. 2001. xii,368pp.
39 Gelderen, Elly van: A History of English Reflexive Pronouns. Person, Self, and Interpretability. 2000.
xiv,279pp.
38 Meinunger, Andr: Syntactic Aspects of Topic and Comment. 2000. xii,247pp.
37 Lutz, Uli, Gereon Mller and Arnim von Stechow (eds.): Wh-Scope Marking. 2000. vi,483pp.
36 Gerlach, Birgit and Janet Grijzenhout (eds.): Clitics in Phonology, Morphology and Syntax. 2001.
xii,441pp.
35 Hrarsdttir, Thorbjrg: Word Order Change in Icelandic. From OV to VO. 2001. xiv,385pp.
34 Reuland, Eric (ed.): Arguments and Case. Explaining Burzios Generalization. 2000. xii,255pp.
33 Pusks, Genoveva: Word Order in Hungarian. The syntax of -positions. 2000. xvi,398pp.
32 Alexiadou, Artemis, Paul Law, Andr Meinunger and Chris Wilder (eds.): The Syntax of
Relative Clauses. 2000. vi,397pp.
31 Svenonius, Peter (ed.): The Derivation of VO and OV. 2000. vi,372pp.
30 Beukema, Frits and Marcel den Dikken (eds.): Clitic Phenomena in European Languages. 2000.
x,324pp.
29 Miyamoto, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 2000.
xiv,232pp.
28 Hermans, Ben and Marc van Oostendorp (eds.): The Derivational Residue in Phonological
Optimality Theory. 2000. viii,322pp.
27 Rika, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. x,206pp.
26 Ackema, Peter: Issues in Morphosyntax. 1999. viii,310pp.
25 Felser, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions.
1999. xiv,278pp.
24 Rebuschi, Georges and Laurice Tuller (eds.): The Grammar of Focus. 1999. vi,366pp.
23 Giannakidou, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. xvi,282pp.
22 Alexiadou, Artemis and Chris Wilder (eds.): Possessors, Predicates and Movement in the
Determiner Phrase. 1998. vi,388pp.
21 Klein, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. x,232pp.
20 Laenzlinger, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and
clause structure in Romance and Germanic. 1998. x,371pp.
19 Josefsson, Gunlg: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. ix,199pp.
18 Alexiadou, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. x,256pp.
17 Beermann, Dorothee, David LeBlanc and Henk van Riemsdijk (eds.): Rightward Movement.
1997. vi,410pp.
16 Liu, Feng-hsi: Scope and Specificity. 1997. viii,187pp.
15 Rohrbacher, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and prodrop. 1999. viii,296pp.
14 Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts (eds.): Materials on Left
Dislocation. 1997. viii,349pp.
13 Alexiadou, Artemis and T. Alan Hall (eds.): Studies on Universal Grammar and Typological
Variation. 1997. viii,252pp.
12 Abraham, Werner, Samuel David Epstein, Hskuldur Thrinsson and C. Jan-Wouter Zwart
(eds.): Minimal Ideas. Syntactic studies in the minimalist framework. 1996. xii,364pp.
11 Lutz, Uli and Jrgen Pafel (eds.): On Extraction and Extraposition in German. 1996. xii,315pp.
10 Cinque, Guglielmo and Giuliana Giusti (eds.): Advances in Roumanian Linguistics. 1995. xi,172pp.
9 Gelderen, Elly van: The Rise of Functional Categories. 1993. x,224pp.
8 Fanselow, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. xvii,232pp.
7 farl, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. xii,177pp.
6 Bhatt, Christa, Elisabeth Lbel and Claudia Maria Schmidt (eds.): Syntactic Phrase Structure
Phenomena in Noun Phrases and Sentences. 1989. ix,187pp.
5
4
3
2
1
Grewendorf, Gnther and Wolfgang Sternefeld (eds.): Scrambling and Barriers. 1990. vi,442pp.
Abraham, Werner and Sjaak De Meij (eds.): Topic, Focus and Configurationality. Papers from the 6th
Groningen Grammar Talks, Groningen, 1984. 1986. v,349pp.
Abraham, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen
Grammar Talks (3e Groninger Grammatikgesprche), Groningen, January 1981. 1983. vi,242pp.
Ehlich, Konrad and Jrgen Rehbein: Augenkommunikation. Methodenreflexion und Beispielanalyse.
1982. viii,150pp.Withmanyphotographicills.
Klappenbach, Ruth (19111977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus
den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beitrge von Helene MaligeKlappenbach. (Written in German). 1980. xxiii,313pp.