Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
in
Role
and
Reference
Grammar
Lilián Guerrero
Sergio Ibáñez Cerda
Valeria A. Belloro
editors
www.filologicas.unam.mx
www.etienda.unam.mx
iifl@servidor.unam.mx
ISBN 978-607-02-0659-7
Introduction. . . . . . . . . . . . . . . . . . . . . . 11
Robert D. Van Valin, Jr. Privileged syntactic arguments, pivots, and con-
trollers . . . . . . . . . . . . . . . . . . . . . . 45
Patrick Farrell. The Preposition with in Role and Reference Grammar . . 179
Brook Danielle Lillehaugen and John O. Foreman. Body parts and the en-
coding of thing and place in Zapotec . . . . . . . . . . . 203
Brian Nolan.The functions, semantics and syntax of the adjective in Irish 269
10 table of contents
Studies on Spanish
Carlos González Vergara. One rule to rule them all: logical structures for
spanish non reflexive se sentences . . . . . . . . . . . 361
Rosa Graciela Montes and Alaide Rodríguez Corte. Subject positioning and
thematic role in children’s narratives . . . . . . . . . . . . 405
In the last few decades, the field of linguistics has had a big development in
México. One of the driving impulses behind this development has come from
the research and teaching realized in the Instituto de Investigaciones Filológicas
within the Universidad Nacional Autónoma de México, especially in the Centro
de Lingüística Hispánica and the Seminario de Lenguas Indígenas. Although the
kind of research done in these two centers differ in terms of the languages under
study —Spanish, in the first case, and a variety of indigenous languages such as To-
tonac, Zapotec, Mixe, modern and classical Nahuatl, Yaqui and Cora, in the second
case—, they share the premise of functionalist approaches: language is essentially a
means of communication and has to be described and analyzed in terms of its func-
tion. However, despite the predominance of functionalist’s studies, the linguistic re-
search done in México has also maintained an open dialogue with different frame-
works and theoretical models, including those that have a clear formalist approach.
The syntactic theory offered by Role and Reference Grammar (RRG, Foley
and Van Valin 1984; Van Valin 1993, 2005; Van Valin and LaPolla 1997) easily
figures at the middle point between formalists and functionalists frameworks. In
contrast to the formal organization of clauses conceived as the result of a variety
of abstract syntactic operations, the view taken by practically most formal models
of syntax, the RRG puts forward the inclusion of language use and meaning as
a central part of the linguistic analysis. Different from those functional models
where grammar can be reduced to discourse, this theory conceives the language as
a structural system in the most traditional sense. RRG identifies itself as a ‘struc-
tural-functionalist’ (Van Valin 1993:1), since it provides a thorough formaliza-
tion for major syntactic issues while incorporating aspects of conceptual semantic
structure and discourse-pragmatic roles.
This theoretical perspective explicitly maintains that syntax cannot be properly
understood without an integrated account of the semantic and pragmatic func-
tions of language. In addition, it claims that a theory of clause structure should
capture all of the universal features of clauses without imposing features on lan-
guages for which there is no evidence, and should represent comparable structures
in different languages in comparable ways. In this vein, the theory posits three
11
12 introduction
The present volume was conceived during the process of organizing the 2007
Conference. The breath of their interests is reflected on the wide range of lan-
guages and grammatical aspects covered in this volume, which collects revised and
extended versions of selected papers from the conference.
The volume is divided into three parts: the first includes the presentations of
the plenary speakers, the second one addresses the study of specific phenomena
in languages of the world, and the third one includes contributions to particular
features in Spanish. In part one, Johanna Nichols examines the juncture-based
split alignment and aspectuality in Ingush (Nakh-Daghestanian), a language that
makes clear formal distinctions between nuclear and core cosubordination. First,
nuclear cosubordination is syntactically ergative while core cosubordination is
syntactically accusative. Second, the richness of aspect-like categories, both overt
and covert, is crucial to refine the inventory of nuclear operators. As a result
Ingush has crucial contributions to make to the general description and theory of
juncture. The paper by Robert D. Van Valin, Jr., revises one of the RRG cen-
tral contributions to the study of grammatical relations: the notion of ‘privileged
syntactic argument’ (PSA). Generally, PSAs have either of two functions: they can
be either the pivot or controller in a construction. The present analysis focuses on
unusual PSAs from English, Barai (Papua-New Guinea) and Liangshan Nuosu
(Southeast Asia) and claims that PSA is a construction-specific relation, not a
general feature of a language like ‘subject’. Ricardo Maldonado’s paper con-
cerns the status of middle voice as a non derived system. Whereas the conceptual
path transitive > reflexive > middle may be adequate for some languages, he argues
for a non-derived middle voice system and he illustrates his claim by describing
data from Yucatec Maya, Porhepecha, Toba, Otomí, Balinese, as well as a wide va-
riety of Spanish se constructions.
Part two contains nine papers all of which examine specific phenomena in lan-
guages of the world, two from a typological view and six in particular languages or
language families ( Japanese, German, English, Modern Irish, Tepehuan, Zapotec
and Uto-Aztecan). Although the subjects are diverse, the papers may be grouped
into two general topics: clause-internal syntax -exploring aspects of the informa-
tion structure, the status of adpositions, and verb morphology-, and studies beyond
the simple clause. In the first group of papers, the syntax-pragmatic interface in
simple clauses is approached in Japanese and German. Mitsu Shimojo discusses
six forms of morpho-syntactic coding for nominative and accusative arguments in
spoken Japanese and presents an analysis of their form-function relationships. The
14 introduction
analysis reveals a need for elaborate discourse representation requiring not only
givenness of referents but also two other discourse-functional properties: manners
of referent specification and saliency of information. Elke Diedrichsen explores
the interrelations between activation status and the choice of RP in a complex
German sentence from natural context. The pragmatic conditions for the choice
of referring expressions are included in the semantics-to-syntax-linking, and the
generalizations provided by an implicational feature hierarchy are put to use for
the description of case marking in German.
The next two papers studies the status of adpositions in particular construc-
tions. Patrick Farell explores the complexity of the English preposition with and
its unrelated uses in clause-modifying PPs. The preposition is considered to be
a marker of a scenario wherein a nominal phrase that is in contention for one of
the semantic macroroles fails to be linked to either. The claim is that with has a
locative central sense (i.e., togetherness in a place), manifested and a bundle of
related having senses that represent more specific, non-locative kinds of together-
ness. Based on data from two variants of Zapotec (Otomanguean), Brook Lille-
haugen and John O. Foreman investigate the semantic characteristics of body
part (BP) terms, which can occur in many types of phrases including referential
phrases, locative phrases, and motion phrases. The authors argue that Zapotec BP
phrases can encode thing and place, as two separate semantic functions. The
data also calls for an expansion and reconsideration of the cross-linguistic types of
prepositional phrases, especially those marking adjunct-arguments.
Dealing with verb morphology, Carmen Conti provides a typological classifi-
cation of benefactive applicatives taking into account the notion of semantic-tran-
sitivity of the verb based on the number of Macroroles, as well as the undergoer
assignment. She demonstrates that the Macrorole-transitivity of the verb stem
is a useful criterion to classify benefactive applicatives across languages. A rep-
resentation of benefactive applicative affixes in a syntactic template is proposed.
By exploring the rich verb morphology in Tlachichilco Tepehua (Totonacan), Jim
K. Watters focuses on the affixes arrangement in relation to the RRG claim in
which the order of verbal affixes with reference to the verb reflects the scope rela-
tions of the operators they express. It is found that morphological processes in-
volved in the lexical vs. syntactic encoding of arguments and adverbial affixes par-
allel syntactic levels and scope relations, and also an apparent counter-example is
discussed and compared with similar claims in other theoretical approaches. In his
paper on the functions, semantics and syntax of adjective in Modern Irish, Brian
introduction 15
Nolan examines the referential and attributive predications, as well as other ad-
jectival structures. Focusing in their morphological properties, he studies the posi-
tion of adjectives within the hierarchical structure of the noun phrase, connected
to certain features in the qualia structure of the head nominal, as well as to the
layered structure of the noun phrase as found in RRG.
Beyond the simple clause, Ranko Matasović proposes a new typology of con-
trol constructions based on an unusual case-marking pattern in obligatory con-
trol constructions in Kabardian (�����������������������������������������������
NW Caucasian�����������������������������������
). Based on three types of control
verbs (intransitives, actor and undergoer control) and in languages with both, ac-
cusative and ergative clause alignment, the author claims that the specification of
the case assignment pattern in control constructions should be added to the other
rules in the linking algorithm in core junctures. Finally, Lilián Guerrero aims
to sort out the difficulties of determining the degree of semantic cohesion among
the matrix predicate and alternative complement structures. She analyses propo-
sitional attitude and cognition predicates in a number of Uto-Aztecan languages,
and concludes that the facts regarding alternative complements for one and the
same predicate led to the revision of one of the semantic sub-hierarchies, the one
related to the participant’s mental disposition.
Part three contains eight papers all of which study particular aspects of Span-
ish. The first group can be divided into two sections: two papers exploring specific
syntactic manifestation based on the lexical aspectual properties of predicates, and
two analyzing aspect as a grammatical category. Sergio Bogard examines the
syntactic and semantic properties of bare NPs on Spanish atelic activity sentences.
He suggests that the entity formalized by the bare NP is not part of the argument
structure of the verb hence it is not a direct object but a ‘pseudo-object’. As a re-
sult, atelic activity constructions are M-intransitive. Carlos González Vergara
studies four types of non-reflexive se constructions: intrinsic se sentences, passive-
reflexives, impersonal-reflexive, and middle sentences. He proposes that the appli-
cation of one and the same simple lexical rule, which states the unspecification of
the highest ranking argument of the logical structure, can explain these sentences
together with factors such as the Aktionsarts, the lexical features of the undergoer
argument, the information structure of the sentence and the operator projection.
Within a more lexical perspective, Rocío Jiménez-Briones elaborates the
RRG logical structure for the Spanish verbs of feeling and proposes an enriched
representations of lexical constructional templates. Unlike logical structures, her
templates codify grammatically salient features (external variables) and semantic
16 introduction
The editors
México, 2009
References
Nolan, B. 2004. Role and Reference Grammar. Linguistic theory and practice: de-
scription, implementation and processing. N. Brian (ed). Dublin: Institute of
Technology Blanchardstown. Available at RRG’s web page.
Van Valin, Robert D., Jr. 1993. Advanced in Role and Reference Grammar. Am-
sterdam: John Benjamin.
— 2001. An introduction to syntax. Cambridge: Cambridge University Press.
— 2005. Exploring the Syntax and Semantic Interface. Cambridge: Cambridge
University Press.
— 2008. Investigations of the syntax-semantics-pragmatics interface. Amsterdam:
John Benjamins.
Van Valin, R., and R. J. LaPolla. 1997. Syntax. Structure, meaning and function.
Cambridge: Cambridge University Press.
Zeitoun, E., and R. Van Valin. 2007. Language and Linguistics. Special issue:
Role and Reference Grammar in Taiwan 8 (1). Taiwan: Institute of Linguis-
tics, Academia Sinica.
Keynote speakers’ contributions
Juncture-based split alignment
and aspectuality in Ingush
Johanna Nichols
University of California at Berkeley
1. Introduction
When Foley & Van Valin 1984 came out I had been working seriously on Ingush
for three years and was grappling with its exuberant and varied clause chaining, in
particular trying to square with the day’s theories of clause combining the fact that
some converbial clauses of Ingush allowed overt reflexive pronouns for coreferen-
tial subjects while others did not, that some converbial clauses were in the scope of
main-clause negation and others were not, that some converbs allowed different-
subject reference and others did not, and so on. Converbs seemed to be a single
morphosyntactic class and the chaining constructions they formed appeared to be
a single syntactic type despite considerable semantic variety. The RRG account of
juncture and nexus clarified the analysis considerably, neatly breaking converbial
clauses down into subordinate vs. cosubordinate with peripheral vs. core juncture
respectively.
Much more has become clear since then. A very large corpus of spoken Ingush
is being collected, transcribed, and annotated (Nichols 1997ff., Sprouse 1997ff.).
The total annotated so far is approaching 100,000 words, not yet a large corpus
but useful for morphosyntactic research. The morphosyntax of Ingush core co-
subordinate clauses, including their marking by verb reduplication and a cross-
The early stages of this project were supported in large part by NSF grant 96-16448.
Collection, transcription, and annotation of texts have been made possible since 2002 by
the Max Planck Institute for Evolutionary Anthropology, Leipzig. The Institute for Slavic,
East European, and Eurasian Studies, University of California, Berkeley, has provided ad-
ministrative support and the Committee on Research, University of California, Berkeley
some further funding. Ronald L. Sprouse has served as technical director of the entire proj-
ect. I thank Sultan Mereshkov for work on the texts and their vocabulary and ethnographic
content.
21
22 juncture-based split alignment and aspectuality in ingush
linguistically rare clitic particle, is well understood (Peterson 2001, Conathan &
Good 2000, Good 2003). RRG work on clause combining has undergone refine-
ment (e.g. Van Valin 1993, Van Valin & LaPolla 1997). A distinct nuclear cosub-
ordinate type has come to light. This paper will use the Ingush corpus and recent
work on clause combining to flesh out the description of cosubordinate clauses in
Ingush. There are three important grammatical points to be made here: Ingush
cosubordination can be considered a type of serialization; there is a clear formal
distinction between nuclear and core cosubordination, with most of the expected
differences in what is and is not shared; and the alignments of the two are differ-
ent, argument sharing in the nuclear type being ergative and that in the core type
accusative.
Ingush is a language of the Nakh-Daghestanian, or East Caucasian or North-
east Caucasian, family traditionally spoken in and near what is now the Republic
of Ingushetia in Russia. Ingush is morphologically ergative, dependent-marking,
mostly head-final, with complex consonant and vowel systems. Its lexicon makes
extensive use of light verb constructions. It has noun genders marked by alterna-
tion of the root-initial consonant on some verbs and some adjectives; the gender
markers are v-, j-, d-, and b-, interlinearized V, J, D, and B here. Gender agree-
ment is ergative in alignment. There is no person-number agreement. The spell-
ing system used here is an all-Latin, diacritic-free approximately phonemic practi-
cal transcription devised for the U.C. Berkeley Ingush Project and used in Nichols
1pin: first person plural inclusive (personal pronoun); 3s: third person singular (personal
pronoun); abl: ablative; all: allative case; adv: adverb (form of nouns); cnd: conditional;
cpz: complementizer; cs: causative; csind: indirect causative; cvant: anterior converb (co-
subordination); cvirr: irrealis converb (subordination); cvseq: sequential converb (cosub-
ordination); cvsim: simultaneous converb (cosubordination, subordination); cvtemp: tem-
poral converb (subordination); dat: dative case; dem: demonstrative; dx: deictic prefix:
hwa- proximal, dwa- distal; erg: ergative case; foc: emph: emphatic clitic or particle; fo-
cus (particle or root consonant gemination); fut: future; gen: genitive case; impf: imper-
fect; impv: imperative; inf: infinitive; ins: instrumental; lat: lative case (directional; some
indirect objects; some adverbial functions); lv: light verb; nw: nonwitnessed tense; obl:
oblique case (in adjectival paradigm); pl: plural; plc: pluractional; pot: potential; ppft:
pluperfect; progr: progressive; prs: present tense; quot: quotative; red: reduplicate (of
verb root); rfl: reflexive; wp: witnessed past; s, a, o, g, t: syntactic roles (g, t: respectively
more goal-like and more theme-like arguments of ditransitive verb).
johanna nichols 23
2004 and other publications. Single vowels have continental values (y roughly as in
Welsh or Polish; a is a central vowel like that of English cup); double vowels are
long; other vowel sequences (uo, ea, etc.) are diphthongs. Consonant digraphs in-
clude ch, sh, zh as in English; c [ts]; gh voiced uvular fricative; ‘ (apostrophe) glot-
tal stop; t’, c’, etc. ejectives; w pharyngeal phoneme (epiglottalized glottal stop) or
(after consonants) pharyngealization, a vocalic or syllabic phonation type. (The
official orthography of Ingush uses Cyrillic as is mandatory in Russia, and is not
phonemic).
cv / cv
In examples, a hyphen indicates easily segmentable affixal boundaries. A period is the
boundary after a gender marker. In interlinears, a period marks less easily segmentable af-
fixes (not segmented in the Ingush line). In the Ingush gender system, male humans are
always V and female humans always J, and this (plus context) informs glosses ‘he’ and ‘she’.
Examples from texts have text number or title indications; others are elicited.
24 juncture-based split alignment and aspectuality in ingush
Simultaneous converb: suffix -zh on present stem. The converb’s action over-
laps with or frames that of the main verb. Interlinear abbreviation: cvsim.
Anterior converb: suffix -aa, -na and other allomorphs on perfect stem. The
action is perfective and sequenced relative to the main verb (usually the con-
verb precedes the main verb in time). Interlinear: cvant.
Sequential converb: suffix -ie on present or perfect stem. Meaning as for an-
terior converb, but used with main verbs in the imperfect tense, the generic
present tense, iterative Aktionsart, and imperative. Interlinear: cvsseq.
Separate assertion in English translations is difficult where an Ingush serialized con-
struction is lexicalized and equivalent to a single English verb, as in (9), (10), (18), and others
below. The semantics of assertion still applies to both components, but this is done with a
single lexeme in English. Many such examples involve conflation of manner and motion
in English, as in (9).
26 juncture-based split alignment and aspectuality in ingush
The simultaneous converb can also have time adverbial subordinating func-
tions, as in (1) above, and both anterior and simultaneous converbs have some
complementation functions, but their most frequent use is in serialization.
The differences between nuclear and core serialization are their different re-
quirements of sharing, different case constraints and alignment in argument shar-
ing, and partly different formal marking.
3. Nuclear serialization
These examples all have intransitive verbs and the two clauses share their sub-
ject. The following sections show that nuclear serialization also involves shared
The anterior and simultaneous converbs also function as participles, modifying nouns,
and serve as bases for tenses using auxiliaries. The sequential converb is mostly homopho-
nous to the irrealis converb used for conditional and concessive subordination.
The second clause is core serialized to a following clause not shown here.
johanna nichols 27
non-subjects and shared tense and time reference, mood, illocutionary force, num-
ber, polarity, and aspectual properties. The alignment of sharing is syntactically
ergative, with a very strong case constraint: nominative arguments must be shared,
and nearly all verbs found in nuclear serialization in a text survey have a nomina-
tive argument (which is shared).
Ingush case paradigms are morphologically ergative, but I use nominative rather than
absolutive as the term for the case marking S/O, since there are enough splits and verbs
with exceptional case government that no term can claim to cover the functions.
In Ingush examples and interlinears, however, the equals sign marks clitic boundaries
as is usual.
For syntactic roles I use S, A, and O following Dixon 1979, and also G (mnemonically
‘goal’) for the less patient-like and most goal-like object of a ditransitive and T for the more
patient-like, less goal-like object of a ditransitive. O is then object only of a monotransitive.
For more on this system of annotation and the level it represents see Bickel & Nichols (in
press).
28 juncture-based split alignment and aspectuality in ingush
Also frequent are transitive clauses sharing both A and O [A=A, O=O], shown
in (15)-(17). The O is most often topical and the evident discourse pivot, suggest-
ing that the [O=O] sharing is the crucial one. Sometimes the A might conceivably
not be shared, e.g. (17) where it is possible that one person might put something
on another’s head for that person to carry.
Other attested patterns are one transitive and one intransitive clause, sharing
[O=S]:
10
The second clause is core serialized to the following one.
johanna nichols 29
In all such examples the overt token of the shared argument takes its case from
the second verb, i.e. it is nominative as yz is in (20).
I surveyed nuclear serialization in a set of texts and other naturally occurring
examples (lexical entries, phrases volunteered by speakers as opposed to elicited
by me) and counted the frequencies of different sharings. The results are shown
in Table 1. The main generalizations to emerge from the survey are, first, that in
nuclear serialization the second clause is usually intransitive. Second, the case of
the shared argument (or one of the shared arguments in [A=O A=O] sharing) is
always nominative in the second clause; there are no exceptions. Third, a large ma-
jority (81%) of the examples have ergative alignment in their sharings, with S or
O shared as S or O. Taken together, these patterns show that a nominative can be
shared with any case and non-nominatives can be shared only with a nominative
(except in [A=A O=O] double sharing, where two ergatives as well as two nomi-
natives are shared). Thus Ingush has a very strong morphological case constraint
on sharing, and that case constraint is ergative in alignment.
Given that converb choice is largely a matter of aspect, is aspect chosen indepen-
dently in the two serialized verbs? Table 2 gives the frequencies of the two aspects
in nuclear serialization in the same survey subcorpus. These results indicate that
the aspects combine freely, i.e. grammatical aspect is not shared. Perfectives are
preferred overall, possibly because the texts surveyed are narrative and the last verb
in a serialization construction tends to move the action forward. Important and
diagnostic of nuclear serialization is that when both verbs are perfective and nei-
ther has a result state, as in (10), they are not sequenced in time but are simultane-
ous or at least non-distinct. The lack of sequencing shows that nuclear serial verbs
do not have independent time reference.
The Aktionsart system of Ingush is not fully analyzed yet, but some preliminary
conclusions can be drawn about Aktionsart sharing in nuclear serialization. The
combinations found in examples used here are shown in Table 3. Various combi-
nations occur, and there is evidently no requirement that Aktionsart be shared.
V1 V2 Example
telic telic (12) viena t’yqeachaav ‘arrived here’
telic progressive (13) jiexaa jaaghar ‘sat ruined’
ingressive ingressive (23) hwabeidda hwabeaxkar ‘came running’
ingressive durative (21) ghattaa lel ‘flies’(‘flown goes’)
durative ingressive (9) udazh dwa-t’ehwa-v.axaa ‘running set off ’
Another such example is (18). Examples where the converb has a result state
and the finite verb is perfective are (10), (17), (19). If these examples are typical
(and I believe they are), it is resultant states, extents, and durations that must be
shared, provided they are part of the aspect and Aktionsart of the verbs. There are
examples where the first verb lacks a result state but is an activity and the second is
perfective and therefore lacks appreciable duration, and with these the first verb is
not literally durative but adds a manner component to the second. An example of
this configuration is (14) above.
23 Ingush verbs make a distinction of singular vs. plural S/O, marked chiefly by
vocalic and/or consonantal ablaut. The survey subcorpus yields one example of
nuclear serialization where both verbs are plural:
There are 19 examples where both are singular, e.g. jedda jexaai ‘ran away’ in
(1). In the remaining examples only one of the two verbs distinguishes argument
number, and in such examples the number-marking verbs are about equally fre-
quent as converb and finite verb. There are no instances of discrepant number, i.e.
singular + plural or plural + singular. It is hard to think of a context where this
might be possible, given that both argument number and the argument sharing of
serialization are ergative in alignment. It could happen with A=S sharing, e.g. (a
constructed example):
But A=S sharing is not frequent in nuclear serialization (see again Table 1)
and there happen to be no examples of it where both verbs distinguish argument
number.
johanna nichols 33
Ingush verbs of motion and transfer may and often do take one of the deictic
prefixes dwa- ‘there, away from speaker’ and hwa- ‘here, to speaker’. Syntactically,
the deictic center is goal or indirect object, and in narrative the orientation can be
that of the narrator, a narrative participant whose viewpoint the narrator takes, or
a speaker in the narrative regardless of viewpoint. Second person is aligned with
first person if the subject is third, but with third person if the subject is first. Deictic
prefixes therefore have a good deal of functional load and serve as the only form
of person marking on Ingush verbs. Not all verbs involve motion, goals, or indi-
rect objects, so not all verbs can take deictic prefixes, and the prefixes are gener-
ally optional even where their conditions are met. It seems that deixis, if overt,
must be identical in nuclear serialization: there are four examples like (23) above
where both verbs have the same deictic prefix and none where they have different
prefixes. Since in these four examples the deictic prefix is overt in both clauses,
this seems to reflect an identity constraint rather than sharing. In most examples
with deixis only one of the verbs has a deictic prefix, usually the second verb as in
(9), (11), (14), (16), (17) above. Usually the other verb has no goal or indirect ob-
ject and could not ordinarily take a deictic prefix, so in that clause deixis is lack-
ing rather than shared.
In nuclear serialization negation is marked only on the main verb. Either the
main verb or the converb can be in the scope of negation, depending on the con-
text, and most often both are in the scope of negation. The same is true of inter-
rogation. (25) shows a questioned nuclear serialization construction with nega-
tive answers, and the two possible answers identify two possible readings, one
with the finite verb in the scope of negation and one with the converb in the
scope of negation.
Other operators also appear to be shared. The following dialog is the begin-
ning of a story. Speaker B interrupts with a clarification of an ambiguous pronoun
in Speaker A’s sentence. The static verb ‘sit’ in Ingush is derived from the punctual
by nuclear serialization: singular wa-xeina d.aagha, plural wa-xeishaa d.aagha lit.
‘having sat down, is/are sitting’. Here Speaker B’s utterance has contrastive focus
on ‘young men’ and topicalizing definiteness on ‘sit’. The definiteness is marked
by emphatic my= together with interrogative =ii, and these are both on the finite
verb. Contrastive focus is marked by fronting, and the verb then moves to imme-
diately follow it, i.e. to clause-second position. When (as here) the verb would im-
mediately follow the fronted phrase anyway, i.e. with no focus, then only the final,
conjugated piece of a verb moves and any prefixes or first elements of compounds
remain in clause-final position. In B’s utterance the converb is clause-final and the
finite verb in second position. That is, the serial construction is treated the same,
for purposes of word order, as a single prefixed or compound verb. This example
shows that definiteness and topicalization semantically apply to the entire serial-
ized sequence, but their formal marking is attached only to the finite verb.
In (20) above the finite verb is followed by a quotative marker, whose scope is
not just the finite verb but the entire nuclear serialization and in fact the entire
johanna nichols 35
sentence. This means that quotative status is shared in nuclear serialization but
not only there.
Core serialization uses the same three basic converbs as nuclear serialization does,
but otherwise differs formally. In core serialization only one argument is usually
shared. It is almost always subject and almost always on an accusative pattern, i.e.
S/A. There are only weak case constraints on sharing. Core serial chains can be,
and often are, longer than two verbs (§15 shows two examples of long chains).
The clearest formal hallmark of core serialization is the serializing and coordi-
nating clitic particle =’a, which is positioned before the last element in the serial
clause (i.e. before the conjugated verb) and is enclitic to the element (word or
prefix) preceding that. This positioning requirement triggers reduplication of the
verb root to provide a host if there is no preverbal word or prefix. The analysis of
this clitic is from Peterson 2001.11 (Examples (27)-(28) just below show the clitic
on a direct object, a deictic prefix, and a reduplicate, in that order.) In addition, the
main clause in a core serial chain usually has verb-initial or at least verb-subject
word order; clear examples below are elicited (27) and (31)-(32) from texts.
The examples in (27)-(33) show some of the variety of sharings that are found
with core serialization. Table 4 summarizes the main patterns.
11
This particle is homophonous with the phrasal coordinator, but the coordinator is
positioned differently: it is enclitic to the last word in its domain, and it occurs on all con-
juncts: e.g. so=’a, hwo=’a, yz=’a ‘me, you, and him’. Its chaining function may well have
arisen through grammaticalization of some sort of coordinating construction, but the dif-
ferences in positioning show that in this function it is not synchronically a conjunction.
This is important for the analysis of core serialization as serialization, since lack of explicit
coordinators is a hallmark of serialization.
36 juncture-based split alignment and aspectuality in ingush
‘They loaded the truck, weighed it, and unloaded it.’ [A=A=A, O=O=O]
(29) Cwan bettaa yshtta q’esta=’a dezh yzh, zha detta=’a dettazh
one month.dat thus separate=& d.lv.cvsim 3p, sheep d.milk=& d.milk.cvsim
naxcha joaxar.
cheese j.take.impf
‘For one month they would separate them (lambs from sheep), milk the ewes, and
make cheese.’ (0216B) [A=A=A]
(30) Yzh t’ex-j.eallalc leatta=’a leattaa,
3p past-J.go.cvuntil red=& stand.cvant
‘They growl, pick it up in their mouths, and run on ahead playing with it.’
The salient temporal difference between nuclear and core serialization is that
core-serialized verbs need not be temporally coreferent, and if both are perfective
they are sequenced. Compare core serials (27) and (28), each of which describes
a sequence of events, with nuclear serials (10), (12), and (16), each of which de-
scribes a single event. On the other hand simultaneous converbs in core serializa-
tion are usually coreferential and overlapping in time with the next verb, e.g. (34)
below, but sometimes sequenced as in (29), which describes a habitual sequence of
non-overlapping events (separate lambs and ewes in the morning, milk the ewes
in the afternoon, make cheese with the milk). The possibility of sequencing shows
that core serial verbs can refer to separate events. Their grammatical tense, how-
ever, is not independent; it is shared, marked only on the finite verb, and even if
sequenced the actions all occur in the same narrative time juncture. The possi-
bility of referring to discrete events distinguishes core serialization from nuclear
serialization.
Though the same complementarity between anterior and sequential converb
applies in core serialization as in nuclear serialization, otherwise converb choice
in core serialization is free and exploited by speakers in order to depict actions
as sequenced vs. simultaneous relative to the next clause. In (34) the spilling of
38 juncture-based split alignment and aspectuality in ingush
the basket is in fact a punctual event but the simultaneous converb makes clear
that it accompanies the boy’s fall rather than preceding it (as an anterior converb
would require). In (35) the sheaves are first all laid down, and the anterior converb
wadaxkaa makes clear that this action is completed; then horses are driven over
the sheaves, and the simultaneous converb xexkazh shows that this continues for
some time.
Unlike nuclear serialization with its shared polarity marked only on the finite verb,
in core serialization either a converb or a finite clause can contain formal negation
and finite-clause negation does not have scope over a converb clause. In (37)-(38)
the converb is negated; in (39)-(40) the main verb is. In each example the only
possible reading is the one where the formally negated clause is denied and the
non-negated one is asserted.
Modals behave the same as polarity: the converb clause in nuclear serialization is
in the scope of a modal main verb, while in core serialization the converb clause is
not in the scope of the modal. In (41), which has nuclear serialization, the scope
of ‘can’ is ‘run’, not ‘go’ (since the two people are already going). In (42), which has
core serialization, the scope of ‘be able’ is ‘buy a car’ and not ‘get my salary’.
40 juncture-based split alignment and aspectuality in ingush
A nuclear serial construction can itself be core serialized to the next verb. Exam-
ples are (16) and (33) above. The formal marking of the core serialization occurs
only on the second verb: in (16) the chaining particle =’a is on the second verb and
in (33) only the second verb is in the oblique anterior converb form. I have not
found any examples where a core serial construction is itself nuclear-serialized to
the next verb (or, put differently, where the first item in a nuclear serial construc-
tion is not just the converb but the converb plus a preceding core-serialized clause).
16. Conclusions
To summarize, in Ingush both core and nuclear serialization use the same con-
verbs and have obligatory argument sharing and tense sharing. In nuclear serial-
ization the aspectual property of result state, polarity, and probably pluractionality
are shared, and deixis, if not shared, must be identical. Not only is grammatical
tense shared, but in addition the time reference of the serialized verbs is not dis-
tinct; even if both are perfective they are not sequenced. Argument sharing appar-
ently involves not just one argument but the entire core valence. The alignment
of argument sharing is ergative, and it is difficult to decide whether ergative align-
ment or nominative case is the operative constraint.
In core serialization, the time reference of the serialized verbs can be discrete,
so core-serialized perfectives indicate a sequence of events; still, grammatical tense
is shared. Result state and polarity are not shared. Argument sharing seems to in-
volve just one argument, and its alignment is accusative.
johanna nichols 41
Ingush thus gives clear evidence of a difference of pure juncture in its serial-
ization types, and a case of juncture-based split alignment. The technical under-
standings of juncture and cosubordination are what make it possible to identify
the constructions treated here as serialization and to capture the distinction be-
tween nuclear and core serialization.
References
Bickel, Balthasar. 1998. Review article: Haspelmath & Koenig, eds., Converbs in
Cross-Linguistic Perspective. Linguistic Typology 2:381-397.
Bickel, Balthasar, and Nichols, Johanna. In press. Case marking and alignment.
In Handbook of Case, eds. Andrej L. Malchukov and Andrew Spencer. Ox-
ford: Oxford University Press.
Conathan, Lisa, and Good, Jeff. 2000. Morphosyntactic reduplication in Chech-
en and Ingush. CLS 36:49-61.
Crowley, Terry. 2002. Serial Verbs in Oceanic: A Descriptive Typology. Oxford:
Oxford University Press.
Dixon, R. M. W. 1979. Ergativity. Language 55. 59-138.
Foley, William A., and Van Valin, Robert D., Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Good, Jeff. 2003. Clause combining in Chechen. Studies in Language 27:113-170.
Haspelmath, Martin, and Koenig, Ekkehard eds. 1995. Converbs in Cross-Lin-
guistic Perspective. Empirical Approaches to Language Typology, 13. Berlin:
Mouton de Gruyter.
— 2004. Ingush-English and English-Ingush Dictionary / Ghalghaai-ingalsii, in-
galsii-ghalghaai lughat. London: Routledge/Curzon.
— 2001. Long-distance reflexivization in Chechen and Ingush. In Long Dis-
tance Reflexives, eds. Peter Cole, Gabriella Hermon and C.-T. James Huang,
255-278. New York: Academic Press.
— 1997ff. Berkeley Ingush Corpus. (Electronic database.) University of Cali-
fornia, Berkeley.
Peterson, David A. 2001. Ingush ‘a: The elusive Type 5 clitic? Language 77:144-
155.
Sprouse, Ronald L. 1997ff. Berkeley Internet Text Collector (BITC). Berkeley:
Department of Linguistics, University of California, Berkeley.
42 juncture-based split alignment and aspectuality in ingush
Van Valin, Robert D., Jr. ed. 1993. Advances in Role and Reference Grammar.
Amsterdam-Philadelphia: Benjamins.
Van Valin, Robert D., Jr., and LaPolla, Randy J. 1997. Syntax: Structure, Mean-
ing, and Function. Cambridge: Cambridge University Press.
The following two longer examples show how Ingush serialization (often inter-
spersed with subordination) functions in natural narrative.
e. [got off and] untied his horse under a tree, (core serial)
f. and lay back crossing his hands under his head. (0408) (core serial)
(42) is fully annotated with shared arguments and syntactic roles indicated. In
(3) a subordinate clause headed by hwaveitaa viecarie ‘if (he) hadn’t gotten (me)
out’ (in (g)) consists of core serialized (a)-(f ) in which the subjects, all A’s, are
shared. Object pronouns are null if they would be absolutive (a, c, f ), overt if da-
tive (d), and all coreferential to the main-clause subject (S) in (h). The shared sub-
jects are indicated by an ellipsis and the null anaphoric objects by an underscore.
Ø marks an ambiguous or unspecified null subject. M = Mahwmad (overt in (f )).
g. … ___ hwa-v.eitaa-v.iecarie
[M.erg] [1sg] dx-V.send-V.ppft.neg.cvirr
A O
44 juncture-based split alignment and aspectuality in ingush
a. taken me in,
b. eating and drinking for a whole week [i.e. fed me for a whole week]
c. put me back on my feet
d. given me money for the road
e. and food
f. rescued me from there
g. and sent me back
h. I wouldn’t be here with you today (0207A.2)
Privileged Syntactic Arguments,
Pivots, and Controllers
Robert D. Van Valin, Jr.
Heinrich Heine University Düsseldorf
University at Buffalo, The State University of New York
1. Introduction
The purpose of this paper is to review some basic notions of the Role and Ref-
erence Grammar [RRG] theory of grammatical relations. RRG started out as a
theory of grammatical relations, with the initial work in the theory challenging
the universality of the traditional, Indo-European-based notion of subject, the
most important grammatical relation (Foley & Van Valin 1977, Van Valin 1977).
In work prior to Van Valin & LaPolla (1997) [VVLP], the notion ‘pivot’ was used
to cover many of the phenomena that ‘subject’ was used for in other theories. This
proved unsatisfactory for a variety of reasons, and in VVLP a more sophisticated
theory of grammatical relations was introduced. The central concept is ‘privileged
syntactic argument’ [PSA]; it is defined as a restricted neutralization of semantic
roles for syntactic purposes in a grammatical construction. A grammatical relation
has two essential properties. First, it must be grammatical or syntactic, not seman-
tic, and this means that it cannot be characterized exclusively in terms of seman-
tic roles, hence ‘the neutralization of semantic roles’ component of the definition.
Second, grammatical relations involve restrictions; for example, finite verb agree-
ment is traditionally viewed as an indicator of subjecthood, when it is restricted
to a single argument type, e.g. the core-initial RP in English or the nominative
RP in German. This is the privileged argument with respect to finite verb agree-
ment. If the finite verb (or auxiliary) were to agree with all of its arguments, as e.g.
I would like to thank Anja Latrouite for comments on an earlier draft and Matthias
Gerner for answering questions about Liangshan Nuosu.
See Van Valin (2005:28) and Van Valin (2008) for arguments for the replacement of
‘NP’ by ‘RP’ in RRG.
45
46 privileged syntactic arguments, pivots, and controllers
(1) Chrisi slapped Patj and then ___ i/*j ran away.
CONTROLLER PIVOT
In this construction there are two PSAs: the RP Chris, which functions as the
controller of the missing argument in the second clause, and the missing argument
in the second clause, which is the pivot. Both are restricted to the first argument in
the core, as the inability of Pat to be the controller in (1) shows, and this is regard-
less of its semantic role, as (2) shows.
This representation should not be interpreted to mean that in the RRG syntactic rep-
resentation of this sentence there is a gap or empty core argument position; the ‘___’ is for
illustrative purposes only, here and throughout the paper. See Van Valin (2005:230) for the
technical representation of (1).
robert d. van valin, jr. 47
(2) a. Patj was slapped by Chrisi and then ___*i/j ran away.
b. Chris ran up and ___ slapped Pat.
c. *Chris ran up and Pat slapped ___.
d. Chris ran up and ___ was slapped by Pat.
In (1) and (2a) the controller is the first RP in the core, regardless whether it
is actor (as in (1)) or undergoer (as in (2a)). Similarly, the missing argument, the
pivot, must correspond to what would be the first RP in the core, if it were pres-
ent; this is shown by the ungrammaticality of (2c). It does not matter whether
the pivot is interpreted as actor (as in (2b)) or as undergoer (as in (2d)). The con-
trast between (2c) and (2d) is particularly important, since the pivot in both is the
undergoer, and the only difference between them is the syntactic position of the
pivot: it is post-nuclear in (2c), which is ungrammatical in this construction, and it
is pre-nuclear, which is required in this construction.
One might argue that these are just subjects by another name, but the intention
of this paper is to show that there are instances of pivots and controllers that do
not correspond to what one would traditionally call subjects. Moreover, the con-
struction-specific nature of PSAs will be highlighted as well. This will be shown
in constructions from three so-called exotic languages: English, Liangshan Nuosu
(Lolo-Burmese, Southwest China; Gerner 2004), and Barai (Papua-New Guinea;
Olson 1978, 1981). In the next section, the purposive construction in English will
be investigated. In the following section, a construction analogous to (1) will be
examined in Liangshan Nuosu, and finally the switch-reference construction in
Barai will be analyzed. Conclusions follow.
(3) a. Pat brought the booki for her sister to read ___i.
b. *Pat brought the book for her sister to read it.
There are two striking features of this construction. First, the obligatory missing
argument, the pivot, is what would be the immediately post-nuclear argument, the
48 privileged syntactic arguments, pivots, and controllers
undergoer,in the infinitival core in (3); the actor,the first RP in the core,is not involved.
The controller in the initial core is also the immediately post-nuclear argument, the
undergoer, not the actor in the traditional subject position. It should be noted that
the pre-nuclear core argument in the infinitival core can be omitted, yielding (4).
(5) a. Pat brought the book in order (for her sister) to read it.
b. *Pat brought the book in order (for her sister) to read __.
See Cutrer (1993) for discussion of the control properties of the two constructions.
robert d. van valin, jr. 49
The verb put takes two post-nuclear complements, an RP core argument and
a PP argument-adjunct. What is significant here is that the obligatory gap may
correspond to either of these post-nuclear core elements. In (6a) the pivot is the
object of the preposition in, while in (6c) the pivot corresponds to the immediately
post-nuclear undergoer. These arguments may be realized by a RP only if the con-
struction is interpreted as a rationale construction and in order is included, as in
(6b) and (6d).
Liangshan Nuosu [LN] (Gerner 2004) illustrates a different aspect of the RRG
theory of grammatical relations, namely the issue of variable vs. invariable PSA
selection. In English, as (1) and (2) illustrate, either the actor or the undergoer of
a transitive verb can serve as PSA, and this selection depends on the voice of the
verb. This type of PSA is termed a ‘variable pivot’ or ‘variable controller’, since the
choice of argument to serve as pivot or controller is not fixed. In many languages,
however, it is fixed: with an intransitive verb the single argument serves as PSA
in various constructions, regardless of whether it is actor or undergoer, but with a
transitive only the actor can function as PSA in the same constructions. This type
of PSA is called an ‘invariable pivot’ or an ‘invariable controller’. Languages of
this type, e.g. Lakhota (North America), Enga (Papua New Guinea) and Warlpiri
(Australia), typically lack voice oppositions. Hence the existence of variable PSAs
in a language normally correlates with the existence of some kind of voice opposi-
tion in the language.
LN appears to be an exception to this pattern. On the one hand, in the equiv-
alent constructions to (1)-(2), it exhibits a commonplace pattern: the controller
and pivot must be in the clause-initial position, similar to the restriction found in
English. On the other hand, LN has variable pivots and controllers yet lacks a for-
mal voice construction of the type found in English, German, Jakaltek, Malagasy
and many other languages. How does the system work?
According to Gerner (2004), there are three patterns of actor-undergoer cod-
ing in LN, two of which depend on the aspectual properties of the verb. When the
verb depicts “on-going action”, then the actor precedes the undergoer, i.e, A U V
order. This is exemplified in (8). When the verb has a resultative meaning, the un-
dergoer precedes the actor, i.e. U A V order, as in (9). Finally, there are verbs which
are aspectually unmarked and can be interpreted as having either A U V or U A V
order, as in (10).
In fact, in VVLP, §6.5, it is argued that this is the most common situation cross-lin-
guistically.
Abbreviations: a: actor; at: actor of transitive verb; art: article; cntrs: contrastive;
coll: collective; conj: conjunction; cont: continuative; contr: controller; def: definite;
diff: different psa; dp: dynamic perfect; ints: intensifier; pp: prepositional phrase; psa:
privileged syntactic argument; pvt: pivot; rp: referential phrase; s: single argument of in-
robert d. van valin, jr. 51
(9) Resultative
vi55ga33 gɯ44su33 a44mo33 thsɨ33 ȶɕ hu44tsḭ33tsḭ33 o44.
clothes art+coll [UT] mother [AT] wash snow-white dp
‘Mother washed the clothes snow-white.’
(10) Ambiguous
ṃ33ka55 ṃ33ko44 ndu21.
Muga Mugo beat
‘Muga beats Mugo.’ or ‘Mugo beats Muga.’
In (8) ȵȡʑɔ33 marks continuative aspect and therefore the verb as denoting on-
going action; if it were missing, then the interpretation would be ambiguous like
that in (10). In (9) o44 marks dynamic perfect and renders the verb perfective; ac-
cordingly the undergoer occurs initially. Finally, in (10) there are two possible in-
terpretations; the first RP can be either actor or undergoer, and likewise the sec-
ond one takes the other role.
As noted earlier, in the LN construction analogous to (1) in English, the con-
troller and pivot in a subsequent clause must be in the clause-initial position. This
means the only certain combinations of clauses in terms of their aspectual proper-
ties are allowed. This constraint is illustrated in the following pair of clause chains
which differ only in the aspect of the verb in the second clause.
transitive verb; same: same psa; spec: specific; stp: stative perfect; top: topic; u: undergoer;
ut: undergoer of transitive verb.
It should be noted that LN has an optional undergoer marking particle in AUV claus-
es and an optional actor marking particle in UAV clauses. Neither occurs in (8), (9) or (10).
Pronouns also have actor and undergoer forms. All of these serve to reduce ambiguity, espe-
cially in complex constructions. See Gerner (2004) for detailed discussion.
52 privileged syntactic arguments, pivots, and controllers
(12) *A-A[resultative]-S
*phu21 su33vɔ55vu33 du44 nɯ33 l33 ̰ ʈɕɔ33 si44 ta33
Pu Mister family top stone.brick take stp
CONTR[AT] [UT]
i21kho33 ___ tsɨ21 ko44ʂ a33 ta33___ i55.
gate block send stp sleep
[UT] PVT[AT] PVT[S]
‘Mister Pu’s family took some stone bricks, ___ blocked up the entrance gate, and
then ___ fell asleep.’
The controller in the first clause in (11) is the actor, and the pivots in the fol-
lowing clauses are an actor and an S argument. The second clause would have
AUV order if it were an independent utterance, and therefore the missing actor,
the pivot, would be the clause-initial argument. In (12), on the other hand, the
second clause is resultative because of the addition of ko44ʂ a33 ‘send’, and accord-
ingly the word order in the clause is UAV. In this case the missing actor would
not be clause-initial, if overt in an independent clause, and therefore the clause
chain is ungrammatical. Note that while ko44ʂ a33 ‘send’ is one of a group of as-
pect markers, not a voice marker, its addition to the verb has the effect of a voice
marker in other languages, given the constraints on word order in LN. Given the
word order variation, it is also possible to have undergoers as controller or pivot,
as in (13) and (14).
(13) U[resultative]-A[on-going]
tsho33ʈɕo55a33ma55 tshɨ33 tsi33 si44 __ ko33 ʂɯ44 bo33 o44.
demon-sorceress 3sgA cheat conj 3U look.for go dp
CONTR[UT] AT PVT[AT] [UT]
‘He tricked the demon sorceressi and [shei] went to look for them.’
robert d. van valin, jr. 53
(14) A[on-going]-U[resultative]
ʈɕɨ33ʈɕɪ33va33ɳɪ33dzṵ33 ... phu21 di44 ___
Djidjevanidzu[name] phu produce PVT[UT]
CONTR[AT] [UT]
In (13) the verb in the first clause is resultative, and therefore the order is UAV;
in the second clause, the verb is in an on-going action form, and therefore the
order would be AUV. The undergoer of the first clause and the actor of the sec-
ond clause, the demon sorceress, are or would be clause-initial, and therefore the
sequence is grammatical. The controller is an undergoer and the pivot an actor. A
grammatical English translation would be ‘The demon sorceress was tricked by
him and went to look for them’. In (14) the opposite pattern occurs: the controller
in the first clause is an actor, and the pivot of the second clause is an undergoer. A
grammatical translation would be ‘‘Djidjevanidzui produced [nothing], [and] was
blown up onto the roof of the house by the demon sorceress’. Both of these trans-
lations involve passivization in English, but there is no formal equivalent in LN,
hence the more literal glosses with the examples.
What happens when there is an ambiguous verb in this construction? Unfortu-
nately, there are no examples of this in Gerner (2004), but he does give the following
description in the text. Given an ambiguous verb in a sentence like (10), followed
by a clause with an intransitive verb, the clause-initial RP will always be the con-
troller, but, as expected, its interpretation will be ambiguous, as illustrated in (15).
The optional actor particle kɯ 21 occurs in (14). Gerner analyzes this as the marker of
a demoted actor of a passive, thereby claiming that there is a passive construction in LN.
However, there is no more reason to analyze the optional actor particle as involving formal
passivization as there is to analyze the optional undergoer particle as involving formal an-
tipassivization. The simplest analysis of each is as an optional role-marking particle which
serves a disambiguating function. If this particle were obligatory in all instances of UAV
order, then the argument for a formal passive construction would be stronger.
54 privileged syntactic arguments, pivots, and controllers
Again, a passive construction gives the best rendering of the second interpreta-
tion into English, but there is clearly no formal passivization involved here, at least
not in the usual sense of the term, but functionally the alternation is similar: there
is variable controller (hence variable PSA) selection in (15).
The RRG theory of PSA selection claims that in many languages there is a
default PSA selection choice, i.e. actor as PSA for accusative constructions and
undergoer as PSA for ergative constructions. However, nothing in the theory re-
quires that a language have default and non-default choices, and in fact in Foley &
Van Valin (1984) it was argued that Tagalog lacks a default voice; rather, there are
a number of voice forms which are used, no one of them being a default choice; Bi-
sang (2006) makes a similar point. If one understands voice in a functional sense,
namely as the correlations between different semantics roles and PSAhood, then
one could say that in LN there are two voices, an actor-as-PSA voice, as in (8), and
an undergoer-as-PSA voice, as in (9). These are not formal voice oppositions. The
marking that distinguishes them is aspectual, and while there are strong correla-
tions cross-linguistically between on-going action-type aspect and actor-as-PSA
voice, on the one hand, and resultative aspect and undergoer-as-PSA voice, on the
other, the primary motivation for the aspectual marking is to characterize states of
affairs in the world, the voice effects being a side effect of the aspectual marking.
This can be seen most clearly with the ambiguous verbs, as in (10), in which both
voice interpretations are possible, and there is no aspectual or other morphology
involved. The optional actor and undergoer particles function as disambiguators
in such situations.
In terms of the RRG theory of voice, such facts are not surprising. RRG ana-
lyzes voice constructions as having two aspects: (i) PSA modulation, which chang-
es the semantic role of the PSA, and (ii) argument modulation, which involves
non-canonical coding of a macrorole argument (see VVLP, §6.4; Van Valin 2005,
§4.5). These may occur independently of each other, as in e.g. Lango and Toba
Batak, and moreover Roberts (1995) argues that when PSA modulation occurs
robert d. van valin, jr. 55
4. Switch-reference in Barai
Barai (Olson 1978, 1981) presents another example of a language with variable
PSAs without a formal voice construction, but it adds a couple of fascinating
twists. The coding of S arguments is exemplified in (16) and (17).
(16) a. Fu difuri.
3sg run
‘He is running.’
b’. E be difuri.
[+SPEC]
‘A (certain) man is running.’
c. Fu-ka difuri.
3sg-ints run
‘He is really running.’
(17) a. Fu visi.
3sg sick
‘He is sick.’
(18) a. Fu na kan-ie.
3sg 1sg hit-1sg
‘He hit me.’
c. Fu-ka na kan-ie.
‘He really hit me.’
���� Ame
e. ije (fu) e ije kan-a.
����������
child def (3sg) person def hit-3sg
‘The child hit the man.’
The word order in these examples is AUV, and the PSA coding properties
found on the S of intransitive verbs are found on the first RP, the actor, in (18).
This is shown most clearly in (18b, c, d, e, f ) and by the ungrammaticality of (18c’,
d’, e’, f ’), in which the PSA coding is found on the second RP, the undergoer. Note
that the verb agrees with the undergoer, not the actor.
��� Ije
c. na visinam-ie.
��������������������������
3sg 1sg sicken-1sg
‘It sickened me.’
The pattern of PSA coding with U-verbs is strikingly different from that with
A-verbs: it is the second core argument position, the one occupied by the under-
goer, that gets the PSA coding. Putting that coding on the first argument position
leads to ungrammaticality, as (19a’, b’, d’) indicate. This is a surprising situation:
the PSA position in the core depends on the class of the verb. This is summarized
in (20).
Since the PSA with an A-verb is an actor, one could argue that A-verbs pattern
accusatively, and correspondingly, since the PSA with a U-verb is an undergoer,
these verbs could be analyzed as having an ergative pattern. That these are the
PSA positions can be seen in the following switch-reference examples.
In (21a) the verb in the first clause is an A-verb, and the controller is the core-
initial RP, ame ije fu-ka ‘the child’; the pivot is the S argument of ko ‘run away’. The
one who ran away can only be interpreted as the child, not the speaker. In (21b),
on the other hand, the verb in the first clause is a U-verb, and the controller is the
preverbal RP, ame ije fu-ka ‘the child’; the pivot is the S argument of barone ‘die’.
The one who died can only be interpreted as the child, not the girl. There is much
more to be said about the switch-reference system, but these initial examples show
clearly that the PSA in clauses with A-verbs is in a different position from the
PSA in clauses with U-verbs.
There is a further complication involving PSA selection, namely that the defi-
niteness status of an argument affects whether it can serve as PSA or not. This is
illustrated in (22) and (23).
��� Fanu
b. ije bu-ka e-be kan-ia.
����������
animal def 3pl-ints person-[-spec] hit-3pl
‘Someone really hit the animals.’
These examples involve the A-verb kan- ‘hit’. In (22a) the actor and undergoer
are both definite, and the word order must be APSAUV. In (b), on the other hand,
the undergoer is definite but the actor is indefinite and non-specific, and in this
case the definite undergoer occurs in the initial PSA-position, while the non-spe-
cific actor occurs in the preverbal position; as (b´) shows, APSAUV word order is
impossible in this situation. Similarly in (c), the undergoer is indefinite but specif-
ic, while the actor is again indefinite and non-specific, and here too the undergoer
must occur in initial position. Similar facts obtain for U-verbs, mutatis mutandum.
Since the PSA position is the preverbal position with U-verbs, the constraint
affects which RP can occur in that position. When the actor and undergoer are
both definite, as in (23a), the order must be AUPSAV. When the actor is definite
and the undergoer indefinite and non-specific, as in (b), then the actor occurs in
the preverbal PSA position, not the undergoer; this is the U-verb analog of (22b).
AUPSAV word order is impossible in this situation, as (b’) illustrates. In (c) there is
canonical word order, since the undergoer is definite and the actor non-specific.
This is the same word order as in (22c’), but because the verb in that sentence is
an A-verb, the example is ungrammatical. Thus the grammaticality of particular
word orders depends on the relative definiteness status of the RPs and whether the
nucleus contains an A-verb or a U-verb.
There is one final complication. Barai has a series of particles which mark new
or contrastive information, and RPs bearing one of these particles can never occur
in the PSA position. This is exemplified in (24).
robert d. van valin, jr. 61
The marker -iebe signals contrastive information, and with an A-verb an actor
marked by -iebe cannot occur in the initial PSA position, as in (24a, a’). Similarly,
with a U-verb an undergoer marked by -iebe cannot occur in the preverbal PSA
position, as (24b, b’) show.
Thus, there are definiteness constraints on the selection of the PSA in Barai.
There is a hierarchy of definiteness, given in (25a), and a general constraint, given
in (25b).
(25) a. Hierarchy of definiteness:
definite > indefinite specific > unmarked > indefinite non-specific
b. Constraint on PSA selection:
RP in PSA slot must not be lower than the other argument on the definiteness hier-
archy.
that it is coreferential with the single argument of the verb ko ‘run away’ in the sec-
ond clause; in such cases, the coreferential argument in the second clause is omit-
ted. In (21b) the controller is the second RP in the clause, the undergoer (also ‘the
child’), and the suffix -na signals that it is coreferential with the single argument
of the verb barone ‘die’, the pivot, which is omitted. The controller with an A-verb
is the argument in what we have identified as the PSA position, while the control-
ler with a U-verb is likewise the argument in what we have identified as the PSA
position. In the first two examples the verb in the second clause is intransitive, and
therefore its single argument is the pivot and PSA. In the next pair of examples,
both clauses have A-verbs.
In both examples, the controller is the actor of the first clause and the pivot is
the actor of the second clause; when they are coreferential, as in (26a), the pivot
argument is omitted, whereas when they are non-coreferential, as signaled by -mo
‘different PSA’, an overt pronoun occurs as the actor of the second clause. This
pronoun is still the pivot of the construction, because it is the argument with refer-
ence to which coreference is signaled.
In all of the switch-reference examples so far the PSA selection has been the
default one; what happens when there is a different selection? This is illustrated in
the following examples; the first is (21a) repeated.
���� Na-ka
b. e-be kan-ie-mo fu ko.
����
1sg-ints person-[-spec] hit-1sg-diff 3sg run.away
‘Someonei really hit me and then hei ran away.’
robert d. van valin, jr. 63
���� Na-ka
b’. e-be kan-ie-na ___ ko.
����
1sg- ints person-[-spec] hit-1sg-same run.away
‘Someone really hit me and then [I] ran away.’
In (27a) the actor is the PSA in the first clause and is the controller. In (27b,
b’), on the other hand, the actor is indefinite non-specific, while the undergoer, a
personal pronoun, is definite. Therefore in terms of (25a, b) the undergoer, not the
actor, must be the controller, the PSA. Hence if the actor of kan- ‘hit’ is also the
one who ran away in (27b), then different-PSA coding is required; this is signaled
by the suffix -mo and by the occurrence of a third-person pronoun fu in the sec-
ond clause. If, on the other hand, it is the undergoer of kan- who ran away, then
because it is in the initial PSA position, it is the controller, and the result is a same-
PSA form, as in (27b’). Thus with an A-verb the controller need not be the actor;
it can be the undergoer, if the undergoer outranks the actor on the definiteness
hierarchy in (25a).
The same contrast can be found in the following examples, both with respect to
the controller in (28a, a’) and with respect to the pivot in (28b, b’).
���� Na i
b. me-mo miane ije na sak-ie.
������������������
1sg work do-diff firestick me bit-1sg
‘I was working and the firestick bit me.’
���� Na i
b’. me-na __ miane sak-ie.
�����������������
1sg work do-same firestick bit-1sg
‘I was working and a firestick bit me.’
In (28a) the actor and undergoer have the same status in terms of (25a), hence
the actor miane ije ‘the firestick’ is the PSA and controller. Since it is the undergoer
in the first clause that is coreferential with the S argument in the second, the verb
64 privileged syntactic arguments, pivots, and controllers
is marked for different-PSA. In (a’), however, the actor RP is unmarked for defi-
niteness, and that means that the pronominal undergoer is the PSA and controller.
It is coreferential with the S argument of the second clause, hence same-PSA cod-
ing on the verb in the first clause. The same situation obtains with respect to the
pivot in the second clause in (28b, b’). In (b) the two RPs are of equivalent defi-
niteness, and accordingly miane ije ‘the firestick’ is the actor and pivot of the sec-
ond clause; the controller in the first clause is the first-person singular pronoun na,
and because these are not coreferential, the verb in the first clause takes -mo ‘dif-
ferent PSA’ and an overt pronoun appears in the undergoer position in the second
clause. In (b’), by contrast, the actor in the second clause is unmarked for definite-
ness and therefore is outranked by the personal pronoun, which serves as the pivot.
Since the controller in the first clause and the pivot in the second clause have the
same reference, the verb in the first clause is marked by -na ‘same PSA’ and there is
no second pronoun in the second clause. Thus, both the controller and the pivot in
these examples are variable PSAs.
All of the transitive verbs in the examples so far have been A-verbs; what hap-
pens with U-verbs? Olson (1978, 1981) contains far fewer examples of construc-
tions with U-verbs than with A-verbs, and therefore it is impossible to get as full
a picture as with A-verbs. In particular, he gives no examples analogous to (27)
and (28) involving U-verbs. Three types of U-verb constructions are given: One in
which both verbs are U-verbs, as in (29), one in which the first clause has a U-verb
and the second clause has either an A-verb or an intransitive verb that would take
an actor argument, as in (30), and one in which the first clause is an A-verb and
the second a U-verb, as in (31).
-ga marks ‘different PSA’ without any specified temporal relationship between the
events in the clauses, while -mo marks it when the actions are sequential.
robert d. van valin, jr. 65
In (29) both verbs are U-verbs, and the undergoer of the first verb is coref-
erential with the undergoer of the second, and accordingly the first verb bears
same-PSA marking. What is different about this example is that there is a pro-
noun in the pivot position in the second clause; this might be due to the occur-
rence of the intensifier -ka, which can only occur attached to a pronoun in PSA
position, as we have seen. In the examples in (30), the first clause contains a U-
verb, and the second clause contains either an A-verb, as in (30a), or an intransi-
tive verb with an actor S, as in (30b). In both sentences the PSA in the U-verb
clause is coreferential with the PSA in the A-verb clause, and yet in both cases
the verb carries different-PSA marking. This shows clearly that there is an addi-
tional principle at work in determining the switch-reference coding. In the final
example, the first clause contains an A-verb and the second a U-verb, and the ini-
tial PSA in the first clause is coreferential with the undergoer PSA in the second,
and the result is same-PSA marking. This is analogous to (28a, a’), in which the
first clause contains and A-verb and the second clause has an intransitive verb
with an undergoer S.
Thus, there appears to be a semantic constraint on the system relating to the
type of verbs in the construction. When the first clause contains an A-verb, the
sentence is compatible with either same- or different-PSA marking regardless of
the type of verb in the second clause: it may contain an A-verb, as in (28b, b’), a
U-verb, as in (31), an intransitive verb with an actor S, as in (27), or an intransitive
verb with an undergoer S, as in (28a, a’). When the verb in the first clause is a U-
verb, on the other hand, same- or different-PSA coding is possible only if the sec-
ond clause contains a U-verb, as in (29), or an intransitive verb with an undergoer
S, as in (21b). In a construction with a U-verb in the first clause and an A-verb or
actor-S verb in the second, only different-PSA marking is possible. This cannot
be due to a constraint against same-PSA marking when an undergoer in the first
clause is coreferential with an actor in the second, because just such an example is
66 privileged syntactic arguments, pivots, and controllers
found in (27b’), and there is same-PSA marking. Hence it appears that the con-
straint is at the level of U-verbs and A-verbs, not at the level of actors and under-
goers. The constraint may be formulated as in (32).
Different-PSA coding therefore has two functions in the system: it can indi-
cate non-coreference between the PSAs in two clauses not affected by (32), or it
may signal a following A- or actor-S verb, when marked on a U-verb.
The Barai PSAs have a number of unusual features. First, as in LN, there are
variable pivots and controllers despite the absence of a formal voice system. With
A-verbs the pattern is accusative and with U-verbs, ergative. Barai thus presents an
interesting example of a split-ergative system; in typological terms, the language
lacks both a formal passive construction (with A-verbs) and a formal antipassive
construction (with U-verbs). Second, and perhaps most unusual, the PSA position
in a clause is determined by the semantic class of the verb. It is of course not at all
unusual for the PSA to be indicated by position in a clause or core, as it is in e.g.
English, Icelandic, Malagasy, or LN, and moreover it is not unusual for there to be
different PSA positions in different constructions; that was illustrated in the Eng-
lish purposive constructions in section 2 as opposed to the English construction
in (1)-(2). What is striking about Barai is that in a simple clause there is variation
in PSA position conditioned by the semantic class of the verb. The first feature, as
discussed in the analysis of LN, is thoroughly compatible with the RRG concept
of voice. The second supports the idea that PSAs are construction-specific. It is
straightforward to analyze clauses with A-verbs as a different construction from
clauses with U-verbs. Then each of these clausal constructions plays a constitutive
role in the switch-reference construction.
Barai PSA-selection principles illustrate another aspect of the RRG theory
of PSAs. One of the earliest claims of the theory was that grammatical relations
involved fundamentally an interaction between semantic role functions, on the
one hand, and discourse-pragmatic functions, on the other.10 An important claim
This is reflected in the name of the theory: ‘role’ refers to the semantic role properties
10
in this regard is that discourse-pragmatic factors can play a role in ‘subject’ selec-
tion in some constructions in some languages. In terms of VVLP and Van Valin
(2005), this is formulated in terms of whether the choice of the argument of a
transitive verb to serve as the PSA in a construction is pragmatically influenced
or not. The importance of the definiteness hierarchy in (25) for PSA selection in
Barai shows that it is definitely pragmatically influenced. It should be noted that
this influence is limited to instances in which there is an asymmetry in the defi-
niteness status of the arguments of a transitive verb; when they are of equal status,
i.e. both definite, then only the default selection is available. This contrasts with
the situation in LN with ambiguous verbs; in (10) the clause-initial PSA can be
interpreted as being the actor or the undergoer.11 There are no such ambiguous
verbs in Barai.12
5. Conclusion
The primary purpose of this paper has been to clarify the RRG notion of ‘privi-
leged syntactic argument’ [PSA] and to distinguish it from the traditional notion
of ‘subject’. PSA is a construction-specific relation, not a general feature of a lan-
guage like subject. This is shown in the English purposive construction in section
2, in which neither the controller nor the pivot corresponds to a traditional subject.
Analyzing grammatical relations in terms of the notion of PSA entails a particular
theory of voice, and the phenomena in Liangshan Nuosu discussed in section 3 are
a problem for standard theories of voice but not to the RRG account; indeed, the
RRG theory of voice allows for just such systems. Barai is the biggest challenge
of all, with its shifting PSA positions dictated by verb semantics and its pragmati-
cally-influenced variable PSAs, which, like those in Liangshan Nuosu, occur in a
language without a formal voice opposition. Here again, RRG is able to provide
RRG started out as a theory of grammatical relations but has expanded to become a general
theory of syntax; the name reflects this origin.
11
The variable PSAs in LN are most likely pragmatically influenced; the controllers
and pivots in clause chaining constructions are paradigm cases of pragmatically-influenced
PSAs. See VVLP, §6.4, Van Valin (2005), §4.3.
12
An unusual feature of the Barai switch-reference system is that it involves variable
PSAs; almost all switch-reference systems monitor invariable PSAs, namely [S, AT]. See
Foley & Van Valin (1984), §7.3.
68 privileged syntactic arguments, pivots, and controllers
an insightful account of complex and unusual data. Barai would be very difficult to
analyze if one did not make the fundamental assumption that grammatical rela-
tions are constituted out of a complex interaction between the semantics of verbs
and their arguments, on the one hand, and the discourse-pragmatic properties of
the arguments, on the other, an assumption on which RRG is based.
References
1. Introduction
This paper attempts to offer a view of middle voice as a basic voice system, a view
that is in contrast with a wide spread tendency to see middle marking as evolving
from reflexive constructions (Kemmer 1993, Lehamn 1995, Faltz 1985 and many
others). While it is true that the development from reflexive to middle can easily
be found in Indo-European languages, a visit to languages from other families will
show that the middle constitutes a category on its own and, in fact, can operate as
the base form for other voice patterns. Based on what we can learn from several
unrelated languages —mostly languages of Mexico and South America— I will
show that even for Spanish and most probably for other Romance languages the
existence of a middle system will provide a coherent representation to capture a
variety of uses that have been considered exceptional or aberrant deviations from
the norm whose best luck is to be listed in the lexicon, as has been the case for tra-
ditional analyses (Aid 1973; Alonso and Henríquez Ureña 1953; Gili Gaya1955;
Goldin 1968; González 1985; Grimshaw 1982; Sells, Zaenen, and Zec 1986; Butt
and Benjamín 2004; and many others) as well as instructional textbooks (Terrell,
Andrade, and Egasse 2006; Canteli Dominicis and Reynolds 1994; King and Suñer
1999; Alonso, Castañeda, Martínez, Miguel, Ortega, and Ruiz 2005 to name a few).
Middle constructions depict actions, events or states pertaining to the subject’s
own sphere. They contrast with active-direct voice in that they designate a process
remaining in the subject’s realm instead of being projected to another participant.
“The middle voice shows that the action is performed with special reference to the
subject” (Smyth 1956:390), in Benvensite’s words the subject “is indeed inside the
process of which he is the agent” (1950:149). It is an action or state affecting to
the subject or her/his interests (Lyons 1968). Voice patterns reflect situation types
that contrast clearly with transitive construals. The transitive active corresponds
69
70 middle as a basic voice system
to situations where two participants (most commonly agent and patient) interact
(Kemmer 1993, 1994). In contrast, middle voice marking corresponds to situation
types implying only the subject. The middle is a construction that focuses on the
subject’s dominion (Maldonado 1992, 1999).
Most current analyses of middle constructions tend to analyze middles as de-
riving from a transitive verb via a reflexive construction. The relationship between
middle and reflexive markers is evident: both construction types refer to the claus-
al subject. Yet there are important differences between them. Reflexives involve an
action where agent and patient are coreferential. Middles refer to actions or states
only involving the subject. The distinction can be observed in the well-known
Russian contrast provided by Haiman
���������������������������������������������������
(1983:796). Example (1a) is reflexive while
(1b) is middle:
b. on utomil-sja.
he exhausted-mid
‘He grew weary.’
This in fact may be the grammaticization path followed by middle markers depicting
The similarity between reflexive and middle constructions has led analyzers
to postulate a path by which middles develop from reflexives. By the distinguish-
ability hypothesis, Kemmer (1993, 1994) suggests two extreme situations with one
participant in one pole and two in the other. Reflexives and middles are placed be-
tween these two poles, as can be seen from the following diagrams:
S S
In the reflexive (2a), my dancing takes place in an image located in the mental
space of my imagination. In the middle (2b), there is only a mental experience oc-
curring inside the subject with no additional mental space. The middle is obtained
Instead of assuming that the middle marker is simply a detransitivation device with no
meaning— as has been suggested by a number of formal approaches —the middle marker
develops an inchoative interpretation from the core middle value as it simply focuses on the
change-of-state undergone by the subject (Maldonado 1992, 1999). Event simplification
develops thus from the core schematic representation of the middle marker.
72 middle as a basic voice system
Recent work on Yucatec Maya (YM) (Martínez 2006, Martínez and Maldonado
in press) has shown that both middles and reflexives develop from active transi-
tive constructions in an independent manner. Reflexive constructions are marked
by the possessive marker inflected for person-number plus a reflexive marker -ba,
as in (3b) which imposes a coreferential reading on a (root or derived) transitive
verb as is ts’ak ‘heal’ in (3a). Middle voice constructions are encoded as a CVVC´
pattern, as in (3c), which is obtained by lengthening the vowel of a transitive active
ricardo maldonado 73
verbal CVC stem ts’ak and imposing a high tone on the first vocalic segment ts’áak
(Ayres & Pfeiler 1997, Briceño 2004, Bricker 1981, Bohnemeyer 2004):
c. Ø-ts’áak-Ø-i(j). Middle
perf.intr-heal.mid-comp-b3s
‘He got better (he became cured).’
In the most transparent cases the reflexive/middle contrast coincides with well
known crosslinguistic patterns. In reflexive constructions the subject acts volition-
ally and with control on the self inducing some change. In contrast, middles high-
light the affectedness undergone by the subject as a consequence of some change
which s/he undergoes. In the middle construction the change-of-state is not vo-
litionally controlled by the subject. Thus the reflexive construction in (3b) des-
ignates the subject’s act of using some medicine to get better, while in (3c) the
subject’s health simply improves with no particular effort.
An outstanding feature of middles in YM is that they always depict an abso-
lute event, one in which energy is not profiled. The event is thus seen as neutral or
spontaneous. In (4) the decrease in energy is observed as we go from the transitive
(4a), to the reflexive in (4b) where the cat stretches out, to the middle construction
in (4c) where the clothes simply sag:
In this paper we will use the following abbreviations; a: a series (ergative), apl: appli-
cative, b: b series (absolutive), caus: causative, comp: completive, dem: demonstrative, dur:
durative, hab: habitual, incomp: incompletive, intr: intransitive, mid: middle, p: posses-
sive, perf: perfective, ref: reflexive, s: singular, trns: transitive, =: bound morpheme, 1, 2,
3: 1st, 2nd, 3rd person.
In brackets we include the date and interview from which examples are extracted.
74 middle as a basic voice system
c. le nook’-o’ Ø-sáats’-Ø-i(j).
dem garment-dem perf.intr-strech.mid-comp-b3s
‘The clothes stretched.’(ConChan: 05/2005)
Emotion middle
Guugu Yimidhirr dumiba‑adhi ‘get a shock or fright’
Mohave mat iθa:v ‘be angry’
Hungarian bán‑kod‑ ‘grieve, mourn’
Emotive speech actions
Latin quero‑r ‘complain’
Cl. Greek olophyre‑sthai ‘lament’
Turkish döv‑ün ‘lament’
Cognition middle
Indonesian ber‑pikir ‘be cogitating’
Pangwa -i‑sala ‘think over, consider’
Spontaneous events
Indonesian ber‑henti ‘come to a stop’
French s’evatiouir ‘vanish’
Hungarian kelet‑kez- ‘originate, occur’
The semantic areas where the middle tends to occur can be characterized in
more general terms: motion (translational, non-translational and change in body
posture), change of mental state (cognition, emotive speech and emotion), spon-
taneous events and self-directed actions (direct and indirect body care). Of all
these categories the one that fluctuates between middle and reflexive marking
across languages is self directed actions. The other three tend to get middle mark-
ing. What is interesting about the middle system in YM is that it shows notable
irregularities from general tendencies of middle marking. First, grooming actions
take reflexive not middle marking as is evident from the ungrammaticality of the
middle construction in (5b):
b. *Ø-cháal-Ø-i(j)
perf.intr-rinse.mid-comp-b3s
‘He rinsed himself.’
Second, in cases of motion, again it is the reflexive, not the middle, the con-
struction employed to signal that the subject controls his movements. In the first
story (6a) the subject’s change of position is a routine action as expressed by the
intransitive verb. In the second story (6b), the reflexive encodes a fast action that
the main character of the story does to avoid being seen by the fox. The middle
construction is precluded from both situations:
ka t-u=xol-kin-t-aj-Ø u=ba.
then perf.trs-a3s=kneel-apl-comp-b3s p3s=ref
‘Then he saw the old fox and he knelt fast.’ (Flojo: 122)
Third, in cases of change of state again the reflexive contrasts with the intransi-
tive form. In the reflexive construction (7b) the subject makes himself sick in order
to avoid going to work. The absolute intransitive is a spontaneous uncontrolled
event:
b. t-u=k’oja’an-kun-t-(aj)-Ø u=ba
perf.trs-a3s=sick-verb-apl-comp-b3s p3s=ref
b. ���������������������������
-xíik-Ø-Ø (l)e booláador-o’.
perf.intr-burst.mid-comp-b3s dem firecracker-dem
‘The firecracker burst.’
mate subject. See Martínez and Maldonado (in press) for an account of accidental reflexives
in YM.
78 middle as a basic voice system
absolutes energetics
Tr/Th Tr/Ag
Lm/Pat
Figure 5
Tr/Ag
Tr/Pat
Reflexive/Causative-Reflexive Volitional
Tr/Th Figure 8
Middles CVVC
Figure 6 Tr/Theme
Figure 9
Given this analysis there is no way to sustain that the reflexive is the source for
the middle. In fact the middle develops from transitive to contrast drastically with
the reflexive construction. In YM the middle is a derived construction indepen-
dent from the reflexive. In the next section we will see other languages where the
middle is a non-derived, a basic construction.
As opposed to YM there are languages where middle voice needs not derive from
any other voice in the language. I will illustrate such a system first from a set of
unrelated languages P’orepecha, Yucatec Maya, Otomi, as well Toba, Amharic and
Balinesse.
ricardo maldonado 79
3.1 P’orepecha
b. chkári tixa‑ra‑s‑Ø‑ti.
wood light‑mid‑perf‑pres‑ind.3
‘The wood lit up.’
a. Marcosï
(11) ��� Dora‑ni exe‑s‑Ø‑ti.
Marcos Dora‑obj see‑perf‑pres‑ind.3
‘Marcos saw Dora.’
80 middle as a basic voice system
b. Marcosï exe‑kurhi‑s‑Ø-ti
������������
Marcos see‑rflx‑perf‑pres‑ind.3
‘Mar saw himself.’
Further evidence of the same situation is exemplified in (14a). The middle suf-
fix ‑ki designates ‘good quality, matureness’ and similar attributes. In the active
transitive construction (14b) the middle marker must be preserved immediately
after the stem for the causative marker -ra to apply. Failure of keeping the middle
marker renders an illegal output as in (14���
c):
These are stems that require a formative marker before getting inflection.
Locative middle markers present exactly the same behaviour. Locative middles
are marked with a rich set of old body-part terms now grammaticized to designate
geometrical schematic locations. As (15) shows, in lack of further specification,
the locative marker designates the subject’s sphere:
As in the case of the attributive middle (16), the transitive locative construction
is derived with the causative marker ‑ta following the middle marker. Thus for
this verb class the middle is more basic than the transitive construction:
It is not the case that for all classes the middle is the basic form. In what Nava
(2004) calls “basic events” (mostly agent-patient interactions) middle and active
are mutually exclusive. In (17a) ‑mi marks liquid objects in intransitive middle
constructions, while -ma does the same job for transitive active ones, as in (17���
b):
These contrasts are quite productive in P’orepecha and they take several con-
trastive duplets (‑ta act/ ‑ra mid; ‑rhi act versus ‑kurhi mid). A further example
is the opposition between ‑ta act, ‑ku mid now with locative middle markers:
The fact that the reflexive -kurhi consistently occupies a position external to
the middle marker in self-benefactive, causative-reflexive and emphatic-reflexive
constructions and in no case does -kurhi constitute the base form for deriving
other constructions attests for the marked character of the reflexive and for the
more basic nature of the middle. A basic Cognitive Grammar representation of
the system is provided in Figure 10:
ricardo maldonado 83
S/A O/P S S
S/A O/P
Reflexive
For some classes the middle is as basic as the active transitive and for others
it constitutes the basic form for either reflexive or transitive constructions. The
reflexive may develop from either the transitive or the middle construction and
under no condition is the reflexive the basic form for the middle.
In this section I provide evidence form unrelated languages showing two points:
that the middles need not develop from reflexives and that middles may be a non-
derived base form. Illuminating work by Messineo (2004) on Toba, a Guaycurú
language from South America has shown that the middle is not only a very pro-
ductive form in verb formation but it also constitutes the base form for reflexive
constructions. A list of verbs that take the middle as the base form is offered in (20):
More important is the fact that in order to make a reflexive construction the
middle marker n- must be present. The base form in (21a) and (22a) licenses the
derived reflexive with -lat in (21b) and (22b). In absence of the middle marker the
output is illegal as in (21c) and (22c):
84 middle as a basic voice system
(21) a. n-awote.
mid-he love
‘He loves/he is in love.’
b. n-awoh- l’at.
mid-he loves-rflx
‘He loves himself.’
c. *awoh- l’at
he loves-rflx
b. n-alawah- l’at.
mid-kill-rflx
‘Commit suicide.’
c. *alawah- l’at
kill-rflx
The special properties of verbs like ‘love’ and ‘kill’ may suggest a preference
for using the middle, however the requirement of having the middle form for the
reflexive rule holds even for unquestionable active transitive verbs, such as rock in
(23b).
(23) a. i-shiwek.
‘He rocks someone.’
b. n-shigue-l’at.
mid-rock-rflx
‘He rocks himself.’
We may conclude that Toba also argues against the traditional view that mid-
dles must develop from reflexives.
Otomi as described by Palancar (2002, 2006) is another language where the
middle is not derived from the reflexive construction. The amount of lexical classes
ricardo maldonado 85
The semantic areas where the middle falls are the expected emotional change
of state and change of position-location verbs. Palancar shows that �������������
46 out of 72
verbs are clearly denominal. However, the semantics leading to the creation of the
novel middle lexeme is not always clear. Here are some clear cases:
(25) Source Middle
a. tsü (in) ‘get frightened’ > n-tsü (in) ‘be afraid’
b. ’ani (tr) ‘ask a question’ > ñ-’ani (in) ‘get informed’
c. pa’t’i (tr) ‘heat something’ > m-pa’t’i (in) ‘warm oneself ’
b. dá=ñ-hand-a=ka=r ñhe
1.pst=m-see-b=loc=sg mirror
Intended reading: *‘I looked at myself in the mirror.’
Actual reading: ‘I was facing the mirror.’
The type of contrast found between middles and reflexives resembles the one
found for P’orepecha where the middle is the unmarked form and the reflexive
ricardo maldonado 87
marks a special situation. Thus (28a) is the routine everyday action of putting
make up while (28b) designates a situation where make up is being used for a spe-
cial occasion and applied on the skin with special care:
The reflexive construction combines freely with all highly transitive verbs. Ex-
ample (29a) is an expected result since the reflexive is used for situations involving
control and volition. Given that the middle has the opposite properties it is ex-
cluded from combining with verbs having high demand of control as in (29b). Yet
the reciprocal construction involving two participants working as a unified entity
in some action requires the middle not the reflexive as in (30):
b. *Wayan ma-jagur
Wayan mid-hit
Amharic provides more evidence against assuming a derivation from the reflex-
ive to the middle. In contrast with Balinese, Amharic does allow for the middle to
combine with verbs of high degree of transitivity. According to Shibatani (2001)
the reflexive develops from a body part noun ras ‘head’ which has grammaticized
into a reflexive. This form is totally unrelated to the middle marker -te. Now the
possibility of using two different markers allows very different construals. As ex-
88 middle as a basic voice system
pected, controlled acts take the reflexive as in (31a) and uncontrolled spontaneous
events take the middle as in (31b):
b. Lemma te-metta.
Lemma mid-hit.pf.3m
‘Lemma was hit.’
*‘Lemma hit himself.’
And again routine actions and reciprocals take the middle, not the reflexive
marker as attested by examples (32) and (33):
Finally, it is the middle not the reflexive the marker that extends to derive an in-
transitive decausative construction, where the agent is allegedly deleted, as in (34):
The data so far seen support the claim that middles may constitute a basic type
of conceptualization. In a variety of languages the middle is a basic, non-deriva-
tive form and it constitutes the basis to derive either reflexive or transitive con-
structions. In other languages it may derive from transitive stems but in none of
the cases seen here does it evolve from the reflexive. As pointed out by Johanna
Nichols (personal communication at the conference) the general tendency to see
ricardo maldonado 89
4. Spanish middles
The idea that the middle may be a basic form has already been pointed out by
Manney (2000) for Modern Greek. A revision of so called “Spanish Reflexives”
(Maldonado 1988, 1993, 1999a, 1999b, 2000) has shown that most uses of the
clitic se (me 1st, te 2nd, nos 1st pl os 2nd pl) receive a natural explanation as middle. The
number of cases where it can actually be claimed that the middle develops from
a reflexive in Spanish is quite reduced. I will limit myself to showing the general
areas where the middle emerges and I will stress the fact that an analysis in terms
of reflexivity or argument loss is inadequate for Spanish. I will suggest that an ap-
proach to se from the perspective of the internal coherence of a middle system will
show that instead of “losing” there is an increase in energy and involvement which
has not benn captured from previous analyses.
I have already suggested that reflexives provide a split representation of the
subject as in (35) and that they are naturally expressed in mental spaces repeated
here for convenience as (36):
(35) Está bien que lo respetes, pero es más importante que te respetes.
‘It is all right that you respect him, but it is more important that you respect yourself.’
I have also pointed out that middle constructions involve no split representa-
tion. They convey an event happening within the subject’s dominion (Maldonado
1992, 1999). An example of imaginar in a middle construction is provided in (37)
where me refers to an internal mental image of a probable action and not to a sepa-
rate representation of the self:
90 middle as a basic voice system
It makes sense that middles don’t take mismo expansions. To the extent that the
middle depicts events that happen within the subject there is no way to split the
representation of the subject and that of the self. Thus becoming sad is something
we undergo not something cause ourselves to undergo. In the same manner, we
don’t use our hand and arms to get ourselves in a standing position. We simply
transmit energy internally to stand up. Thus the ungrammaticality of the examples
in (40) is predicted:
The syntactic consequences of the two construals are also predicted. Instead of
having an agent acting on a patient the subject is an experiencer, that in most cases
enacts actively some change of state. The following cognitive grammar representa-
tion is offered to illustrate the contrast. The reflexive has the same basic represen-
tation of the transitive construction with the exception is the dotted line of corre-
spondence connecting subject and object to signal coreferentiality. In contrast the
middle has only one participant undergoing some change.
ricardo maldonado 91
These cases have been treated in Spanish grammars as exceptional or even aber-
rant cases from a general reflexive pattern (Alonso and Henríquez Ureña 1953, Gili
92 middle as a basic voice system
Gaya1955, Goldin 1968, González 1985, King and Suñer 1999). The question is
whether actually the middle develops conceptually from the reflexive construal.
As recently shown by Gutiérrez Ordóñez (ms.) the historic evolution is quite dif-
ferent. Gutiérrez Ordóñez shows that the use of the term “passive” for the Latin
suffix –r in verbs like amor, amabar, amabor, etc. is wrong since such suffix served
at least four functions passive, middle deponent verbs, reflexive (corporal self care)
and impersonal.
As is well known the –r suffix was lost as all the final consonants were dropped
in the evolution from Latin to Romance. There was a considerable number of
adjustments in the system to cope with such loss. The passive took the Latin
periphrastic construction which entered the Spanish language as ser ‘be’ + past
participle.
The old reflexives vestior, tondeor, lavor, etc. took the active form with the per-
sonal and the reflexive pronouns: Ego me lavo, Tu te lavas, Ille se lavat. Crucially, the
old middles took the same pattern. Thus the use of the reflexive pronoun was a
strategy that covered the loss of both reflexives and middles as shown in Table 2:
Moreover the reflexive was also used for passive and impersonal constructions.
Given these facts it cannot be claimed that the reflexive construction itself was
established as such and then it extended to middle and then to other functions.
No one would claim for example that the reflexive was first established and then
it extended to passive or to impersonal. What can be observed is that the reflexive
was a strategy employed to cover a wide range of functions as the multifunctional
suffix –r was lost. Consequently it equally covered reflexives, middles, non-peri-
phrastic passives and impersonals. The internal organization of each construction
may be independently motivated.
An outstanding feature of the middle construction in Spanish is that it covers
basically all the range of situations described by Kemmer in her typological study
(1993) Compare Table 3 with Table 1 above:
ricardo maldonado 93
Yet such rule is not applicable for cases of motion like (44), emotional reaction
like (45) or for what I have called full exploitation middles (Maldonado 2000) as
in (46):
Motion * Ø FN2 se V
(44) Juan subió la montaña. > Juan se subió a la mesa.
‘Juan went up the mountain.’ > ‘Juan got on the table.’
Full exploitation
(46) Paulette sabe francés. > Paulette se sabe la canción.
‘Paulette knows French.’ ‘Paulette knows the song by heart.’
In none of these constructions is it the case that an argument has been deleted.
Subject and object remain and yet the clitic se is inserted to derive a middle mean-
ing. Now in Role and Reference Grammar a much more refined analysis has been
proposed by González Vergara (2006 and this volume). According to González
Vergara se is the morphemic manifestation of a lexical change where the sentence
logical structure is modified by diminishing the prominence of the actor, the most
Privileged Sytantic Argument. Consequently, the undergoer if present is profiled.
All uses of se are accounted for by this general rule. The generalization is powerful
and well motivated. It applies in different ways depending on the verb Aktionsart.
For reasons of space I cannot do enough justice to such creative proposal. I will
limit myself to providing the general way in which the rule applies and to show
the need for an alternative view (see the author’s contribution in this volume for
ricardo maldonado 95
a fuller representation of the RRG analysis). The general rule applies across the
board. It is a general instruction to underspecify the higher element in the hierar-
chy in logical structure. The difference between this and other formal approaches
is subtle but significant. It is not that the argument that is deleted it is only left un-
derspecified. Rule (47) is the most schematic representation of more specific rules.
It actually accounts for all constructions in (48).
(48) a. reflexive passive: Se construyeron muchos edificios. ‘Many buildings were built.’
b. reflexive impersonal: Se acusó a Pedro. ‘�����������������������
��������������������������� Someone accused Pedro��.’
c. middle: Esa puerta se cierra fácilmente. ‘The door closes easily.’
d. intrinsic se: Pedro se levantó. ‘Pedro stood up.’
Juan se acostumbró. �������������������������������������
‘Juan got used to (doing something).’
One can infer that in verbs like bajar ‘go down’ and caer ‘fall down’ the rule se-
lects the goal. Now what is not clear is how the rule underspecifies an argument in
(49) or, for that matter, in (50) where all the elements either from the transitive or
the intransitive source are still present. Pedro is no less the subject-actor in any of
these examples.
nager ‘swim’, voler ‘fly’ and so on. This type of contrast can be seen in a variety of
languages.
The energetic nature of Spanish middles comes from the fact that the event
is compressed to the crucial moment of change. Event compression may take
place in several ways. Most commonly we don’t see the subject acting on the ob-
ject, only the object change-of-state is being focused. Here the coincidence with
González Vergara is considerable. Now to the extent that our attention centers on
the change-of-state undergone by the only participant in the clause the event is
seen as short, rapid or even abrupt. Contrast the absolute intransitive uses of the
following motion verbs with their energetic middle counterpart. The reading for
(51b) is one of abruptness:
That the absolute is not short or abrupt can be seen from the fact that natural
and slow processes as that of the sun coming out cannot take middle marking (52b):
The gradual trajectory along a path is also construed as absolute (53a) while
the energetic middle focuses on the actual moment where the subject leaves some
location (53b), a source focus as suggested by Ibañez (2005) and González Vergara
(2006):
The same explanation takes effect in the dominion of time. For dormir ‘sleep’
the long sleeping path along the night is coded by the absolute while the energetic
change-of-state is marked by the middle:
It is rather evident that this is a matter of construal. The middle marker needs
not be used in all situations. From (55) it can be seen that the option is context
dependent. We can either see the whole process of change from sleep to vigil or,
alternatively, we may focus on the crucial moment of change. Without further
specification both options are possible, yet only the dynamism of the energetic
middle can highlight the crucial point in which the (sudden) change from sleep
to vigil takes place. Crucially the dynamic reading of abruptness is more natu-
rally associated with the middle construction not so clearly with the intransitive
in (55):
One problem with González Vergara’s analysis is the overwhelming tendency to ana-
lyze most verbs as deriving from a transitive source even in cases where the intransitive
source is a closer relative. For instance dormir ‘sleep’ and despertar ‘awake’ are not treated
in the same manner. While dormir is a decausative obtained from the transitive by the
general rule, despertar is an aspectual derivation by the second application of the rule. The
implication is that dormir has a transitive counterpart while despertar does not. His argu-
ments to defend such contrast are not convincing. The behavior of both verbs is exactly the
same: the middle use contrasts with the intransitive base form not with the causative one.
The contrast between dormir and despertar is related to the natural length of dormir and
the shorter period of despertar. Yet the import of the middle marker is exactly the same.
ricardo maldonado 99
The focusing function of the middle energetic as contrasting with the intransi-
tive absolute is represented in Figures 13 and 14. The middle as centered in the
subject participant condenses the event to depict an event reduced in time as de-
picted by the small square:
S S
There is a further extension obtained from the focusing function of the middle
marker. Events marked for the middle can provide a reading of counter-expec-
tations. While (57a, b) depict natural events as rain falling or a ball falling in a
basketball game, the case of (57c) contradicts the natural conceptualization of hu-
mans as normally being in a standing position. The unexpected loss of control is
marked by the middle energetic marker:
c. El borracho se cayó.
‘The drunkard fell down.’
The issue of expectations can be seen in further examples. Dying can be seen
either as a natural biological phenomenon as in (58a,b), or it can correspond to an
event that the speaker did not expect to happen as in (59):
100 middle as a basic voice system
(59) Don Keño se murió sin que su hijo pudiera hablar con él.
‘Don died before his son could talk to him.’
A strong argument proving that the middle encodes the speaker’s expectations
is that the use of the dative marker le which represents the subjective affectedness
of Juan –the external possessor of papá—requires the use of the middle marker as
in (60a). In its absence the output is ungrammatical, as in (60b):
The dative construction indicates that the experiencer simply suffers some af-
fectedness, the middle highlights that s/he actively enacts an emotional change. In
(62a) what kids do is what is important, while in (62b) the way I react at kids is
really what matters.
Whether the middle contrasts with the intransitive, whether it is in opposition
with the dative of emotional reaction the middle imposes an increase either in
participant involvement, or in speed. It may also incorporate the speaker’s expec-
tations. Either from the intransitive (Figure 17) or from the emotional reaction
construction (Figure 15) we move to the middle (Figure 16). In the middle con-
struction we have a compressed event where the change-of-state is in focus and
the degree of participation increases.
Figure 15. Emotional reaction Figure 16. Energetic Middle Figure 17. Absolute
Another area where the middle involves increase of energy is the case of the
full exploitation middle, a construction where the subject maximally exploits the
object (Maldonado 2000). In verbs of consumption comer ‘eat’, fumar ‘smoke’, beber
‘drink’, tragar ‘swallow,’ etc. and its abstract manifestation of consumption predi-
cates saber ‘know’, aprender ‘learn’ the subject not only brings the object into her/
his dominion as other verbs do (tomar ‘take’, agarrar ‘grasp’, comprar ‘buy’, con-
seguir ‘get’) but also exploits it in different ways. Crucially, the use of the clitic se
highlights the fact that the whole object is consumed. Thus, in (63b) and (64b),
the respective utterances entail that Victor ate the whole portion of meat and that
102 middle as a basic voice system
Adrian has read the whole paper. It is also implied that they both accomplished
it in a short span of time. The transitive construction without se remains neutral
about both the total affectedness of the object and the time efficiency of the mid-
dle construction:
The contrast is parallel to the lexical difference in English between drink and
drink up, where the particle entails full exploitation. One could claim that the
clitic se is nothing but an aspectual marker that changes activities into accomplish-
ments. This conclusion would be wrong, however, since the meaning imposed by
se is quite more specific. It can take imperfect as well as perfect and the restrictions
follow the properties of highly transitive constructions.
Notice that the clitic itself is not responsible for the aspectual change. From the ex-
amples in (iia, b) it can be seen that there are accomplishments without se:
The middle marker not only indicates that the object is of good quality but also
that the subject is highly involved as in (65a) and highly expressive as in (65b).
The involvement increase is also clearly attested in the Spanish middle system
not only for transitive verbs of consumption but for emotion. The clitic se con-
sistently designates a higher degree of subject involvement as in (66b). Without
se the feeling is neutral as in (66a). Notice that adverbial phrases reducing the
subject’s involvement would be incompatible with the use of the middle marker,
as in (66c):
on the object noun are stringent, aspect is more flexible. The event tends to be perfective as
in (63b) or (64b) but imperfective events are also possible, as in (iiia, iiib):
(iii) a. Se estaba comiendo la carne cuando oyó un disparo.
‘He was eating his meat when he heard a shot’
b. Don Nico se bebía su tequila antes de comer.
‘Don Nico would drink his tequila before supper’
As for the object, it must meet every property expected for accomplishments. For a
whole thing to be totally affected it must be clearly identified, isolatable and easy to ma-
nipulate. The object must be bounded and individuated although it needs not be specific,
thus mass nouns and generics are out (*Se tomó café ‘He drank up coffee’, *Se comió tortillas
‘He ate up tortillas’)
104 middle as a basic voice system
Finally the ground is set to account for deponent verbs which in traditional
grammars are simply listed as lexical reflexives. While there is no question that
these verbs can only be construed with the middle marker, the meaning is consis-
tent with what the middle designates. Spanish deponent verbs designate energetic
readings. From (67a-c) it can be observed that the subject deeply participates in the
emotional act and the absence of a middle marker the output is not grammatical:
These verbs were all intransitive in early Spanish until the xvii century where
the optional use of se became obligatory as the high degree of the subject’s in-
volvement lexicalized in the verb. That the physical manifestation of an emotional
feeling is not only volitional but intense can be seen from the fact that using man-
ner adverbials like con desgano ‘unwillingly’, sin realmente sentirlo ‘without really
feeling it’ render the construction unacceptable:
Moreover the intense involvement of the subject can be seen from the fact that
the middle construction can take volitional modifiers:
These facts argue against the claim that the clitic se is a reflexive marker or
that it can be reduced to an argument deletion marker. Instead they show that the
middle marker depicts a high degree of energy which may be presented in terms
of subject involvement in the event. In cases where the participant involvement
increase is not present either the event is dynamic as it signals speed and sudden-
ness or it introduces the speaker’s expectations. The increase is now on the part of
the conceptualizer.
5. Conclusions
In this paper I have tried to show that middle voice can be a basic construction in
different languages of the world. I have argued against the widespread assumption
that the middle necessarily develops from a reflexive construction. From a set of
unrelated languages I have provided evidence that the middle may be a basic con-
struction. As such, it can be the basis for the development of other constructions.
One perfectly good candidate to derive from the middle is precisely the reflexive
construction as shown by Toba, P’orepecha and Amharic. In other languages the
middle may in fact develop from transitive roots as in Yucatec Maya and Otomi
but this development has no relation with the reflexive construction at all. Balinese
and Amharic corroborate the view that reflexives and middles are totally indepen-
dent being the middle a more basic construction. I have also tried to show that the
idea that the middle develops from the reflexive is not only an Indo-European
story but more importantly a simplified misinterpretation of Indo-European lan-
guages. Based on Spanish I have shown that the rule by which either the subject or
the object are deleted by the clitic se to obtain a decausative or a reflexive construc-
tion is a gross overgeneralization that loses the core function of the middle marker.
While the subject may be downplayed, as nicely put forward by González Vergara
(2006), what is crucial about the clitic se is that it fulfills the expected middle
functions found across the languages of the world. As I have proposed elsewhere
(Maldonado 1992, 1999) the middle compresses the event into one participant
and it focuses on the crucial moment of change-of-state to obtain an energetic
event. The increase in speed and suddenness of the action is but one manifesta-
tion of the energetic projection of the middle event. The higher degree of involve-
ment of the experiencer in full involvement transitive constructions and in verbs
of emotion and emotional reaction is another manifestation of this phenomenon.
106 middle as a basic voice system
References
Canteli Dominicis, M., and Reynolds, J. 1994. Repase y Escriba. Curso Avan-
zado de Gramática y Composición. ���� New ������������
York: Wiley.
Faltz, L.M. 1985. Reflexivization: a Study in Universal Syntax. New York: Gar-
land.
Fauconnier, G. 1985. Mental Spaces: Aspects of Meaning Construction in Natu-
ral Language. Cambridge: MIT Press.
Foster, M. 1969. The Tarascan Language. Berkeley and Los Angeles: University of
California Press (publications in linguistics, 56).
Friedrich, P. 1970. Shape in grammar, Language. Vol.46, Núm.2, pp. 379-407.
—-. 1971. The Tarascan suffixes of locative space. Meaning and morphotactics. (Lan-
guage Research Monograph, 9). Bloomington: University of Indiana Press.
Gili Gaya, S. 1955. Curso superior de sintaxis española. Barcelona: Editorial Spes.
González Vergara, C. 2006. Las construcciones no reflexivas con se. Una pro-
puesta desde la gramática del papel y la referencia. PhD dissertation. Madrid:
Universidad Complutense de Madrid.
Gutiérrez Ordoñez, S. ms. Activas, pasivas, impersonales y medias. Real Aca-
demia de la Lengua Española. Apuntes para la Academia.
Grimshaw, J.1982. On the lexical representation of Romance reflexive clitics.
In Mental Representations of Grammatical Relations, J. Bresnan (ed.). Cam-
bridge: mit Press.
Haiman, J. 1983. Iconic and Economic Motivation. Language 59:781-819.
Hill, J. 1969. Volitional and Non-volitional Verbs in Cupeño. Papers from de Re-
gional Meeting of the Chicago Linguistic Society 5: 348-356.
Ibáñez, S. 2005. Los verbos de movimiento intransitivos del español. Una aproxi-
mación léxica-sintáctica. México: Instituto Nacional de Antropología e His-
toria y dgep-unam.
Kemmer, S.1993. Middle Voice. Amsterdam/Philadelphia: John Benjamins Pub-
lishling Co.
—-. 1994. Middle Voice, Transitivity and Events. In Voice Form and function, B. Fox
and P. Hopper (eds.). Amsterdam/Philadelphia: John Benjamins. 179-230.
King, L. and Suñer, M. 1999. Gramática española. Análisis y práctica. New York:
Mc Graw Hill.
Langacker, R. 1987. Foundations of Cognitive Grammar Vol. 1. Theoretical Pre-
requisites, Stanford: Stanford University Press.
—-.1991. Foundations of Cognitive Grammar Vol. 2: Descriptive Application, Stan-
ford: Stanford University Press.
108 middle as a basic voice system
Lehmann, C. ������
1995. Thoughts on grammaticalization, Studies in Theoretical Lin-
guistics 1, München – Newcastle: Lincom Europa.
Lyons, J. 1968. Introduction to theoretical linguistics. Cambridge: Cambridge Uni-
versity Press.
Maldonado, R. 1988. Energetic Reflexives in Spanish. Proceedings of the Annual
Meeting of the Berkeley Linguistics Society. 14: 153-165.
—-. 1992. Middle Voice: The Case of Spanish se. Ph.D. Dissertation University of
California, San Diego.
—-. 1999. A media voz: problemas conceptuales del clítico se en español. México: In-
stituto de Investigaciones Filológicas, unam.
—-. 1993. Dynamic
�������������������������������
construals in Spanish. Studi italiani di linguistica teorica e ap-
plicata, XXII-3.
—-. 1995. Reflexividad y niveles de activida. Memorias del II Congreso Nacional
de Lingüística de la Asociación Mexicana de Lingüística Aplicada. ������ México.
AMLA. 43-65.
—-. 2000. Conceptual Distance and Transitivity Increase in Spanish Reflexives. In
Reflexives Form and Function, Frajzyngier, Z. (ed.). Amsterdam: Benjamins.
Maldonado, R. and Nava, F. 2002. Tarascan Causatives and Event Complexi-
ty. ���
In The grammar of Causation and Interpersonal Manipulation, Shibatani,
M. (ed.) Amsterdam: Benjamins. Typological Studies in Language 48,
157-195.
Manny, L. 2001. Middle Voice in Modern Greek: Meaning and Function of a
Morphosyntactic Category, Amsterdam, Benjamins.
—-. 2002. The Reflexive Archetype and its Various Realizations in Modern
Greek. Studies in Language 22.1, 1-48.
Martínez, I. 2006. Reflexiones sobre voz media en el maya yucateco. Maestría en
Lingüística Indoamericana. Centro de Investigaciones y Estudios Superiores
en Antropología Social.
Martínez, I. and Maldonado, R. in press. Middles and Reflexives in Yucatec
Maya. In Fieldwork and Linguistic Theory in American Indigenous Languages,
University of New Mexico.
Messineo, C. 2002. La marcación verbal activa/inactiva en toba (gauycurú) y sus
motivaciones. LIAMES (Revista de Lingüística Indígena americana) 2. Campi-
nas: Instituto da Linguagem, Departamento de Lingüística, unicamp. 38-50.
Nava, F. 2004. Voz media en tarasco. Ph.D. Dissertation. Universidad Nacional
Auntónoma de México.
ricardo maldonado 109
1. Introduction
Japanese is known for its array of argument encoding types, ranging from a rich
stock of post-nominal markers to variations in constituent ordering. In this study,
I propose unified descriptions of the discourse-pragmatic properties of a range
of argument coding types, including those commonly found in spoken Japanese
—the two case markers ga (nominative) and ga (accusative), the so-called “topic”
marker wa, the zero marker (i.e. absence of overt post-nominal marking), and
post-predicative placement of arguments (a.k.a. postposing). These morphosyn-
tactic forms have been discussed from a variety of angles in the literature; how
ever, the previous studies have failed to describe these forms in a unified frame-
work with respect to how the paradigmatic and syntagmatic choices are made by
the speaker in production and how the form-function relationships are repre-
sented in comprehension. For the purpose of the present study, I discuss results
from quantitative analyses of spoken Japanese to probe the form-function rela-
tionships and describe the findings in RRG. The proposed RRG representations
utilize discourse representation structures and focus structure projections as well
as interactions of linking with clause structures, and further elaborate the current
formulation of discourse representation to properly describe the observed proper
ties of the argument codings.
There are other possible ways to mark argument NPs, such as -mo ‘also’ and -tte (quota-
tive); however, these markers denote more concrete meanings and thus exhibit more trans-
parent properties in light of speakers’ choices of the markers. Thus, the present study limits
its scope to the five post-nominal forms given.
113
114 focus structure and beyond: discourse-pragmatics in rrg
2. Discourse-pragmatics in RRG
There are two parts of RRG representations which are directly linked with dis-
course-pragmatics. The “focus structure projection” (Van Valin 1993, Van Valin
& LaPolla 1997: 201-218) represents the information structure of a sentence by
way of the “potential focus domain” and “actual focus domain”. The “Discourse
Representation Structure” (Van Valin 2005: 170-174) represents referents and
propositions in terms of presuppositions and assertions (Lambrecht 1994). It has
been proposed (Van Valin 2005) that the focus structure projection of a sentence
is derived from the discourse representation structure of the sentence, as the ac-
tual focus domain of a sentence in the focus structure projection is determined by
the presupposition/assertion discourse representation structures of the sentence,
which represent the given discourse context. An example of a simple sentence is
given in Figure 1.
x x, y
In what follows, I discuss the use of six different morphosyntactic forms for
nominative and accusative arguments in spoken Japanese and propose RRG rep-
resentations of the observations to demonstrate a model of how the form-function
relationships can be captured by this theory of syntax. As will be discussed later,
I propose that proper representation of Japanese morphosyntax requires further
elaboration of discourse representation structure by incorporating a greater range
of discourse-pragmatic properties beyond presupposition and assertion.
The following abbreviations are used for the examples: acc=accusative, cl=classifier,
The present study discusses six argument forms in Japanese. Two case markers
(nominative and accusative), two adverbial markers (“topic” and the zero particle),
zero anaphora, and post-predicative placement of an argument, as shown in the
following examples.
There are five possible forms for an argument. The case marker ga or o, de-
pending on the argument role, wa, the zero marker, and zero anaphor represent
the paradigm, out of which a particular form is chosen for a given argument (as
shown in (a)-(d) examples). In contrast with the default verb-final ordering in
the (a)-(d) examples, the (e) examples show the marked ordering in which the
mitsuaki shimojo 117
argument is placed after the verb. In these cases, the same paradigm of post-
nominal marking (except for zero anaphor) applies to the post-predicative ar-
gument NPs. For this reason, post-predicative arguments exhibit an interesting
interplay between the functional property of particular NP forms and that of the
marked word ordering. This will be discussed later with respect to the defocusing
of referents.
The spoken Japanese data used for the present study consists of two-party infor-
mal conversations by eight pairs of native speakers, which were recorded at the
University at Buffalo in June 2002. Each pair’s conversation is 30 minutes long,
and the entire data set consists of four hours of conversation in total. The native-
speaker participants in the recordings consist of ten females and six males, whose
ages ranged from 19 to 36, and they were mostly from the Kanto area of Japan
—the eastern central region with Tokyo in the center. The paired participants were
mutual friends and they used the plain (i.e. casual) form of predicates consistently
without the -desu or -masu (distal) form of the predicates, except for one pair, in
which the distal forms were mixed with the plain forms.
The recorded conversations were transcribed and segmented into clausal units.
In this procedure, utterance boundaries were identified according to pause and the
transcribed text was divided accordingly. With respect to clausal units in which
nominative and/or accusative arguments followed the predicate, only those cases
where there was no intervening pause between the predicate and the following
Although examples for nominative and accusative arguments are given separately here,
transitive sentences contain both nominative and accusative arguments, each of which rep-
resents the paradigmatic choice for the NP form. For example, all sentences given in (4)
contain an ellipsed nominative NP (i.e. a zero anaphor). Also, it is structurally possible that
both nominative and accusative arguments appear after the verb, though it is rare in actual
discourse.
The adverbial particle wa is commonly labeled as a “topic” marker in the literature, and
the present study uses this traditional label also. However, it should be noted that this label
may be misleading if “topic” is taken as its discourse property because wa-marked referents
do not necessarily represent a continuing discourse theme. As will be discussed later, this
study claims contrastiveness as the primitive property of wa.
118 focus structure and beyond: discourse-pragmatics in rrg
5. Givenness
In (6), the nominative argument in the answer can be marked with ga or the
zero particle (though the zero particle would make the utterance more colloquial
sounding), but it cannot be marked with wa or ellipsed all together in this par-
ticular context (as indicated by #). In this sentence, the whole sentence including
the nominative argument represents actual focus, and it is reasonable to assume
that ‘car’ is unidentifiable and nonactive in the preceding context but can be either
shared or unshared (i.e. the existence of the car may be known or unknown to the
hearer prior to the utterance). In (7), the nominative argument represents argu-
ment-focus, and in this context, ga is the only possibility. Although ‘car’ is activat-
ed and shared prior to the utterance, it is not identifiable for the hearer (who asked
the question) with respect to the proposition ‘X broke down’ as the hearer does not
know which broke down. The answer in (8) represents predicate-focus, and it is
most natural to ellipse the nominative argument. The zero-marked NP makes it
repetitious but acceptable (hence, the question mark), and ga or wa are less than
acceptable in this context.
The following set of data points to the same pattern with respect to accusative
arguments, though there is a slight difference from the nominative case when the
referent is identifiable.
120 focus structure and beyond: discourse-pragmatics in rrg
Zero Zero
Givenness type GA O WA
anaphor particle
Unidentifiable no yes yes yes/no no
Identifiable yes no (yes)* (yes)* no
Nonactive no yes yes yes no
Active yes/no yes/no yes yes no
Shared yes/no yes/no yes yes no
Unshared no yes/no yes/no yes/no no
*For identifiable referents, -o and the zero particle are acceptable but zero anaphor is preferred.
The observations suggest the following. First, not all NP forms are linked with
givenness. The zero particle may be used in any givenness condition, whether the
referent is previously given or new, regardless of givenness type. On the other
hand, wa is not acceptable in any of the examples. Thus, it is reasonable to assume
mitsuaki shimojo 121
that givenness is not a relevant index to describe the use of the zero particle and
wa, and whatever property that makes the use of wa acceptable is not present in
any of the examples above. Secondly, not all givenness types are relevant to de-
scribe the properties of the NP forms. In fact, it is only identifiability that shows
a regularity with respect to the use of zero anaphor and the case markers. Zero
anaphor is clearly the choice for identifiable referents and ga for unidentifiable
referents. A similar pattern is found with o since it is not the best choice for iden-
tifiable referents, though it is acceptable. However, it is noteworthy that the case
markers are the only possible choice if the argument represents narrow argu-
ment-focus, when the purpose of the utterance is referent identification, as in the
case of (7). Activation of referents (i.e. saliency of referents in consciousness) and
shared knowledge do not seem to be directly relevant to the givenness associated
with zero anaphor and ga.
In order to discuss the discourse data quantitatively in terms of givenness, I
present the results from referential distance (RD) measurements below. RD is a
linguistic distance in clausal units measured backward to the most recent repre-
sentation of the coreferential expression (see Givón 1983). For example, an RD
of 1 indicates that the referent in question was represented in the immediately
preceding clausal unit, which presumably makes the referent more given (i.e. more
readily identifiable and more activated) than a referent of RD 20, for example, for
which the most recent previous reference was given 20 clauses ago in the preced-
ing context. The measurement results are presented in Table 2.
O and ga contain cases of argument-focus (in which there is a recent previous co-refer-
ence but it is unidentifiable); therefore, these cases bring down the mean RD closer to the
other three forms.
122 focus structure and beyond: discourse-pragmatics in rrg
The results reflect the observations discussed earlier. Zero anaphora is found
at the lowest end of RD and ga at the highest end, which suggests a natural pair-
ing of these two forms in terms of givenness. Similarly, the accusative marker
exhibits a pattern similar to the case of ga, in which there are more tokens in
the range of larger RDs, but it shows more tolerance toward smaller RDs at the
same time, as predicted by the observation discussed earlier (Table 1). Notice that
the distribution of wa and the zero particle is relatively ambivalent in terms of
givenness, being distributed more equally between the two ranges of RDs. The
ambivalence is also shown by the mean RD; wa falls at about halfway in the 20-
point RD scale.
6. Contrastiveness
In search of a property to describe wa and the zero particle, I take up the notion
of contrastiveness because there has been a claim to associate wa with contras-
tiveness (Kuno 1973, Clancy & Downing 1987). As a starting point for the dis-
cussion, I use Clancy & Downing’s (ibid) characterization of contrastive context,
which is given in (12).
b. Action/state reaction:
A contrasted state of affairs is presented in reaction to a state of affairs present-
ed previously, typically by a different speaker.
The first type of direct contrastive context “parallel activities/states” is one that
fits the template of “A does/is X but B does/is Y” given by the same speaker. In the
other type “action/state reaction”, a previously given referent is contrasted with a
newly given referent. Table 3 shows the token counts for each type of contrastive
context in the present spoken Japanese data. The result shows a clear tendency
for wa to be used in a contrastive context (82% of the total), and the zero particle
typically appears in a non-contrastive context (91% of the total).
mitsuaki shimojo 123
b. Perceptual judgment:
Speaker’s description of temporary states of affairs
The terms “experiential” and “perceptual judgment” are used by Uchida (1989). See also
Shibatani (1990) for discussions of the two judgment types with respect to topic marking
in Japanese.
The zero particle tokens here include only nominative arguments because the notion of
experiential judgment relates to topicalized sentences and it is not clear whether the zero-
marked arguments are topicalized NPs (which modify the entire sentence), particularly in
the case of accusative arguments, whose topicalization is much less common.
124 focus structure and beyond: discourse-pragmatics in rrg
Contrastive Non-contrastive
Experiential judgment 122 .36 30 .09
To summarize, wa and the zero particle reflect a mirror image in terms of con-
trastiveness, as the two forms are overwhelmingly associated with contrastiveness
and non-contrastiveness respectively. This empirical observation correlates with
the inherent properties of the two forms, which may be labeled contrastive speci-
fication and absolute specification of referents respectively (Lee 2002). If an overt
marking denotes a particular pragmatic meaning which more or less represents
some kind of relationship with other referents/propositions (explicit contrast by wa,
exclusive identification by ga, etc.), then with the absence of an overt marker, the
sentence subdues the tone which would be achieved by an overt particle and refers
to the target referent in an absolute manner which makes no reference to others.
7. Saliency (importance)
Pre-predicative Post-predicative
Zero Zero
O WA GA
anaphor particle
Mean RP-f 3.2 2.8 2.7 2.3 1.9 1.9
Mean RP-d 2.1 3.0 2.6 2.0 1.1 0.7
Table 6. The six forms by mean RP-f and RP-d (Shimojo 2005: 114)
In terms of mean RP-frequency, the zero particle and the post-predicative ar-
guments exhibit the least referential persistence, and this pattern is even clearer in
terms of mean duration of uninterrupted persistence. There is almost a whole step
down from ga to the zero particle and the post-predicative arguments. The results
here suggest that the zero particle and post-predicative placement of arguments
share the property of referential non-persistence, i.e. cataphoric defocusing of ref-
erents. Notice that zero anaphor is at the top of the scale of referential frequency,
which in turn is associated with greater identifiability of referents (see the earlier
discussion of givenness).
To summarize the data, we have seen three types of functional properties, given-
ness, contrastiveness, and saliency. Givenness and contrastiveness are associated
126 focus structure and beyond: discourse-pragmatics in rrg
with the morphological means of Japanese, and saliency with the syntactic means,
as shown in Table 7.
Each of the functional properties may be described as follows. From the sen-
tence production point of view, if the referent is identifiable at the moment of ut-
terance, no instantiation of the corresponding NP is necessary in the clause struc-
ture. If the referent is not identifiable, then the NP is marked explicitly with an
appropriate case marker. If the referent is to be presented in contrast with others,
the corresponding NP is wa-marked, regardless of identifiability of the referent.
If the referent is to be presented in the absolute manner, then no overt marking
is used. If the referent is to be defocused in the cataphoric context, then the cor-
responding NP is placed after the predicate. From the comprehension point of
view, the mapping is reversed to link the forms with their corresponding func-
tional properties, and in this process, the functional properties must be properly
represented in the contextual representations of the sentences. In what follows, I
will describe the form-function relationships outlined above using RRG represen-
tations.
9. RRG representations
tions, and representation of production and comprehension of the forms. This sec-
tion discusses the former, and the latter will be discussed in the description of
linking in the next section.
The sentence in (14c) has three ellipsed arguments, which are all retrievable
from the preceding discourse. The RRG approach to zero anaphora is character-
ized by direct linking of discourse representation structure and logical structure,
bypassing the simplified clause structure, as shown in Figure 2.
In the sentence above, it is only the verb that represents assertion; thus it is the
only overt element in the clause structure among the elements that appear in the
discourse representation structure. Hence, in the linking from syntax to semantics,
all arguments need to be retrieved from the discourse representation. In this way,
all elements in the logical structure are properly linked; thus, the completeness
constraint is satisfied.
Following the representation of the Mandarin example above, I capture zero
anaphora in Japanese as shown in Figure 3, along with a sentence that represents
128 focus structure and beyond: discourse-pragmatics in rrg
Zero anaphora is not limited to arguments in Japanese. See Shimojo (2008) for discus-
sion of zero-verb and its RRG representation.
mitsuaki shimojo 129
9.3 Saliency
10. Linking
Having described the RRG representations for the six morphosyntactic forms in-
dividually, in this section I integrate them using the linking algorithm and demon
mitsuaki shimojo 133
strate how the form-function relationships are represented in the linking from
and to semantic representations.
The linking from semantics to syntax and discourse representation structure
corresponds with the speaker’s production of the sentence, and the representation
of the six forms in the linking procedure needs to capture the speaker’s selection
of structurally and pragmatically appropriate forms. The procedure relevant to the
present study is outlined in (15).
In the linking algorithm above, the most significant steps are (15.3b-4), where
the morphosyntactic coding of the arguments and syntactic templates are deter-
mined. I illustrate the argument coding procedure with the examples given earlier
in (6)-(8), which are repeated below.
Following the linking procedure given in (15.3b), the argument coding in (16)
is predicted as follows. First there would be no need for contrastive specification for
‘car’ given the preceding context (i.e. ‘What happened?’); however, if absolute speci-
fication is appropriate, the zero particle is used. If not, then overt case marking is
used since ‘car’ is not identifiable from the preceding context. In (17), since the argu-
ment represents narrow argument-focus, the case marking overrides the others. The
example in (18) is an interesting case since the zero marking is not totally unaccept-
able. The preceding context of “What happened to your car?” (or any other readily
imaginable context based on the question) does not require contrastive specification
for ‘car’; however, absolute specification for the referent may be appropriate, par-
ticularly if the speaker does not relate the referent to others in the present discourse.
Otherwise, the default choice here is zero anaphora since the referent is identifiable.
Given the illustration above, one may wonder when wa is ever used. The fol-
lowing example would be useful in this regard.
The first part of the answer in (19) clearly requires contrastive specification for
‘car’ being OK, in contrast with the motorcycle that broke down (also the sentence
does not represent narrow argument-focus). Hence, the linking step in (15.3b.i)
assigns wa to the argument. Notice that the second part of the answer represents
narrow argument-focus on ‘motorcycle’, hence, ga overrides wa here.
One may argue that the first part of the answer renders ‘motorcycle’ in the second part
identifiable since the proposition given by the first part would imply that it is the motor-
136 focus structure and beyond: discourse-pragmatics in rrg
As in the linking from semantics to syntax, the linking procedure from syntax to
semantics needs to incorporate discourse representation structures. Furthermore,
in the linking process, the discourse-pragmatic properties associated with particu-
lar argument forms need to be so-registered in the corresponding discourse repre-
sentation structures. The procedure relevant to the present study is outlined in (20).
The linking steps in (20.4-5) reflect the viewpoint that these particular forms
in the argument coding help the hearer construct proper mental discourse repre-
sentations. In this respect, the presence or absence of contrastive links for related
referents and the index of saliency affect how information is organized and stored
in the mental representations of discourse and how the mental text develops in the
subsequent discourse. In this view, zero anaphora and the case markers, as well as
the other coding forms, play a role as mental processing instructions for the hear-
cycle that broke down. Although this may be the case in a step-by-step description of a
given state of affairs (hence, wa for ‘motorcycle’ would be better there), it is fair to assume
that in (19) both parts of the answer collectively represent a speaker’s single ‘thought unit’
as a reply to the question.
10
This is required for zero anaphora for verbs (Shimojo 2008).
mitsuaki shimojo 137
Pre-predicative Post-predicative
Zero Zero
O WA GA
anaphor particle
Zero anaphora, the case markers, and wa are all associated with saliency in
one way or another. Zero anaphora represents continuation of the same reference,
which renders the referents globally salient. The case markers are used for prag-
matic focus in the introduction of previously unidentifiable referents; hence, these
referents are rendered locally salient, even if they are not always rendered globally
salient. Wa establishes a contrastive relationship with other referents and therefore
raises accessibility of the given referent via the contrastive link (thus, facilitates co-
hesion in discourse). On the other hand, the zero particle and post-predicative ar-
guments sit on the other side of the saliency scale. The zero particle represents “no
search” for contrastive relationships and therefore dissociates the NP it attaches to
from other referents. Post-predicative arguments are placed in the “outcast” posi-
tion that follows the predicate, a critical point of sentence processing in verb-final
languages, as the semantic representation of the predicate makes it possible to link
138 focus structure and beyond: discourse-pragmatics in rrg
up the arguments put in queue.11 The characterization of each form outlined above
is summarized in Table 9.
Focusing of referents
Zero anaphor Represents continuing reference (globally salient)
wa Raises accessibility of referents via contrastive link (globally
salient)
Case markers Represents pragmatic focus (locally salient)
Defocusing of referents
Zero marker Cancels contrastive link
Post-pred. argument Delays processing of the referents
Table 9. The six forms in terms of (de)focusing of referents
11
Kuno (1987) claims a “Flow of Information Principle” which associates the immedi-
ately pre-verbal position of a sentence with the most important information in Japanese.
See Kim (1988) for a similar claim.
mitsuaki shimojo 139
12. Conclusion
This study has examined six forms of morphosyntactic argument coding com-
monly found in spoken Japanese and demonstrated that all these forms are closely
connected with discourse-pragmatic properties. These properties are linked with
all levels of RRG representations, which demonstrates that discourse-pragmatics
is not only required for grammar but also plays a role in every aspect of grammar.
Furthermore, I hope to have demonstrated that information structure defined
by the notion of givenness is insufficient to capture the dynamics of the argu-
ment coding system and that proper description of the observed form-function
relationships requires an extended functional characterization, including man-
ners of referent specification and saliency of information. For this reason, gram-
mar requires proper representations of discourse which are capable of incorpo-
rating all necessary discourse-functional properties that influence sentence forms.
While givenness of information in discourse representations may be a universally
required discourse-pragmatic property, more complex representations are required
for a language such as Japanese. Thus, the discourse representation structure de-
veloped recently in RRG provides a proper means to accommodate such language
specificity.
140 focus structure and beyond: discourse-pragmatics in rrg
References
Clancy, P., and Downing, P. 1987. The use of wa as a cohesion marker in Japa-
nese oral narratives. In Perspectives on Topicalization: the Case of WA, J. Hinds,
S. K. Maynard, and S. Iwasaki (eds.), 3-56. Amsterdam and Philadelphia:
John Benjamins.
Givón, T. (ed.). 1983. Topic Continuity in Discourse: a Quantitative Cross-linguistic
Study. Amsterdam and Philadelphia: John Benjamins.
Kim, A. H-O. 1988. Preverbal focusing and type XXIII languages. In Studies in
Syntactic Typology, M. Hammond (ed.), 147-169. Amsterdam and Philadel-
phia: John Benjamins.
Kuno, S. 1973. The Structure of the Japanese Language. Cambridge, Mass.: MIT
Press.
— 1978. Danwa no Bunpoo (The Grammar of Discourse). Tokyo: Taishukan.
Lambrecht, K. 1994. Information Structure and Sentence Form. Cambridge:
Cambridge University Press.
Lee, D.-Y. 2002. The function of the zero particle with special reference to spoken
Japanese. Journal of Pragmatics 34: 645-682.
Prince, E. F. 1981. Toward a taxonomy of given-new information. In Radical
Pragmatics, P. Cole (ed), 223-255. New York: Academic Press.
Shibatani, M. 1990 The Languages of Japan. Cambridge: Cambridge University
Press.
Shimojo, M. 1995. Focus Structure and Morphosyntax in Japanese: Wa and Ga,
and Word Order Flexibility. Ph.D. dissertation, University at Buffalo, The
State University of New York.
— 2005. Argument Encoding in Japanese Conversation. Hampshire and New
York: Palgrave Macmillan.
— 2008. How missing is the missing verb? The verb-less numeral quantifier
construction in Japanese. In Investigations of the Syntax-Sematics-Pragmatics
Interface, R. Van Valin (ed.). Amsterdam: John Benjamins.
Uchida, M. 1989. Shugo o meguru joshi no yoohookubun nitsuite (On the usage
classification of particles for subject). In Nihongogaku no Shintenkai (A new
trend in Japanese linguistics), S. Kuno and M. Shibatani (eds.), 29-44. Tokyo:
Kuroshio.
Van Valin, R. D., Jr. 1993. A synopsis of role and reference grammar. In Ad-
mitsuaki shimojo 141
vances in Role and Reference Grammar, R. D. Van Valin., Jr. (ed.), 1-164. Am-
sterdam and Philadelphia: John Benjamins.
Van Valin, R. D., Jr. 2005. Exploring the Syntax-semantics Interface. Cambridge:
Cambridge University Press.
Van Valin, R. D., Jr., and LaPolla, R. J. 1997. Syntax: Structure, Meaning and
Function. Cambridge: Cambridge University Press.
Exploring the role of pragmatics in the interface:
On the interrelations between activation status,
choice of RP and syntax
Elke Diedrichsen
University of Düsseldorf
Part of the research carried out for this paper has been supported by a grant from the
German Research Foundation (DFG). I am grateful for this support and for the valuable
discussions with the participants of the 2007 RRG conference in México-City. In par-
ticular, I would like to thank Brian Nolan, Michael Silverstein, Robert D. Van Valin, jr.,
Dietmar Zaefferer and the anonymous reviewers of this article for helpful suggestions. Any
errors are my own. Furthermore, I thank Cornelius Puschmann for advice concerning the
use and citation of internet blogs and forums for scientific purposes.
Abbreviations are as follows: acc: accusative, act: actor, asp: aspect, aux: auxiliary, cl:
clause, clm: clause linkage marker, dat: dative, dec: declarative, def: definite, deic: deictic,
dem: demonstrative, dir: directional, f: feminine, if: illocutionary force, imp: imperative, in-
def: indefinite, inf: infinitive, ingr: ingressive, loc: locative, ls: logical structure, m: mascu-
line, mod: modality, n: neuter, nasp: nominal aspect, neg: negative, nom: nominative, num:
number, nuc: nucleus, part: particle, past: past (tense), pastp: past perfective, perf: perfect,
pl: plural, pp: prepositional phrase, pocs: postcore slot, prcs: precore slot, pred: predicate,
pres: present (tense), proc: process, prop: proper, psa: privileged syntactic argument, pstp:
past participle, purp: purposive, qnt: quantifier, rp: referential phrase, sg: singular, sta: status,
tns: tense, und: undergoer.
143
144 exploring the role of pragmatics in the interface
structure that is based on the Aktionsart of the verb to a complex syntactic rep-
resentation, including rules for word order, case marking and the combination of
clauses to complex sentences. The source of the linking process is a universal se-
mantic predicate-argument-structure. The linking algorithm is supposed to be
carried out in two directions: The linking from semantics to syntax is used in or-
der to represent language production, while the linking from syntax to semantics
is applied to represent language comprehension/parsing.
This paper will be concerned with the linking from semantics to syntax. The
linking algorithm for this procedure, as proposed in Van Valin (2005:136 ff.) is a
process that has to be carried out in five steps for each individual clause, and there
are some substeps to be accounted for, depending on the kind of clause and, in
particular, on the language.
The single steps will be summarized here, and they will be dealt with more ex-
tensively in the subsequent sections.
In step 1, the semantic representation of the sentence is to be constructed, de-
pending on the logical structure of the verb. It involves the selection of referential
expressions for the nucleus and the arguments, and the determination of LSC-
and RP-operators. It also represents the activation statuses of the referents. In
step 2, the Actor-Undergoer assignment for the arguments is to be determined,
according to the semantically based Actor-Undergoer-Hierarchy given as figure
4.4 in Van Valin (2005: 126). In step 3, the morphosyntactic coding of the argu-
ments is determined. This involves the selection of the Privileged Syntactic Argu-
ment, if there is one, and the assignment of case-marking, verbal agreement and
adpositions. In step 4, the syntactic templates have to be selected. This is done in
accordance with the syntactic template selection principle stated as (5.2) in Van
Valin (2005:130). As a last step, the arguments are assigned to positions in the
syntactic representation of the sentence. Towards the last step, there is an increas-
ing number of language-particular features to be accounted for, and accordingly,
step 5 involves very many substeps, from which the appropriate steps have to be
chosen. Van Valin and Diedrichsen (2006) provide the linking algorithms for a
set of German sentences; those are formulated with respect to the German case
system and the particular word order rules. The analysis carried out here is based
on this paper.
RRG representations of sentences involve pragmatic information in various
ways. There is a focus structure projection that depicts the potential focus domain
in a given language and the actual focus domain in a given sentence. Many of the
elke diedrichsen 145
syntactic rules are directly connected to the facts of focus structure attested in
a given language. Further pragmatic information is supplied in RRG-represen-
tations with respect to the aforementioned “activation”-statuses that referents of
nominal expressions in argument positions have in discourse.
The following activation statuses are distinguished:
The activation statuses of the referents are realized in the semantic representa-
tion of the sentence. Furthermore, in the layered structure of the referring expression
(RP), an RP-operator is assigned with respect to definiteness and deixis. So, the
discourse status of the respective RP is displayed in the operator projection as well.
In the semantics-to-syntax-linking, the activation statuses are presented as a
result of step 1 and, accordingly, as part of the semantic representation (cf. Van
Valin 2005:137). The linking diagrams involving activation statuses do not make
clear, however, on which grounds those activation statuses are assigned (cf. for ex-
ample Van Valin 2005:140 f.). The rule for the “selection of referential expressions”
that is to be carried out in step 1.c. of the linking algorithm does not account for
the criteria that influence this selection. In particular, the pronoun-noun alterna-
tive, that is available in every language, is not mentioned here. The consequences
of the activation statuses in the linking process are partly touched upon in an ap-
proach using discourse representation structures that are based on the distinction
between presupposition and assertion (cf. Van Valin 2005, chapter 5.4.1). As the
activation statuses in the example are stipulated, however, their relation to the fo-
cus structure and to the choice of RPs remains rather unclear.
The terms ‘NP’ and ‘RP’ are under consideration in the RRG framework (cf. Van Valin
For an investigation of the role of the activation statuses with respect to the
choice of RPs and the linking algorithm, it would be necessary to analyze sentenc-
es from a stretch of discourse, taken from “naturally occurring language”.
As it is commonly known, the activation status of a referent in discourse deter-
mines the kind of nominal expression that is used to refer to it in certain respects
(cf. for example Ariel 1991, Gundel et al. 1993, Givón 1989). This is quite clear
in the case of pronouns, less clear in the case of definite RPs. Pronouns, especially
those referring to speaker and addressee, are inherently “active”; they display the
“activeness” of the referent in the situation. Such kinds of expressions have been
called “indexicals” in language philosophy (cf. the writings of C.S. Peirce, Bar-Hil-
lel 1970, Silverstein 1981).
As for the pragmatic organization of syntactic structures, there are construc-
tions that serve to realize “topic chaining” in discourse, for example conjunction
reduction (Van Valin 2005). Furthermore, arguments with “active” referents are
very likely to be chosen as Privileged Syntactic Arguments (Van Valin and La-
Polla 1997:421 ff.).
So, if the pragmatic conditions under which a sentence is uttered are to be taken
into account by the linking algorithm, one firstly has to assume that the activation
statuses precede the semantic representation logically. The choice of the nominal
expressions for the arguments takes place in the lexicon; it is a consequence of the
activation statuses of the referents (this is also suggested in Van Valin 2005:182,
figure 5.40). Thus, the semantic representation is based on the logical structure of
the predicator, on the one hand, and on the activation statuses of the referents, on the
other hand.
The activation statuses of referents of nominal expressions can have syntactic and
morphosyntactic effects, some of which will be explored in this paper.
In many languages, case marking is assigned depending on the type of nominal
expression. This is true for split ergative languages (Dixon 1994), but also for the
occurrence of a distinctive accusative marker in accusative languages like German.
The distribution of case markers within and across languages can be explained by
the feature hierarchy established by Michael Silverstein (1976).
elke diedrichsen 147
(1) The
�������������������������������������������������������������
Hierarchy of features and ergativity (Silverstein 1976); simplified
���������������������������
depiction after
Dixon (1994)
1st person (speaker)
+definite/referential/specific
2nd person (addressee) Highest probability of accusative marking
3 (pronoun)/demonstrative Lowest probability of ergative marking
rd
Dresdner Bank.
Dresdner Bank.sg.Ø
‘The weather (forecast) is presented to you by the Dresdner Bank.’
Thus, it seems that the German case-split revolves around the contrast be-
tween speech-act participants and third person nominals, on the one hand, and
the gender distinction between masculine and feminine, on the other hand. For
third person pronouns and common nouns, the cut-off point for distinctive accu-
sative marking lies between masculine singular and feminine/neuter/plural.
The ordering of the positions on the hierarchy is generally explained with re-
spect to cross-linguistic tendencies concerning the argument positions in transi-
tive sentences: Nominals from the top of the hierarchy tend to occur as Actors,
while those from the bottom of the hierarchy rather occur as Undergoers. In most
Ø=No indication of case marking; nom being the default unmarked case.
elke diedrichsen 149
constructions across languages, nominals from the top of the hierarchy tend to
be the Privileged Syntactic Argument (Van Valin and LaPolla 1997, chapters 6
and 7). The order of nominals on the hierarchy has been interpreted in terms of
“degrees of indexicality”; as with the use of pronouns, their referents are presup-
posed in discourse, while this is not necessarily true for other nominals (Silver-
stein 1981). For the sake of processing ease, “the thing the proposition is about” is
preferably chosen among those entities that are part of the immediate discourse
situation (Diedrichsen 2006, Lambrecht 1994). Other approaches interpret the
hierarchy in terms of degrees of “animacy” of the listed positions, in that the upper
positions are said to be “better agents” than the lower ones (Dixon 1994). The rea-
son why active participants tend to be Actors could be that there is an inclination
to code the “trigger” or the “source” of the event as the starting point of the utter-
ance (Diedrichsen 2006).
As for the masculine/feminine distinction with common nouns in German, it
has been observed that “conceptual closeness” to human beings is often coded in
the masculine, while the feminine is used to code more abstract notions (Zubin
and Köpcke 1996, Eisenberg 1999/2001). Furthermore, the feature cluster [mas-
culine + animate / human] seems to be decisive for the stability of the membership
in the only declension class that has distinctive accusative marking (Köpcke 1994,
1995). Thus, there is a motivation for case-marking distinctions with first and sec-
ond person pronouns in terms of topicworthiness due to presupposition and acti-
vation in discourse, on the one hand, and for masculine pronouns and nouns with
respect to “concreteness” and “agentworthiness” on the other hand.
In this paper, one sentence out of German discourse will be analyzed in terms
of a semantics-to-syntax linking. The procedure of the linking will be oriented
towards the linking algorithm presented in Van Valin and LaPolla (1997), Van
Valin (2005) and, in particular, the semantics-to-syntax linking for German pro-
posed in Van Valin and Diedrichsen (2006). The sentence stems from an internet
forum that is open to the public, but the writers are not professionals, neither can
they expect their contributions to be read by very many people. In general, they are
anonymous and give themselves nicknames. Readers of the forums are generally
invited to write comments on the contributions, and, in some cases, start conver-
sations with the authors. Contributions in internet forums represent interesting
pieces of naturally occurring discourse, as they are generally neither elicited nor
otherwise “observed” by the researcher. Furthermore, they are easy to access, as the
forums are open to everyone, and the contributions are saved in archives. Forum
150 exploring the role of pragmatics in the interface
contributions necessarily appear in written language, but they display many fea-
tures of oral language (Schlobinski and Siever 2005). For a non-involved observer,
the background knowledge that is necessary for understanding single sentences is
gained from the previous discourse, which means the information provided by the
author himself. In this paper, the activation statuses of the RPs in the sentence will
be assigned with respect to this background knowledge, on the one hand, and with
respect to the information provided by the sentence itself. The activation statuses
will be taken into account in the first step of the linking, so that the semantic rep-
resentation will be based on the logical structure of the predicator and on the acti-
vation statuses of the referents. Note that the representations will not account for
the focus structure projection. Assigning the focus structure to a complex sentence
from naturally occurring discourse is a very doubtful matter, and even though the
activation statuses give some hints at the locations of the “new information”, this
discussion will be left out here for the sake of simplicity.
The paper will proceed as follows: Firstly, the sentence to be analyzed will be
introduced with respect to the forum where it has been uttered and the broader
context it appears in. The following section provides a lexicon and the syntactic
templates for the sentence and specifies some general RRG-based principles of
German grammar that are needed for the analysis of the sentence. After that, the
analysis of the sentence parts in terms of RRG is carried out. For each of the three
clauses the sentence is composed of, a semantics-to-syntax-linking will be given.
The results of these analyses will be depicted in three single tree-structures with
the respective constituent- and operator-projections for each of the clauses.
The paper closes with a short summary. The final section will also contain a
complete tree structure of the complex sentence with constituent and operator
projections.
1.3. Material
The sentence that is analysed here is an attested example from the internet forum
ciao.de, where customers are invited to evaluate products, services and companies.
In the respective stretch of discourse, somebody who calls himself “logge” reports
that he wanted to buy a medallion for his girlfriend at www.neckermann.de
�������������������������
(a
�������
big
German catalogue company that also offers internet shopping) and tells the read-
ers about his experiences.�����������������������������������������������������������
(3)
����������������������������������������������������������
is an excerpt from: “Erfahrungsbericht von logge über
elke diedrichsen 151
zu durchforsten.
part search.through.inf
‘searching the entire jewelry category for the wanted piece.’
The methodology for the presentation of the grammatical rules and requirements
underlying the sentence in (3) is adapted from the “Bonsai Grammar for German”
(Van Valin and Diedrichsen 2006). It is called “Bonsai Grammar” because it pro-
152 exploring the role of pragmatics in the interface
vides a “mini model” of German Grammar that consists of the grammatical rules
that are necessary to produce the sentences analyzed there. Equally, the lexicon,
the syntactic inventory and the semantics-to-syntax-linking presented here are
only those that account for sentence (3). In a way, this sketch of a grammar “pre-
tends” that a language consist of the words and rules that enable speakers to pro-
duce this particular sentence. Thus, it is a “mini-grammar” of a “mini-language” in
its own right.
Sections 2.1 to 2.3 account for the lexicon, the syntactic inventory and general
principles of the “language” in which the sentence can be produced. Section 3
gives the semantics-to-syntax linkings for each of the clauses that constitute the
sentence. All of these sections give representations based on RRG principles, but
those are replenished with information from other functional approaches.
Firstly, the activation statuses of the referents are realized in the semantic repre-
sentation of the clauses in that the choice of the referring expressions is explicitly
tied to the activation statuses of the referents in discourse. Furthermore, the Ger-
man case marking rules are given with respect to the feature hierarchy established
by Michael Silverstein (1976, 1981), an implicational hierarchy that is applied to
describe case-splits cross-linguistically.
2.1. Lexicon
LSC Operators:
< < < < < < < < >>>>>>>>
IF TNS STA NEG MOD DIR ASP LS
RP operators:
< < < < < < (x)>>>>>>
DEIC DEF QNT NUM NASP
Pronouns:
Ich – 1.sg.N
Nouns:
Bock m – ‘male sheep’; informal for ‘desire’
Schmuck m – ‘jewelry’
Kategorie f – ‘category’
Stück n –‘piece’
elke diedrichsen 153
Verbs:
landen – ‘land’
sein – ‘be’ (can be AUX)
haben – ‘have’ (can be AUX)
suchen – ‘look for’
durchforsten – ‘search through’ (informal)
Adjectives/adverbs:
gesamt – ‘whole, entire’
gesucht – ‘searched, looked for, wanted’ (see verbs: suchen): PSTP’s can be used as adjec-
tives in prenominal position.
Prepositions:
auf (requires dat with stative locations)
nach (requires DAT)
Quantifier (negated):
kein
Clause linkage marker:
und – ‘and’
The following syntactic templates are available in the syntactic inventory for the produc-
tion of the sentence in (3).
Figure 6. PP template.
Figure 7. RP template.
elke diedrichsen 155
2.3. RRG-based rules and principles involved in the linking from semantics to syntax
German has an accusative case system, but overt case marking is only used for
certain types of nominals. There is thus a case split with respect to the realization
of the accusative case. The type of case marking found in German can generally
be described according to the feature hierarchy that has also been used to describe
case splits in Australian languages (Silverstein 1976, 1981, Dixon 1994, Blake
1994). It is given in (1) above.
There are the following cut-off-points for German accusative marking (see
section 1.2):
1. pronoun vs. noun; criteria: activation in discourse; topicworthiness
2. masculine singular vs. others; criteria: humanness, animacy, agentworthiness
158 exploring the role of pragmatics in the interface
(7) Case assignment rule for German prepositions which assign both dative and accusa-
tive case:
a. Assign dative case to the first argument of be-loc´ (y, x)
b. Assign accusative case to the first argument of BECOME/INGR be-LOC´ (y, x).
In this section, the linking rules for the German clauses will be provided. They are
based on the linking algorithm worked out for German grammar in Van Valin and
Diedrichsen (2006). Those general linking rules are repeated for some of the steps
in order to make the design of the procedure more comprehensible. The steps to
be executed for the particular clauses are then spelled out in turn.
3.1. Semantics-to-Syntax-Linking for (3.a): Ich bin also auf www.neckermann.de ge-
landet
Step 1. In the lexicon, construct the semantic representation of the sentence, based
on the LS of the predicator.
a. Access LS for landen and select prepositional LS to fill be-LOC , slot in LS, auf:
do, (xi [PROC be-on, (y, x) & INGR be-on, (y, x)])
b. Determine the value of the operators to be expressed: For the sake of clarity,
the logical structure is given for each of the argument positions.
<IF DEC <TNS PRES <ASP PERF < do, (xi [PROC be-on, (y, x) & INGR be-on, (y, x)])>>>>
<IF DEC <TNS PRES <ASP PERF < do, (xi [PROC be-on, (y, x) & INGR be-on, (y, x)])>>>>
c. Select the referring expressions to fill the variable positions in LS, according to
the activation statuses of the referents.
- Active (ACT): actively under consideration in the discourse,
Use pronouns for active referents. In order to refer to yourself as speaker, use the
pronoun ich. In order to refer to your addressee, use the pronoun du. If you feel that
your relationship to this addressee requires respectful distance, use the pronoun Sie.
In order to refer to people and things which are actively under consideration in
the discourse by means of direct mention, use the pronouns er, sie or es; depending
on the gender of the entity you refer to.
160 exploring the role of pragmatics in the interface
- Accessible (ACS):
In order to refer to people and things that are not actively under consideration,
but readily recognized by the addressee due to world knowledge, occurrence in the
situation, or can be inferenced from something previously mentioned, use a full
noun + a definite article der, die or das, depending on the gender of the entity you
refer to.
- Inactive (INA):
In order to refer to people or things that are inactive, i. e. previously mentioned but
not actively under consideration and not assumed by the speaker to be recognized
by the addressee, use a complex lexical expression. As the referent should be iden-
tifiable by the lexical information, use the definite article.
In order to refer to people (and, sometimes, things like companies) that are
commonly known or are at least well known to you and your recipient(s), use a
proper name.
- Brand-new-anchored (BNA):
In order to refer to something that is not previously mentioned but related to
something already mentioned or accessible, use a complex lexical expression. If the
referent can be identified through the relation, use the definite article.
Cf. Prince 1981, Chafe 1987, Ariel 1991, Gundel et al. 1993, Fretheim and Gundel
1996 for terms and definition of the activation statuses. Many possible relations between
the activation status of a referent and the use of the definite article are exemplified in a cor-
pus study carried out in Diedrichsen (2006). See also Vater (1984), Löbner (1987).
elke diedrichsen 161
The semantic representation below accounts for the activation level of the dis-
course participants that is reflected in the kinds of referring expressions chosen for
them.
<IF DEC <TNS PRES <ASP PERF < do’ (ichACV [PROC be-on’ (www.neckermann.
deina, ich) & INGR be-on’ (www.neckermann.de, ich)]) >>>>
Step 2. Determine the actor and undergoer assignments, following the Actor-Un-
dergoer Hierarchy and the Default Macrorole Assignment Principles.
Step 4. Select the syntactic template(s) for the sentence, according to the follow-
ing general rules from Van Valin and Diedrichsen (2006). They are repeated here
for convenience:
a. For main declarative clauses and W-questions, select the clause template with
PrCS.
b. For embedded clauses, select the subordinate clause template.
c. For the core template, follow the core template selection principles.
d. Select the nucleus template.
e. For RPs, select the appropriate template depending upon whether the RP is
pronominal, a common noun or a proper noun.
f. Select a periphery template for all adjunct modifiers.
162 exploring the role of pragmatics in the interface
ad Step 4. Here, the substeps of Step 4 are carried out for (3.a).
a. Select the PrCS template, which is obligatory in main declarative clauses.
b. Does not apply.
c. Select a two-place core, one place for the nucleus and one for the proper name.
d. Select the nucleus template.
e. Select a pronoun template, a PP template and a proper name template.
f. Does not apply.
This is a reformulation of the rule from Van Valin and Diedrichsen (2006); see Die-
drichsen (to appear) for justification.
elke diedrichsen 163
This second clause is part of a topic chain that is realized by a clausal juncture (co-
subordination). The position before the finite verb is lexically unfilled; the miss-
ing argument of clause (3.b) is obligatorily coindexed with the PSA of the pre-
vious logical structure. So, the referent of this variable is recoverable from the
construction as a whole. Note that this “conjunction reduction” is only possible
when the unfilled element is the PSA of the clause it appears in (cf. Van Valin
2005: 229 ff.). The special conditions imposed by this construction type are sum-
marized in a constructional schema. It is given for V2-clauses, because only they
are relevant here.
164 exploring the role of pragmatics in the interface
rp pp nucleus
Figure 15. Tree structure with constituent and operator projections for 3.a.
elke diedrichsen 165
syntax:
Juncture: clausal
Nexus: Cosubordination
Construction type: conjunction
[CL[PrCSRP][CORE [NUC...]...]...]1,[CL [CORE [NUC...]...]...]2,...CLM
[CL[CORE [NUC...]...]...]n
Unit template (s): cf. template selection principle in (4)
PSA: Clause 1: variable syntactic controller = pragmatic controller
Clause 1 + n: variable syntactic pivot = pragmatic pivot
Position of the PSA: immediately after the conjunction.
No element may intervene between the conjunction and the finite verb.
The conjoined clause will not contain a nominative, following the “Accessibility to
PSA selection principles” in Van Valin (2005: 100 (4.15))
Linking: controlled argument in clause 1 + n = pragmatic pivot
morphology:
clm: coordinating conjunction or disjunction
semantics/pragmatics:
Sequence of events sharing a common primary topical participant
pragmatics:
Illocutionary force: shared across all conjuncts
Focus structure: predicate focus in all conjuncts
Figure 16: Constructional schema for 3.b (cf. Van Valin 2005: 231)
Linking rules for (3.b): (…) und __i habe keinen Bock,
c. Select the referring expressions to fill the variable positions in LS, according
to the activation statuses of the referents.
The topical referent that is highly active due to appearance in the conjoined
clause is recoverable from the construction as a whole (see the constructional
schema above). So, its position in LS is unfilled, but obligatorily coindexed
with the PSA of the previous logical structure. For the abstract concept ‘de-
sire’, use a full RP as in connection with the negation, it is “brand-new un-
anchored”.
<IF DEC <TNS PRES <NEG< have’ (__iACV, BockBNU) >>>>
Figure 17. Tree structure with constituent and operator projections for 3.b.
elke diedrichsen 169
Van Valin 2005:244, 7.33b; section 2.3): The occurrence of a core as the linked core
in a non-subordinate core juncture reduces the number of core slots by 1. Being a
linked core without a finite verb, the zu-infinitive-construction does not involve a
Precore Slot (cf. Van Valin and Diedrichsen 2006). The special conditions imposed
by this construction type are summarized in the following constructional schema:
morphology:
No CLM; infinitive with the particle zu
semantics:
Psych-action, causative/jussive, commissive, directive, speech acts
pragmatics:
Illocutionary force: unspecified
Focus structure: unspecified
Figure 18: Constructional schema for 3.c (cf. Van Valin 2005: 250)
Linking rules for (3.c): (…) die gesamte Schmuck-Kategorie nach dem gesuchten Stück
zu durchforsten
170 exploring the role of pragmatics in the interface
<DEF+ <QNT E < NUM SG <NASP COUNT < be’ (Kategorie (y), [entire’])>>>>>
<DEF+ <QNT E < NUM SG <NASP COUNT < be’ (Stück (z), [searched.for’])>>>>>
c. Select the referring expressions to fill the variable positions in LS, according
to the activation statuses of the referents.
As the linked core is “subjectless”, the position of the x-argument is lexically
unfilled. As the things referred to are not actively under consideration, but
readily recognized by the addressee due to mention in a previous utterance
(das gesuchte Stück ‘the piece I am looking for’), or can be inferenced from
something previously mentioned (die gesamte Schmuck-Kategorie, ‘the entire
jewelry category’), they are both accessible, so use a full noun + definite arti-
cle. As Schmuck-Kategorie is feminine, use the article die. As Stück is neuter,
use the article das.
Semantic representation:
[do’(xiACV, [search.through’(xi, die gesamte Schmuck-KategorieACS)] PURP
[INGR see’(xi, das gesuchte StückACS)])]
elke diedrichsen 171
Figure 19. Tree structure with constituent and operator projections for 3.c.
elke diedrichsen 173
4. Conclusion
The work on this paper has shown that it is possible to integrate more prag-
matic facts into the RRG-based linguistic description than it has been done in
previous work.
The RP-operators can be seen as indicators of the activation statuses that will
appear in the semantic representation: RPs with deictic operators are always active,
and those with a positive definiteness value cannot be “brand-new unanchored”.
Furthermore, the topic-chaining function of certain construction types, as for ex-
ample conjunction reduction, becomes apparent when a piece of naturally-occur-
ring discourse is examined. In the examined sentence, the PSA in each of the
clauses is the active element. This supports the general view that “activeness” in
discourse correlates with syntactic PSA-status.
The Nominal Hierarchy turns out to be a very valuable tool for the description
of case marking in an approach that tries to give language-specific rules based on a
cross-linguistic framework. According to the present findings and considerations,
the hierarchy can be motivated both on semantic and on pragmatic grounds, and
this supports the general concept of RRG, that grammars of all languages can be
described in terms of a Syntax-Semantics-Pragmatics-Interface.
References
Van Valin, R. D., Jr. and E. Diedrichsen. 2006. A Bonsai Grammar for Ger-
man. Available on the RRG web page: http://linguistics.buffalo.edu/people/
faculty/vanvalin/rrg/BonsaiGrammarGerman.pdf
Vater, H. 1984. Determinantien und Quantoren im Deutschen. Zeitschrift für
Sprachwissenschaft 3:1, 19-42.
Zubin, D. and Köpcke, K.-M. 1996. Prinzipien für die Genuszuweisung im
Deutschen. Lang, E. and G. Zifonun (eds.): Deutsch – typologisch. Berlin/
New York: De Gruyter, 473-491.
The Preposition with
in Role and Reference Grammar
Patrick Farrell
University of California, Davis
1. Introduction
The English preposition with has several diverse and seemingly unrelated uses for
marking verb complements and adjunct modifiers in clauses. The following ex-
amples illustrate the general instrumental use.
The with phrases in these examples specify the enabling thing(s) used by the
primary acting participant in an event. Without much change in meaning, the PPs
can be replaced in paraphrases employing the verb use, e.g., Paula broke the window
using {a bat/her hand}. The examples in (2) specify something manipulated by the
agent in the performance of an action, but with less of an instrumental implication
—to varying degrees— as evidenced by the differing degrees of awkwardness of a
use paraphrase, e.g., ?The farmer loaded the wagon using hay, *The army supplied the
soldiers using guns.
Although, note that at least use paraphrases of the type Paula used this wine to fill the
glasses are acceptable with at least some verbs, other than those of the supply type. I return to
this matter in section 3.4.
179
180 the preposition with in role and reference grammar
What these examples have in common is that they describe events in which the
object of with is a thing that is conceived of as ending up in, on, or in the posses-
sion of the referent of the verb’s object. Even less like the examples in (1), those
in (3) have with phrases that specify a co-participant in an event, as evidenced by
their ability to be paraphrased with sentences with nearly the same meaning con-
taining a conjoined NP, and possibly the adverb together, e.g., Jack and Jill went
to the store [together]. By contrast, examples like those in (1) and (2) have no such
paraphrase, e.g., *Paula and wine filled the glass [together], *Paula and a bat broke
the window [together].
That these are locative uses of the preposition is indicated by the possibility of
a where question (Q: Where did you put the CDs? A: with the DVDs, Q: Where do
you live? A: with my mother), unlike with the other types of with construction iden-
tified above (e.g., *Q: Where did Jack go to the store? A: with Jill). This meaning is
naturally extended to more abstract situations, as in the case of (7a), which em-
ploys a routine “conceptual metaphor” (Lakoff & Johnson 1980), i.e., ‘a romantic
relationship is a place,’ and (7b), in which togetherness in a mental space is at issue.
(7) a. Is Paula still with Jake, or does she have a new boyfriend?
More systematically, the central sense is conceptually related to the more ab-
stract scenario of togetherness in having relationships of various kinds, as illustrat-
ed by the examples in (8), in which with phrases are used as noun modifiers. The
key claim is that the use of with in clauses such as those from (1) to (4) involves
only the engagement of the related senses of with displayed in (5), (6), and (8).
Foley & Van Valin (1984) first propose the idea in Role and Reference Grammar
(RRG) that English prepositions can not only express meanings but also mark as-
pects of the procedure for linking arguments in the conceptual semantic structure
182 the preposition with in role and reference grammar
of verbs with the semantic macroroles of Actor (A) and Undergoer (U), which
figure prominently in the determination of grammatical relations and in syntac-
tic phenomena of various kinds. Employing a predicate-calculus decomposition
of verb meaning of the kind articulated in Dowty (1979), with the objective of
providing formal semantic distinctions that yield aktionsart classes (state, activity,
accomplishment, etc.) and thematic relations (patient, location, theme, etc.), RRG
relates the assignment of the A and U roles to argument positions in the lexical
“logical structure” (LS) of verbs. By way of example, the verb clear, as in Tom cleared
the dishes from the table means something like ‘x does something to y; because of
this y is not on z anymore.’ This kind of meaning is expressible in the LS formal-
ism in the way shown in Figure 1, which also shows the linking to thematic rela-
tions, semantic macroroles, and grammatical relations.
Under this approach, the use of certain prepositions is partially predictable from
LS details. For example, the preposition from is used for the first argument (i.e.,
location-type argument) of BECOME NOT be-loc’, where loc’ stands for any
locative preposition (on, in, etc.), or BECOME NOT have’, provided that it is
not linked to a macrorole. Examples include Bill took the money from Tom (the LS
of which includes […BECOME NOT have’ (Tom, money)…]), I bought the car
The default privileged syntactic argument (PSA) of RRG corresponds roughly to the
subject relation of other theories. The U macrorole itself, in a syntactically transitive clause
with an A, corresponds to the object or primary object of other theories.
patrick farrell 183
from Tom, and I removed the stuff from the refrigerator. It turns out that the argu-
ment with the theme relation does not have to be expressed as the U with the verb
clear, as evidenced by the paraphrase Tom cleared the table of dishes, in which the
table is the U. The RRG analysis is that the LS of the verb is invariant. What dif-
fers is that the U macrorole is linked with the location argument rather than the
theme. This linking is said to be a marked one because the argument whose role
is closer to the U end of the Actor-Undergoer Hierarchy (shown in Figure 2) is
ranked higher for U. In any case, the rule of preposition assignment, which is as in
(9), captures the regularity concerning the use of of, accounting not only for clear
the table of dishes but also for a range of semantically parallel cases such as those il-
lustrated in (10).
Figure 2 is based on the Actor-Undergoer Hierarchy as it is presented in Van Valin
(1993). The discussion here leaves aside the implications of the recent proposals concerning
the A-U hierarchy presented in Guerrero and Van Valin (2004) and Van Valin (2005: 127),
according to which the choice of location as U would not be more marked. The thematic
relation labels are convenient names for classes of semantic roles defined in terms of argu-
ment positions in LSs. The term theme (including percept, possessed, stimulus, consumed,
and implement, for example, in addition to the traditional undergoer of movement) refers
to the second argument of a two-place state predicate (be-at’, have’, see’ know’, etc.) or
activity predicate (eat’, use’, etc.); location, or locative, (including source, goal, and possessor)
and experiencer refer to the first argument of such a predicate; patient refers to the argument
of a one-place state predicate (dead’, broken’, exist’, etc.); effector refers to the first or only
argument of do’. In Van Valin & La Polla (1997), the A-U Hierarchy is portrayed directly
in terms of positions in LSs, rather than in terms of thematic relation labels, which are used
here because of their wide recognition across theories.
184 the preposition with in role and reference grammar
Building on insights from Foley & Van Valin (1984), Jolly (1993) proposes a uni-
fied account of the use of with in cases such as those illustrated in (1)-(3).This
account is adopted and further refined in Van Valin & La Polla (1997: 378–381).
The key idea is that with, as the main argument-marking non-locative preposition
in English, is chosen to mark an argument of a verb when it is a potential A or U
that is at least as good a choice for one of these roles as any other argument, based
on status on the A-U Hierarchy (Figure 2), and happens not to be selected for the
macrorole in question. By way of illustration, fill, as in Paula filled the vase with
flowers, has the LS and linking to grammatical relations shown in Figure 3. Since
the location argument is linked to the U macrorole and the theme argument is a
less marked choice (and therefore at least as good a choice in this particular con-
struction), with is assigned to the non-U theme. The rule for with (Van Valin &
La Polla 1997: 380) is as follows.
As less marked choices, a theme argument is of higher rank than a location argument
with respect to the U macrorole and an effector argument is of higher rank than a theme ar-
gument with respect to the A macrorole. I only consider here the simple version of the pro-
posed rule, leaving aside the complex attempt to collapse it with a rule for assigning with
to postonominal adjunct NPs, as in the man with a beard. For present purposes, it suffices to
show that even the simple version of the rule is problematic.
patrick farrell 185
Figure 3. Linking from LS to grammatical relations for Paula filled the vase with flowers
As the examples in (2) share the property of having what can be characterized
as a non-U theme and a U ranked lower for U-linking, this rule is appealed to for
these and various other similar cases.
For comitative uses of with, as in the examples in (3), the idea is that the LS
contains co-arguments that could either be expressed as one syntactic phrase,
with and conjoining two NPs, or as two syntactic phrases, only one of which is
assigned a macrorole. Thus, the analysis of Jack went to the store with Jill is as
shown in Figure 4. Since both Jack and Jill express arguments that have the ef-
fector thematic relation, they are both eligible for A assignment. If such were to
occur, the result would be Jack and Jill went to the store. If only Jack is linked to
the A macrorole, Jill, which is an equally good choice for A, is marked with with,
following (11).
Figure 4. Linking from LS to grammatical relations for Jack went to the store with Jill
The operator ^ in LS representations signifies the same thing as &, but with an addi-
For the instrumental use of with, as in Paula broke the window with a bat, the
proposal is that an implement used in the accomplishment of some action is, in
effect, an intermediary co-effector in a complex causal chain. The LS of this ex-
ample and the linking with macroroles and grammatical relations is as shown in
Figure 5. Because a bat has the effector thematic relation, it is ranked equally on
the A-U Hierarchy with Paula and is therefore marked with with, following (11),
as it is not chosen as A. Its potential for the A macrorole, based on status on the
A-U Hierarchy, is masked by a linking principle that requires the highest effec-
tor in a causal chain to be linked to the A macrorole. However, the claim is that it
shows up as A if the highest effector argument and the associated sub-LS [do’
(x …] CAUSE) are suppressed from the LS of break, in which case A bat broke the
window is possible.
(12) a. She left her son at the park with considerable trepidation.
Considering examples like (12e), Van Valin and La Polla suggest that there is
a rule, similar in spirit to (11), that assigns with to manner adverbials if they are
expressed as nouns rather than in their unmarked form as adverbs (cf. He did his
homework enthusiastically). However, not only is it unclear that enthusiastically is
a less marked option than with enthusiasm, but this additional rule has no appar-
ent relevance for (12a-d), in which there is no noun/adverb alternation, e.g., *She
left her son at the park (considerably) trepidationally, *Paula showed up in class today
questioningly about the reading, and which express additional information about
something of relevance to the event that the acting participant has, rather than
anything about manner per se.
Another set of difficulties arises in connection with instrumental with, since,
to begin with, it shows up not only when the implement is a thing with potential
for A linking (in the absence of a superordinate action sub-LS), but also in various
other cases, including those in which the positing of an intermediate action sub-
LS is semantically implausible (13c-d) or evidence for one is lacking by virtue of
the impossibility of suppressing the posited superordinate action sub-LS (13a-b).
(13) a. Paula broke the window with her hand. vs. *Her hand broke the window.
c. Chris made a snowman with the shovel. vs. *The shovel made a snowman.
d. She looked at the star with a telescope. vs. *A telescope looked at the stars.
Recognizing the problem with cases such as (13d), Van Valin and La Polla sug-
gest that in the case of activity verbs, such as look, that do not designate an event in
which an implement is necessarily involved, the with phrase is more like a manner
adverb. The LS and linking for (13d) are as shown in Figure 6. In order to engage
the linking-based with-marking rule to account for the appearance of with, it has
to be assumed that the second argument of use’ in the PP is marked with with be-
cause, as the theme of use’, it is at least as good a choice for the U macrorole as the
theme argument of see’, but is not chosen as U.
188 the preposition with in role and reference grammar
But, it is far from clear that a with phrase in the case of the verb break, as in
Figure 5, has a different status (i.e., as an argument of the verb, rather than an ar-
gument of an adjunct sub-LS built on use’) and a meaning difference of the sug-
gested magnitude. If there were such a meaning difference, it is unclear why both
cases are equally well paraphrased with a simple use + infinitive structure: Jack used
a telescope to look at a star and Paula used a bat to break the window. Moreover, all
types of instrumental with phrases have the same behavior with respect to tests
for adjunct-hood (see Farrell 2005: Ch. 1). For example, it is well known that only
adjuncts can be left out of the in-focus phrase in the pseudocleft construction, as
illustrated by the examples in (14), which show that a temporal adjunct can be
omitted from the in-focus phrase (in bolds) with the verb clear, but neither its lo-
cation/source nor theme arguments can.
(14) a. What Paula did after dinner was clear the table of dishes.
b. *What Paula did the table (of dishes) (after dinner) was clear.
c. *What Paula did of dishes (after dinner) was clear the table.
The following examples show that different kinds of instrumental with phrases
behave like adjuncts with respect to this phenomenon:
(15) a. What Paula did with {the bat / her hand} was break the window.
b. What Jack did with the telescope was look at the stars.
patrick farrell 189
There are two kinds of potential motivation for different analyses for break +
with instrument and look + with instrument. First, it is easier to conceive of the
instrument in a breaking act as itself acting on the patient than it is to conceive
of the instrument in a looking act as itself doing anything to anything. Second,
as already noted, A bat broke the window is possible and *A telescope looked at the
star isn’t. But, these two kinds of potential motivation, taken together, can give
conflicting results. In the case of He gunned the neighbor down with a 45, clearly
the referent of a 45 can be conceived of as acting on and causing a change of state
in the referent of the neighbor, which should indicate that there is an interme-
diate action sub-LS. But, *A 45 gunned down the neighbor is impossible, which
should indicate that there is no intermediate action sub-LS. The problem is that
the meaning of gun down includes “use a gun” as one of its components. Since a
45 cannot be conceived of as using a gun, it cannot be the effector argument of
gun down. The default PSA/Actor possibilities of verbs are constrained by vari-
ous semantic and world knowledge factors. *A telescope looked at the stars, for ex-
ample, is semantically anomalous because the effector argument of look must be
conceived of as seeing something, which animate beings can do and telescopes
cannot. There is no reason to believe that the possibility of an inanimate A pro-
vides any evidence for intermediate action sub-LSs for only certain verbs or cer-
tain kinds of verbs. It is simply unsound to reason that the mere possibility of an
inanimate A with an accomplishment verb like break implies an underlying LS
with a suppressed superordinate effector that is conceived of as using the A argu-
ment as an instrument, given the possibility of sentences such as A falling branch
from that tree broke the window, for which there is no possible conceptualization
of a suppressed superordinate effector or of any use of the falling branch by any-
one or anything.
To make matters worse, consider a causative accomplishment verb, such as put
together, in a sentence such as The boy put together the bike with a manual, in which
the referent of the manual is as much an instrument as in other instrumental with
sentences (cf. The boy used a manual to put together the bike). The manual cannot
be conceived of as doing anything to the bike in the event designated by this sen-
tence. Moreover, the manual cannot be the default PSA of put together (*The man-
ual put together the bike). All putative evidence for an intermediate action sub-LS
converges on the conclusion that there cannot be one. The LS for this sentence
and the linking to macroroles and grammatical relations must be something like
what is shown in Figure 7. Crucially, since the U macrorole goes to the patient ar-
190 the preposition with in role and reference grammar
gument, which is not equal to or lower than the theme for U linking on the A-U
hierarchy, (11) cannot be engaged to account for the appearance of with.
In any case, the generalization that the linking rule approach misses with re-
spect to instrument-containing sentences is that the second argument of use’ is
systematically marked with with (if it is not linked to a macrorole).
Perhaps most importantly, it is unclear how (11), or any other rule based on
linking to macroroles or marked argument expression of any kind, could be en-
gaged to account for the appearance of with in examples such as those in (5). Con-
sider, for example, Paula is with the angels. In order to engage (11), one would have
to assume that this example involves some version of comitative with: Paula and
the angels express co-arguments with the same thematic relation in some LS. But,
there is no plausible LS for this sentence representing this state of affairs. (16a)
seems to be the only reasonable LS that might give rise to with marking by (11);
but such an LS should be able to be expressed as (16c) and entails the existence of
LS (16b), which should be expressed as (16d).
b. be’ (Paula)
d. *Paula is.
Rather, it seems that Paula is with the angels is no different than Paula is on
the bed or Paula is near the angels. The primary conceptual content in the be + P
patrick farrell 191
construction resides not in the verb be but in the P. If one is to say that be-near’
(angels, Paula) is the LS of Paula is near the angels, so must one say that be-with’
(angles, Paula) is the LS of Paula is with the angels.
This claim is further strengthened by the locative use of with in sentences such
as those exemplified in (17a).
The verb put, for example, has an LS such as [do’ (x, Ø)] CAUSE [BECOME
be-loc’ (y, z)]. Among the choices for the PP goal/location argument are various
locative prepositions, as illustrated by (18a). Crucially, the goal/location argument
is not equal to or lower than the theme argument for U-linking. Moreover, an LS
such as [do’ (x, Ø)] CAUSE [BECOME be-loc’ (Ø, y) ^ be-loc’ (Ø, z)] fails to
express the correct meaning of something like I put the CDs with the DVDs, since
it is not the case that both the CDs and the DVDs end up in an unspecified loca-
tion; they are conceived of as ending up in the same location and only the CDs
are conceived of as moving there. The right meaning is expressed by BECOME
be-with’ (DVDs, CDs). Similarly, with is not a manifestation of comitative with
in cases like (17b), since the meaning of this sentence is not the same as that of
Jack and his mother live (= live’ ( Jack) ^ live’ (mother)). Rather, the lexical LS of
live, whose complement is necessarily conceived of as a location, must be live’ (x,
y). The location argument can be expressed by an NP marked with a variety of
locative prepositions, as indicated by (18b). Given these considerations, it should
be clear that rule (3) does not account for the possible appearance of with in such
cases as (17). The only explanation for the appearance of with as the marker of the
location argument of put and live is that with is one of the prepositions that the
first argument of the predicates be-loc’ and live’ can be marked with. The over-
all implication that emerges from both a consideration of cases such as (17) and
Paula is with the angels is that with is itself a meaningful preposition, rather than
a mere marker of a kind of scenario in a procedure for linking arguments in an LS
192 the preposition with in role and reference grammar
3. An alternative analysis
One way of figuring out the central sense of a polysemous preposition is to con-
sider its earliest meaning (Tyler & Evans 2003). Although the early meanings of
with are interesting with respect to its historical development, unlike with many
prepositions they do not happen to be the same as its current locative sense, or,
indeed any current sense. According to the Oxford English Dictionary (OED), the
preposition with in Old English could express such locative meanings as ‘along-
side’, ‘towards’, and ‘against’ and could be used in various figurative extensions of
these. Only later did with come to have most of its current senses, by taking over
the semantic ground of and turning obsolete the preposition mid. Beyond their
contribution to the development of the current senses, the early meanings of with
only survive in such opaque compounds as withstand, where the ‘against’ sense was
at play. One apparent development plausibly involves a kind of bleaching, wherein
the more specific sense of ‘against’, as in against the fence, or ‘alongside,’ turns into
simple ‘near’, as in The weather drove the ship with the coast of Sicily (see OED sense
5b). The locative sense that occurs in current English, as in Paula is with the angels
or The apples are with the oranges, involves even more bleaching, as the proximity
component of both ‘alongside’ and ‘near’ is gone. The apples can be conceived of
as being with the oranges by virtue of both being in a walk-in refrigerator within
which they are not necessarily near or alongside each other. The conceptualizer
need only conceive of the location of the oranges as relevant to knowing the loca-
tion of the apples. The relationship that with designates is closest in terms of con-
ceptual content to together, for which reason The apples and oranges are together has
roughly the same meaning as The apples are with the oranges. The locative meaning
of with is simply sameness of place for two things. Since this is the meaning that
Jolly (1993) only notes in passing that something like this meaning must be recognized
for predicative uses of with. Coseriu (1978), in an effort to identify a single meaning for
Spanish con, the uses of which are similar to English with, posits, in essence, what is claimed
patrick farrell 193
(19) a. Is Paula still with Jake, or does she have a new boyfriend?
c. When it comes to classic cars, the ’56 Thunderbird is right there with the ’57 Chevy.
There are at least four related senses of with, which can be characterized as its
having-sense cluster of meanings and which can be seen as more specific instan-
tiations of the central locative sense, inasmuch as having is a kind of togetherness
relationship. That is to say, if I have something as a part, for example, me and
this part are “together.” It cannot be an accident that other languages also use the
same preposition for the sameness of place relationship and having relationships.
In Portuguese, for example, estar com ‘be with’ is used both to predicate sameness of
place and as a simple synonym of the verb ter ‘have’, as illustrated by the following
examples.
here to be the central locative sense. The analysis developed here goes further, by fleshing
out the details of the polysemy of with and the associated consequences.
194 the preposition with in role and reference grammar
For whatever reason, in English the verb have is used in clauses that express
having relationships but is replaced by with in corresponding PPs with the same
meaning, as illustrated by the following examples.
c. A vase that has flowers (in it) A vase with flowers (in it).
d. A table that has candles (on it) A table with candles (on it).
The basic network of the senses of with in English can be depicted as in Figure
8, which shows the meanings of with in both its central togetherness sense and its
having senses when it is used in a PP functioning as a postnominal modifier. The
key claim of this paper is that all uses of with can be characterized as instantiations
of this small set of related senses, or metaphorical extensions thereof.
In idiomatic expressions such as come up with, come down with, and end up with, the
appearance of with is motivated by the meaning of the verbs, which designate a
change of state whereby the A (in the case of come up with) or the U (in the other
cases) is construed as coming to have something. The LS of Paula came up with a
solution, for example, is something like do’ (Paula, Ø) & BECOME have’ (Paula,
solution). As such, the use of with for the second argument of have’ is an instan-
tiation of sense B1.
Similarly, the use of with in adverbial phrases like those illustrated in (12) is
a simple manifestation of sense B1, as already noted above. In all such cases, a
paraphrase with the verb have expresses the same meaning, although it may be
somewhat awkward due to the preference for the with phrase alternative (He did
his homework {with/having} enthusiasm, I’m heading to Las Vegas {with/having} a lot
of money in my wallet, etc.).
One virtue of the proposed analysis is that comitative with phrases can be ana-
lyzed as an instantiation of the central sense of with. If someone does something
with someone else, the doing participants are conceived of as being together in a
group, which is a simple metaphorical extension of the concept of togetherness
in a place. Thus, a sentence such as Jack went to the store with Jill can be given a
different analysis than Jack and Jill went to the store, as shown in Figure 9. The
appearance of with is motivated by its inherent meaning. At the same time, this
analysis avoids the problems associated with the standard RRG move of allowing
a macrorole to be linked to only one conjunct of an argument (see Jolly 1993), al-
lows a straightforward account of why comitative with phrases typically have the
It is worth noting that, given such a LS for come up with, it would not be possible to use
linking-rule (11) to account for the appearance of with in Paula came up with a solution. This
is a single macrorole predicate, as evidenced not only by the appearance of with but also by the
impossibility of passivization (*A solution was come up with by Paula). Moreover it clearly has
a do’ component in its LS, as evidenced by the possibility of adding an adverb such as care-
fully or actively. Since the theme argument, which is marked with with, is not at least as good a
choice for A as the effector, (11) is not applicable.
196 the preposition with in role and reference grammar
distributional properties of adjuncts (Farrell 2005: Ch. 1), and still accounts for
the meaning parallel with conjoined NP sentences.
(22) a. do’ (Jack, Ø) ^ NOT do’ (Jill, Ø) & BECOME be-at’ (store, Jack) NOT be-at’ (store, Jill)
b. do’ (Jack, Ø) ^ NOT do’ (Jill, Ø) & BECOME be-at’ (store, Jack) NOT BECOME be-at’
(store, Jill)
(22b) means that Jack did something and Jill did not do something and that
Jill, unlike Jack, does not come to be at the store. But the sentence in question en-
tails neither that Jill did not do something nor that she did not come to be at the
store, as evidenced by the coherence of Jack went to the store without Jill but met up
with her there. It only entails that she was not with Jack in his store-going action.
The use of with to mark the theme (i.e., the moving or transferred thing) in change
of place and transfer scenarios can be attributed to the fact that one of the having
senses of with is invariably engaged. In cases such as The army supplied the soldiers
with guns, the designated event is clearly one in which the U (the recipient argu-
ment) is conceived of as coming into possession of something. Although there are
aspects of the supply concept that are difficult to elucidate with the LS decomposi-
tion system, the relevant part of the LS of supply is clearly similar to that of give
and must be something like [do’ (army, Ø)] CAUSE [BECOME have’ (soldiers,
guns)]. The use of with is a simple manifestation of sense B1 in Figure 8, i.e., ge-
neric having.
In the case of Paula filled the vase with flowers, for example, the optional PP
headed by with fills in details about the end state of the patient argument of fill.
The verb fill designates an event in which the acting participant does something
that results in a change of state whereby this thing comes to have something in it
and becomes full. A plausible LS for the sentence in question is [do’ (Paula, Ø)]
CAUSE [BECOME have-in’ (vase, flowers) & BECOME full’ (vase)]. Since the
vase ends up being one with flowers (in it), with is a motivated choice of preposi-
tion. This is a manifestation of its containment sense, i.e., sense B3. One nice con-
sequence of this analysis is that it makes it possible to include the BECOME full’
aspect of the meaning of this verb in its lexical LS, in such a way as to account for
the fact that the location argument (the first argument of have-in’) is also the pa-
tient argument, i.e., the sole argument of full’, which explains why it must be the
198 the preposition with in role and reference grammar
There are essentially two kinds of instrumental with phrases. In one kind a cru-
cially active (body) part of the acting participant in an event is made explicit in an
adjunct PP (e.g., Paula opened the bag with her teeth). In the other kind, an imple-
ment in the dominion and under the control of the acting participant is expressed
in the PP, as in Paula broke the window with a bat or The boy put together the bike
with a manual. Given that such with PPs can be paraphrased with use phrases (e.g.,
It is also worth noting that the pseudocleft test for adjuncthood, which only allows the
verb and all its arguments to be focused on, shows that the with-marked “instrument” is an
adjunct but not the with-marked theme (What I did with a funnel was f ill the glasses with
the wine vs. *What I did with the wine with a funnel was f ill the glasses). The fact that the
with-marked NP can be an adjunct just in case there is no other instrument (What I did
with the wine was f ill the glasses) indicates that the alternative quasi-instrumental concep-
tualization is presumably possible. With verbs like supply, for which the theme argument is
obligatorily syntactically expressed and is not amenable to an instrumental construal, such
is not possible (*What the army did with the uniforms was supply the soldiers).
patrick farrell 199
Paula broke the window using her hand), all that is needed to account for the choice
of with to express this meaning is an appropriate decomposition of use. It is not
surprising that use can have as its U argument the same kinds of things that the
complement of with can, i.e., a part of the acting participant or an object in the
dominion of and under the control of the acting participant (e.g., Paula used her
teeth to open the bag or Paula used a bat to break the window). Given that use desig-
nates an event in which having something (as a part) enables someone (or some-
thing) to do something, it is reasonable to consider use to have something like the
following lexical LS.
(23) use: [ have(-as-part)’ (x, y)] CAUSE [be-able’ (x, [do’ (x, Ø)]
The proposal, then, is that an adjunct expressing the core of this meaning of use
is manifested as a with PP, by virtue of the fact that it instantiates one of the hav-
ing senses of with, as shown in Figure 10. Because the theme argument introduced
by the LS predicate have(-as-part)’ is not an argument of the lexical LS of the
verb of the clause, the with phrase is expressed as a syntactic adjunct. An LS with
essentially the same content could be derived from the lexical LS of use’, by plac-
ing the LS of an infinitival phrase in the place of the Ø argument of do’, for which
reason Paula used a telescope to look at a star is a paraphrase of Paula looked at a star
with a telescope.
The key claim of this analysis is that what all instrumental with phrases have in
common is that the referent of the NP they contain is conceived of as something
that enables an action by virtue of being a part of or being in the dominion of the
acting participant. They may also sometimes be conceived of as an intermediary
acting participant, as in the case of Paula broke the window with a bat. But, this
conceptualization emerges from simple world knowledge: if a bat that Paula has
enables her to break a window it is because what people do with bats is swing
them and this makes them contact and have an effect on things. The exact nature
of the interaction between people, instruments and things that are acted upon
is something that simply varies across event types. In the case of something like
The boy put together the bike with a manual, the referent of the manual, clearly, does
not act on anything. It nevertheless can be said that the boy could do what he did
to the bike because he had a manual. One virtue of the proposed analysis is that
it captures the meaning commonality across the full range of instrumental with
phrases.
[
Fig. 10. Meanings and linking for sentences with instrumental adjuncts
patrick farrell 201
4. Conclusion
Within RRG, with has been viewed as a special preposition because it has been
considered to not be semantically locative and to be used primarily as a marker of
scenarios of alignment between the theory-internal constructs of positions in LS
and the macroroles of Actor and Undergoer. The claim is that because the align-
ment scenarios giving rise to with marking can arise in multiple places in LSs,
with has the special property of being able to occur multiple times in a clause, e.g.,
Mario filled the glasses with wine with a funnel with Jack with enthusiasm. The
claim I make in this paper is that with is just another basically locative preposition,
with a highly schematic central meaning: togetherness in a place. One big advan-
tage of this stance is that it allows a straightforward analysis of what the RRG
linking rule for with fails to yield any account of, i.e., the existence of such routine
sentences as The CDs are with the DVDs and I put the CDs with the DVDs. By
recognizing this togetherness in a place meaning and its sanctioning relationship
to other non-locative senses of togetherness in various having relationships, the
proposed analysis portrays with as a rather ordinary preposition, i.e., one with a
locative central sense and a network of related, possibly non-locative, senses. Thus,
with is not unlike in, for example, which can have both a central containment
sense and a related temporal sense, both of which are manifested in I put the turkey
in the oven in time.
The proposed account also elucidates the conceptual motivations for the vari-
ety of uses that with has in clauses. The reason that it can occur multiple times in
a clause is the same reason that for, for example, can, as in the case of He headed for
the hills for three hours for his brother for no good reason. For can be used in its cen-
tral locative sense for the goal argument of the verb head, for example, and can be
used in its extended temporal, purposive, and benefactive senses in different kinds
of adjunct phrases, of which there can be more than one. Because the meaning of
with is such that it can indicate having relationships of various kinds as well as dif-
ferent senses of togetherness and there can be multiple phrases indicating such re-
lationships within any given clause, the possibility of multiple PPs headed by with
within a clause is unsurprising.
By systematically revealing and explicating the semantic motivations for the
various uses of with and by accounting for a wider range of data, the proposed
analysis constitutes a significant improvement over the traditional RRG ap-
proach.
202 the preposition with in role and reference grammar
References
Brugman, C. M. 1988. The Story of Over: Polysemy, Semantics and the Structure of
the Lexicon. New York: Garland.
Coseriu, E. 1978. Gramática, Semántica, Universales. Madrid: Gredos.
Dowty, D. R. 1979. Word Meaning in Montague Grammar: The Semantics of
Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht:
D. Reidel.
Farrell, P. 2005. Grammatical Relations. Oxford: Oxford University Press.
Foley, W. A. and Van Valin, R. D., Jr. 1984. Functional Syntax and Universal
Grammar. Cambridge: Cambridge University Press.
Guerrero, L. and Van Valin, R. D., Jr. 2004. Yaqui and the analysis of primary
object languages. International Journal of American Linguistics 70: 290-319.
Jolly, J. A. 1993. Preposition assignment in English. In Advances in Role and
Reference Grammar. R. D. Van Valin Jr. (ed.), 275-310. Amsterdam: John
Benjamins.
Lakoff, G. and Johnson, M. 1980. Metaphors We Live By. Chicago; University
of Chicago Press.
Tyler, A. and Evans, V. 2003. The Semantics of English Preposition: Spacial Scenes
and Embodied Meaning and Cognition. Cambridge: Cambridge University
Press.
Van Valin, R. D. Jr. 1993. A synopsis of role and reference grammar. In Ad-
vances in Role and Reference Grammar. R. D. Van Valin Jr. (ed.), 1-164. Am-
sterdam: John Benjamins.
— 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge Uni-
versity Press.
Van Valin, R. D. Jr. and La Polla, R. J. 1997. Syntax: Structure, Maening and
Function. Cambridge: Cambridge University Press.
Body parts and the encoding of thing
and place in Zapotec
Brook Danielle Lillehaugen
iifl-Universidad Nacional Autónoma de México
John O. Foreman
Utica College
1. Introduction
This paper investigates the semantic characteristics of body part (BP) terms in
Zapotec drawing on data from two diverse braches of Zapotec: Valley Zapotec,
represented by data from Tlacolula de Matamoros Zapotec (TMZ), and Northern
Zapotec, represented by Macuiltianguis Zapotec (MacZ).
Zapotec languages belong to the Otomanguean language stock. TMZ is spo-
ken in the Tlacolula Valley, which is located approximately 30 km to the southeast
The Ethnologue and ISO codes refer to TMZ as San Juan Guelavía Zapoteco (Gor-
don 2005), but since this is also the name of one of the speech varieties within the Tlacolula
Valley, we refer to the Ethnologue’s ZAB group as Tlacolula Valley Zapotec to prevent con-
fusion, and to the individual varieties within by the name of the pueblo in which they are
spoken. MacZ belongs to the Ethnologue’s ZAA group, Sierra Juárez Zapotec.
We are grateful to our wonderful Zapotec consultants for sharing their time and lan-
guage with us. We owe a special thanks to Roberto Antonio Ruiz, Josefina Antonio Ruiz,
and Juana Ramos Jimenez for providing the TMZ data and to Margarita Martínez for the
MacZ data. We would like to thank the editors of this volume for their helpful comments.
This work has also benefited from the comments of Michael Galant, Marcus Smith, and
especially to Pamela Munro, to whom we owe a special thanks for all of her help and in-
sight. All errors are, of course, our own. This research was partially funded by the UCLA
Department of Linguistics and the Instituto de Investigaciones Filológicas – UNAM
(Lillehaugen) and by a Utica College Summer Fellowship (Foreman).
To hear TMZ, visit the online digital Archives for the Indigenous Languages of Latin
America (www.ailla.utexas.org) where TMZ materials are being archived, and search for
“Tlacolula de Matamoros Zapotec”.
203
204 body parts and the encoding of thing and place in zapotec
of Oaxaca City. Oaxaca City (labeled Oaxaca) can be located roughly south of
Mexico City in Figure 1; the Tlacolula Valley is presented in Figure 2. MacZ is
spoken in the pueblo of San Pablo Macuiltianguis, which is approximately 70 km
to the north-northeast of Oaxaca City (Figure 3).
The use of body part (BP) terms to express location is an areal feature of Meso-
American languages (Campbell, Kaufman, and Smith-Stark 1986) and has been
specifically documented for Otomanguean in general (Kaufman 1974), and for
Zapotecan in particular (e.g. Butler 1980; MacLaury 1989; Munro and Lopez, et
al. 1999; Jensen de López 2002; Lillehaugen 2003, 2006; Munro 2007; and Pickett
1959, 1974). The typology of BP locatives in Zapotec is explored in Lillehaugen
and Munro (2006, 2007, and in press).
BP terms can occur in many types of phrases in Zapotec, including referential
phrases (1), locative phrases (2), and motion phrases (3). In examples (1) – (3), the
BP phrases are in bold and their function is indicated to the right. (Throughout
this paper, we mark each Zapotec example with an abbreviation for the language
it comes from.)
206 body parts and the encoding of thing and place in zapotec
The TMZ data is presented in the orthography developed and described in Munro
and Lopez, et al 1999. The MacZ data is presented in the orthography developed and de-
scribed in Foreman 2006. The following abbreviations are used in the glossing: 1sg: first
person singular, 1sgg: first person singular genitive, 2sgg: second person singular genitive,
3: third person, 3d: third person dative, 3f: third person formal, 3g: third person genitive,
at : general locative, bas: phonological base, c: completive, dist: distal, h(ab): habitual, in-
vis: invisible, neg: negative, neu: neutral, perf: perfective, pl: plural, poss: possessed, prep:
prepositional applicative, prog: progressive, and s: stative.
brook danielle lillehaugen and john o. foreman 207
their syntactic contrast with nouns in Section 3. For more detail on this topic, we
refer interested readers to Lillehaugen (2003, 2006). (For those readers interested
in work on spatial language and expressing location in general, we refer you to
Levinson 1996, 2003; Pederson et al. 1998; Talmy 2000, to name just a few.)
In this paper, we focus on how Zapotec BP terms can encode the semantic no-
tions of both thing and place. We argue that these represent distinct semantic
functions of the BP terms and that neither function is synchronically derived from
the other. In particular, we show that the place interpretation is not synchronic-
ally derived from the thing interpretation.
For example, loh me’es and lle’e etthu’nà’ in (1) are thing denoting, referring to
‘the tabletop’ and ‘the inside of the pot’ respectively. But in the sentences in (2),
the same sequences are place denoting, yielding the interpretations ‘on the table’
and ‘in the pot’. A natural question arises as to whether these phrases could carry
the same interpretations in both sentences or whether at least one interpretation
might be synchronically derived from the other. In particular, it seems plausible
that the phrases could be thing denoting in all contexts or at the very least that
the place interpretation derives directly from the thing interpretation. Such
proposals have been made for other varieties of Zapotec as in Pérez Báez (2007)
for Juchiteco, Isthmus Zapotec.
So, if loh me’es and lle’e etthu’nà’ were always thing denoting, the locative sen-
tences in (2) would have the literal interpretations ‘the dog is standing (on/at) the
tabletop’ and ‘the beans are (in/at) the inside of the pot’. While an initially attrac-
tive idea, we argue in this paper that for the varieties of Zapotec under consider-
ation, this hypothesis cannot be maintained. In TMZ and MacZ, BP phrases are
not consistently thing denoting. They also can denote places, which have deno-
tations distinct from and not derived from the thing interpretation. We base our
arguments on the following observations: some BP phrases can encode things
but not places, some BP phrases can actually encode places but not things, and
some BP phrases denote places that are not directly related to the thing denoted
by the related referential NP.
We present these arguments in the following manner. In Section 2, we pres-
ent the referential and locative uses of BP words; in Section 3 we give an overview
of the syntactic differences between the two. Then, in Section 4, we present two
hypotheses regarding the semantic characteristics of referential and locative BP
phrases, and in Section 5 we argue that referential and locative BP phrases are of
different semantic types, namely that referential BP phrases are thing denoting
208 body parts and the encoding of thing and place in zapotec
As expected, BP words in general can encode things. They can be used to refer to
human and animal body parts as seen below in (4):
In addition to these referential uses, many BP words can also occur in locative
expressions as in the following examples.
The noun phrase contained inside the BP phrase does nothing to disambiguate
between these interpretations. In both cases, it immediately follows the BP word,
which is consistent with the behavior of both inalienable possessors and preposi-
tional objects in these head-initial languages. Pronouns are just as ambiguous:
Detailed phonetic analysis of such phrases is, however, lacking. We believe that it would
be very fruitful to look for intonational differences, for example in the fronting or question-
ing of BP phrases.
210 body parts and the encoding of thing and place in zapotec
b. dets=a’. (TMZ)
behind=1sg
‘Behind me.’
b. luita’=nì. (MacZ)
beside=3g
‘Beside him/her/it.’
This is true even in MacZ, which has a few modest case distinctions among
the clitic pronouns. For instance, third person neutral clitic forms have a distinct
genitive form =nì which contrasts with nominative/accusative/dative =nà (Fore-
man 2006). But as we can see in (10) above, the BP word licenses a genitive NP
regardless of whether the BP expression receives a referential interpretation (10a)
or a locative interpretation (10b).
Despite the superficial similarities, there are syntactic differences between the
referential and locative expressions. Referential BP phrases exhibit the behavior of
NPs, while locative BP phrases exhibit the behavior of PPs. The syntactic evidence
for these distinct uses is detailed in Lillehaugen (2003, 2006). In the next section,
we provide a brief overview of this evidence.
In this section we briefly summarize some of the syntactic evidence showing that
referential BP and locative BP words are of different syntactic categories, namely
that referential BPs are nouns and locative BPs are prepositions. Evidence from
adjunction, categorical selection, and modification are presented below. This sec-
tion is not meant to be comprehensive, but only to give the reader an idea of the
types of syntactic tests used. (For a more detailed analysis, see Lillehaugen 2003
and 2006).
brook danielle lillehaugen and john o. foreman 211
3.1. Adjunction
Evidence can be taken from adjunction that BP locatives are not nouns. Intransi-
tive verbs such as ‘sing’ require no complement (11a). Some types of phrases are
allowed as adjuncts, such as làa’any yu’uh ‘in the house’ (11b), but others are not,
such as yu’uh ‘house’ (11c).
The difference in grammaticality between (11b) and (11c) suggests that làa’any
yu’uh ‘in the house’ and yu’uh ‘house’ are not of the same syntactic category. Làa’any
yu’uh ‘in the house’ can function as an adjunct in the sentence while the noun
phrase yu’uh ‘house’ cannot, even though one might expect that yu’uh ‘house’ could
semantically function as a location for the singing. Our explanation for this be-
havior is a syntactic one: namely that làa’any yu’uh is a prepositional phrase and
can function as a locative adjunct, while yu’uh is a noun phrase, and cannot func-
tion as a locative adjunct.
Another piece of syntactic evidence that BP locatives are not nouns comes from
the categorical selection (c-selection) of verbs, i.e. from the verb’s ability to require
the complement it selects to be of a certain grammatical type. Consider (12a), be-
low, where the verb zòob ‘sit’ selects a complement, which is in bold. Note that zòob
‘sit’ requires such a complement (12b).
212 body parts and the encoding of thing and place in zapotec
Bad with any meaning; e.g. cannot mean ‘The dog is sitting.’
What is the syntactic category of the complement ni’ih me’es in (12a)? Follow-
ing MacLaury’s (1989) assumptions about the status of BP locatives, one could
hypothesize that this phrase is a noun phrase, ‘the table’s foot’.
We can test this hypothesis in the following way. If the category of the com-
plement of the positional verb in (12a) is nominal, then this would suggest that
the verb ought to be able to c-select other noun phrases as complements. But, as
shown in (13), this is not possible: me’es ‘table’ alone cannot function as a comple-
ment to the positional verb.
Bad with any meaning; e.g. cannot mean ‘The dog is sitting by / at the table.’
Possibly, the crucial difference between me’es ‘table’ and ni’ih me’es is that ni’ih
me’es looks like a possessed noun phrase ‘the table’s foot’ and me’es is just a simple
noun. One could hypothesize at this point, that perhaps the verb has some way to
specify that its complement must be a possessed noun phrase. Although this seems
unlikely, it is also easy to show that this type of selection won’t work, because then
other possessed noun phrases should also occur as complements. As we can see in
(14), this is not possible: x:me’esa’ ‘my table’ cannot function as a complement to
the positional verb:
Bad with any meaning; e.g. cannot mean ‘The dog is sitting by / at my table.’
We have seen, then, that the positional verb zòob ‘sit’ allows the BP locative
phrase ni’ih me’es ‘under the table’ as a complement (12a), but does not allow sim-
brook danielle lillehaugen and john o. foreman 213
ple noun phrases (13) or possessed noun phrases (14). This shows that the BP loca
tive phrase is functioning differently in the grammar from noun phrases, whether
these are possessed or not. In order to maintain the hypothesis that the BP loca-
tive phrases were noun phrases, one could hypothesize further that only phrases
beginning with a closed set of nouns can occur as complements of positional verbs.
But such a constraint would be stipulative and miss an obvious generalization: the
BP locatives function differently from other BP words in the syntax. Our expla-
nation for this pattern of data is that me’es ‘table’ and ni’ih me’es ‘under the table’
are not of the same syntactic category (nor semantic type, as we will show later).
We claim that me’es ‘table’ is a noun phrase and ni’ih me’es ‘under the table’ is a BP
prepositional phrase.
3.3. Modification
One final type of evidence showing a contrast between referential BP words and
locative BP words comes from modification. When BP locatives are used in lo-
cational constructions, they cannot be modified in the ways that nouns can: they
cannot be modified with the plural marker, quantifiers, numerals, or adjectives.
Below, this is exemplified with the use of the plural marker.
In TMZ, the plural marker da precedes the noun which it modifies, as can
be seen in (15). A noun may sometimes be interpreted as plural without being
marked with da, but marking a noun with da requires that it have a plural inter-
pretation.
The plural marker can precede a BP term when that word is functioning as a
noun, as in (16). Here loh me’es is the noun phrase ‘tabletop’.
It is also possible for the BP preposition loh ‘on’ to take a noun phrase comple-
ment that begins with the plural marker da (17):
One might imagine that this sentence could be grammatical if loh were the
noun ‘face’ here. Via metaphor the sentence could mean something like ‘I put the
cups at / by the faces of the tables’. However, this sentence is ungrammatical. Spe-
cifically, the ungrammaticality lies in the fact that the plural marker da cannot
modify the word loh, because in this sentence, loh ‘on’ is a preposition.
If the BP locatives are analyzed as syntactic nouns, these results are surprising.
However, if BP locatives are analyzed as syntactic prepositions, the ungrammati-
cality of BP prepositions modified by adjectives, numbers, quantifiers, and the plu-
ral marker becomes expected, as prepositions in the language cannot be modified
in this way.
It has been argued for Juchiteco (Isthmus Zapotec), that there is no semantic dif-
ference in the function of referential BP phrases and locative BP phrases (Pérez
Báez 2007). Pérez Báez argues that both types of BP phrases function to name
things, and specifically, that locative BP phrases do not function to name places,
but rather are “thing denoting” (Pérez Báez 2007:6). The examples in (19) show
her analysis of both a referential use of ‘face’ (19a) and a locative use (19b), both
glossed as ‘face’ and analyzed as thing denoting.
216 body parts and the encoding of thing and place in zapotec
(19) a. fa!n gu-ndaa* lu^ mezha!7 (Juchiteco; Pérez Báez 2007: 4, ex 10)
John c-break face table
‘John broke the tabletop.’
We, however, argue that clauses like that in (20) are not interpreted as proposed
in the thing denoting hypothesis in (21). Rather, we argue for the place denot-
ing hypothesis presented in (22), in which the BP locative functions as a preposi-
tion, mapping a thing to a place:
(22) place denoting hypothesis
Bèe’ecw zuu loh meés. (TMZ)
dog neu.stand on table
‘The dog is standing on the table.’
BP phrases in TMZ and MacZ can function to name both things and
places, but these are separate semantic functions in Zapotec, and it is the place
For information on the orthographic conventions used in the citation of the Juchiteco
�
data, we refer the reader to Pérez Báez (2007) . The glossing has been slightly modified to
conform to the glossing conventions used in this paper.
brook danielle lillehaugen and john o. foreman 217
function that occurs in (20). In the next section, we provide semantic evidence
against the thing denoting hypothesis and in support of the place denoting
hypothesis.
We present three types of semantic evidence indicating that referential and loca-
tive interpretations of BP phrases represent mutually independent functions: (i)
there are BP phrases that can be used to encode things, but are infelicitous as
denoting places, (ii) there are BP phrases which can encode places but cannot
denote things, and (iii) there are BP phrases which denote places not composi-
tionally related to the thing denoted by the related referential NP. These facts are
discussed in turn.
Certain BP phrases can encode things, but are infelicitous in denoting places.
For example, x:quihche’ehcya’ in TMZ denotes ‘my hair’, a thing, as in (23a), but
cannot refer to a place *‘in/on/at/among my hair(s)’ as shown in (23b):
A second argument is that there are certain BP phrases that can encode places
but not things. In TMZ the phrase ni’ih ca’rr is felicitous as a locative ‘under the
car’, but never can denote a thing:
Thus, although ni’ih ‘under’ is used in the locative expression in (25a), there
is no referent ni’ih ca’rr ‘the car’s foot’ (25b), although this is syntactically well
formed. The speaker rejected the possibility that the wheels or tires might be
called ni’ih ca’rr ‘the car’s feet’, instead saying that this part of the car could only
be referred to as ya’annt ‘tires’ or rrueed ‘wheels’. In this case, ni’ih ‘under’ cannot be
functioning by identifying a component part of the Ground, since the car has no
part that can be referred to as ni’ih ca’rr ‘the car’s feet’.
We find similar facts in MacZ. For example, the BP term loo ‘face, on’ can oc-
cur with yaa ‘tree’ to denote a place ‘in/on the tree’, but there is no thing which
Interestingly, when the speaker was asked to translate ni’ih ca’rr out of context he said
“abajo del coche [under the car]”. It is our experience that out of context BP words are
translated with their referential meaning. This is consistent with the fact there is no physi-
cal referent ni’ih ca’rr.
brook danielle lillehaugen and john o. foreman 219
can be identified as loo yaa ‘face of the tree’. Thus, while a bird can be located loo
yaanà’ ‘in/on the tree’ (26a) there is no part of the tree which can be identified as
its face (26b), even though loo ‘face’ can serve to denote things in other contexts
(26c):
In fact, there are several prepositions which exhibit the morphological and
syntactic characteristics of BP prepositions but which surprisingly never encode
things though perhaps they historically did so. Synchronically, however, they can
only denote places. Here, we provide two examples from MacZ: chuuba’ ‘between’
(27a) and gaxxa ‘near’ (27b):
Despite the fact that the pronominal complements of these prepositions mor-
phologically look like possessors, they are not. Chuuba’ and gaxxa cannot be ana-
lyzed as thing denoting nouns. They cannot encode things as can be seen in the
following examples:
This contrasts nicely with another BP term laagwi’ ‘center, in the middle of ’
which can be both thing denoting (30a) and place denoting (30b):
The existence of such place denoting prepositions like chuuba’ ‘between’ and
gaxxa ‘near’ clearly indicates that formal similarity to a possessed NP does not
necessarily indicate a thing denoting NP. Furthermore, such words show that
hypothesizing distinct thing and place denoting nature of certain BP terms does
brook danielle lillehaugen and john o. foreman 221
not require any semantic machinery nor lexical type not already present in the
languages.
Finally there are BP prepositional phrases which denote places not directly relat-
ed to the thing denoted by the related referential NP. In order to consider these
examples, we first need to be explicit about what things can be denoted by the
NP loh me’es ‘face of the table > ‘tabletop’. The part of the table that is referred to
as loh me’es ‘tabletop’ is based on the composition of the table itself, and it does not
change depending on the orientation of the table. This is shown below, where (31)
can be used to describe the tabletop (i.e. the circled part) in Figure 4 where the
table is in canonical orientation, and in Figure 5 where the table is on its side.
Although the NP loh me’es ‘tabletop’ can refer to the tabletop regardless of the
orientation of the table, it only coincides with the topmost part of the table, if
the table is in canonical orientation, as in Figure 4. Thus although the circled area
in Figure 6 below is the topmost part of the table in the picture, the NP loh me’es
‘tabletop’ cannot be used to describe the area circled, as indicated by the #.
222 body parts and the encoding of thing and place in zapotec
The data is consistent, however, with the place denoting hypothesis (22) that
we have been arguing for here: that BP noun phrases denote things and that BP
prepositional phrases denote places. Thus, any particular BP phrase may, depend-
ing on the BP term which heads the phrase, either denote things or places or
both. As we can see in the summary below in Table 2, contrary to the expectations
of the thing denoting hypothesis, not all BP phrases denote things (as with ni’ih
ca’rr, and the things denoted by BP noun phrases can be distinct from the places
denoted by the corresponding BP prepositional phrase (as with loh me’es). This is
precisely the predictions of the place denoting hypothesis.
Note that these prepositions are obligatory; omitting the prepositions from
(36) results in ungrammatical sentences (37):
So while loh is required in (36), it is not expressing the direction of the motion,
which seems to be completely conveyed through the verb. The same pattern can
For the most part, BP prepositions are not optional when they are present in a con-
struction. There are two notable exceptions, (i) the use of locative preposition can be op-
tional before certain large locations, like restaurants (Munro in prep.) and (ii) the use of the
non-locative preposition loh appears to be optional in one of its uses, i.e. the direct object
marking use (see Lillehaugen 2004b, 2006).
226 body parts and the encoding of thing and place in zapotec
be seen in (38), for MacZ. Here we see the BP term ru’a ‘mouth’ used with both a
goal (38a) and a source (38b).
7. Conclusions
Within the Role and Reference Grammar (RRG) framework, there are two main
types of adpositions: predicative and non-predicative (Van Valin and LaPolla
1997). Predicative adpositions function like predicates in that they contribute sub-
stantive semantic information to the clause in which they occur, both in terms of
their own meaning and the meaning of the argument that they license. Non-pred-
icative adpositions are said to be those that do not add any substantive semantic
information to the clause and do not license the argument they mark. Rather, their
argument is licensed by the main predicate, i.e., the argument of a non-predica-
tive adposition is considered a core argument. Such non-predicative adpositions
are viewed as resultant of semantics of the predicate and are in effect case markers
assigned by the main predicate. However, there is a third type of adposition that is
brook danielle lillehaugen and john o. foreman 227
recognized within the theory: namely, one that can mark an argument of the verb
but nevertheless contribute meaning to the clause (Van Valin 2005: 21-22). These
three types of adpositional phrases are summarized and exemplified below, follow-
ing Jolly (1993):
Beyond this initial typological distinction of adpositional types and how these
types are represented in the constituent projection of the layered structure of the
clause, adpositions seems to be under-studied within RRG, although it is a focus of
current work, such as Ibañez (this volume) and Farrell (this volume). Ibáñez (this
volume) comments on the need to expand the typology of prepositional phrases
within RRG. For example, he posits that in Spanish there are PPs which function
as oblique core arguments, i.e. they are headed by a preposition that cannot be sys-
tematically assigned from a specific structural context. Farrell (this volume) claims
that the standard RRG analysis for the English preposition with has to be modi-
fied. He points out that as the RRG adpositional typology stands, there is no way
to account for the fact that with has basic semantic features which are present in
the meaning of the preposition, regardless of which particular type (i.e., (41a-c))
it manifests as in a particular construction. This is contrary to the analysis in Van
Valin and Lapolla (1997), which claims that the preposition with has no semantic
content, but only a functional meaning.
The Zapotec data that we present here, also calls for an expansion and recon-
sideration of the cross-linguistic types of prepositional phrases. Note, for example,
that while in English on the table in (42a) is an adjunct, in Zapotec a locative phrase
in this type of sentence (i.e. a sentence with a positional verb, see e.g. Lillehaugen
228 body parts and the encoding of thing and place in zapotec
2006: Chapter 5) functions as a complement of the verb (42a). For example, the
omission of the locative phrase results in an ungrammatical sentence (42b).
b. *Bèe’ecw zuu.
dog neu.stand
Bad with any meaning, e.g. cannot mean ‘The dog is standing,’ or ‘The dog is
standing (somewhere).’
It is our hope that the data and analysis provided here can stimulate addition-
al research into adpositional phrases within RRG. In conclusion, we have shown
that BP phrases in Zapotec can encode both things and places, and demon-
strated that these are necessarily separate semantic functions in Zapotec. We have
presented some initial data which suggest that BP terms, though required in many
source and goal phrases, appear not to encode source or goal, although this is
an area for future research.
References
Lillehaugen, B.D. and P. Munro. 2007. Component Part Locatives and Frames
of Reference. Presented at the annual meeting of the Society for the Study of
the Indigenous Languages of the Americas, Anaheim, CA.
MacLaury, R.E. 1989. Zapotec body-part locatives: Prototypes and metaphoric
extensions. International Journal of American Linguistics 55(2): 119-154.
Munro, P. In preparation. Expressing Location Without Prepositions in Valley
Zapotec. In Expressing Location in Zapotec, B. D. Lillehaugen and A. H. Son-
nenschein (eds.), ms.
— 2007. From parts of speech to the grammar. In Perspectives on Grammar Writ-
ing, T.E. Payne and D. Weber (eds.), 71-111. Amsterdam: John Benjamins.
Munro, P., Lillehaugen, B.D. and Lopez, F.H. In preparation. Cali Chiu: A
Course in Valley Zapotec. ms. ucla / unam.
Munro, P. and Lopez, F.H., with Méndez [Martínez], O.V., Garcia, R. and
Galant, M.R. 1999. Di’csyonaary X:tèe’n Dìi’zh Sah Sann Lu’uc (San Lu-
cas Quiaviní Zapotec Dictionary / Diccionario Zapoteco de San Lucas Quiaviní).
Los Angeles: (ucla) Chicano Studies Research Center Publications.
Pederson, E., E. Danziger, D.Wilkins, S. Levinson, S. Kita, and G. Senft. 1998.
“Semantic Typology and Spatial Conceptualization.” Language 74: 557-89.
Pérez Báez, G. 2007. The encoding of locative and path relations in locative con-
structions in Juichiteco. Presented at the annual meeting of the Society for the
Study of the Indigenous Languages of the Americas, Anaheim, CA.
Pickett, V. B. et al. 1959. Vocabulario zapoteco del Istmo : Castellano-zapoteco, za-
poteco-castellano. Serie de vocabularios indígenas “Mariano Silva y Aceves”,
3. Mexico: Instituto Lingüístico de Verano. [2nd ed., revised and enlarged
(1965); republished (1968, 1971)].
— 1974. Zapoteco del Istmo, Juchitan, Oaxaca. (Archivo de lenguas indigenas
del Estado de Oaxaca, 1). Mexico: Instituto de Investigacion e Integracion
Social del Estado de Oaxaca.
Talmy, L. 2000. Toward a Cognitive Semantics (2 vols). The MIT Press, Cam-
bridge.
Van Valin, R. and LaPolla, R. J. 1997. Syntax, Structure, Meaning and Func-
tion. Cambridge: Cambridge University Press.
Van Valin, R. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cam-
bridge University Press.
Virtual Mexico. http://www.virtualmex.com/map.htm, accessed November 22,
2002.
Applicatives and Role and Reference Grammar
Carmen Conti
Universidad de Jaén
1. Introduction
This paper deals with the theoretical aspects of benefactive applicatives within RRG,
rather than with the description of examples. See Conti (2006) if you are interested in a
descriptive work on benefactive applicative constructions across languages. I am employing
the label of beneficiary to mean two types of semantic relations: the non-required recipient
of verbs of creation and preparation, like in John baked a cake for Mary and John peeled an
apple for Mary; and “plain beneficiaries”, according to Van Valin & LaPolla (1997: 38), like
in John sang for the students.
The abbreviations of this paper are as follows. abs: absolutive, appl: applicative, aux:
auxiliary, ben: benefactive, e: epenthetic vowel, hab: habitual, impf: imperfect, m: macro-
roles, med: medial, obj: object, pl: plural, pref: prefix, wit: witness.
The prefix mem- occurs with all M-transitive verbs (see Purwo 1997, and Cole & Son
2004 for the gloss).
231
232 applicatives and role and reference grammar
The sample consists of 100 languages from all the genetic macro-phyla described
in Ruhlen (1987). Within these languages, 24 hold applicatives to code the beneficiary.
These languages are the following ones: Abaza (North Caucasian; spoken in Russia),
Awa (Trans-New Guinea; spoken in Papua-New Guinea), Barasana (Tucanoan; spo-
ken in Colombia), Barupu (Trans-New Guinea), Bukiyip (Torricelli; spoken in Papua-
New Guinea), Chichewa (Niger-Congo; spoken in Malawi), Chukchi (Chukotko-Ka-
mchatkan; spoken in Russia), Halkomelem (Salishan; spoken in Canada), Indonesian
(Austronesian; spoken in Indonesia, Java, Bali), Gadsup (Trans-New Guinea; spoken in
Papua-New Guinea), Kinyarwanda (Niger-Congo; spoken in Rwanda), Lango (Nilo-
Saharan; spoken in Uganda), Meithei (Sino-Tibetan; spoken in India), Maasai (Nilo-
Saharan; spoken in Kenya), Noni (Niger-Congo), Noon (Niger-Congo; spoken in Cam-
eroon), Huallaga Quechua (Quechuan; spoken in Peru), Sahaptin (Penutian; spoken in
U.S.A.), Swahili (Niger-Congo; spoken in Tanzania), Tinrin (Austronesian; spoken in
New Caledonia), Tukang Besi (Austronesian; spoken in Indonesia, Sulawesi), Tuscarora
(Iroquoian; spoken in Canada), Tzotzil (Mayan; spoken in Mexico), West Greenlandic
(Eskimo-Aleut; spoken in Greenland). The languages in the sample do not pertain to
carmen conti 233
Regardless the theoretical view, most scholars agree that applicatives are in-
creasing valence morphemes. For example, Donohue (2003) remarks that applica-
tive morphemes in languages like Barupu add an argument in the clause that was
not subcategorized by the verb stem. In my opinion, this conception of applica-
tives is basically correct at least for benefactive applicatives, as shown by the data
in my sample (see Conti 2006). However, it is necessary to clarify in which sense
benefactive applicatives change the valence of the verb stem, as we will see latter.
Applicatives are not only defined in relation to the valence of the verb stem,
but also to the syntactic properties of the entire clause. As pointed out by Payne
(2000), the prototypical applicative construction is the result of a syntactic promo-
tion: an adjunct or an oblique is promoted to object in the syntactic hierarchy. This
means (i) applicatives usually alternate with an adpositional frame —see example
(1) from Indonesian—, and (ii) the applied object plays a relevant semantic role in
the predicate (e.g., undergoer) or a relevant syntactic role in the sentence, whatever
the theoretical model is (e.g., it is the final direct object, it gets structural Accusa-
tive Case, etc.).
In sum, the prototypical applicative is presumably a derivational verb mor-
pheme that changes the verb stem valence by adding a new core argument, which,
in addition, will be the undergoer of the predicate and, as a consequence, will fit
the morphosyntactic and structural diagnostics of undergoers of two-place verbs
(for instance, the benefactive applied object usually controls the object agreement,
it is the privileged syntactic argument in passive, and so on).
From now on, I will intend to demonstrate that benefactive applicatives will
change the verb valence in a different way depending on the transitivity of the
verb stem. All languages in my sample allow benefactives with transitive verb
stems. In particular, the beneficiary applied object seems to behave as undergoes of
any specific morphosyntactic type. As pointed out by Peterson (1999), languages attest-
ing applicatives can be either head-marking or dependent-marking; they may show either
an accusative alignment, an ergative alignment or a stative-active alignment (among
others); and they can be either primary object languages, DO/IO languages or split-di-
transitivity languages (split-ditransitivity languages are not mentioned in Peterson 1999,
but must be added here).
See Conti (2006) for these diagnostics in the case of benefactive applicative con-
structions.
234 applicatives and role and reference grammar
simple transitive verbs. In some of these languages (e.g., Tzotzil, according to Ais-
sen 1987), the benefactive applicative is not allowed with intransitive verb stems.
In other cases, the benefactive applicative can attach to intransitive verb stems, but
it quits the benefactive meaning (cf. Cole & Son 2004). In the case of those lan-
guages allowing benefactive applicatives with intransitive verb stems, the resulting
valence of the derivative verb varies from language to language, as we will see in
the following paragraphs.
These data made me think about the question of whether the M-transitivity of
the verb stem can help us to classify benefactive applicatives across languages. To
demonstrate the pertinence of the semantic valence status of verbs, let us see what
happens in the languages of the sample when a benefactive applicative attaches
to an intransitive verb. I will briefly analyze the examples from Abaza (O’Herin
2001), and from Halkomelem (Gerdts & Kiyosawa 2005). Abaza is an ergative
language with casual system. It has different types of applicatives, amongst which
we find the benefactive applicative, which can attach to either transitive or intran-
sitive verb stems. According to O’Herin (2001), when the stem is intransitive, it
does not undergo the same changes than the transitive ones. As shown in (2), the
actor of the benefactive applicative construction is marked by the absolutive prefix
on the verb y-, instead of the ergative prefix, even though the latter is the expected
marker for actors of transitives in an ergative language:
The example in (2) reveals that the benefactive applicative has not affected the
verb valence. In other words, the benefactive applicative seems not to impact on
the syntactic transitivity of the verb stem. In Halkomelem the situation is differ-
ent. According to Gerdts & Kiyosawa (2005), the applied object of intransitives
—regardless it is benefactive or not— fits most of the morpho-syntactic diagnos-
tics of objects of simple transitive verbs. For instance, the applied object controls
Keep in mind, however, that Payne (2000) considers that Abaza does not have applica-
tives.
carmen conti 235
the object agreement on the verb, it is the privileged syntactic argument in passive
and it may be reflexivized and reciprocalized like the undergoer of simple transi-
tive verbs.
The examples from Abaza give rise to a more complicated question of whether
applicatives with intransitive stems are true applicatives in this language. Of course,
if we define application in a narrow sense, as an increasing valence operation that
must add an undergoer, benefactive applicatives with intransitives are only puta-
tive applicatives. However, if we define application as an increasing valence change
that add a new direct core argument, but not necessarily an undergoer, benefactive
applicatives with intransitives in Abaza and other similar languages are true ap-
plicatives. For different reasons I cannot argue here, I will accept that the affixes of
the type illustrated in (2) are applicatives.
Thus, I will assume that the affixes in the examples (1) and (2) belong to the
class of applicative morphemes. It is true that applicative affixes constitute, at least
in my view, a heterogeneous class of morphemes across languages: they are deri-
vational affixes that attach to a verb, but may be prefixes or suffixes; they do not
usually exhibit agreement features (though there are also exceptions, as occurs in
P’orhépecha, as noticed by Capistrán 2006), etc. They all share, however, the ca-
pability of modifying the verb valence by adding a new direct core argument. Spe-
cifically, I will argue that benefactive applicatives account for different types of
morphological operations depending on the semantic transitivity of the verb stem:
(i) if the stem is M-transitive, then the applicative re-assigns the undergoer to the
new argument; (ii) if the stem is M-intransitive, then the applicative may add an
Undergoer core argument —resulting in a M-transitive—, or may add only a new
core argument without increasing the semantic valence.
In order to account for these facts within RRG, my argumentation will be as
follows. Firstly, I will present the logical structure of sentences with a beneficiary.
Secondary, I will enunciate the rules capturing the morphological operations that
According to Gerdts & Kiyosawa (2005: 337-338), in Halkomelem the applicative
verb from intransitive stems cannot hold the transitive suffix –nxw (expressing an acciden-
tal, unintented action) or the antipassive affixes –m and –els.
We might wonder if the type of morpheme is relevant to the classification of applica-
tives, that is, if the difference between prefixes and suffixes play a role in the classification of
applicatives. In my account, this morphological categorization is not relevant, at least from
a cross-linguistic perspective, but there are other views of the matter (see Baker 1996).
236 applicatives and role and reference grammar
applicatives seem to give rise to. And thirdly, I will propose a basic inventory of
templates for applicative affixes.
In order to account for the linking from semantics to syntax in benefactive appli-
cative constructions, it is necessary to offer, first of all, a logical structure for appli-
cative sentences. I will assume that applicative morphemes, like other derivational
affixes, are entries in the lexicon of languages. In fact, beneficiaries are not core
arguments of simple stems in any of the languages of the sample, so it is neces-
sary to incorporate them into the logical structure of the predicate by means of the
adjunction of a predicate. In particular, benefactive applicatives, like benefactive
adpositions, are predicates in the logical structure.
In addition, I will assume that the logical structure of a predicate with a ben-
eficiary is exactly the same for the applicative frame, like the one we saw in (1a),
and the adpositional frame in (1b). I have found no evidence for proposing two
different logical structures for these alternate frames. Thus, benefactive applica-
tive morphemes and benefactive adpositions are represented by means of the same
predicative structure. Specifically, the predicative structure I propose is based on
the analysis of Jolly (1991: 126) and adopted in Van Valin & LaPolla (1997: 383).
Jolly represents the English preposition for as a purposive predicate of the type
shown in (3):
According to Van Valin & LaPolla (1997: 383), we may translate the predica-
tion containing a beneficiary PURP [BECOME have’ (Mary, cake)] as one action
being done with the intent of realizing another state of affairs.
carmen conti 237
In the case of one-place verbs with a beneficiary, like in the example from
Abaza or in the English sentence John sang for the students the logical structure
in (3) must be modified. I propose a preliminary alternative logical structure as
shown in (5).
The only difference between the logical structure in (3) and the one in (5) is
that the second argument of have’ (y, z) in the latter is an empty set referring to
a general activity: in the example in (6), the students are the intended goal of the
activity of singing.
At this point of the analysis, benefactive applicatives are not different from
other markers of beneficiaries. Then, how to capture that, in spite of the seman-
tic representation, benefactive applicatives add a new core argument and usually
give rise to a new selection of the undergoer? In opposition to simple three-place
verbs in some languages, which allow the recipient argument to be assigned the
undergoer (see, for instance, the case of give in John gave Mary a book), the new
assignment of the undergoer in applicative constructions is not associated with
a certain lexical class of verbs (normally, verbs of transference), but with a mor-
phological process. In effect, the assignment of the undergoer macrorole to the
beneficiary is determined by the attachment of a derivative affix, which activates
a specific principle to select U. Due to the fact that affixes should not give infor-
mation about macroroles assignment (macroroles assignment must depend only
on verbal predicates), the assignment of the undergoer in applicative constructions
should rest on an independent operation associating application with the principle
that assigns the undergoer macrorole. In order to account for benefactive applica-
tive constructions across languages, I will propose two independent morphological
operations that take place in different stages of the linking algorithm: benefactive
application can be either a morpholexical operation or a morphosyntactic operation.
The splitting into different types of operations seems to depend on the Macro-
roles-transitivity of the verb stem as well as on to what extent the attachment of
benefactive applicatives is constrained by one type of verb valence or another. In
238 applicatives and role and reference grammar
fact, those benefactive applicatives that allow the assignment of the undergoer
macrorole to the beneficiary usually attach to M-transitive verb stems, whereas
those other benefactive applicatives that add a beneficiary as a new non-macrorole
direct core argument seem to attach only to M-intransitive verb stems. What I will
argue from now on is that the former type of applicative expresses a morpholexical
operation, whilst the latter is the expression of a morphosyntactic operation.
I understand that a morpholexical operation takes place in a pre-syntactic level
and can be expressed by means of different morphological processes, like compo-
sition, incorporation and derivation. Morpholexical operations create new words
in the lexicon of languages and may affect the stem in two senses: they can change
the semantics of the stem (giving rise to a new meaning as well as to a new seman-
tic representation of the word, as occurs with the prefix dis- in disqualify) or they
can change the number of arguments of the stem (e. g. the affix se in abrirse ‘to
open [anticausative]’) or the semantic relationship among them (e. g. the prefix co-
in co-occur). In general, application is a derivative process that frames within the
latter type of operation: applicatives do not seem to change the meaning of the verb,
but just its valence. Even though both types of morpholexical operations occur in a
pre-syntactic level, each one seems to take place at different moments: the former
type is previous to the establishment of the Logical Structure of the new word,
whilst the latter type must be operative at the moment of assigning the macroroles
(that is, in the linking from semantics-to-syntax). On the opposite, morphosyn-
tactic operations do not change the semantics of the stem nor the number of the
verb arguments. Instead, they modify the morphosyntactic coding of arguments.
As we will see latter, there are some applicatives that also belong to this type.
Then, benefactive applicatives are markers of a morpholexical operation when
they give rise to the assignment of the undergoer macrorole to the beneficiary.
Macroroles assignment rests on two bases: the Logical Structure of the predicate
and the position held by the core arguments. In a two-place verb, the actor
macrorole is assigned to the left-most core argument, whilst the undergoer is as-
signed to the right-most argument of the Logical Structure. In three-place verbs,
however, the assignment of the undergoer may be more complex, as noticed by Van
Valin (2002), Guerrero & Van Valin (2004), and Van Valin (2005). In effect, some
languages allow recipients to be undergoers or even both recipients and themes
See the volume edited by Spencer & Zwicky (1998) for morpholexical operations and
morphosyntactic operations
carmen conti 239
The M-transitivity of the applicative verb remains intact when the verb stem
is M-transitive: both the stem and the derivational verb have two macrorole argu-
ments (actor and undergoer). Unlike the verb stem, which assigns the undergoer
to the right-most argument in the Logical Structure, the undergoer in the bene-
factive applicative construction has been assigned by means of Principle B.
From a cross-linguistic perspective, benefactive application does not give rise to
the same type of operation when the M-transitivity of the verb stem is intransi-
240 applicatives and role and reference grammar
As was mentioned below, rules (8) and (9) should take place in different steps
in the algorithm: rule (8) should apply at the moment in which the macroroles are
assigned (in the linking from semantics-to-syntax), whilst rule (9) should apply
when arguments are coded, just inmediately after the macroroles assignment.
As other functional affixes (derivational affixes modifying the number of verb ar-
guments) and inflectional affixes, benefactive applicatives attach to the nucleus,
as illustrated in Figure 1.10
Other morphological processes, such as noun incorporation and agreement,
may co-occur with benefactive application.11 In the sample, I have found the fol-
lowing order tendencies across languages:
10
In my opinion, the attachment of other derivational affixes —like those giving rise to
new entries in the lexicon and modifying the semantics of the stem (but not the morpho-
syntactic coding of arguments)— should not be represented in the syntactic template.
11
I will not take into account those cases in which more than one applicative may co-
occur in the clause, nor the attachment of aspectual affixes.
carmen conti 241
• The benefactive applicative is attached first to the verb stem than the object
agreement affixes, as illustrated in (11) below. Again, if we accept that affixes
ordering reflect the order of morphological processes, agreement is more ex-
ternal than benefactive application in verb formation.
12
Keep
�������������������������������������������������������������������������������������
in mind that the attachment of the morphemes to the right or to the left of the
verb will vary from language to language.
13
Keep in mind, however, that noun incorporation can be more external than applica-
tion in some languages, like Ainu. As noticed by Kaiser (1999), Ainu allows applicatives to
be attached to the verb stem before noun incorporation. In order to explain the unexpected
pattern of Ainu, Kaiser (1999) proposes that nouns can be incorporated either into the verb
root or into the applicative stem if the applicative is considered the head of the word.
242 applicatives and role and reference grammar
These order tendencies correspond with the general assumption that morpho-
lexical operations (like incorporation and a certain type of application) are more
internal than morphosyntactic operations in word formation. Or, in other more
familiar words, the morphological processes of composition, incorporation, and
derivation are more internal than inflection. However, it is necessary to remark
that the order of affixes in verb formation does not permit to establish any differ-
ence between benefactive applicatives as result of a morpholexical operation and
benefactive applicatives as result of a morphosyntactic operation. In both cases,
applicatives occur in the same position on the verb. In sum, it seems that the order
of applicatives in verb formation and the order of morphological processes do not
correlate in the case of morpholexical application and morphosyntactic application.
Morphological processes affecting the verb valence, like incorporation and ap-
plication, are transparent to syntax; that is to say, they go beyond the limits of the
word. In effect, benefactive applicatives do not only create a new word, but also af-
fect the code of core arguments and, as a consequence, the structure of the clause.
In my opinion, that is the reason why benefactive applicatives have to be repre-
sented in the syntactic template.
In those cases in which the non-macrorole core argument —that is, the theme
or the patient of the verb stem— is a noun incorporated into the verb, as occurs
in example (9) from Chukchi, the incorporated noun attaches to the node pred.
The benefactive applicative, as was noticed above, attaches to the nucleus. The fact
that pred is a more internal node than the nucleus reflects that incorporation is a
more internal process in word formation than application. The template for a co-
occurrence of incorporation and application will be as follows:
nucleus
appl pred
appl- v n
4. Conclusions
References
Van Valin, Robert D. 2002. The Role and Reference Grammar analysis of three-
place predicates. Available at http://wings.buffalo.edu/linguistics/rrg.
— 2005. Exploring the Syntax-Semantics interface. Cambridge: Cambridge Uni-
versity press.
Van Valin, Robert D. & Randy J. LaPolla. 1997. Syntax. Structure, Meaning,
and Function. Cambridge: Cambridge University Press.
Tepehua verb morphology, operator scope,
and the encoding of arguments
James K. Watters
Summer Institute of Linguistics International
1. Introduction
A claim made early in the development of Role and Reference Grammar (RRG)
was “that the order of verbal affixes with reference to the verb reflects the scope
relations of the operators they express” (Foley and Van Valin 1984: 223). Foley and
Van Valin were primarily referring to operators of tense, aspect, and mode. A simi-
lar claim is made in more recent works in RRG:
Operators are ordered with respect to each other in terms of the scope principle…
with the verb or other predicating element in the nucleus as the anchorpoint, and
thus the ordering restrictions on the morphemes expressing the operators are uni-
versal. Hence nuclear operators are closest to the nucleus, while clausal operators are
farthest away from the nucleus (Van Valin and La Polla 1997:49).
This paper includes material first presented in a cursory way in my dissertation (Wat-
ters 1988) which described Tepehua morphosyntax within the model of Role and Refer-
ence Grammar.
247
248 tepehua verb morphology, operator scope...
(1) [(adverb/verb)[verb]]
(2) a. xwilili-ch’uk’u-y
around-cut-impf
All‘S/he cuts Yexamples
Tepehua circularly.’
are written in a modified form of the practical orthography,
based on Mexican Spanish characters, e.g., x = /S/, j = /h/, ch = /tS/, etc.
All Tepehua examples are written in a modified form of the practical orthography,
2
The following abbreviations are used in the glosses. 1, 2, 3: first, second, third person;
based on Mexican
art: article; caus: Spanish
causative;characters, e.g., x = //,
com: comitative; comp:j = /h/, ch = /t/,cont:
completive; etc. continuative; dat:
3
The following abbreviations are used in the glosses. 1, 2, 3: first, second, third person;
art: article; caus: causative; com: comitative; comp: completive; cont: continuative; dat:
james k. watters 249
b. paaxkah-laqts’in
love-see.impf
‘S/he is grateful to Y.’
c. sqolih-tapaatsaa-y
whistle-work-impf
‘S/he works whistling.’
2. Morphological layers
Beyond the basic distinction between verb stem and affixes, there are distinct
structural layers within the Totonac-Tepehua verb, the most basic of which cor-
responds to the distinction between inflection and derivation. While there is no
universal set of criteria for distinguishing the two levels cross-linguistically, we will
use a rather simple distinction here.
Tepehua (and Totonacan in general) has a number of nominalization processes
that turn verbs into nouns. For our purposes, derivational morphology includes all
affixes that fall within the domain of nominalization. Thus you can have a nomi-
nalized passive form in Tepehua that serves a labeling function, as in (3):
The nominalized verb (the string within the brackets) includes the verb and
two overt affixes: a form of the instrumental prefix and the passive/reflexive suffix.
dative; desid: desiderative; dir: direction toward; fut: future; imm: immanent; impf: im-
perfective; inf: infinitive; ingr: ingressive; inst: instrumental; irr: irrealis; nom: nominal-
izer; pass: passive; pl: plural; pos: possessor; pf: perfect; pfv: perfective; prep: preposition;
pt: past tense; recip: reciprocal; refl: reflexive; rep: repetition; ret: return; seq: sequen-
tial; sg: singular; sub: subject; uo: unspecified object; via: via/means.
250 tepehua verb morphology, operator scope...
Following the criterion used here, then, both the instrumental and the passive/re-
flexive are non-inflectional (derivational) processes in Tlachichilco Tepehua.
Those affixes that interact directly with verbal syntax —aspect, directionals,
mode, tense, number and person— are the categories of the outer, inflectional
morphology. The primary affixes of the outermost layer, the domain of inflec-
tion, can be seen in Table 1 and Table 2 below. Again, these affixes are those most
typical of inflectional affixes, marking aspect, mode, tense, direction, person, and
number.
The irrealis ka- and the prefix kin- ‘1obj’ cannot co-occur.
james k. watters 251
within RRG. The claim in RRG regarding the ordering among affixes within the
same level is not absolute:
Within a group of operators at the same level, there is some variation in ordering.
Aspect is normally inside of nuclear directionals, but in a few languages it occurs
outside of them. Tense and status vary in their position relative to each other across
languages, but they are always inside of evidentials and illocutionary force (Van Valin
2005:11).
ta-
ingr
Body-part prefixes
lexical syntactic
core-level clause-level
encoding encoding
-ni -nVn [Stative- ‘oho -putun -choqo -pal -kan
dat uo Aspectual comp desid rep again pass/refl
Verb (Unmarked order -’alhi
Compound] alternate order cont
allowed) -’ela
seq
----- INNER ----- ----- OUTER -----
These include affixes that affect the lexical or syntactic encoding of arguments
as well as adverbial elements.
(4) t’aa-mi-l
com-come-pfv
‘S/he came with Y.’
(5) kii-tapaatsaa-l
ret-work-pfv
‘S/he went to work and returned.’
(6) puu-mi-l
via-come-pfv
‘S/he came by means of it (car, horse, etc).’
There is structural evidence that there are further morphological layers within
this span of non-inflectional morphology. First, only the Dative and Unspecified
Object suffixes may occur on the first member of a verb-verb compound:
james k. watters 253
b. [st’aa-nah]-[‘an]-ta
[sell-uo]-[go]-pf
‘S/he goes selling.’
b. an-ka-lh
���� ix-’ii-ka kasolinaj
���go-pass-pfv 3pos-get-pass gasoline
‘Someone went to get gasoline.’
The adverbial affix -choqo is usually found on the tensed verb but may occur on
the infinitive:
254 tepehua verb morphology, operator scope...
The other adverbial suffixes occur on the tensed verb in the construction:
In summary, the suffixes that are ordered closest to the verb stem occur on the
infinitive in such constructions, the passive-reflexive suffix is found on both the
infinitive and the tensed verb, and the adverbials are found on the tensed verb.
Turning our attention to the encoding of arguments, the “innermost” affixes are
those that affect what I will call the “lexical encoding” of arguments: the dative
applicative suffix and the unspecified object suffix, and the causative prefix. The
latter two of these operations affect the semantic structure of the clause by chang-
ing the mapping of arguments in the logical structure of the verb onto semantic
macroroles: one by adding a “causer”, the other by detransitivizing the verb (i.e.,
erasing the Undergoer position). Table 5 displays this two-way categorization of
the valence-affecting affixes.
These affixes along with the simple or derived verb stem are what Aschmann referred
as the “inner layer” of verb formation in early work in Totonac (Aschmann 1952; Aschmann
and Wonderly 1952).
james k. watters 255
The dative suffix —apart from its use to license a benefactive— corresponds to
what Van Valin and LaPolla (1997: 52ff ), following LFG terminology, call non-
predicative adpositions. It commonly serves to allow an indirect core argument of
the verb to occur as Undergoer. As an indirect argument of the verb, the semantic
role of the argument is determined by the verb. Otherwise the argument is as-
signed the default “benefactive” reading:
���
b. staq-ni-lh xkaan
give-dat-pfv water
��������������������������
‘S/he gave him/her water.’
c. st’aa-lh
sell-pfv
‘S/he sold it.’
d. st’aa-ni-lh
sell-dat-pfv
‘S/he sold it to him/her.’ or ‘S/he sold it for him/her.’
256 tepehua verb morphology, operator scope...
b. lhilhiy t’aa-milhpaa-y
daily com-sing-impf
b. puu-chiwiini-lh mikrofono
via-speak-pfv microphone
‘S/he spoke in/by means of a microphone.’
Like other affixes affecting the syntactic encoding of arguments, the passive-
reflexive suffix has no effect on the macrorole assignments or the verb’s Aktion-
sart; it only affects the morphosyntactic argument of the clause, i.e., which argu-
ment appears as subject:
(16) ik-st’aa-ni-ka-lh
1sub-sell-dat-pass-pfv
����������������
‘I was sold it.’
In sum, the affixes that determine the lexical encoding of arguments —the
causative, dative and unspecified object suffixes— are involved in the semantic
structure of the clause and occur closer to the verb stem. The applicative prefixes
and the passive-reflexive suffix affect the mapping from Actor-Undergoer onto
james k. watters 257
the PSA and syntactic object positions. The morphological ordering of these af-
fixes reflects their level of morphosyntactic operation.
When there are multiple adverbs in a sentence, they are constrained by the layers
of the operator projection, in that adverbs related to more outer operators occur
outside of adverbs related to more inner operators. In the simplest case, ‘outside of ’
means ‘farther from the verb’ (1997:165; Van Valin 2005:20).
In this section, I will attempt to show that the Tepehua data support an exten-
sion of this syntactic claim to the ordering of morphological adverbials. I will pres-
ent evidence for core-level and clause-level adverbial affixes.
At first glance, it may seem unusual to consider the following three suffix as
core-level adverbials -’ojo ‘completive’, -choqo ‘again’, and -putun ‘desiderative’.
However, as in all constructions, the simple glosses can be deceiving. All three
of these suffixes affect the internal semantics of the cores in some way, relating
directly to properties of individual arguments and their relation to the state or
event.
Thus, -’ojo ‘completive’ serves not only an aspectual function —it may refer to
the totality of members of the set designated by the subject, by the object, or to the
totality of the action itself. Similarly, -choqo refers to the fact that some state or ac-
tivity is predicated as re-occurring, and -putun marks volition on the part of one of
the arguments within the Logical Structure [LS] in regard to the event.
The completive suffix -’ojo ~ -’oo (or its cognate) occurs throughout Totonac-
Tepehua, though with variant ordering restrictions. In the highland dialect of To-
tonac it has become the marker of third person plural subject. However, in other
Totonac variants and in Tepehua variants it regularly marks something like the
totality of the action. Thus in Tepehua -’ojo can be roughly described in the fol-
lowing way: with an intransitive verb it may denote the total set of members of a
plural subject or, with a singular subject, completion of the action. With a transi-
tive verb it generally refers to the total set of members of a plural object. See the
examples presented below.
258 tepehua verb morphology, operator scope...
(17) chiwiin-’oo-lh
talk-comp-pfv
‘S/he finished talking.’
(20) lak-st’aa-ni-’oo-lh
3pl.obj-sell-dat-comp-pfv
‘S/he sold Y to all of them.’
(21) a. chiwiin-choqo-lh
talk-rep-pfv
‘S/he talked again.’
b. ki-xtaq-ni-choqo-lh
1obj-give-dat-rep-pfv
‘S/he gave it back to me.’
As noted, -choqo may precede or follow the desiderative suffix -putun, but often
with different readings (at least for some speakers). The two possible orders are
given different interpretations reflecting the difference in scope of the adverbials:
(22) a. k- wayn-choqo-putun
1sub-eat-rep-desid.impf
‘I want to eat again.’
james k. watters 259
b. k- wayn-putun-choqo-y
1sub-eat-desid-rep-impf
‘Again I want to eat.’
While no such alternate orderings are found with affixes encoding operators, it
is striking that alternate ordering of affixes is allowed in such adverbial affixes (see
Beck 2007 and McFarland 2007 for more extended discussion of such variable
ordering in Totonac involving other affixes). These morphological facts parallel
observations regarding the rigid order of operators versus the variable ordering of
adverbs in syntax, a point we will return to at the end of this paper.
For examples of what is traditionally called the desiderative suffix in Totonac
and Tepehua studies, see the examples in (23) and (24). Note that the “desidera
tive” in these instances is not a feature of the speaker but rather marks the desire of
a referent in the clause.
(23) tapatsaa-putun
word-desid.impf
‘S/he wants to work.’
(24) k-xtaq-putun-kan-aa-w
1sub-give-desid-pass-impf-1plsuab
‘We (excl) want to be given away.’ or ‘Someone wants to give us (excl) away.’
While there’s nothing remarkable about (23), the occurrence of the desider-
ative with the passive, as in (24) produces an interesting ambiguity. This ambiguity
shows the semantics of the desiderative construction is very distinct from want to
constructions in English as well as from verb-verb compounds in Tepehua. In pro-
ductive verb-verb compounds, the two verbs must have the same subject; however,
as (24) shows, in the desiderative construction, there is no requirement that the
Actor/Experiencer of the main verb be the one with the desire. Rather the prop-
My use of the term “desiderative” follows the tradition in Totonacan studies (e.g. Mc-
Quown [1940]1990:183). As in much of the Americanist tradition, it signals desire on the
part of the subject of the clause, not desire on the part of the speaker. The latter use of “de-
siderative” is found in the literature on modality (e.g. Palmer 1986: 10,105), referring to an
operator that has scope over the entire clause.
260 tepehua verb morphology, operator scope...
erty of “desiring” the event can be understood as applying to either the Actor or
Undergoer of the verb stem, providing further evidence that -putun functions as a
core-level adverbial.
Turning to the clause-level adverbials, there is a central semantic factor they
have in common: they each designate some temporal relation between the state or
event expressed by the clause and the context (e.g., to other states or events that
may or may not be explicitly mentioned). The clause-level adverbials modify the
entire logical structure associated with the verb stem; i.e., they don’t have any in-
ternal effect on the interpretation of the core. It is this function that marks all three
as clause-level adverbials in contrast to the core-level adverbials we just looked at.
All three of these suffixes are less frequent in the Tlachichilco dialect than the
core-level adverbials just discussed. Their functions are often taken up by either
full Tepehua adverbs or by Spanish loans.
The form -pal(a), like -choqo, is usually best translated ‘again’, signifying the event
or state referred to by the core has attained at some previous time. It often occurs
with ‑choqo. While the two suffixes are synonymous in some constructions, there
are two kinds of differences that show up with certain verbs and in certain contexts.
The first difference is a matter of semantic scope and has only been noted by my
Tepehua consultants when occurring with the motion verbs an ‘go’ and min ‘come’.
The difference is conveyed by the translations given the forms in (25) and (26):
(25) a. ka-min-choqo-yaa
irr-come-rep-fut
‘S/he will come again (=come back).’
b. ka-min-pal-aa
irr-come-again-fut
‘Again, S/he will come.’
(26) a. ka-’an-choqo-yaa
irr-go-rep-fut
‘S/he will go again (=return)’
b. ka-’an-pal-aa
irr-go-again-fut
‘Again, s/he will go.’
james k. watters 261
There are some data that suggest the translations in these examples reflect dif-
ferences in at least the Tlachichilco dialect of Tepehua, readings that are presented
in (27) and (28).
(27) a. ka-min-choqo-yaa
[BECOME NOT be.at’ (x, there)] & [BECOME again be.at’ (x, here)]
b. ka-min-pal-aa
again [BECOME NOT be.at’ (x, there)] & [BECOME be.at’ (x, here)]
(28) a. ka-’an-choqo-yaa
[BECOME NOT be.at’ (x, here)] & [BECOME again be.at’ (x, there)]
b. ka-’an-pal-aa
again [BECOME NOT be.at’ (x, here)] & [BECOME be.at’ (x, there)]
But the distinctions between the (a) and (b) forms are subtle and are compli-
cated by apparent differences in speaker judgments. The suggested logical struc-
tures reflect the fact that (27a) is appropriately used when the subject is a local
resident, while (27b) is appropriately used for a resident that that has gone away,
say to work, and will come back. Similarly, (28a) more appropriately applies to one
who has established his residence elsewhere than does (28b).
The second difference in the readings associated with these two suffixes has to
do with the speaker’s viewpoint. Unlike -choqo, the suffix -pal(a) has acquired an
evidential reading of doubt or, in other cases, conveys sarcasm. These elements are
clearly understood in the readings of (29) and (30) in Tlachichilco Tepehua:
While the -pal(a) suffix is quite common in both Huehuetla and Pisaflores
Tepehua, it is more unusual in Tlachichilco and most often is found in the kind
262 tepehua verb morphology, operator scope...
(32) ka-wayn-el
irr-eat-seq
‘Eat first!’
(33) a. milpaa-’ela-lh
sing-seq-pfv
‘S/he sang first.’
(before some event, e.g. before Y sang or before S/he did something else)
b. p’unah milpaa-lh
first sing-pfv
‘S/he sang first.’
The last of these clause-level suffixes -(q)’alhi signifies ‘always’, i.e., that the
event or the state continuously occurs.
(36) skaka-’alhi-ta
get.hot-cont-pf
‘S/he is always hot.’
(37) xix-’alhi-ta
get.dry-cont-pf
‘S/he is always dry.’
(38) kuh-’alhi-ta
wake.up- cont-pf
‘S/he is always awake.’
The examples in (34) and (35) are in the imperfective and, as achievements,
give the iterative readings. Similarly, if the (36)-(38) were in the imperfective, the
translations would be something like “S/he is always getting hot”, “It is always
getting dry”, and “s/he is always waking up”. However, the perfect aspect with
achievement verbs in Tepehua indicates the resultant state holds:
(39) a. ta-’on
3plsub-get.fat.impf
‘They get fat.’
b. ta-’on-ta
3pl.sub-get.fat-pf
‘They are fat.’
Similarly, if an achievement verb with -’alhi occurs in the perfect it has a reading
that is characteristic of statives. In such a construction with -’alhi in the perfect, the
“always” reading has scope over the resultant state reading, suggesting the perfect
suffix is within the scope of the temporal adverbial -’alhi. This is as we would ex-
pect if -’alhi is, in fact, a clause-level adverbial. However, the scope is not reflected
in morphological order of these two suffixes since the perfect suffix follows -’alhi.
264 tepehua verb morphology, operator scope...
It is clear that this is an instance where the structural distinction between deri-
vational and inflectional morphology skews the match between affix order and
scope. Assuming there is a real difference between the derivational and inflection-
al components, we can account for this mismatch of suffix order and suffix scope.
The perfect aspect is a nuclear operator but it must occur outside the clause-level
operator ‘alhi, because it is inflectional while the latter is within the derivational
component. That is, as in (40) and (41), the linear order is determined by the
structural distinction between inflection and derivation, while the scope is deter-
mined by the semantics of the elements.
Simply based on data from Tepehua (if we had time we could draw on similar
data in Totonac, as well) it is clear that the claim that morpheme orders reflect
the scope or levels of operators has to be tempered by structural concerns within
the morphology, in this case the distinction between derivational and inflectional
affixes.
5. Conclusion
There are two points to be made in conclusion regarding the way Role and Refer-
ence Grammar moderates the claim that morphological order reflects semantic
scope. This is best seen by comparing the strong claim made by Cinque (2004)
in his cross-linguistic study, regarding the ordering of both adverbs and operators
cross-linguistically:
Much as inflectional morphology, functional particles, and auxiliaries were at the time
[1999] considered to be the overt manifestation, in head format, of the functional
portion of the clause, AdvPs, I argued, could be seen as the overt manifestation of the
same functional distinctions in specifier format. The main evidence for their belong-
ing to the functional make-up of the clause was the observation that cross-linguisti-
james k. watters 265
cally the number and type of the different classes of AdvPs and their relative order
appears to exactly match the number, type and relative order of functional heads mor-
phemes (Cinque 2004:683-4; my emphasis).
Note that the strong prediction of exact match between the relative orders of
operators and adverbs is required by Cinque’s model which treats adverbs as the
specifiers of functional projections. The claim in RRG is not as strong. First, there
is what could be called thresholds of operators:
It should be noted that within a group of operators at the same level, there is some
variation in ordering. Aspect is normally inside of nuclear directionals, but in a few
languages it occurs outside of them. Tense and status vary in their position relative to
each other across languages, but they are always inside of evidentials and illocution-
ary force; universally, illocutionary force is always the outermost operator over the
clause (Van Valin 2005:11).
A further disclaimer is made by Van Valin and La Polla (1997) regarding the
ordering of adverbs, which can be applied to morphological adverbials as well:
Finally, it seems that work by Cinque and others, in which adverbs are treated
as specifiers of functional projections, is motivated at least in part by the fact that
adverbs seem to share some of the characteristics of operators. In RRG, the paral-
lel scope relations of adverbs and operators reflects this similarity, while the pres-
ence of adverbs in the constituent projection and their variable orderings lead us
to expect the kinds of differences between operators and adverbs expressed in the
quote above. The presence of variable ordering of some adverbial affixes in the
morphology is parallel to the variable orderings of adverbs found in syntax.
In conclusion, the ordering of affixes in Tepehua verb morphology support
mappings between semantics and morphosyntax proposed by RRG. First, there
are differences in structure and ordering between those affixes that involve the
lexical encoding of arguments and those that mark the syntactic encoding of argu-
266 tepehua verb morphology, operator scope...
ments. Second, the data support the relation established within RRG between the
scope relations and the ordering of operator affixes in the morphology of the verb.
Furthermore, there is a similar, although more complex, relation between the order
of adverbial affixes and scope.
References
1. Introduction
269
270 the functions, semantics and syntax of the adjective in irish
crua ‘hard’, bog’ soft’, láidir ‘strong’, lag ‘weak’, glan ‘clean’,
e) Physical
salach ‘dirty’, te ‘hot’, fuar ‘cold’, géar ‘sharp’, maol ‘blunt’
Additional properties
m) Cardinal numbers aon ‘one’, dhó ‘two’, céad ‘first’, deireanach ‘last’
Adjectives have a referential use and as such can serve either for referring to the
property that they denote or to the person or object to which the property belongs.
When used predicatively, adjectives function to modify the referent of the noun
phrase. This function is different from that of attributively used adjectives that
modify the reference (or meaning) of the head nouns rather than the referents. A
characteristic of adjectives when used attributively is that they are morphologi-
cally marked for agreement with the noun they quality.
brian nolan 271
Pustejovesky notes that “we have little insight into the semantic nature of adjec-
tival predication, and even less into the semantics of nominals. Not until all major
categories have been studied can we hope to arrive at a balanced understanding of
the lexicon and the methods of composition” (1995: 7). He considers the Dixon’s
taxonomic classification of adjectives by the semantic-field associated with the ad-
jective to be useful (1995: 22). However, he also believes that it is important to ex-
amine the relational and logical properties of the adjectives in their usage. It would
appear that, in fact, the qualia constitutive (material, weight, parts and compo-
nents) and formal (orientation, magnitude, shape, dimensionality, colour and posi-
tion) roles are essentially subsumed within Dixon’s typology of adjectival semantic
types to constitute a subset of Dixon types, in Table 3.
We can observe that the taxonomy of semantic properties of Dixon and Ai-
khenveld (2004) is more comprehensive than those noted by Pustejovsky (1995).
In addition, their list is both typologically motivated and cross-linguistically at-
tested. Indeed, many of these adjectival semantic properties simply have no equiv-
alent in the Qualia Theory model.
The lexical entry for a nominal in Role and Reference Grammar (RRG) is
partially based on Pustejovsky’s theory. This qualia theory attempts to capture the
attributes and behaviours of nominals and verbs as constitutive, formal, telic, and
agentive. We are concerned here with adjectives and how they can qualify nouns.
The qualia theory requires that the lexical entry for a noun contains a set of qualia
{QC, QF, QT, QA}, which represent its primary semantic properties, much like a logi-
cal structure represents the semantic properties of a verb.
A more complete semantic representation is available when the verb and noun
lexical entries are combined (Van Valin 2005: 50ff ). Adjectives bind with the noun
then according to their compatibility with the constitutive and formal properties
of the host noun.
brian nolan 273
In the layered structure of the noun phrase (henceforth LSNP, Van Valin 2005:
24), there is a nominal nucleus and a nominal core consisting of the nucleus itself
and associated arguments. The NP level corresponds to the clause / sentence lev-
els of the layered structure of the clause. Each of these has a periphery, as shown
in Figure 1. Nominal aspect concerns whether the referent entity is an individual,
parts of an individual, a set of individuals, or a sortal kind. CoreN operators are to
do with quality with respect to number, qualification and negation while the NP
operators (2) relate to locating the NP in a discourse situation.
Figure 1. The general schema for a layered structure of the noun phrase
(Van Valin 2005:24ff )
274 the functions, semantics and syntax of the adjective in irish
(2) NP, CoreN and NuclearN operators (Van Valin 2005: 24ff)
NuclearN operator
o Nominal aspect (count-mass distinction, classifiers in classifier languages)
CoreN operators
o Number
o Qualification (quantifiers)
o Negation
NP operators
o Definiteness
o Deixis
Irish is a VSO language and adjectives occur immediately after the noun that
they qualify. As in many languages (Dixon and Aikhenveld 2004: 10), a noun in
Irish may be modified by more than one adjective. Generally, there is a preferred
order in which the adjectival semantic types will occur. In languages where adjec-
tives precede the noun, the unmarked order is value, dimension, physical property,
speed, human propensity, age, and colour.
In Irish, where the adjectives follow the noun, the Rijkhoff proposal would
predict that the order is the reverse of that, as indicated in (3). This will be tested
later and, if correct, will validate both the claim of Rijkhoff (2004: 224) and the
RRG conceptualisation of the layered structure of the noun phrase. Curiously, this
preferred unmarked order does not follow the linear core and peripheral order
presented by Dixon and Aikhenveld (2004: 44) in their taxonomy of adjectival
semantic types.
Figure 2. The symmetry in the underlying structure of the clause and
the noun phrase (Rijkhoff 2004: 224)
276 the functions, semantics and syntax of the adjective in irish
Prior to Van Valin (2005) in RRG, adjectives were treated as nuclearN opera-
tors. Present thinking (ibid: 24ff ) is to consider adjectives as analogous to adverbs
in the clause. That is, they are constituents of the nuclearN periphery and their
position is constrained according to the iconicity principle. This principle predicts
that the adjectives must occur x closer to the nominal nucleus than the coreN and
NP-level operators and modifiers. Additionally, the unmarked order of adjectives,
their preferred order, will apply, that is, they will be adjacent to the N in the phrase.
A partial projection for the Irish LSNP in (4) is given in Figure 3. We can note
that the gender is marked on the noun and the adjective. The adjective is also
marked for plural number.
In this paper we adopt the following abbreviations. adj: Adjective; adv: Adverb; conj:
Conjunction; cop: Copula; det: Determiner; fut: Future; gen: Genitive; gend: Gender;
impers: Impersonal; lit: Literally; ls: Logical Structure; n: Noun; np: Noun Phrase; num:
Number; pass: Passive; past: Past Tense; pl: Plural; pn: Pronoun; poss: Possession; prep:
Preposition; pred: Predicate; pres: Present; rrg: Role and Reference Grammar; sg: Single;
subv: Substantive verb of ‘to be, exist’; v: Verb; va: Verbal Adjective; vn: Verbal Noun.
brian nolan 277
np
coren
nucn perypheryn
n adj
An dá theach mhóra
coren
coren num
qnt coren
def np
Figure 3. The layered structure of the Irish NP in (4)
b. Attributive
Tá an bhean láidir ag damhsa
be.subv.pres the.det woman.n.f strong.adj at.prep dancing.vn
‘The strong woman is dancing.’
do‘ (woman, [dance’ (woman)]) ^ be’ (woman, [strong’])
c. Identificational
Tá sé anois ina shagart paróiste
be.subv.pres he.pn.m now.adv in.prep.his.poss priest.n.m parish.n.m
278 the functions, semantics and syntax of the adjective in irish
in Oileán Ciarraí
in.prep Island.n.m Kerry.n.m
‘He is now a parish priest in Kerry Island.’
be’ (he, [a parish priest])
d. Specificational
Is buaiteoir Chomórtas na mBan í.
be.cop winner.n.m competition.n the.det women.n.f she.pn.f
‘She is the winner of the Women’s Competition.’
be’ (she, [the winner of the Women’s Competition])
Tá sé ina buaiteoir.
be.subv.pres he.pn in.prep:his.pn-poss winner.n.m
‘He is the winner.’
be’ (he, [the winner])
e. Equational
Tá Sean ar an duine is cliste
be.subv.pres Sean.n-m on.prep the.det person.n.m be.cop clever.adj
sa rang
in.prep:the.det class.n.m
‘Sean is the person who is cleverest in the class.’
equate’ (Sean, the person who is cleverest in the class)
Adjectives maith ‘good’, olc ‘bad’, beag ‘small’, bán ‘white’, dubh ‘black’, etc.
Possessives mo ‘my’, do ‘your’, a ‘his, her, their’, ár ‘our’, bhur ‘your-PL’
dhá ‘two’, trí ‘three’ ceithre ‘four’, etc
Numeral chéad ‘1st’, dara ‘2nd’, triú ‘3rd’ (thing not person), etc.
duine ‘1st’, beirt ‘2nd’ triúr ‘3rd’ (person not thing) etc.
Indefinites áirithe ‘any’, amhán ‘any’, céanna ‘same’, éigin ‘some’, eile ‘other’, etc.
Interrogatives cá ‘what’, cé ‘who’
Demonstratives seo ‘this’, sin ‘that’, úd ‘there’
Definite
an ‘the-Sg’, na ‘the-PL’
determiner
Table 4. Types of items that can quality the noun phrase
These items, listed in Table 4, may be examined within the layered structure of
the noun phrase. We will concentrate on adjectives and their co-occurrence with
these possessives, numerals, indefinites, interrogatives, demonstratives and defi-
nite determiners in relation to the noun phrase. In Irish, an adjective can be used
predicatively or attributively. When used predicatively (6), it qualifies the noun or
pronoun indirectly as predicate or part of the predicate. When used attributively, it
qualifies the noun directly (7).
b. Tá mé tuirseach.
be.subv.pres I.pn tired.adj
‘I am tired.’
feel’ (me, [tired’])
280 the functions, semantics and syntax of the adjective in irish
c. Tá mé fuar.
be.subv.pres I.pn cold.adj
‘I am cold.’
feel’ (me, [cold’])
Attributive use
b. Tá lá deas ann.
be.subv.pres day.n.m nice.adj there.dem
‘It is a nice day.’
be’ (day, [nice’])
(9) a. Ba bhreá an lá é.
be.cop great.adj the.det day.n.m it.pn.m
‘It is a great day.’
With predicative use of an adjective, the preposition go ‘to’ is frequently, but not
always, used, as shown in the following examples. Typically, this use of the prepo-
sition is deployed when an opinion or value judgment is expressed.
(10) a . Tá mé go maith.
be:subv-pres I:pn to:prep good:adj
‘I am good.’
b. Tá an sceal go holc.
be:subv-pres the:det story:n.m to:prep bad:adj
‘The story is bad.’
We mentioned earlier that Irish is a VSO language and that adjectives occur im-
mediately in after the noun that it qualifies. A noun in Irish may, of course, be
modified by more than one adjective. Generally, there is a preferred order in which
the semantic types will occur, as mentioned before in (3) and repeated here for this
discussion.
We now examine the co-occurrence of the adjective with the NP operators, start-
ing with demonstratives and following with indefinites, possessives, number and
interrogatives.
4.2.1 Demonstratives
In these examples, the demonstrative is placed closer to the noun than the
adjective. The function of a demonstrative is to locate a referent in relation to
a speaker and is therefore deictic in nature. In the layered structure of the noun
284 the functions, semantics and syntax of the adjective in irish
4.2.2. Indefinites
Some indefinites appear after the noun (13), in the same manner as attributive
adjectives but, importantly, with the adjective immediately following the noun
and before the indefinite. Others indefinites appear before the noun (14). In both
these instances with indefinites, the position immediately closest after the noun is
reserved for the attributive adjective.
c. Gach re lá te
every:indef second/other:num day:n.m hot:adj
Every second hot day
In addition, possessives (15a), numerals (15b) and interrogatives (15c) appear before the
noun, with the adjective following the noun.
In terms of the preferred priority order to qualify the noun, we can see that the
position closest to the noun is reserved for the attributive adjective. Note in (16b)
that Cailín ‘girl’ is actually a noun with morphological gender of masculine rather
than the feminine marking expected, in virtue of the form of the word ending with
-ín. We discuss the determination of gender in the next section.
286 the functions, semantics and syntax of the adjective in irish
This data provides evidence that validates the RRG conception of the layered
structure of the noun phrase. In particular, the evidence from the examples con-
cerning the demonstratives and adjectives clearly shows that the demonstrative
functions deictically as an isomorphic mapping into a set of individuals to isolate a
member of the set for qualification by the attributive adjective. As such, it validates
the scope of the operators and that Irish adheres to the model.
The occurrence of agreement markers for number, gender and case (or different
combinations of these) is considered to be another defining characteristic of adjec-
tives. Agreement markers provide a link between adjectives and their head nouns
and, hence, they may be regarded as indicating the dependence of adjectives upon
the head nouns. Adjectival agreement is widespread and productive in Irish. An at-
tributively used adjective is marked morphologically for number, gender, and case,
according to the noun that it qualifies. (An exception to this occurs when the noun
is qualified by a number (NUM) operator). Some examples are shown in Table 5.
We next need to discuss numbers and their implications for the N. First, draw-
ing on Ó Siadhail (1996:46), we can note that in Irish a number is always fol-
lowed by an associated N. Secondly, the numbers aon ‘one’ to sé ‘six’ triggers leni-
tion on the initial consonant on the N, for example, sé bhád ‘six boats’. The noun
bád becomes bhád as a result of lenition occurring. Thirdly, the numbers seacht
‘seven’ to deich ‘ten’ cause eclipsis, for example, deich n-éan ‘ten birds’ (Ó Siadhail
1996:46ff ). That is, the noun éan becomes n-éan following this process within this
context. The morphologically singular form of the N follows these numbers; some
examples are in Table 8.
brian nolan 289
We will now continue our discussion of numbers and relate these first to the
determiner and then, following, to the adjective within the NP. We will again draw
on Ó Siadhail (1996:53) in our discussion. We will discuss first the use of the
definite determiner with numbers. The number two is dhá in lenited form and dá
in non-lenited form used after determiners. The plural form of the determiner is
required with all numbers over two but can be employed optionally with dá ‘two’.
We can see examples of these in (17).
b. An dá fhuinneoig
the:det-sg two:num window:f.sg
‘The two windows.’
c. Na dá fhuinneoig
the:det.sg two:num window:f.sg
‘The two windows.’
d. Na trí fhuinneoig
the:det-pl three:num window:f.sg
‘The three windows.’
e. Na cheithre shagart
the:det-pl four:num priest:f.sg
‘The four priests.’
The use of adjectives after numbers is as follows. After aon ‘one’ the adjective is
singular (18a). After the numbers dhá ‘two’ to deich ‘ten’ the adjective is plural and
lenited on the initial consonant (18b-d).
b. An dá fhuinneoig mhóra
the:det.sg two:num window:n.f.sg big:adj.pl
‘The two big windows.’
c. Na dá fhuinneoig mhóra
the:det.sg two:num windown:f.sg big:adj.pl
‘The two big windows.’
d. Na trí theach bheaga
the:det-pl three:num window:n.m.sg little:adj.pl
‘The three little houses.’
The Irish adjective, therefore, does not agree in number with its nouns when
it is qualified by dhá ‘two’, trí three’, etc. or beirt ‘two (people)’ (Christian Brothers
1997:68). The examples in (19) show gender marking agreement on the noun and
its modifying adjective while the NUM operator has scope over the NP. Number
marking for plural also occurs on the adjective. The adjective has the common plu-
ral marking (typically -a, see Table 5). Notice how the noun and the adjective all
show the lenition agreement marker.
b. An dá bhád bheaga
the:det.sg two:num.thing boat:n.m.sg small:adj.m.pl
‘The two small boats.’
d. Tríbhád mhóra
three:num boat:n.m.sg big:adj.m.pl
‘Three big boats.’
292 the functions, semantics and syntax of the adjective in irish
The numerical operators on the noun —the numbers dhá ‘two’, trí ‘three’, and
so on— trigger morphological changes in the form of lenition on the first conso-
nant of the associated noun (e.g., NUML N) as in (17), (18) and (19). Overall, this
evidence further lends support for the RRG conception of the layered structure of
the noun phrase in that we can clearly see that the NUM operator has scope over
noun + adjective number agreement.
Gender and case agreement marking, of course, are not impacted by the NUM
operator as is to be expected given they have to do with nominal agreement prop-
erties and grammatical function respectively.
According to Bhat and Pustet (2000: 758), all languages that have adjectives as a
distinct category also have morphological or syntactic constructions of adjectives
for indicating one of more types of degree modifications. Accordingly, we can re-
gard degree modification as an important part of the distinctive criteria of proto-
typical adjectives. The basis of this characteristic is the fact that adjectives denote a
single property. Nouns, for example, denote a cluster of properties characterized by
the respective adjectives. They also denote an object or entity that possesses those
properties. In turn, the possibility of modifying the meaning of a noun by adjec-
tives as attributes presupposes that adjectives indicate a single suitable property.
Such an indication is also necessary for adjectives to function as the basis of degree
or comparison with respect to that property.
brian nolan 293
Irish adjectives can be used for “comparison for equality” and “comparison for
degree”. The former uses the constructional schemata in (21) while the latter uses
the schemata in (26). Languages generally use several types of constructions -
morphological as well as syntactic- for denoting degree modification and Irish is
no different in this regard. In addition, the “superlative” employs the form of the
adjective used in comparison for degree but with a different schema.
We now examine adjectives used in comparison for equality. Irish has four sche-
mata that may be deployed to this function (21).
Examples for each schema are below. In these examples, we find usages of
the existential verb ta ‘to be’. This is usually called the substantive verb ‘to be’ in
Celtic linguistics. We gloss this verb as subv to distinguish it from the copula
cop is ‘be’.
leatsa
with:prep+you:pn-emphatic
‘He is not as good a boatman as you.’
sé riamh.
he:pn before:adv
‘He is as bad as he ever was.’
b. Tháinigh mé chomh gasta agus a thiochfadh liom
came:v.past I:pn as:part quick:adj and:conj as:prep come:v.fut with:prep+me:pn
‘I came as quick as I could.’
Adjectives used to express comparative degree can utilize any of the following
three constructional schemata (26). When used in these constructions, the mor-
phological form of the adjective in the examples (27)-(29) is different from the
base adjectival form in Table 10.
brian nolan 295
sa rang
in:prep+the:det class:n.m
‘Sean is the cleverest person in the class.’ (lit: ‘Sean is the person who is cleverest
in the class.’)
sa rang
in:prep+the:det class:n.m
‘Sean is among the cleverest (people) in the class.’
296 the functions, semantics and syntax of the adjective in irish
c. an buachaill is óige
the:det boy:n.m be:cop youngest:adj
‘The youngest boy.’ (lit. the boy who is the youngest)
and derivational processes are regular and productive. Adverbs can also be derived
from adjectives. In Irish we can find derivational processes involving the following
strategies:
An intensifier can prefix an adjective with the resulting form being another adjec-
tive. That is, the adjectival sortal type is retained.
(32) a. láidir
strong
‘Strong’
b. ró-láidir
too-strong
‘Too strong’
Adjectives can be derived from verbs in several languages by changing the verbs
into participles or verbal adjectives. These participles generally retain some of the
tense-aspect distinctions and are less prototypical as adjectives. Adjectives mor-
phologically derived from verbs are very productive in Irish and are used widely
as participles in the perfective passive, one of a number of passive constructions
found in the language. We can see an example of the perfective passive and sche-
ma in (33). A interesting point worth noting in regard to (33) is the status of agam
‘at me’. This is called a prepositional pronoun and it is a part of speech or word
class particular to Irish and the Celtic languages. In this productive word class, a
298 the functions, semantics and syntax of the adjective in irish
preposition and a personal pronoun conflate together and inflect for person and
number. Some examples for the same preposition include: agat ‘at you.sg’, aige ‘at
him’, aci ‘at her’, againn ‘at us’, agaibh ‘at you.pl’, acu ‘at them’. The result may have
a somewhat irregular morphological shape. A full discussion of these, however, is
beyond the scope of this paper.
In the same way that multiple adjectives may co-occur with a noun, an adjec-
tive may be used productively with a verbal adjective. We see several different ex-
amples of this in (34) where the (verbal) adjectives, derived originally from a verb,
are deployed in adjectival function. Table 11 shows some of the verbal and verbal
adjective forms.
Adjectives can compound with other adjectives where the resulting form is still an
adjective.
ar an dreisiúr.
on:prep the:det dresser:n
‘She put a fresh ornament on the dresser.’
timpeall na n-oileán.
around:adv the:det islands:n.m
‘For a couple of years one made very profitable fishing around the islands.’
Adjectives can form compounds with nouns where the resulting form is a nominal
(36). The function of qualifying a noun may be accomplished by either using an
independent adjective with the noun or by forming an adjective-noun compound.
Most of the adjectival meanings are denoted by states that are then compounded
with the noun in this manner. When the adjective is compounded with a noun,
the adjective always appears as a prefix on the noun.
orm.
‘The old folk’s prudence made me very resourceful.’ (lit. ‘The prudence of the old
person put huge reserves on me’)
8. Conclusion
In this paper we have looked briefly at how adjectives are considered in the general
functional linguistic literature. We have identified a set of semantic types and re-
302 the functions, semantics and syntax of the adjective in irish
lated these to the work of Pustejovsky in Qualia Theory and the lexicon in RRG.
We considered the status of the adjective within the RRG layered structure of the
noun phrase, its operators and a related functional model from Rijkhoff. We tested
the prediction made by Rijkhoff (2004: 224) that adjectives will appear closer to
the noun than nominal aspect or lexical number, or possessor marking and in do-
ing so we were able to confirm that it and the RRG layered structure of the noun
phrase, and its operators, are compatible with the adjectival and other data from
Irish. The preferred adjective occurrence for Irish was confirmed to be (repeated
from (3) above):
References
Bhat, D.N.S and Regina Pustet (eds.). 2000. Adjectives. Morphology: An inter-
national handbook on inflection and word-formation. Volume 1. Berlin: Mouton
de Gruyter.
Christian Brothers. 1997. New Irish Grammar. Dublin: C.J. Fallon-Mount
Salus Press.
Dixon, R.M.W. 1977. Where have all the adjectives gone? Studies in language.
1:1, 1-80.
Dixon, R.M.W. and Alexandria Y. Aikhenvald (eds.). 2004. Adjectival Classes: A
cross-linguistic typology. Oxford: Oxford University Press
brian nolan 303
Matasovic, Ranko. 2002. Adjectival phrases. Available at the RRG website: http:
linguistics.buffalo.edu/people/ faculty/vanvalin/rrg/Matasovic.pdf
Ó Siadhail, Mícheál. 1996. Learning Irish. New Haven & London: Yale Univer-
sity Press.
— 1989. Modern Irish. Cambridge MA: Cambridge University Press.
Pustejovsky, James. 1995. The Generative Lexicon. MA: MIT Press.
Rijkhoff, Jan. 2004. The Noun Phrase. Oxford: Oxford University Press.
Schwartz, Linda. 1993. On the syntactic and semantic alignment of attributive
and identificational constructions. Advances in Role and Reference Gram-
mar, R. Van Valin (ed). Amsterdam/Philadelphia: John Benjamins Publish-
ing Co.
Thompson, Sandra, A. 1990. A Discourse Approach to the Cross-Linguistic Cat-
egory ‘Adjective’. Explaining Language Universals, J. Hawkins (ed). Cambridge
MA: Blackwell.
Van Valin, Robert D. 2005. Exploring the Syntax-Semantics Interface. Cambridge:
Cambridge University Press.
Van Valin, R. and R. LaPolla. 1997. Syntax: Structure, meaning and function.
Cambridge: Cambridge University Press.
A new typology of control constructions
within the framework
of Role and Reference Grammar
Ranko Matasović
University of Zagreb
1. Introduction
Ergative/Oblique [erg]: -m
Nominative/Absolutive [nom]: -r
For a description of Kabardian and for transliteration conventions see Matasović
(2007).
Abbreviations used in this paper. a: actor, abs: absolutive, af: affirmative, caus: caus-
ative, clm: noun class (gender) marker, comp: complementizer, dir: directional, erg: erga-
tive/oblique, impf: imperfect, inf: infinitive, neg: negation, nom: nominative, pl: plural,
pres: present, pret: preterite, refl: reflexive, sg: singular, u: undergoer.
305
306 typology of control constructions within the framework of rrg
0-yǝ-rǝ-rǝy-ǝa-t-xa
3-3.sg-3s.g-3sg-cause-give-pl
‘The man makes the woman give the books to the girl.’
The shared argument (ś’āla) is in the Nominative case in (4) and (6), but, rather
unexpectedly, in the Ergative case in (5). How is the case of the subject of x oayǝn
‘want’ determined? This verb is, in RRG terms, M-intransitive, taking a single
macrorole argument. This means that its subject (the only macrorole) should be
in the Nominative; this is indeed the case, when its complement is a NP in the
Nominative:
The examples (4), (5), and (6) show that the “subject” of the verb in the depen-
dent (linked) core determines the case of the subject of the main (matrix) verb.
Thus, if the dependent verb is M-transitive, its Undergoer is in the Nominative,
and its Actor in the Ergative; that actor is co-referent with the single macrorole
argument of the matrix verb, which is also in the Ergative (5). If the dependent
verb is M-intransitive, its single macrorole (Undergoer) will be in the Nominative,
and, by virtue of co-reference, the single macrorole of the matrix verb will also be
in the Nominative (4) and (6).
Let us address the first question first. What we need to know is whether there
are languages in which the argument structure of the linked verb in a control con-
struction determines the case assignment of the argument it shares with the ma-
trix verb. We shall call this pattern of case assignment “Dependent first (DF)” to
distinguish it from “Head first (HF)”, the more usual pattern in which the matrix
verb assigns the case marking to the shared argument. Indeed, we shall see that
such constructions do exist elsewhere, but they are constrained by two factors: the
type of clause alignment, and the type of the verb in question.
Our first generalization is that DF-pattern of case assignment with actor con-
trol verbs, as well as with verbs having a single macrorole argument, is found only
in ergative languages. An example of this pattern comes from Enga.
(8) Enga (Papua NG, from Van Valin & LaPolla 1997: 580)
a. Baa mená méndé nya-la pe-ly-á-mo
3.sg pig det get-inf go-pres-3sg-dec
‘He is going to get the pig.’
In (8a) the single macrorole argument of the matrix clause baa ‘he’ is unmarked,
as the subject of the intransitive verb pe ‘go’. In (8b) baa receives the Ergative case,
although the matrix verb is intransitive, because the co-referent unexpressed actor
of the linked verb would be in the Ergative.
Of course, the question of which core (in a core juncture) is responsible for assigning
case to the shared argument only makes sense in languages in which the core (rather than
the clause) is the domain of case assignment (Van Valin 2005: 257-260).
A similar DF pattern of case assignment is found in purposive constructions in Newari,
also an ergative language (see Van Valin & LaPolla 1997: 580), as well as in some construc-
tions with modal verbs in Ingush ( Johanna Nichols, p. c.), and in Adyghe, Kabardian’s clos-
est relative. Both Ingush and Adyghe are ergative languages.
ranko matasović 309
The DF pattern of case assignment with actor control verbs is logically impos-
sible in accusative languages: in those languages, actors are always assigned the
same case (usually called “Nominative”). Since the actor of the matrix verb has
to be the “subject” of the linked verb, and the case of the subject is Nominative,
both orders of case assignment would yield the same result: the controller and the
controlee will both get the Nominative. Even if the case-assignment procedure
starts with the controlee, the accusative-type case-assignment rules will assign it
the case of the Actor, which is the Nominative.
However, the DF pattern is logically possible in accusative languages with verbs
that have undergoer control. With those verbs, the undergoer of the matrix verb
can also be the actor of the linked verb, and the question arises how its case should
be determined. If the DF pattern is applied, the argument structure of the linked
verb determines the case of the shared argument. This is attested in Korean (cp.
Kumaxov & Vamling 1998: 291):
It is not the case, however, that only accusative languages have the DF pattern
with undergoer control verbs, because ergative languages can also have it. This is
again found in Kabardian:
An apparent exception involves verbs which take “Dative subjects”, such as Latin pla-
cere or German gefallen ‘like’. In Latin Mihi placet cantare ‘I like to sing’, the shared argu-
ment is a dative pronoun (mihi) but its case is assigned by the matrix verb placere. This
seems to be the rule in all languages that have “Dative subjects”, although it is conceivable
that there might be languages in which the case of the shared argument in core junctures
with verbs that ordinarily take “Dative subjects” is dependent on the case that argument
would take in the linked core, i.e. if the “subject” of linked core can also be in a case other
than the Nominative.
This sentence would also be possible with the Undergoer of the main verb yenghi, in
the accusative, i. e. with the default “head-first” strategy of case assignment, cp. Polinsky and
Potsdam (2006: 5), also Kwon & Polinsky (2006) and Monahan (2003).
With a different word order, in this construction it would have been possible to put l’ǝ
‘man’ in the Nominative in this sentence (with the default, “head-first” strategy of case as-
signment, see below).
310 typology of control constructions within the framework of rrg
In (10), the matrix verb ġada?oan ‘persuade’ is transitive, and its subject fəz
‘woman’ should take —accordingly— Ergative case, while its direct object l’ə ‘man’
should be in the Nominative. The direct object is marked, however, as Ergative
because it is co-referent with the transitive subject of the linked verb ś’ən ‘make’.
Kabardian also has the DF pattern with actor control verbs (Kumaxov & Vamling
1998: 210-212), ġagoəġan ‘promise’:
In (11), the matrix verb ġag oəġən ‘promise’ is transitive, and its subject l’ə ‘man’
should be in the Ergative case. Again, it is Nominative because the dependent verb
q’ak’ oan ‘come’ is intransitive, and its single macrorole argument (in traditional
terms, its “subject”) is co-referent with the actor of the matrix verb. In (12), where
the dependent verb is transitive, the shared argument appears in the Ergative case:
I conclude, then, that the order of case assignment in matrix and linked clauses
is a typological parameter independent of clause alignment and the type of control
verb. This brings us to the following typology of case assignment in control con-
structions in Table 1. The symbol (+) means that the DF pattern is attested, while
(-) means that it is impossible.
Verb type
Language type Actor control
Undergoer control
(and intransitive matrix verbs)
Ergative + +
Accusative - +
Table 1. Typology of case assignment in control constructions
ranko matasović 311
Although the Korean example in (9) shows that the DF pattern is possible in
accusative languages, it is important to note a principled asimmetry between ac-
cusative and ergative languages with respect to this pattern: namely, in accusative
languages the transitivity of the linked verb cannot play any role in the case as-
signment to the shared argument, which is always assigned the Nominative case
by the linked verb. In ergative languages, on the other hand, the linked verb can
assign the Ergative case to the shared argument, if it is transitive, or the Absolutive
case, if it is intransitive. Another thing to note is that the DF strategy is usually an
alternative strategy to the more widespread HF strategy. Besides the DF pattern
seen above, the HF patter is also possible in Korean:
It appears that the two possibilities of the case marking of the shared argu-
ment in (9) and (13) arise because the structural position of the shared argument
Yenghi, is ambiguous: the HF pattern is used if Yenghi serves as an argument in
the matrix core, while the DF pattern is used if it is in the linked core. In other
words, the parsing for (9) would be Chelswu-ka [Yenghi-ka hakkyo-lul ttena-tolok]
seltukhayssta, while the parsing for (13) is [Chelswu-ka Yenghi-lul [ hakkyo-lul ttena-
tolok]seltukhayssta].
Similarly, according to Van Valin and LaPolla (1997: 580), the DF pattern ob-
served in Enga is only the preferred strategy, but the HF pattern is possible as well,
i.e. the sentence (8) would have been also possible with an absolutive shared argu-
ment. The same is true in Kabardian, but there the choice of the case-assignment
strategy crucially depends on the word order. Thus, (14a), (14b), and (14d) are
OK, but (14c) is ungrammatical:
This means that l’ə ‘man’ is in the Nominative only when it is in the same
clause as fəz ‘woman’, i. e. that the structure of (14a) is: [l’ə-m wəna-r yə-ś’ə-nwə]
[fəzə-m X q’-yə-ġada? o-ā-ś] where X is the unexpressed actor of the linked core.
The structure of (14b) is, on the other hand, [fəzə-m q’-yə-ġada? o-ā-ś l’ə-r] [X
wəna-r yə-ś’ə-nwə]. This could lead us to believe that there are two syntactic slots
in the sentence, and that the case assignment takes place before the unexpressed
argument (X) is deleted. Such an approach is advocated by the generative gram-
marians, especially Polinsky’s work on “backward control” within the Minimalist
framework (Polinsky & Potsdam 2002, 2006). In fact, under Polinsky’s analysis
both the controller and the controlee get the appropriate case marking, and are
present at some syntactic level, but then one of the co-referent arguments is de-
leted. The sentences (14a) and (14b) would therefore have the following represen-
tation:
Similarly, the pattern of case assignment depends on the position of the argu-
ment in constructions with the actor control verb ‘promise’ in (15a). However, if
the co-referent argument is in the matrix core, then it has to be Ergative, because
it is the subject of the transitive matrix verb, as shown in (15b). Finally, if the posi-
tion of the co-referent argument is ambiguous (i.e., if it can be both in the matrix
and in the linked core), then both case markings are possible, such as in Korean.
Namely, the construcion in (15c) can be analyzed in two ways, both as l’ə-m [q’a-
k’ oa-nwə sə-q’-yə-ġag oəġ-ā-ś] and as [l’ə-r q’a-k’ oa-nwə] sə-q’-yə-ġag oəġ-ā-ś.
These examples show that, with this actor control verb, the DF pattern of case
assignment is limited to instances where the co-referent argument is in the linked
core, and not in the matrix core.
We saw that the choice of the case assignment pattern in Kabardian largely de-
pends on word order, which determines the structural position of the shared argu-
ment in a control structure. There is some evidence that word order in Kabardian
is, in turn, largely determined by information structure of the sentence, since fo-
cal NPs are normally preverbal (see Kumaxov and Vamling 2006: 111-120). This
means, then, that the choice of the case assignment pattern probably also depends
on the information structure of the sentence, but this hypothesis remains to be
tested in a forthcoming investigation. We must note, however, that in other lan-
guages, the case assignment pattern does not appear to depend on word order. For
example, in Tsez, a Nakh-Dagestanian language, both the HF and the DF pat-
terns of case assignment are possible with the verb -oqa ‘begin’ without any change
in the word order, or constituent structure (examples from Potsdam & Polinsky
2002: 248-9):
314 typology of control constructions within the framework of rrg
The two constructions are grammatical and are synonymous. The only differ-
ence between the two lies in the case marking on the argument kid ‘girl’ shared by
the matrix and the linked verb. In (16a), it is in the Ergative, because the linked
verb is transitive, and its subject must be in the Ergative case. On the other hand,
in (16b) it is in the Absolutive, because the matrix verb (-oqa ‘begin’) is intransi-
tive, and its subject must be in the Absolutive case.
3. Typological implications
One question, however, remains unanswered: why is there such a clear cross-lin-
guistic preference for the Head-first pattern over the Dependent-first pattern of
case-assignment in control structures? As Kabardian data show, and Korean ex-
amples confirm, even languages having the DF strategy also make use of the HF
pattern as an alternative possibility, whereas the reverse does not seem to occur in
human languages, i.e., there are no languages in which the DF pattern is either
the sole possibility, or the preferred alternative in all constructions. A priori, there
does not seem to be any reason why verbs in the linked core should be incapable of
assigning case to the arguments they share with the verbs in the matrix core, nor
why constructions with the more common DF pattern of case assignment should
be easier for processing.
There are, in my opinion, two general principles that could be invoked to ac-
count for the asymmetry of the two logically possible case marking patterns. The
first principle involves typological universals of word order. In control construc-
tions, either the argument of the matrix core, or the argument of the linked core
can be left unexpressed under co-reference. Patterns of equi-NP deletion in other
constructions teach us that it is regularly the second of the two co-referent argu-
ments which is left unexpressed, e. g. in sentences like John hit Mary and ran
ranko matasović 315
away the co-referent subject of the second clause is unexpressed. Since linked
cores in control constructions usually appear in the syntactic position of objects
of matrix verbs, they will naturally occur in those positions of the sentence that
are reserved for objects, which means that they will be placed after the subject
of the matrix verb, except in the cross-linguistically very rare language types in
which the object regularly precedes the subject (the OSV, OVS, and VOS types).
In sentences like John wants to hit Mary the linked core will occur before the
matrix core only in languages in which *John to hit Mary wants, or *John Mary
to hit wants are regular or unmarked word orders. It is clear that such languages
are significantly rarer than those in which the matrix core precedes the linked
core. This means that at least the DF pattern of case assignment in actor (and
single argument) control verbs will be disfavored cross-linguistically, and we have
already seen that the DF pattern of case assignment in actor control verbs is logi
cally impossible in accusative languages (cf. Table 1). In ergative languages it
may occur only if word order is relatively free (as seen in Kabardian), or if linked
clauses regularly precede matrix clauses. That is, three logically independent con-
ditions must be fulfilled for the DF pattern with actor and single argument
control verbs to arise:
It is clear, then, that languages in which all three conditions are fulfilled are
bound to be rare. Another reason why the DF pattern is typologically rare might
have to do with the diachrony of clause linkage. I believe that patterns of gram-
maticalization of infinite verbal forms explain why, by default, they cannot assign
case in linked cores. Infinitives in more familiar languages —including all Indo-
European languages known to me— develop from nominal forms, i.e., from verbal
For the development of infinitives in Indo-European see Disterheft (1980). The thesis
that infinitive markers usually develop from case-markers or adpositions governing case is
independently confirmed by Heine & Kuteva (2002: 247-248).
316 typology of control constructions within the framework of rrg
nouns, which do not assign case, but rather receive it from matrix verbs. When
they become infinitives, i.e., when they lose their nominal features such as case, they
still remain in syntactic constructions in which only the matrix verb assigns case
to arguments it shares with the linked verb. This can be clarified if we look at the
Latin example (18).
4. Conclusion
To sum up, we have argued that a full typology of control constructions should
include the distinction between Head-First and Dependent-First patterns of
case assignment to the shared argument. This is relevant to RRG inasmuch as
the specification of the case assignment pattern in control constructions should
be added to the other rules in the linking algorithm in core junctures (Van Valin
2005: 239-259). In Kabardian, the choice of the pattern cannot be fully predicted
ranko matasović 317
from the structural position of the shared argument, although it is partly depen-
dent on it.
We have shown that the DF pattern is attested with all three types of control
verbs (intransitives, actor and undergoer control verbs) and in languages with both
accusative and ergative clause alignment, but that there are limitations on the oc-
currence of the Dependent-First pattern in the accusative languages. We have also
tried to provide an explanation of the fact that the Head-First pattern is much
more common cross-linguistically, but more research is needed in order to sub-
stantiate the preliminary hypotheses suggested in this paper.
References
1. Introduction
For the study of complementation, most theories agree on the fact that the degree
of syntactic integration is established straightforwardly based on specific struc-
tural properties, while trying to determine the degree of semantic cohesion among
the main predicate and its complement is somewhat more complicated. Take for
example the set of predicates expressing the cognitive notion of thinking. These
predicates share a number of semantic and conceptual features that partly condi-
tion the morpho-syntactic properties of the construction in which they can ap-
pear; still, several members of this class may combine with more than one type of
complement, making it difficult to predict the overall meaning of the construction.
This paper provides evidence for a revised hierarchy based on the participant’s
mental disposition, the epistemic values of the predicate as well as the speaker’s
degree of subjectivity, which helps motivate the semantic side of the Interclausal
Relation Hierarchy as proposed by Role and Reference Grammar. The analysis
focuses on the semantic notion of thinking as realized in a number of Southern
Uto-Aztecan languages.
The information is organized as follows. Section 2 briefly reviews some pro-
posals dealing with the form and function of complementation, and draws the at-
tention to the semantic dimension. Section 3 lays out the basic principles of clause
union in Role and Reference Grammar. Section 4 analyses the morpho-syntactic
and semantic properties of complement clauses involving mental predicates in a
number of Uto-Aztecan languages. Section 5 provides arguments in favor of the
revised participant’s mental disposition scale which, together with the temporal
scale, and the necessarily shared participant hierarchy, serve to better explain the
semantic extensions covered by a basic mental predicate and their correlation with
its syntactic manifestation. Section 6 concludes this study.
319
320 on the semantic dimension of complementation
Dixon (2006: 27-33) overtly distinguishes between main predicates –those that func-
tion as a main verb in simple and complement clauses- from “secondary concepts” –those
predicates that usually appear with a verbal complement only (explicitly or not). The latter
may receive a special treatment in particular languages, i.e. they may be realized as an unin-
flected particle, a verbal affix or a lexical verb.
lilián guerrero 321
such as the reference of the subject, the time reference, and the aspect or mood
values of the dependent unit (Cristofaro 2003: 111).
Modality verbs:
want, begin, finish, try, etc
Manipulation verbs:
make, tell, order, ask, etc
Perception-Cognition-Utterance
(PCU) verbs:
see, know, think, say
Weakest bond…….…. Strongest bond
of both verbs as one complex predicate (Givón 1980: 371). Secondly, the semantic
scale helps to determine which predicates may allow the omission of certain in-
formation within the dependent unit, i.e. “Syntagmatic economy” (Haiman 1985),
omission of obligatory shared arguments (Haspelmath 2003); the “Principle of
information recoverability” (Cristofaro 2003). Since verbs like try, want, begin pre-
determines the identity of the subject and certain TAM values in the complement,
it is more likely such information is omitted. And, finally, semantic scales also al-
low a natural explanation for the fact that languages are more likely to undergo
diachronic changes that lead to iconic and/or economic patterns rather than vice
versa (Haiman & Thompson 1984; Ohori 1992; Givón 2006).
The establishment of more elaborated semantic scales together with the evalu-
ation of different degrees of syntactic integration has resulted in a better under-
stating of the relation between the main predicate and its complements. Com-
plications arise, however, when trying to formalize the linking between certain
semantic functions and specific syntactic coding devices of the complement con-
struction. The third approach explicitly proposes a theory of clause linkage consid-
ering both the semantic and syntactic dimension of complementation.
Role and Reference Grammar (RRG; Foley & Van Valin 1984; Van Valin & La-
Polla 1997 (henceforth VV&LP); Van Valin 2005) proposes three main compo-
nents for the study of clause union: the theory of juncture, the theory of nexus,
and the theory of the interclausal semantic relation. The theory of juncture deals
with the units which make up complex sentences: nucleus, core and clause. In a
nuclear juncture, there is a single core containing two nuclei taking a single set of
core arguments. In a core juncture, there is a single clause containing more than
one core, each with its own set of arguments. In a clausal juncture, whole clauses
are joined and each clause may be fully independent of the others. The theory of
nexus concerns the syntactic relationship between the units in the juncture, subor-
dination, coordination, and co-subordination, each type distinguished on the basis
of structural or operator dependencies as illustrated in Figure 1. In subordination,
the linked unit functions either as an argument (as in complementation), or as a
modifier. In coordination, the two units are ‘added together in a sequence’ in a rela-
tionship of equivalence and independence at the level of the juncture. Cosubordi-
324 on the semantic dimension of complementation
nation shows properties of both: there are two equivalent units joined together (as
in coordination), but one unit depends on the other (as in subordination), in terms
of operators. What distinguishes cosubordination from coordination is operator
sharing: in the former, the linked verb must be dependent upon the matrix verb for
expression of one or more operators at the level of juncture, whereas in the latter,
the two verbs can, but do not need to be independently specified for the relevant
operators.
These syntactic combinations are organized into the Syntactic Relations Hierar-
chy in Figure 2, ranked in terms of their syntactic tightness. The linkage types at
the bottom are combinations of whole clauses constituting sentences, e.g. clausal
coordination. As one goes up the hierarchy, the linked units lose more and more
features of an independent clause until they are reduced to a bare nucleus or pred-
icate, e.g. nuclear co-subordination. It is important to keep in mind that these
juncture-nexus types are abstract linkage relations, not grammatical constructions
types. This means that each of these linkage types may be realized by more than
one grammatical construction in a particular language, and vice versa, the same
grammatical construction may involve different linkage types.
The syntactic combinations express a wide variety of semantic relations be-
tween the units in the juncture. For the semantic side, RRG adopts the work of
Silverstein (1976) and Givón (1980) in relation to the degree of semantic integra-
tion and iconicity within a large set of sematic notions. RRG juxtaposes the syn-
tactic and the semantic hierarchies to create the Interclausal Relations Hierarchy
in Figure 2.
lilián guerrero 325
Strongest Closest
Causatives[1]
Nuclear cosubordination Phase
Manner
Nuclear subordination Motion position
Means
Nuclear coordination Psych-action
Purposive
Core cosubordination Jussive
Causative [2]
Core subordination Direct perception
Indirect perception
Core coordination Propositional attitude
Cognition
Clause cosubordination Indirect discourse
Direct discourse
Clause subordination Circumstances
Reason
Clause coordination Conditionals
Concessive
Sentential subordination Simultaneous actions
Sequential actions
Sentential coordination Situation-Situation: unspecified
Weakest Loosest
Syntactic relations Semantic relations
Because there are fewer juncture-nexus types than distinct semantic relations, a
language invariably has some syntactic linkages expressing more than one seman-
tic relation. It is also the case that a given semantic relation can be conveyed by
more than one juncture-nexus type, and also a single mental verb may be capable
of taking more than one juncture-nexus combination (see Van Valin & Wilkins’s
1993 analysis for the English verb remember). Since there is no one-to-one relation
326 on the semantic dimension of complementation
between the type of semantic relation instantiated by the predicate and the link-
age chosen to encode it, RRG explicitly states a general principle reflecting some
strong universal constrains on which semantic types can be realized by which syn-
tactic linkage: “the tightest syntactic linkage realizing a particular semantic rela-
tion should be higher than or as high on the Interclausal Relations Hierarchy as
the tightest syntactic linkage realizing semantic relations lower on the Interclausal
Relations Hierarchy (VV&LP: 483).”
Regarding the semantic side of complementation, the RRG theory proposes
a more elaborated Semantic Relation Hierarchy (1) which seeks to reflect the
interaction of a set of semantic sub-hierarchies including, but not limited to,
those in (2).
b. Causal hierarchy:
physical > verbal > underspecified [non-defeasible] > inferred [defeasible]
Roughly speaking, direct causation and phase predicates involve a high degree
of semantic bond, since they attest the highest values on each of the semantic
sub-hierarchies in (2). The general principle would predict that these semantic
relations at the top end should be realized by the strongest syntactic combination
(e.g. nuclear/core junctures & cosubordination/subordination nexus). At the other
end, perception, cognition and discourse predicates which are at the bottom of the
lilián guerrero 327
scale, should be realized by the loosest syntactic linkage (i.e. core/clausal junctures
& subordinate/coordinate nexus).
Focusing on mental verbs, psych-action predicates can be defined as a mental
disposition regarding a possible action on the part of a participant in the state
of affairs (e.g. Nicole wants to play in the park). This type of semantic notion is
characterized by selecting the higher values on the necessarily shared participant
[NSP] and the participant’s mental disposition [PMD] sub-hierarchies (3a). In
turn, propositional attitude predicates lexicalize the participant’s attitude, judg-
ment or opinion regarding a state of affairs (e.g. I think Nicole is in the park). These
predicates have mixed values on the scales (3b). Finally, cognition predicates, as
the expression of knowledge or mental activity (e.g. I realized that Nicole has gone
to the park) have the lowest values on the sub-hierarchies (3c). The general princi-
ple also predicts about the types of syntactic-semantic relations which should not
be attested in any language: if psych-action can be realized by core subordination,
for instance, it should not be possible for, say, propositional attitude predicates to
be realized by some type of nuclear juncture.
Attitude and cognition verbs are only distinguished by the PMD scale; psych
action verbs are distinguished from the two because of the necessarily shared (ac-
tor) participant value. Actually, as is, the PDM sub-hierarchy reflects the assump-
tion that assertion verbs and those that reports assertions are alike across languages
(Traugott & Dasher 2002; Dahl 1997; Noonan 2007). But languages usually have
syntactic (or pragmatic) strategies distinguishing the private vs. public cognitive
domain involved in mental and speech act verbs (Shinzato 2004: 861).
For a detailed discussion on a different approach of the semantics of complex clauses
328 on the semantic dimension of complementation
In order to better account for the relations between the mental predicate and
its complement, we need a semantic scale based on both the inherent epistemic
properties of the predicate and the notion of subjectivity or speaker’s relatedness.
The revised version of the participant’s mental disposition hierarchy in (4) was
originally presented in Guerrero (2004, 2006), and part of the motivation that
lie behind it is the idea that any mental state/activity predicates presupposes
that the participant is, at least at some degree, conscious or aware at the time
she engages in this mental experience (Fortescue 2001: 17). The scale captures
properties previously established for mental predicates, such as whether the verb
(i) presupposes/implies the truth of the complement, (ii) encodes weak or strong
assertion (Karttunen 1971; Hopper 1975; Guitart 1978; Ransom 1986, among
others) and, crucially, (iii) reflects the speaker’s influence, commitment, closer/
remote attitude toward the content of the complement (Givón 1982; Langacker
1990, Wright 1995).
Guerrero (2008) provides evidence for the first two values based on desider-
ative (psych-action) expressions. Now, the analysis focuses on the next three posi-
tions evoking experiences generated in the speaker’s mind. In what follows, the
basic predicates of “thinking” in Southern Uto-Aztecan languages are analyzed. In
these languages, one basic word covers more than one central domain on the epis-
temic/cognitive scale in (4).
The discussion focuses mainly on the semantic side of the predicate-comple-
ment combinations, but comments on the syntactic linkages are also included. In
the analysis, syntactic tightness is based on the use of certain clause linkage mark-
ers, operator dependency, the syntactic status of the complement, and coreferen-
in RRG, see Ohori (1992, 2001, 2005). The author establishes the feature anchoring (the
process of locating a predicate in a mental model having its own epistemic status, e.g. tense
and mood), meaning that propositional complements are defined as p1 comp p2, where p1
is fully anchored while p2’s anchoring is relative to p1 (2005: 10). Since Anchoring en-
codes “[features] predetermined by the main predicate” (2005: 11), the distinction among
psych-action, attitude and cognitive predicates is, again, lexically determined.
lilián guerrero 329
tiality vs. argument sharing (for a better comprehension of the syntactic side see
Guerrero 2004). The exposition begins with the Tepiman sub-branch, and follows
with the Yaqui language (Taracahita sub-branch).
Several Southern Uto-Aztecan languages make use of some cognates of the proto-
Uto-Aztecan *’i(r)a ‘feel, think, whish’ (Miller & Hill 2003) to express the seman-
tic notion of thinking. In most of the languages, this predicate may combine with
more than one complement type. In Eudeve (extinct), for instance, the cognate era
takes two types of complements, expressing two related but not identical senses.
The construction in (5a) expresses desire or personal intention; the two cores ob-
ligatorily share a semantic argument (the actor ne’e ‘I’), and the dependent verb
is marked by the desiderative suffix –ba. The construction in (5b) expresses the
cognitive action of thinking; here, each core expresses its own actor, the dependent
subject is non-nominative, the dependent verb takes a tense suffix and the linked
unit, as a whole, is marked by the accusative suffix –ta. For simplification, in the
examples I use “think” as a general gloss only (i.e. not a semantic primitive). The
complement is between square brakets.
The Uto-Aztecan family is one of the largest and most widely distributed in North
America. The family is grouped into a Northern and a Southern branch, the latter consist-
ing of four subgroups: Aztecan (classical and modern Nahuatl), Pimic (Tohono O’odham,
Southern and Northern Tepehuan, Tepecano and Pima), Corachol (Cora and Huichol)
and Taracahita (Tarahumara, Guarijío, Yaqui and Mayo). Because of the space, this study
focuses on Tepiman and Taracahita languages.
Abbreviations used in this paper. acc: accusative, com: comitative, compl: completive,
clm: clause linkage marker, dir: directional, gen: genitive, impf: imperfective, irr: irrealis,
loc: locative, neg: negation, nom: nominative, nmz: nominalizer, ns: non-subject, pass:
passive, pastc: past continuative, pfv: perfective, pl: plural, pot: potential, pres: present,
presu: presumable, refl: reflexive, S: subject, sg: singular.
330 on the semantic dimension of complementation
The Tepiman languages share the cognate ilid ~ ilídyi (*’i(r)a plus the applicative
-d yi ’). This lexical predicate also appears in an array of constructions expressing
mental activities: the participant’s intention and wishes (6a), the expression of a
state of affairs as a thought (6b), and some sort of mental process or evaluation of
an external situation (6c).
The situations expressed in (6a-b) convey a wish on the part of the main par-
ticipant, that the event encoded in the complement be realized; in the two, the de-
pendent event would be brought about by the experiencer herself. The meaning of
volition in (6a) is only possible when the main and dependent subjects are corefer-
ential, and so the dependent subject is omitted; the two cores are adjacent, and the
dependent verb must be unmarked for TAM operators, all factors which indicate a
core cosubordination linkage. Although the actor NP is also shared in (6b), the de-
pendent verb is marked by a TAM operator; hence core coordination. The clause
subordination in (6c) expresses an external situation in which the participant con-
sciously thinks about a state of affairs in which she is not involved. Here, the con-
tent serves as a syntactic and semantic argument of the complement-taking verb.
The examples below are from Pima Bajo. Again, in (7a-b), the person who has
a mental disposition towards some action is involved in doing it. The examples in
(7c-e) express a more elaborated mental experience (remembering, recalling, and
thinking). In Pima, the experiencer is always expressed twice in the main predicate;
thus, the literal meaning of (7a) would be something like ‘I think on my staying.’
All the examples are from Estrada (1988: 87-89).
In Yaqui, there are, at least, five grammatical complement types, and the predi-
cate ’ea is the only one which seems to combine with all of them. While clause
coordination and subordination correspond closely to the most traditional no-
tions, the two differ significantly when applied to nuclear and core junctures,
since the syntactic structures may not be recognized as coordinate or subordinate
lilián guerrero 333
in most approaches. Let’s begin with the tightest syntactic linkage. In (8) the
mental predicate and the dependent verb are morphologically linked together.
Yaqui data come from oral texts.
d. enchi ne yoi-t-’ea-n
2.sg.acc 1.sg.nom yori-clm-think-pasc
‘I thought you were a yori (white man, foreigner).’ (Hilario: 249)
Syntactically, no referential restrictions hold between the linked unit and the
main predicate which is immediately adjacent to the non-matrix event. When the
subjects are coreferential, the dependent subject is omitted or a reflexive pronoun
appears (8a-b), otherwise the subject must be accusative (8c-d), as it is the case for
most dependent (non-main) subjects in Yaqui. There is also no restriction on the
relevant TAM operators on the dependent verb (8d); the occurrence of the CLM
-t between the two cores is optional. Although a semantic argument of the matrix
predicate, the complement and the main predicate within this syntactic combina-
tion do not figure a subordinate nexus relationship, among other reasons, because
the main verb cannot be passivized, e.g. *hearing me speaking Yaqui was enjoyed
(8c). Hence core coordination.
Most regularly, core junctures makes use of two less common CLMs -bae-kai
(9a) as well as -benasi (9b-c), where the dependent unit appears embedded before
the main predicate, i.e., in the preferred position for direct core arguments. In (9a)
there is an obligatorily argument sharing and there is also obligatory sharing of
the core operators, these factors yield a core cosubordination linkage. Other verbs
334 on the semantic dimension of complementation
taking this syntactic linkage are those meaning ‘hope’, ‘dream’ and ‘imagine’, all
encoding some sort of self-intentional, future-oriented activity, involving a strong
commitment on the part of the participant about the realization of the event. In
(9b-c) the subjects must be different, the dependent verb can be independently
marked by the relevant operators and the clausal complement as a whole serves as
a core argument of the matrix predicate, resulting in a core subordination struc-
ture. In fact, it is possible to encode the source of such mental activities as a core
argument in (9c) e.g. ‘I think about you, because I have seen you, that you enjoy to
be here’. That is, the tighter construction involves a self-oriented interpretation,
whereas the less tight linkage expresses the speaker’s attitude about a proposition
involving another party.
a. ne
(10) ��� junuen’ea-Ø [enchi kari-ta tute-ne-’u]
�� 1.sg:nom thus.think-pres 2.sg:acc house-acc clean-pot-clm
‘I wish (agree) that you would clean the house.’
lilián guerrero 335
Although most of the senses expressed by ’ea are related in some way, it is not
always easy to establish the nature of such relationship. By exploring the contexts
in which these constructions appear in texts, it is possible to motivate certain dis-
tinctions based on the speaker’s attitude. For instance, in (8b) the speaker was sick
and tired of looking for a cure, so he went to visit the main healer to ask for help;
although he is making a request, he attempts to soften the force of it in order to
receive a positive answer. In (8d), after a series of incidents between the speaker
(narrator) and his interlocutor, the later tries to apologize for his harsh attitude ar-
guing that he just realized he is not a yori. In both cases, the construction expresses
the interlocutor’s attitude towards the truth of the propositional content. In con-
trast, in (9b-c) ’ea expresses a mental experience as a reaction from an external
source, i.e. thinking as the result of a corporal sensation, having seen some facts, or
having made some inferences from external circumstances. As part of two derived
epistemic predicates, ’ea not only encodes the speaker’s positive opinion about the
content of the complement —as in junuen-’ea ‘think like this, wish this, agree to’ in
(10a)—, but it also can mean that such information is (or is not) in the speaker’s
mind at this time, without any degree of commitment, judgment or evaluation –as
in ju’unea ‘know about’ (lit. ‘think-this’) in (10b).
The semantic meanings of ’ea may be also expanded by way of phrasal col-
locations to highlight the participant’s awareness. Differently from the previous
examples, the propositional content in (11) is expressed as a report, meaning the
linked unit shares most of the features of an independent clause. The position of
the complement-taking predicate in relation to the dependent unit has changed:
instead of clause finally, the main predicate appears clause initially; the dependent
subject is nominative rather than accusative, and there is no CLM. All these fac-
tors together yield clausal coordination. In relation to its semantics, the presence
of the particle jumak ‘maybe, possibly’ seems to reduce the degree of the speaker’s
involvement by expressing the current thought as an opinion, cautious comment.
Interestingly enough, all data I have using this syntactic structure involve first per-
son singular present tense, suggesting the mental verb is acting as an epistemic
formulaic marker (Thompson 2002).
336 on the semantic dimension of complementation
In fact, Yaqui provides good evidence for the cognitive shift from internal
mental experiences to external speech (cf. Traugott & Dasher 1987; Dahl 1997;
Shinzato 2004; Givón 2001; Cristofaro 2003), since ’ea has literally adopted the
structure of indirect quotations used by verbs like jiia ‘say’ (12a) to express some
sort of ‘verbalized thoughts’. In (12b), the dependent subject is nominative, the
propositional complement comes first, followed by the citative CLM –ti.
In (12b), the mental predicate does not express worry, concern, judgment, or
a cautious opinion regarding the content of the proposition, but it purely reports
the thought as an assertion. Actually, most occurrences of ’ea involve first person
singular in simple present tense, i.e. “egocentric” clauses (Dahl 1997). This prefer-
ence and the use of the indirect discourse structure suggest the predicate here may
function as a subjectivity marker (Scheibman 2002:167).
Therefore, in contrast to the Tepiman use of the mental predicate in favor of
volition and internal experiences on the part of the participant, the Yaqui cognate
favors the expression of the participant’s attitude, commitment, reasoning or opin-
ion in relation to a state of affairs or a proposition. In any case, the cognates evenly
encode different values within a cognitive /epistemic domain.
5. Discussion
As a way of summary, Table 3 (page 339) reports the juncture-nexus types found
in these Uto-Aztecan languages, as well as the interaction of the relevant semantic
lilián guerrero 337
sub-hierarchies. On the syntactic side, column 1 indicates the linkage type; col-
umn 2 reflects whether the dependent subject must be left implicit or be marked
accusative or nominative; column 3 indicates the operators dependency, whether
the dependent verb must be unmarked, marked by a special form (potential, ir-
realis, subjunctive) or fully marked; column 4 indicates the occurrence of a clause
linkage marker. The more arguments and operators are shared between the two
units, and the more restricted the use of CLMs, the tighter the predicate-comple-
ment construction will be.
On the semantic side, column 5 specifies the temporal relationship among the
events. Simultaneous actions must be actual (i.e. ongoing events), but at least the
second action may be actual or potential for all other temporal relations; unspeci-
fied temporal ordering of the events represents the loosest semantic cohesion be-
tween the units. Column 6 indicates whether or not there is a necessarily shared
participant [NSP]; the ‘no’ value does not mean they cannot be coreferential. And
the last column conveys the revised version of the participant’s mental disposition
scale: intention > internal/direct experience > mental experience: commitment >
mental experience: reasoning > non-mental experience: report.
The closest semantic notions are realized by the tightest syntactic linkages.
When the mental predicate encodes intention and internal experiences regarding
a state of affairs —the highest values in the PMD—, the construction conveys the
participant’s mental disposition to act in the event brought about by herself, and
the main and dependent events are sequential (usually but not necessarily a fu-
ture-oriented event). When the speaker expresses her attitude, commitment, judg-
ment of an event involving herself, or evaluates the propositional content of the
complement, the next value in the PMD, languages tend to allow more than one
option depending on the degree of speaker’s commitment. As soon as the predi-
cate entails the experiencer’s consciousness (i.e. reasoning, evaluating, understand
something previously experienced or learn), the syntactic linkages get less tight, as
in the case of core subordination, clausal subordination. A final shift from mental
experiences to external speech is the reporting of a previous thought or assertion
adopting the structure of indirect discourse. Then, by including the revised par
ticipant’s mental disposition scale as part of the other semantic sub-hierarchies,
the Role and Reference Grammar theory of clause linkage may account for such
correlations involving mental predicates.
338 on the semantic dimension of complementation
6. Conclusion
Southern Uto-Aztecan languages provide good evidence for the general principle
governing the interaction of the semantic and syntactic relations observed in men-
tal complement constructions. The essential claim was that all these senses codify
what originates in the subject’s mind: from personal desires and less conscious
internal experiences to mental activities as the result of observation, evaluation
or judgment, to the reporting of thoughts. The closer semantic relations are em-
bodied by the tightest syntactic linkages in the particular language. Indeed, the
Uto-Aztecan family is not unique, since the semantic extensions of “thinking”
verbs and the multiple alternatives for complement types, have been attested else-
where. Similar constructions involving mental predicates have been observed in
other languages, i.e., English (Thompson & Mulac 1991; Brinton 1996); Japanese
(Shinzato 2004), Swedish (Dahl 1997), German and Dutch (Nuyts 2000), Spanish
(García-Miguel & Comesaña 2004, Vázquez Rozas 2006), as well as on the cross-
linguistic study of thinking terms by Goddard (2003).
Syntactic integration Semantic integration
Dependent
Juncture-nexus TAM CLM Temporality NSP PMD
subject
Eudeve era
core coordination (5a) ø Special ø Sequential Yes Intention
clausal subordination (5b) acc Full -ta Unspecified No Mental Experience: reasoning
N. Tepehuan ilídyi
core cosubordination (6a) ø Unmarked ø Sequential Yes Intention
core coordination (6b) ø Special iš- Sequential No Internal experience
O’odham ilid
clause subordination (6c) nom Special m- Unspecified No Mental Experience: reasoning
Pima Bajo ilid
nuclear coordination (7a) ø Special ø Simultaneous Yes Intention
core coordination (7b) ø Special ø Sequential Yes Intention
core subordination (7c) nom Special ø Sequential No Internal experience
clause coordination (7d-e) nom Full ko Unspecified No Mental experience: reasoning
Yaqui ’ea
core cosubordination (9a) ø Unmarked -baekai Simultaneous Yes Internal experience
core coordination (8a-d) acc Special ø, -t Sequential No Mental experience: commitment
core subordination (9b-c) acc Full -benasi Unspecified No Mental experience: commitment
clause subordination (10) acc Full -’u Unspecified No Mental experience: reasoning
clause coordination (11) nom Full ø Unspecified No No-mental experience: report
sentence subordination (12) nom Full -ti Unspecified No No-mental experience: report
Table 3. Some syntactic-semantic correlations involving the mental complement-taking predicate
340 on the semantic dimension of complementation
References
Traugott, E., and R. Dasher. 1987. On the historical relation between mental
and speech act verbs in English and Japanese. Papers from the Seventh Inter-
national Conference on Historical Linguistics, Giacalone Ramat, A. et al. (eds.).
Amsterdam: John Benjamins, 561–573.
Van Valin, R. D., Jr. 2005. Exploring the syntax-semantic interface. Cambridge:
Cambridge University Press.
Van Valin, R., and R. J. LaPolla. 1997. Syntax. Structure, meaning and function.
Cambridge: Cambridge University Press.
Van Valin, R. D., Jr. and D. Wilkins. 1993. Predicting Syntactic Structure
from Semantic representations: remember in English and its equivalents in
Mparntwe Arrernte. Advances in Role in Reference Grammar, R.D. Van Va-
lin Jr. (ed). Amsterdam: John Benjamins 499-534.
Vázquez R., V. 2006. Construcción gramatical y valor epistémico. El caso de su-
pongo. Actas del XXXV Simposio Internacional de la Sociedad Española de
Lingüística. M. Villayandre (ed). León: Universidad de León.
Wright, S. 1995. Subjectivity and experiential syntax. Subjectivity and subjec
tivization: linguistic perspectives. D. Stein & S. Wright (eds.). Cambridge:
CUP, 151-172.
Studies on Spanish
Spanish atelic activity sentences
complemented by a bare noun phrase.
Two macroroles or just one?
Sergio Bogard
El Colegio de México
1. Introduction
In Spanish, as in many other languages, there is a verbal class with two alternative
argument structures, one transitive and another intransitive, and there is no me-
diation, in neither case, of any formal mark in the verb to establish the difference
between their respective sentences, except the presence or not of the direct object
noun phrase. I am talking about the ambitransitive verbs, also named “labile” by
Dixon (1994: 18), such as estudiar ‘study’, cocinar ‘cook’, correr ‘run’, nadar ‘swim’,
etc., that can be the lexical nuclei of transitive sentences like those in (1), or of in-
transitive sentences like those in (2).
347
348 spanish atelic activity sentences complemented by a bared np
d. Teresa nadó con gran energía en las competencias del año pasado.
‘Teresa swam with great energy during last year’s competitions.’
On the one hand, the examples in (1) express the meaning of an activity clearly
bounded because of the presence of a highly individuated object, whose reference
establishes the end point of that activity and interrupts its continuation (Smith
1999: 480-481). Those sentences express a telic and perfective meaning related
with accomplishments.
On the other hand, the examples in (2) express the meaning of an ongoing
activity, that is not bounded by the presence of an end point that prevents its con-
tinuation; in other words, that kind of activity refers to a situation with an arbi-
trary end point, so that it can be finished at any given moment (Smith 1999: 481).
These sentences express an imperfective aspectual meaning related with activities,
but with a special type of activities: atelic activities (Bogard 2005, and ms.). Be-
tween the cases in (1) and (2), we find sentences like
In these examples the verb expands in the first place, and the only one we are
concerned for in this paper, with a bare noun phrase, traditionally analyzed as di-
rect object. We can appreciate that this kind of noun phrase expresses a non-ref-
erential meaning, clearly contrasting with the sort of reference of the direct object
noun phrases of the examples in (1).
sergio bogard 349
of the entity depicted by a complement noun phrase, whose status of direct object
we have strongly questioned elsewhere (Bogard 2005, and ms.). Under these cir-
cumstances, the complement noun phrase does not represent a limit point for the
definite conclusion of events depicted by their respective sentences.
Consequently the transitive accomplishments and the atelic activities expanded
by a noun phrase involve two different values in the semantics of their sentenc-
es, namely, the internal temporal nature and the degree of individuation of the
complement phrase. Both are strictly associated with one property: in the case of
transitive accomplishments, they must appear clearly delimited from the discourse
context, as opposed to what happens in the case of atelic activities. This means
that transitive accomplishments are built through a direct relationship between
a bounded, perfective aspectual sense and a phrase with a referentially countable
direct object, while atelic activities are built through a direct relationship between
a non bounded, imperfective aspectual sense and a complement noun phrase of
non-countable reference —mass nouns and bare plurals. This parallelism, that has
already been observed and presented, in terms of temporal and spatial delimita-
tion respectively, by Taylor (1977: 210-11), Mourelatos (1981: 203-07) and Tenny
(1994: 24-5), among others, ultimately suggests that the reference of complement
noun phrases has a great deal of influence on the aspectual interpretation of at
least the two types of sentences involved in this analysis.
Let us now deal with the complementary bare noun phrase of atelic activity
sentences. Is this bare noun a direct object, and hence a macrorol?
If we take the argument structure of the verb as a point of reference, we can ac-
cept that the minimal possible sentence in a language must be composed by the
verb and the elements that formalize the arguments derived from its lexical sense,
which leads us to assume that these elements are not optional but required. Hence,
knowing that the direct object materializes one of the arguments of transitive
verbs, it follows that this is a required element, as we can see in sentences (4a) and
(5a), with a sense of accomplishment, in contrast with those in series (b):
On the basis of these examples, we can suggest that the noun phrases cerveza
‘beer’ and zanahorias ‘carrots’, present in examples (6a) and (7a), do not have an
argument value because of their optional nature, and therefore they are not direct
objects. This suggestion is supported by the fact that these noun phrases can not
be turned into the single argument of passive construction (8), nor can they be the
subjects of the absolute participial construction (9), in contrast with (10). Finally,
these noun phrases cannot be replaced by the accusative clitic, as we see in (11):
The fact that these noun phrases cannot be replaced by the accusative clitic, il-
lustrated in (11), is due to the fact that these clitics replace noun phrases with defi-
nite or known reference, as we can see in examples (12) and (13):
b. Rosa cultivó zanahorias antes de encontrar un mejor trabajo y siempre las vendió a buen
precio.
‘Rose planted carrots before finding a better job and always sold them at a rea-
sonable price.’
We can see in (12) that, in a context where the accusative clitic has previously
appeared, the grammatical answer, using the phrase coreferential with the clitic,
has a clearly bounded and definite reference. This contrasts with the non-refer-
ential answer, which is ungrammatical. In (13), on the other hand, the accusative
clitics have a non-referential noun phrase as antecedent (‘beer’ and ‘carrots’), but,
in their own sentence, they refer to an element of definite and known reference,
given that in their discourse context, the referents of such phrases have been previ-
ously presented.
If the consequence of the previous arguments is that the noun phrase that ex-
pands the verbal nucleus in the sentences with a sense of atelic activity is not a
direct object, then we have to admit that, even in languages as Spanish, that type
of sentence is intransitive.
We have concluded that the noun phrase that expands the verb in the sentences
of atelic activity in a language like Spanish does not form part of the argument
structure of the verb, and consequently is not a direct object but, if anything, a
“����������������������������������������������������������������������������
pseudo-object”, as named by Ramchand (1997: 115). A similar position has al-
ready been suggested by Lazard (1984 and 1998), and can be inferred from studies
within the perspective of Role and Reference Grammar, as Van Valin (1990) and
Van Valin and LaPolla (1997).
Putting aside Lazard’s position, let us consider this matter from the perspective of
an analysis presented in Van Valin (1990��������������������������������������������
)�������������������������������������������
. Van Valin states that the semantic roles
356 spanish atelic activity sentences complemented by a bared np
derived from the verbs meaning can be grouped into two macroroles, actor and
undergoer. These are considered the primary arguments of a transitive predica-
tion, and one or the other can constitute the only argument of an intransitive
verb. According to the theory, the number of macroroles depends on the logical
structure of the verb and varies from zero to two as a maximum. As macrorole as-
signment depends on the “semantic transitivity” of the predicate, the number of
syntactic arguments does not necessarily have to coincide with that of the mac-
roroles.
Regarding verbs of accomplishment, Van Valin argues that since two argu-
ments can appear in their logical structure, they can take two macroroles, actor
and undergoer. On the other hand, with respect to activity verbs, it is stated that
even though they can present more than one argument in their logical structure,
they can only have one macrorole, that of actor. (The fact that their only macro-
role is actor is shown by their compatibility with agentive complements). More-
over, given that activity verbs inherently formalize unbounded dynamic values
(which cannot codify states or conditions), and that they are incompatible with
the semantic role of patient, it follows that activity verbs cannot assign the role of
undergoer (Van Valin 1990: 225-8).
Within the perspective of Role and Reference Grammar, (semantic) transi
tivity is defined in terms of the number of macroroles that a verb takes: if it is two,
the verb is transitive; if one, intransitive; if none, intransitive. Considering this,
as well as what has already been said here, and recalling that we are only deal-
ing here with ambitransitive or labile verbs, we must come to the conclusion that
these verbs are transitive when functioning as verbs of accomplishment, and in-
transitive when functioning as atelic activities. In the latter case, this is independ-
ent of the number of arguments present in their logical structure.
It is necessary to note that, with this concept of transitivity, a verb may con-
tinue to be (semantically) intransitive even when expanded with a direct object.
This is suggested in Van Valin����������������������������
’s analysis
�������������������������
of the sentence Bill ate pasta, where it
is claimed that “pasta is direct object but not undergoer, because eat here func-
tions as an activity verb (Bill ate pasta for / *in ten minutes), and it was pointed out
above that activity verbs as a class do not take undergoers” (Van Valin: 1990: 229).
If atelic activity verbs do not take undergoers, their logical structure has only
one macrorole, and hence we are dealing with (semantically) intransitive verbs.
On the other hand, it seems necessary to redefine the concept of direct object if
we are willing to accept that it can syntactically codify an optional participant,
sergio bogard 357
and not be derived from the lexical semantics of the verb. Remember the logical
structure of activity verbs, represented as in (14).
Van Valin recalls one consistent behavioural property of direct objects, which
has been mentioned before: “if a language has a passive construction, then the
NP which would be the direct object in the active voice can appear as the subject
in the passive voice” (Van Valin 2001: 60). Without considering for the moment
the nature of direct objects of possession verbs like tener ‘have’, which cannot be
turned into passive subjects or be part of an absolute participial construction (al-
though can be replaced by accusative clitic), the bare noun phrases I have been
dealing with, analogously, cannot be turned into passive subjects, as shown in (8),
since their verbs cannot be passivized; they cannot be the single argument of abso-
lute participial constructions, as shown in the contrast between (9) and (10); and
neither can they be replaced by accusative clitic, as shown in the contrast between
(11) and (12).
Interestingly, Van Valin and LaPolla (1997: 148-150) display a somewhat dif-
ferent analysis based on the Italian verb mangiare ‘eat’, whose semantic and syn-
tactic behaviour is equivalent to the Spanish verbs at issue, including comer ‘eat’.
With respect to mangiare, they observe that this verb has two apparently transi-
tive uses for two different Aktionsarts, namely, active accomplishment and activity,
with two syntactic arguments. In the case of active accomplishment mangiare, its
second syntactic argument can be turned into a passive subject, and can also be
transformed into an absolute participial, options which are banned with the ac-
tivity reading. Their conclusion is that active accomplishment mangiare has two
syntactic arguments and therefore takes two macroroles, whereas the second ar-
gument of activity mangiare, being non-referential, only characterizes the type of
action expressed by the verb, Consequently, it is an inherent argument and cannot
be undergoer. Thus, they add, activity verbs like mangiare behave like intransitives.
I would like to argue that they don’t merely behave as intransitives, but are indeed
intransitives.
As it has been claimed, I agree with an intransitive analysis of atelic activities verbs
in Spanish, although I have to leave the discussion of telic activities for the future.
A last conclusion I would like to draw has to do with the syntactic status of the
bare noun phrase I have been talking about, and which in Van Valin and LaPolla’s
358 spanish atelic activity sentences complemented by a bared np
References
Smith, C. 1999. Activities: states or events? Linguistics and Philosophy 22: 479-
508.
Taylor, B. 1977. Tense and continuity. Linguistics and Philosophy 1: 199-220.
Tenny, C. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht:
Kluwer Academic Publishers.
Van Valin Jr., R. 1990. Semantic parameters of split intransitivity. Language 66:
221-260.
Van Valin Jr., R. 2001. An Introduction to Syntax. Cambridge: ‘ Cambridge Uni-
versity Press.
Van Valin Jr., R. and LaPolla, R. J. 1997. Syntax. Structure, meaning and func-
tion. Cambridge UK: Cambridge University Press.
Verkuyl, H. J. 1972. On the compositional nature of the aspects. Dordrecht: Reidel.
Verkuyl, H. J. 1993. A theory of aspectuality. The interaction between temporal and
atemporal structure. Cambridge GB: Cambridge University Press.
One Rule To Rule Them All: Logical Structures
For Spanish Non Reflexive se Sentences
Carlos González Vergara
Pontificia Universidad Católica de Chile
This paper is based on my doctoral dissertation, that I developed under the direction of
Dr. Ricardo Mairal. I want to express my gratitude to him, Robert Van Valin, Valeria Bel-
loro and María Inés Prado for their support and generosity.
The abbreviations used in the paper are as follows: 3 = third person, acc = accusative,
agx = agreement index, asp = aspect, crea = Corpus de Referencia del Español Actual
(‘Modern Spanish Reference Corpus’), hra = highest ranking argument, mod = modal-
ity, nom = nominative, np = nominal phrase, nuc = nucleus, pl = plural, pp = prepositional
phrase, pred = predicate, psa = privileged syntactic argument, refl = reflexive, s = sentence,
sg = singular, und = undergoer, v = verb.
361
362 logical structures for spanish non reflexive se sentences
morirse (‘to die’), the particle se is named “diafasic se” because this word without
se (morir) seems more formal. Finally, se in the expression érase una vez is named
“narrative se” because this construction evokes the literary gender of fairy tales, as
an equivalent of “once upon a time”. We will not be concerned with this sociolin-
guistic or stylistic uses in this text, but the detailed classification put forward by
Contreras is important because it shows us the complexity of this field of study in
Spanish.
One of the most widely known classifications of the meanings of se usually dis-
tinguishes six kinds of se sentences (adapted from Hernández 1966):
a) Reflexive and reciprocal se sentences. As we can see in (2), this kind of sen-
tence can have both a reflexive reading (‘my friends hit themselves’) or a recip-
rocal reading (‘my friends hit each other’).
b) Passive reflexive sentences. As we can see in (3), a sentence of this kind has a
passive meaning, not a reflexive one. The traditional name “passive reflexive” is
only motivated by the presence of se and opposes to the “periphrastic passive”
built with the verb ser (‘to be’).
(3) Se construyeron muchas escuelas.
refl built.3pl many schools
‘Many schools were built.’
b. Pedro se fue.
Pedro refl go.3sg.past
‘Pedro went away.’
b. Pedro se levantó.
Pedro refl rose.3sg
‘Pedro got up.’
c. La puerta se cerró.
the door refl closed-3sg
‘The door closed.’
carlos gonzález vergara 365
(8) Given any kind of logical structure, unspecify the argument x of the predicate.
a. States: pred’ (x, y) pred’ (Ø, y)
b. Activities: do’ (x, [pred’ (x, (y))]) do’ (Ø, [pred’ (Ø, (y))])
c. Active accomplishments:
c’. do’ (x, [pred1’ (x, y)]) & INGR pred2’ (y) do’ (Ø, [pred1’ (Ø, y)])
& INGR pred2’ (y)
c’’. do’ (x, [pred’ (x)]) & INGR be-LOC’ (y, x) do’ (Ø, [pred’ (Ø)])
& INGR be-LOC’ (y, Ø)
d. Accomplishments and achievements: BECOME/INGR pred’ (x, y)
BECOME/INGR pred’ (Ø, y)
e. Semelfactives:
e’. SEML pred’ (x, y) SEML pred’ (Ø, y)
e’’. SEML do’ (x, [pred’ (x, (y))]) SEML do’ (Ø, [pred’ (Ø, (y))])
f. Causatives:
f ’. [do’ (x, Ø)] CAUSE [(BECOME/INGR) pred’ (y)] [do’ (Ø, Ø)] CAUSE
[(BECOME/INGR) pred’ (y)]
f ’’. [do’ (x, Ø)] CAUSE [do’ (y, [pred’ (y)])] [do’ (Ø, Ø)] CAUSE [do’ (y, [pred’ (y)])]
The application of the rule in (8) can have different syntactic consequences,
and these consequences correspond to the varied traditional se sentences. All of
them, nevertheless, have in common that the semantic argument that originally
would have been expressed as privileged syntactic argument (PSA) of the sentence
is not available to be selected after the application of the rule.
For instance, we can have a transitive “non-se” sentence with a causative logical
structure, as in (9).
If we apply the rule on (8) to this logical structure, we get one where the high-
est ranking argument has been unspecified, as can be seen in (10). This exclusive
logical structure can materialize as four kinds of sentences: an intrinsic-se sentence
(10a), a passive-reflexive sentence (10b), an impersonal-reflexive sentence (in this
carlos gonzález vergara 367
case, in order to obtain this meaning, we have to change the inanimate argument
camisa for a human argument like Juan) (10c), and —finally— a middle sentence
(10d).
a. La camisa se ensució.
the shirt refl stained.3sg
‘The shirt got dirty.’
b. Se ensució la camisa.
refl stained.3sg the shirt
‘The shirt was stained.’
c. Se ensució a Juan.
refl stained.3sg to Juan
‘Juan was stained.’
In Figure 1, we can see how the linking process works for the original transi-
tive sentence in (9). In the logical structure, we can observe the basic causative
structure. The highest ranking argument (Pedro, the first argument of the activity
predicate) takes the actor macrorrole, and the argument camisa takes the undergo-
er macrorrole. Then, the actor is selected as PSA of the sentence and takes nomi-
native case. The undergoer takes accusative case. We proceed with the selection
of the syntactic template, that critically shows two positions in the core and has a
nucleus with an AGX node. The PSA materializes as the first core NP, and as the
personal suffix in the AGX node. The undergoer is expressed as the second core
argument. The verb in active voice, finally, links to the predicate node.
The AGX is a notion proposed by Belloro (2004: 43): it is a node dependent of the
nucleus, and it receives the agreement specifications of all core argument positions present
in the logical structure.
368 logical structures for spanish non reflexive se sentences
SENTENCE
CLAUSE
CORE
NP NUC NP
PRED AGX
What happens to the linking process if we apply the lexical rule in (8) to this
logical structure? The result can be seen in Figure 2. Here, we notice that the high-
est ranking argument has been unspecified. I mark this with the label HRA=Ø. It
is important to see that this label and the dotted line are not really part of the link-
carlos gonzález vergara 369
ing. They only represent that the logical structure has suffered a lexical phenom-
enon. Therefore, we don’t violate the completeness constraint (Van Valin, 2005:
129). The argument camisa takes the undergoer macrorole and, because of the un-
specification of the argument x, it is selected as the PSA of the sentence. This
argument takes nominative case and is linked as a core NP and as the personal suf-
fix of the verb. The morpheme se appears in the AGX node as the morphological
manifestation of the lexical rule. The sentence we obtain (la camisa se ensució) has
a predicate focus structure and its meaning is ‘the shirt got dirty (by accident or
naturally)’. From a traditional point of view, this would be an intrinsic-se sentence.
SENTENCE
SPEECH
ACT
CLAUSE
CORE
NP NUC
AGX PRED
la camisa se ensuci -ó
PSA: NOM
In the following text, taken from CREA (Corpus de Referencia del Español Actu-
al ‘Modern Spanish Reference Corpus’) we can see an equivalent intrinsic-se sen-
tence. It describes a scene in the novel Los aires difíciles. In this context, the clause
el aire se ensuciaba (‘the air got dirty’) is perceived as a part of a natural process.
(11) Aún no había podido dormirse, pero no dijo nada. [...] Nadie más parecía
darse cuenta, pero ella lo veía, lo sentía con tal nitidez que cerraba los ojos cuan-
do los muros empezaban a combarse, a inclinarse entre sí, y el aire se ensuciaba,
se enturbiaba en el presentimiento de la polvareda que armarían los cascotes al
caer como una lluvia gruesa y mortal sobre sus cabezas.
‘She wasn’t asleep yet, but she didn’t say anything. [...] Nobody else seemed
to notice, but she could see it, she felt it so sharply that she closed her
eyes when she felt the walls wobbling inwards and leaning one against the
other, and the air got dirty, it got dirty because she could feel the dust that
would rise when the rubble fell like thick and deadly rain over their heads.’
If we maintain the same logical structure of Figure 2, with the highest rank-
ing argument unspecified, we can also obtain a sentence with a different meaning:
se ensució la camisa (‘the shirt was stained (by someone)’), as can be seen in Fig-
ure 3. The only difference between this diagram and Figure 2 is the information
structure. Se ensució la camisa has a sentence focus structure, contrasting with the
intrinsic-se in (10a), where the NP la camisa acts as a topic. This favors a reading
similar to ‘someone stained this shirt’, that has been traditionally related to the
passive-reflexive sentence (see figure 3, next page).
In (12) we have an example taken from CREA where we can see this meaning.
This is a discussion about television programs and their stars. In this context, the
phrase se ensucie la personalidad del rival (‘the rival’s personality be stained’) has a
meaning where an unspecified person is the instigator of the state of affairs.
Almudena Grandes, 2002: Los aires difíciles, Barcelona: Tusquets.
carlos gonzález vergara 371
SENTENCE
SPEECH
ACT
CLAUSE
CORE
NUC NP
AGX PRED
se ensuci -ó la camisa
PSA: NOM
(13) En ese marco, quiero que enérgicamente nosotros respondamos, pero no con una
mera respuesta de prensa —que creo debemos hacer—, sino de tal forma que estas
cosas no vuelvan a pasar, porque cuando se ensucia a los ediles en la prensa lo
que queda no es lo que uno repara después, sino la denuncia que fue hecha.
‘Within this context, I want us to answer energetically, but not with only
an answer through the press —that I think we must do—, but also in a way
that this things won’t happen again, because when the mayors get stained
in the press what remains is not what you can mend afterwards, but what
was stated in the first place.’
Let us examine finally how middle sentences are derived. We can start with the
same logical structure and specify the presence of two operators: modality (pos-
sible) and aspect (imperfect), as well as a new predicate: easy’. In this case, the
highest ranking argument is unspecified and this triggers the presence of se. The
argument camisa, on the other hand, takes the undergoer macrorole and is se-
http://www.juntamvd.gub.uy/actas/indexados/sesion_060310.htm
carlos gonzález vergara 373
SENTENCE
CLAUSE
SPEECH
ACT
CORE
NUC PP
AGX PRED
se ensució a Juan
ACC
lected as PSA. Therefore, it takes nominative case and is linked to the core (and
its pronominal features to the AGX). The verb is in active form, and the predicate
easy’ is realized as a peripheral adverb. The resulting sentence la camisa se ensucia
fácilmente (‘the shirt gets dirty easily’) is a typical middle sentence in Spanish. The
linking process for this sentence can be seen in Figure 5.
SENTENCE
SPEECH
ACT
CLAUSE
CORE S
modality:
possible CORE
AGX PRED
aspect:
NUC
imperfect
V
V
PSA:NOM
< mod POSSIBLE < asp IMPERFECT (easy’ ([do’ (Ø, Ø)] CAUSE [BECOME dirty’ (camisa)]))>>
(14) Estas pinturas están compuestas por una resina sintética (vinílica o acrílica) que
se halla emulsionada en agua. Presentan una alta impermeabilidad, son muy
poco transpirables y, aunque se pueden lavar, se ensucian rápidamente.
‘These paints are composed by a synthetic (vinylic or acrylic) resin in the
form of an emulsion based in water. They present high impermeability,
perspire very little and even though they can be washed, they get dirty
quickly.’
The analysis for middle sentences in Spanish put forward in Figure 5 is based
on a proposal by Felíu Arquiola (2008). I don’t follow here the analysis of Van
Valin and LaPolla (1997: 417), who say that the logical structures of the middle
sentences correspond to an attributive state, where the adverb acts as attribute.
This is because in Spanish the presence of the adverb, although frequent, is not
mandatory in middle sentences. We can see this in examples like la tuberculosis se
cura (‘tuberculosis can be cured’) in (15), taken from CREA, whose meaning is
also middle.
In Table 1 I show an example of these schemata: the proposal for the properties
of the Spanish middle construction. In Figure 6, we can see how these properties
affect the linking process: (1) the presence of the AGX node in the nucleus, (2) the
argument modulation, (3) the PSA modulation, (4) the omission of the actor, (5)
the active voice of the verb, (6) the presence of the imperfect aspect operator, (7)
the presence of the morpheme se in the AGX node, (8) the presence of the modal-
ity operator POSSIBLE, and (9) the predicate focus structure.
Syntax
Template: AGX in nucleus
Argument modulation: central positions are reduced in 1
PSA modulation: variable
- Undergoer argument [-human]: the undergoer argument is selected as PSA
- Undergoer argument [+human]: no argument is selected as PSA
Morphology
Verb: active voice, imperfect aspect
Morpheme se in AGX node
Semantics
Property interpretation, related to the modality operator POSSIBLE
PSA is not the instigator of the state of affairs, but it is affected by it
Pragmatics
Focal structure: predicate focus (default)
Illocutionary force: unspecified
It is evident that in this paper I have not explained, or merely started to ana-
lyze, many complex aspects of the Spanish se constructions and their idiosyncratic
meanings. However, I hope I have shown how the Role and Reference Grammar
theoretical framework allows us to explain that the occurrence of the marker se in
all the kind of sentences presented can be based on a single lexical rule.
Figure 6. Properties of the Spanish middle construction and their influence in the linking
378 logical structures for spanish non reflexive se sentences
References
1. Introduction
The study of the lexico-semantic properties of verbs and their relation with syn-
tax has long been common ground for projectionist and constructionist theories.
Neither, however, has been able to provide a thorough explanation of the nature of
lexicon, grammar and their relationship.
From our point of view, projectionist theories like Role and Reference Gram-
mar (RRG; Van Valin & LaPolla 1997; Van Valin 2005), although they have pro-
vided a rich system of lexical representation for the decomposition of both single
predicates (cf. remember in Van Valin & Wilkins 1993) and lexical classes like the
verbs of saying (Van Valin & LaPolla 1997: 115-118; Van Valin 2005: 46), their
mainstream current representations primarily deal with the grammatically-salient
aspects of lexical meaning. On the other hand, construction-based approaches,
like Construction Grammar (CxG; Goldberg 1995, 2002, 2006), focus mainly on
the representation of constructions, sometimes overlooking the kind of constraints
Financial support for this research has been provided by the DGI, Spanish Ministry of
Education and Science, grant no. HUM2004-05947-C02-01/FILO,
���������������������������������������������
grant no. HUM2005-
02870/FILO, and grant no. HUM2005-01728/FILO. The ����������������������������������
research has been co-financed
through FEDER funds. This work is an extended version of the paper delivered at the
International Conference on Role and Reference Grammar held in Mexico D. F., 2007. I
would like to thank the audience at the Conference and the anonymous reviewer(s) of the
article for their insightful comments and suggestions. All usual disclaimers apply.
Abbreviations employed in this article: acc ‘accusative’, col ‘colloquial’, cxg ‘Construc-
tion Grammar’, flm ‘the Functional Lexematic Model’, fml ‘formal’, infml ‘informal’, ls
‘logical structure’, lt ‘lexical template’, mtt ‘Meaning and Text Theory’, nsm ‘Natural Se-
mantic Metalanguage’, lcm ‘the Lexical Constructional Model’, rrg ‘Role and Reference
Grammar’, sb ‘somebody’.
381
382 lexical templates for the spanish verbs of feeling
that explain how and why constructions interact with specific verbal predicates
(Ruiz de Mendoza & Mairal 2007a: 27). In this paper the Lexical Constructional
Model (Mairal & Van Valin 2001; Mairal & Faber 2002, 2007; Mairal & Ruiz
de Mendoza 2006; Ruiz de Mendoza & Mairal 2007a, 2007b) and its notion of
lexical template will be presented as the meeting point for both paradigms. The
elaboration of these templates will be exemplified with a�������������������������
sub-domain of verbs of
‘feeling’: the
���������������������
Spanish verbs of anger.
The organization of the paper is as follows: section 2 introduces the theoretical
premises of the Lexical Constructional Model. In section 3, the lexical templates
themselves will be presented, focusing on the syntactic (3.2.1), the lexico-semantic
(3.2.2) and the pragmatic features (3.2.3) they codify. Finally, the conclusion will
be provided in section 4.
2. The LCM and the paradigmatic organization of the Spanish anger verbs
External
Unification process and internal
constrains
Semantic interpretation
The LCM takes for granted that (i) constructions are vital to account for the
cases of multiple argument realization, and that (ii) the unification of the argu-
ment structure of a verb and a construction should be ruled in the grammar by
means of a number of external and internal constraints. Semantic interpretation
is then arrived at by the unification of the lexical template of the predicate under
analysis and the constructional template in which it appears. �������������������
This paper focuses
on the former, i.e. the content and relevance of Spanish lexical templates within
the LCM, leaving the latter for further research. The interested reader is referred
to the work carried out by Ruiz de Mendoza and Mairal mentioned in the Refer-
ences section below, as well as the papers included on the website of the LEXI-
COM project.
One of the landmarks of the LCM is its strong semantic concern, giving much
more emphasis to the lexico-semantic properties of verbs than RRG or CxG. This
is due to the fact that the LCM stems from the Functional Lexematic Model
(henceforth FLM; cf. Martín Mingorance 1998; Faber & Mairal 1999), which
has also been partially integrated into the new model. By using the main tenets
of Dik’s Functional Grammar (1997a, 1997b) and Coseriu’s Lexematic Theory
(1981), the FLM organized both the English and the Spanish verbal lexicons
paradigmatically and syntagmatically into a series of coherent semantic classes or
lexical domains such as: existence, change, possession, speech, emotion, ac-
tion, cognition, movement, physical perception and manipulation.
Each of these lexical domains is obtained by means of a process of exhaustive
semantic factorization, working upwards from various dictionary entries of the
possible predicates belonging in each class (Faber & Mairal 1999: 83-101). Thus,
the dictionary definitions are broken down into two meaning components for
each verb:
For an updated account of the LCM, we refer the interested reader to the LEXICOM
Causar enfado
(‘to cause anger’)
1. Enfadar: causar en alguien un estado de enfado
(‘anger: cause in sb a state of anger’)
1.1. Enojar: enfadar a alguien, causándole enojo (formal)
(‘anger: anger to-acc sb, causing-him/her anger’)
1.2. Enfurruñar: enfadar a alguien suavemente o por mimo (informal)
(‘cause sb to sulk: anger to-acc sb gently or because of too much care’)
1.3. Molestar: enfadar a alguien ligeramente, causándole inquietud del ánimo
(‘annoy: anger to-acc sb a little, causing-him/her to be restless’)
1.3.1 Pinchar: molestar a alguien repetidamente con la intención de que se
enfade
(‘needle: annoy to-acc sb repeatedly so that s/he gets angry’)
1.3.2 Incomodar: molestar a alguien, causándole incomodidad
(‘bother: annoy to-acc sb, causing-him/her to feel uncomfortable’)
1.4. Disgustar: enfadar a alguien, causándole disgusto o pesadumbre
(‘upset /displease: anger to-acc sb, causing-him/her annoyance or sorrow’)
1.4.1 Contrariar: disgustar a alguien, causándole contrariedad o dificultades
(‘vex: upset to-acc sb, causing-him/her vexation or difficulty’)
1.4.2 Desazonar: disgustar a alguien, causándole desazón
(‘disturb: upset to-acc sb, causing-him/her uneasiness’)
1.5. Enemistar: enfadar a alguien, haciendo que deje de ser su amigo
(‘antagonize: anger to-acc sb, causing their friendship to break off ’)
Since we want to maintain the lexico-semantic and pragmatic parameters that define
this Spanish sub-domain, the glosses translate each definition into English more or less
literally in Table 1 as well as in the examples used. If
�����������������������������������������
the equivalent English verbs were pro-
vided, those parameters would be lost.
rocío jiménez-briones 385
Hyponymy is the semantic relationship that percolates down through the verb
class of Table 1 above. The nuclear meaning of the sub-domain is lexicalized in
enfadar ‘anger’, since all the verbs are its direct or indirect hyponyms. On the other
hand, parameters such as manner, instrument, purpose, reason, time, degree and
register encode a different type of specification of the more generic term enfadar
‘anger’, resulting in the rest of hyponyms. For example, (1.3) molestar ‘annoy’ gives
content to degree –ligeramente, ‘a little’–, whereas (1.3.1) pinchar ‘needle’ instanti-
ates manner –repetidamente ‘repeatedly’– and purpose –con la intención de que se en-
fade ‘so that someone gets angry’. It is precisely this idiosyncratic component that
we employ to distinguish each of the verbs that belongs to this class, constituting
the basis for the internal variables of the lexical templates analyzed in section 3.2.2.
The term lexical template (henceforth LT) is used here to refer to the formalized
construct that functions as a metaentry, containing the core meaning and the syn-
tactic information relevant for the verbs of a lexical class or sub-domain.
The RRG logical structures (LSs) are the starting point of the LTs developed
by the LCM. However, whereas LSs only capture those features that have a di-
rect role in the mapping into syntax, the LCM ����������������������������������
incorporates an enriched semantic
component by using all world-knowledge
������������������������������������������������������
elements �����������������������������
previously identified by the
FLM; as a result, ���������������������������������������������������������������
a complete domain of verbs is certainly defined (cf. Table 1)��.
As this new semantic component includes relevant aspects of word meaning, LTs
become methodologically closer to CxG semantic frames (Mairal & Faber 2007:
141). Furthermore, since LCM templates are set out to be universally and typo-
logically valid -unlike frames-, they employ a metalanguage based on semantic
primitives, lexical functions and Aktionsart distinctions. With this in mind, LCM
templates are made up of the following modules or components:
T�����������������������������������������������������������������������������
he syntactic component, on the one hand, is realized by means of the LSs pro-
posed in RRG and the predicate classes identified there: states, activities, achieve-
rocío jiménez-briones 387
I�����������������������������������������������������������������������������
nterestingly enough,���������������������������������������������������������
the FLM ������������������������������������������������
nuclear terms coincide, to a great extent, with
Wierzbicka’s inventory of primitives identified in the Natural Semantic Meta-
language framework (NSM; Wierzbicka 1996, 1999; Goddard and Wierzbicka
2002, 2005, 2007), which has been shown to be valid for over a hundred languages.
The semantic primitives identified so far in the NSM are presented in Table 3 (in
next page).
It is worth mentioning that all predicates used in the syntactic module of the
LCM templates, unlike the ‘canonical’ RRG LSs, have been taken from the FLM’s
inventory of nuclear terms (cf. Table 2) or from the NSM’s list of semantic primes
below. Besides, since primes are language-neutral in the NSM, Spanish predicates
have been employed in the elaboration of the LTs for the Spanish verbs of anger.
For the NSM theory, “any natural language is adequate as its own semantic meta-
language” (Goddard 2002: 5). Thus, the hypothesis is that the set of irreducible
semantic primes identified for English must have exponents not only in Spanish,
as illustrated in Table 3, but in almost all the languages one sets up to study. It
388 lexical templates for the spanish verbs of feeling
Grammatical
NSM semantic primes Spanish exponents
category
i, you, someone/person,
yo, tú, alguien/persona,
Substantives people, something/thing,
gente, algo/cosa, cuerpo
body
this, the same, other/
Determiners esto, lo mismo, otro
else
one, two, some, all, many/ uno, dos, algunos, todo,
Quantifiers
much mucho
does not mean, however, that semantic primes are expressed identically in all the
world’s languages: exponents of primes may be single words, phrasemes or bound
morphemes, depending on the language under consideration. They may even have
different morphosyntactic properties and belong to different word-classes, as long
as the prime conveys the same requisite meaning in each language (Goddard &
Wierzbicka 2007: 800). The numerous NSM studies (cf. Goddard & Wierzbicka
2002) carried out in languages as different as Polish, Lao, Russian, or French, to
name just a few, provide sufficient evidence that a language-neutral metalanguage
is a valid tool to account for the lexico-semantic properties of predicates in maxi-
mum detail.
In the LCM, this core of semantic primitives is in turn combined with the op-
erators or lexical functions proposed by Mel’cuk (1989) and his colleagues (Alonso
Ramos 2002) in Meaning and Text Theory (MTT) to explain the lexical colloca-
tions of a language:
Semantic function
MTT lexical functions Definition
(with their application adapted
to paradigmatic structure)
anti Antonym/negation
caus Cause
cont Continuity/duration
390 lexical templates for the spanish verbs of feeling
Table 4. MTT & LCM functions used in the Spanish sub-domain of anger-verbs
3.2. The elaboration of lexical templates for the Spanish verbs of anger
���� The dog scared the boy: [do’ (dog, ∅)] CAUSE [feel’ (boy, [afraid’])]
c.
(Van Valin 2005: 47)
Unlike the theories above, the LCM conceives verbal meaning as an enhanced
representation made up of a pragmatic, a semantic and a syntactic component
which makes use of a metalanguage based on an inventory of primes, semantic
functions and Aktionsart distinctions: the lexical template. This format reduces
circularity and gives templates a typological dimension that is impossible with
language-specific representations.
The specific LCM templates for the definitions previously introduced in Ta-
ble 1 are presented in great detail in the next three sub-sections. As a general-
ization to all LCM templates, the reader must bear in mind that two types of
variables are employed to differentiate the syntactic and the semantic modules
that make up each template: internal variables and external variables. The for-
mer are signaled with Arabic numerals (1, 2, 3) and appear as subscripts��������
������������������
of the
semantic����������������������������������������������������������������������
functions they are arguments of, whereas the latter are marked in Ro-
man letters (x, y, z) and will always be mapped into syntax. The default link-
ing between external and internal variables is x = 1, y = 2, and z = 3, although it
can be specified if necessary. Let us now analyze the particular features of the
syntactic, semantic and pragmatic components of the lexical templates for the
Spanish verbs of anger.
As detailed in section 3.1, the syntactic module of LCM templates, unlike LSs,
employs predicates drawn from the FLM’s inventory of nuclear terms (cf. Table 2)
or from the NSM’s list of semantic primes (cf. Table 3). Furthermore, since each
language is described on its own in the NSM, without imposing the categories of
the linguist’s native language, usually English, Spanish predicates have been used
in the LTs of the Spanish verbs of anger. On the grounds of typological analysis,
the LCM is thus able to solve the two major drawbacks of RRG structures: the
delimitation of the decompositional chains and the usage of English as the only
language of lexical representation (Ruiz de Mendoza & Mairal 2007a: 30; Ruiz de
Mendoza & Mairal, 2007b).
The RRG LS of (5) is inherited by all the hyponyms of enfadar ‘anger’, so it is
not included again in the LTs, unless specified differently, like in (1.7) ofenderse
‘take offense’ and (1.14.1) escandalizarse ‘get outraged’, which exhibit the follow-
ing LS:
The Spanish predicate ofenderse ‘take offense’ and escandalizarse ‘get outraged’
are semantically related to the feeling of anger when used as accomplishments
only. According to the Spanish dictionaries consulted for this study, the causative
accomplishments related to these verbs, that is, ofender ‘cause offense’ and escandal-
izar ‘outrage’, are closer to causing someone to lose respect and to causing a shock,
respectively, rather than to causing anger. Therefore, they are included in this sub-
domain with their accomplishment LS, instead of the ‘canonical’ LS in (5).
Neither does the Spanish predicate (1.5) enemistar ‘antagonize’ fully inherit the
causative accomplishment LS of (5), as presented below:
(7) [(do’ (x, ∅)] CAUSE [BECOME sentir’ (y, [enfado’ (z)])]
We are aware that reciprocal verbs are represented differently in RRG (cf. Van Valin
rocío jiménez-briones 393
b’. [do’ (your words, ∅)] CAUSE [BECOME feel’ (my friends, [anger’ �������������
(my friends)]
c’. [do’ (your words, ∅)] CAUSE [BECOME feel’ (my friends, [anger’ ����
(?)]
and plus in Table 4, require one argument only; binary functions like manner,
result, sympt, because, purp, instr, and manif select two arguments. Argu-
ments may appear as subscripts of the function that operates on them and/or at
the same level of the function they are related to. For instance, in the semantic
module of (1.3) molestar ‘annoy’, a unary function and two binary functions are
present. This is illustrated in (9).
(9) a. (1.3) Molestar: enfadar a alguien ligeramente, causándole inquietud del ánimo
(‘annoy: anger to-acc sb a little, causing-him/her to be restless’)
As a way of exemplifying, the templates of (10) reflect the fact that the event of
angering, bothering, upsetting, vexing, disturbing, irking, pissing off, irritating and
enraging someone in Spanish results in a new feeling: enojo ‘formal anger’, incomo-
didad ‘discomfort’, disgusto /pesadumbre ‘annoyance/sorrow’, contrariedad/dificultad
‘vexation/difficulty’, desazón ‘uneasiness’, fastidio/hastío ‘nuisance/disgust’, cabreo
‘piss off ’, ira ‘irritation’ and rabia ‘rage’.
In other predicates, the semantic function sympt can serve as the argument of
anti, a unary function that negates it. In other words, the combination of sympt
and anti is used to express that a friendship relationship has ended –as in (1.5)
enemistar ‘antagonize’- or that composure and patience have been lost –as in (1.10)
descomponer ‘disturb’, and (1.13) exasperar ‘exasperate’.
Another unary function like magn may take the hypernym of the class enfadar
‘anger’ as its argument and thus define notions such as: (1.11) enrabietar/enrabiar
‘nettle’, (1.13) exasperar ‘exasperate’ and (1.14) indignar ‘incense’, each of one en-
coding a higher degree of anger. Furthermore, the function plus can operate on
magnenfadar to express the highest degree of anger in this class, as it is lexicalized
in Spanish in (1.15) enfurecer ‘infuriate’, (1.15.1) sulfurar ‘infuriate very much’, and
(1.16) encolerizar ‘enrage’.
As illustrated below, semantic functions like manner, because, purp, culm,
cont, instr, and manif are also helpful to further delineate the rest of the predi-
cates in the class under study:
d. ����
1.8 Mosquear: <col> �����������������
[enfadar & MANNER2repentino & MINUSCONTtiempo]
Example (11a) inherits the LS of its hypernym –(5)-, which is then modified
by the semantic component coded in brackets. This semantic part, minusenfadar
& because2sympt2mimo, has the following interpretation: enfurruñar ‘cause sb to
sulk’, used in an informal register, means to anger somebody a little because the
second participant has the symptom of (= feels) too much care.
In the same way as in (11a), the templates of (11b), (11d), and (11e) inherit
the causative accomplishment structure and show a semantic description of the
idiosyncratic properties of each verb. In (11b), the semantic part is interpreted
as follows: reventar ‘rile’, a predicate used colloquially, is fastidiar ‘irk’ to a higher
degree (magnfastidiar), with the purpose of the first participant (purp1) of mak-
ing the second one reach the highest point of anger (culmsympt2 anger). This
formalization neatly mirrors the dictionary definitions of this verb as presented in
Table 1: (������
1.6.1) Reventar: fastidiar mucho a alguien con la intención de que se enfade
(‘rile: irk somebody a lot so that they get angry’). ���������������������������������
In (11d), the second participant
is suddenly angered (manner2repentino) for a short time (minusconttiempo),
whereas in (11e) the irritation of 2 is carried out by the first participant, who says
or does something (instr1say/do������������������������������������������������
) with the purpose of making the second partici-
pant reach the highest point of anger (culmsympt2 anger).
As pointed out in section 3.2.1, both (11c) and (11f ) contain an accomplish-
ment LS in this sub-domain. In (11c), this syntactic structure is modified with
a semantic representation of the following type: �[because1sympt1insulto/des-
precio)]. This indicates that the participant involved in the accomplishment gets
angry because s/he feels insulted or despised. In (11f ), the only participant gets
incensed (culmindignar) and that anger can be seen because s/he looks shocked
(manif1sympt1horror/escándalo).
rocío jiménez-briones 397
As formalized in the examples (9), (10) and (11), the kinds of arguments that a
function can operate on are the following:
b.
����������������������������������
Function + internal variable:
(1.14) Indignar: (…) causándole (…) �[result2]
(‘incense: (…) causing-him/her (…)�)
c. Function + function:
(1.14) Indignar: (…) causándole indignación
������������ [result2sympt2indignación]�
(‘incense: (…) causing-him/her indignation�)
d. Function + prime:
(1.12.1) Provocar: irritar a alguien con palabras u obras (…) �����������
[irritar & instr1decir/
hacer]
‘provoke: irritate to-acc sb with words or deeds (…))
A word is needed for (12e). Since defining all the predicates of a language with
only sixty-five primes in such a concise format as the LT would be highly un-
likely, the LCM contemplates using natural language words in its representations.
Such words would be stored in an ontology of nouns and adjectives, and retrieved
whenever a kind/type of symptom, manner, place, etc., is specified. These non-
primitive elements would eventually be defined in terms of the primes already
employed in LTs.
In the semantic module, the inheritance system of semantic information is also
captured. As specified in the examples above, templates
��������������������������������
make use of the amper-
sand symbol —‘&’, which is a connective meaning ‘and’— followed by the im-
mediate hypernym to show that hyponyms inherit the properties of their super-
ordinate terms. This unified format allows us to build up LTs that are simpler and
easier to read, since we do not have to repeat the previous semantic information
already given for the hypernym.
398 lexical templates for the spanish verbs of feeling
Before explaining the last component of templates, viz. the pragmatic module,
let us summarize the LCM templates previously explained in sections 3.2.1 and
3.2.2. Table 5 below accounts for them.
Causar enfado
(‘to cause anger’)
1. Enfadar [(do’ (x, Ø)] CAUSE [BECOME sentir’ (y, [enfado’]]
[BECAUSE1SYMPT1insulto/desprecio)] [BECOME
1.7 Ofenderse
sentir’ (x, [enfado’])]
[MAGNenfadar&BECAUSE2motivoleve&MINUSCO
1.11 Enrabietar/enrabiar
NTtiempo]
[MAGNenfadar&RESULT2ANTISYMPT2paciencia/
1.13 Exasperar
aguante]
(13) a. the informative function: it is the type of discourse a word appears in (legal, reli-
gious, etc.)
400 lexical templates for the spanish verbs of feeling
Within the sub-domain of the Spanish verbs of anger, the register features
identified are formality (<fml>), informality (<infml>) and colloquialism (<col>).
They play a role when differentiating predicates like enojar ‘formal anger’, enfur-
ruñar ‘cause sb to sulk’�, reventar ‘rile’, mosquear ‘bug’, cabrerar ‘piss off ’ and sulfurar
‘infuriate very much’, which their templates reflect accordingly in (10a), (11a),
(11b), (11d) and (10g).
4. Conclusion
References
1. Introduction
In this article we address the problem of variable word order in sentences in Span-
ish, with special attention to the variable positioning of the subject NP and discuss
variable orders with respect to the thematic role of the subject (Van Valin and La-
Polla 1997) and its placement along the Actor-Undergoer Hierachy (AUH) (Van
Valin 2001). We examine this topic using as a basis the language of children in
three age-groups (3, 5 and 9) as they respond to a narrative task.
In languages where word order of elements in the clause is not rigidly fixed,
as in Spanish, the topic of variable word order has drawn considerable attention
from researchers. A main concern has been to determine if there are conditioning
factors, whether syntactic, semantic or pragmatic, which play a part in determin-
ing or predisposing a particular ordering of clause constituents. Our research falls
along these lines, and our questions are oriented towards examining various fac-
tors that have been identified as influencing or constraining the disposition of ele-
ments in a clause, in particular, the subject positioning with respect to the verb. In
this particular study we will be focusing on the thematic role of the subject NP in
the clause constellation. We want to examine each thematic role with respect to
its placement along the Actor-Undergoer Hierarchy as proposed within Role and
Reference Grammar (RRG) (Van Valin and LaPolla 1997; Van Valin 2001) to see
if there is any evidence to support our hypothesis that the identification of a role
The data on which this study is based was obtained through the support of conacyt
(México) for the research project 26152-H and through the support of the Vicerrectoría
de Investigación y Estudios de Posgrado (buap, México) to Projects V08-05/EDH/G and
08EDH-06-G. A previous version of this article was painstakingly read by anonymous re-
viewers who made very solid and useful suggestions. We thank them for their efforts. Errors
and imprecisions that remain are of course ours.
405
406 subject positioning and thematic role in children’s narratives
Spanish has variable alternation between SV and VS order. The grammatical sub-
ject can occur either pre-verbally (1) or post-verbally (2) as in the following sen-
tences:
b. y un búho salió
and indef owl emerged
‘And an owl came out.’ (3FAIK)
Abbreviations we will be using in the translation of the examples are as follows: def =
definite article, indef = indefinite article, dem = demonstrative, dat-cl = dative clitic pro-
noun, refl = reflexive pronoun.
The code used identifies the different transcripts in the Puebla “Frog Story” database.
The first digit indicates the age of the child, followed by M or F for gender and a three let-
ter code to identify each individual speaker.
rosa graciela montes and alaide rodríguez corte 407
The notion of semantic roles has been proposed and used to capture the semantic
relations holding between a predicate and its arguments. Two structurally identi-
cal sentences, as for example (3a) and (3b), may show different underlying seman-
tic relationships between the participants and the main verb.
The Subject NP Juan is projected differently by the verb in each of these sen-
tences, a distinction which may be captured by positing a different semantic role
for each, so that Juan functions in an agentive role for paint but is an experi-
encer for see. In the same way we can see that the object NP la mesa (‘the table’)
has a different degree of affectedness in each of the sentences. Again, we could try
and capture the difference by considering it as a patient or affected entity in (3a)
but as the source or stimulus of a perception in (3b). This semantic distinction
between structurally identical NPs is captured by the notion of semantic or case
roles for the verb (Fillmore 1968).
rosa graciela montes and alaide rodríguez corte 409
The types of semantic roles for verbs can be specific to each verb or verb type or
they may be more abstract constructions ranging over a group of predicates. Ob-
serve the subject NP in the following two sentences:
If we stay close to the semantics of each particular verb, Juan could be seen as
the walker in (4a) or the runner of (4b). In addition, since both sentences in-
volve verbs of motion, we could capture a generalization by indicating that Juan is
the mover for this verbal class, and even more generally we can posit the thematic
relation of actor, a role which can be found across a wide range of different verbs
(e.g. esperar ‘wait’, toser ‘cough’, comer ‘eat’ etc) (Van Valin and La Polla 1997, Van
Valin 2001). In this study, we propose that it is this more abstract level of role rela-
tion which correlates with the position of the subject.
Only two studies dealing with variable word-order have made some reference
to semantic roles as a factor affecting ordering. The first one, Contreras (1983)
discusses a linear organization for clause constituents. When establishing or-
der, the ‘rheme’ (new information) occurs in clause final position and the ‘theme’
is placed towards the front. But when several constituents could be considered
equally rhematic, Contreras postulates that final placement follows a hierarchical
organization related to the role or function of each constituent. The hierarchical
order that he proposes is as follows: 1. instrument, manner adverbial, strong ad-
verbials (temporal, locative); 2. target; 3. complement, source, location, temporal,
beneficiary, identifier; 4. patient; 5. agent, cause, possessor, experiencer; 6. weak
time and place adverbials. According to this hierarchy, if two items are equally rhe-
matic, a beneficiary or a patient would be more likely to occur in final position
rather than an agent. The author’s purpose in establishing this hierarchy appears
to be different from what we are interested in. In addition, the elements included
in it are of different types and include both nuclear and peripheral elements in the
clause structure. Nevertheless, the hierarchy is enlightening and helped orient the
410 subject positioning and thematic role in children’s narratives
present study with respect to the role that semantic relations play in predisposing
preferential ordering schemas.
More recently, Gutiérrez Bravo (2002), working within the framework of Op-
timality Theory, addresses the question of word order in Spanish, seeking to find
structural explanations for the different word orders. Although most of his argu-
mentation falls outside the scope of this paper there is one point which is par-
ticularly relevant to our purposes here. He posits the notion of “pole” to indicate a
structural position in the clause which will usually be filled and which will usually
occur pre-verbally. This may (but need not) coincide with the subject of the sen-
tence. He establishes a hierarchy of elements that are the most likely to fill the
“pole” position: agent > experiencer > theme > location > manner > time
> reason. This hierarchy then explains why in sentences like (6), the dative NP
rather than the subject occupies the initial position.
lexical decomposition which permits the break-down of a lexical verb into its
component, atomic predicates and permits the principled identification of the
thematic roles of the verb arguments. In contrast with semantic roles presented by
other models, which correspond to states of affairs in the world and are relatively
open-ended, the thematic relations posited by RRG are relatively constrained,
corresponding to the type of predicate identified, its compositional structure and
the type of relations it accepts. In general terms, all verbs can be semantically de-
composed into combinations of two basic predicate types ‘states’ (pred’) and ‘ac-
tivities’ (do’), with two basic argument positions do’ (x, y) and pred’ (x, y) which
can be reduced to just one for monadic predicates.
The Thematic Relations that RRG identifies correspond to the type of predi-
cate, whether ‘state’ or ‘activity’, and the position occupied by the argument, wheth-
er (x) or (y). In addition, the RRG framework argues that in addition to thematic
roles, one can posit a more general type of semantic relation labeled as a ‘semantic
macrorole’ (Van Valin & LaPolla 1997, Van Valin 2001). Only two macroroles are
postulated: actor and undergoer, based on the basic opposition presented in
the two arguments of a transitive construction.
Pretheoretically, the essential insight motivating the postulation of two seman-
tic macroroles is that despite the profusion of thematic relations or related notions
that can be argued for, there is nevertheless a fundamental opposition between
what have been called the two cardinal arguments of a transitive predication, an
agent-like role and a patient-like role, and it is these two arguments that many
morphosyntactic phenomena are keyed to (Van Valin 2001).
The different thematic roles identified with types of predicates can be placed
along a continuum which has been labeled the Actor-Undergoer Hierarchy de-
pending on how actor-like or undergoer-like they are. Placement along this
hierarchy seems to correlate with pre-verbal or post-verbal position for the subject.
sition based on semantic role distinctions. The guiding questions behind the study
are the following:
(7) What are the factors favoring preverbal (SV) or postverbal (VS) subject positioning?
a. How does the type of verb play a part?
b. Does the thematic role of the subject influence positioning?
c. Are there differences in the language use of the children in the three age groups of
our study that may be accounted for by language development?
The data for this paper are taken from a child language corpus in which the sub-
jects respond to a narrative elicitation task. The children are shown a picture-book
and are asked to narrate the story depicted in the book. In order to document
changes in language development, we took data from three different age-groups
of 3, 5 and 9 year-olds.
The language samples were obtained through individual video-taped inter-
views with each child -either at their home or at their school- in which the child
was presented a story-elicitation task using Mercer Mayer’s wordless graphic story
Frog, where are you? (Berman and Slobin 1994). The use of a common stimu-
lus instrument permitted the elicitation of samples which are similar and readily
comparable.
The interviews were transcribed following the conventions established in the
CHAT (CHILDES) transcription system (MacWhinney 1995). This task was
fairly straightforward, although there were some utterances which required mak-
ing some particular coding decisions. In all cases we accepted and used what ap-
peared to be the child’s final version, as for example (8b) which was taken as the
Initially we had planned to have 24 children in the sample, 4 boys and 4 girls in each of
three age groups (3, 5 and 9 years). Since there were only two available transcripts for three-
year-old boys, we thought to complete the set by including data from two young 4-year-
olds (age: 4;3). However, when examining the distribution data we found that the 4-year-olds
pattern more closely with the older children rather than the younger, and were skewing the
results for the 3-year-olds. For this reason, when presenting results concerning the distribu-
tion of forms by age group we omit the data from the two 4-year-old boys, although these
data are included in the global discussions.
rosa graciela montes and alaide rodríguez corte 413
child’s final version for an utterance such as (8a) which shows a false-start and
hesitations:
(8) a. <el niño se> [//] el [/] el wawá@ se va a cael [%com: caer].
<the boy refl> [//] the [/] the doggie refl go to fall
b. el wawá se va a caer.
the doggie refl go to fall
‘The doggie is going to fall.’
The selection criteria for the clauses to be analyzed included the following: (i)
the clause had to have an explicit subject (nominal o pronominal) and a finite verb;
(ii) the clauses could be either main or subordinate clauses (as long as there was an
explicit subject), each was counted separately and only once; (iii) we included only
complete clauses (omitting abandoned or interrupted utterances). We then regis-
tered for each clause all information regarding the subject position, the thematic
role, and the verb type, as well as information regarding definiteness and animacy.
The complete sample obtained consisted of 687 clauses with an explicit subject. The
distribution of the sample between age groups is presented in Table 1:
In the CHAT transcription conventions, the symbol [//] indicates that what precedes,
included in angle brackets, was a false-start, while [/] indicates a possible hesitation with
the same item repeated before and after the symbol. The @ symbol attached to a word
marks it in some way as non-standard. Finally, [: text] or [%com: text] indicates an editorial
comment of some sort. The transcriptions in our data base are full, detailed transcriptions
as in (8a). However, when showing examples we will eliminate unnecessary details and
present them as in (8b).
414 subject positioning and thematic role in children’s narratives
4.2. Results
Table 2 illustrates the results regarding the position of the subject NP in the
corpus.
Clauses SV VS
687 453 (66%) 234 (34%)
Table 2. Positioning of the Subject NPs
We can see that SV-order predominates and this concurs with what has been
found in other studies. Silva Corvalán (1982) and Bentivoglio and Weber (1986)
present quantitative results based on data from adult populations of Mexican-
American and Venezuelan speakers respectively. Rodríguez-Corte (2007) presents
results from a corpus of spontaneous conversational interactions of Mexican chil-
dren (4 to 12 years old). As is shown in Table 3, in all cases SV order is greater than
VS order.
Author Clauses SV % VS %
Silva Corvalán (1982) 338 194 57.40 144 42.60
Bentivoglio &Weber (1986) 398 239 60.00 159 40.00
Rodríguez Corte (2007) 423 308 72.80 115 27.20
Table 3. Subject positioning as reported in previous studies (Rodríguez Corte 2007)
However, when we separate out results for the three age-groups in our sample,
we find differences between the children’s language use. Compare Table 2 which
shows the distribution of subject position based on the total sample with the data
from Table 4 which separates results by age-group:
Age Clauses SV % VS %
3-4 228 106 46 122 54
5 228 173 76 55 24
9 231 175 76 56 24
Table 4. Distribution of subject positioning by age-groups
rosa graciela montes and alaide rodríguez corte 415
Children in the two older groups (5 and 9-year olds) pattern similarly in their
use of SV versus VS order, with SV order being favored in 76% of the cases over
VS order (24%). However, for the younger children, VS-order is noticeably fa-
vored (54%).
As was mentioned previously, since we only had six 3-year-olds we had ini-
tially completed the sample with transcripts from two young 4-year-olds. How-
ever, upon closer examination the 4-year-olds were found to pattern with the older
children, favoring SV over VS order, as Table 5 shows.
Age Clauses SV % VS %
3 198 82 41 116 59
4 30 23 77 7 23
Table 5. Distribution of subject positioning in the two younger age-groups
Because of these results we decided not to group the 4-year-olds together with
the 3-year-olds when carrying out comparisons by age-group.
5. Positioning of the subject with respect to type of verb and thematic role
Our main objective was to examine how the type of verb in the clause and the the-
matic role of the Subject NP interacted with subject positioning. The bulk of the
verbs found in the data fall into the expected verb classes given the content of the
story stimulus used to elicit the speech samples. Thus, 230 verbs (33%) refer to
movement through space (directed movement) and 188 (28%) refer to the place-
ment or location of referents in the story space (localization, existence, posture/
position). In addition, we find 106 verbs (15%) referring to the characters’ “mental”
acts, whether of perception, volition or other internal states (perception, sensory,
volition, cognition), 40 verbs (6%) are verbs of communication and 33 (5%) refer
to states of consciousness, as for example being awake or asleep, which fit into the
semantic classes of physiology and physiological modification. Table 6 shows the
The story tells of a young boy who wakes up one morning to find that his pet frog has
disappeared. He embarks on a journey, with his dog, to search for the frog, looking in various
locations, encountering a number of characters and experiencing a series of (mis)adventures
along the way, until he eventually finds the missing frog.
416 subject positioning and thematic role in children’s narratives
The procedure that was followed is the one established in Van Valin and LaPolla
(1997:116). Space considerations limit the amount of detail we can present in this paper.
rosa graciela montes and alaide rodríguez corte 417
of each thematic role with the two semantic macroroles, actor and undergoer
which are depicted in the top line of the table. Thematic roles organized towards
the left of the Table are seen to be more actor-like while those towards the right
show more undergoer-like characteristics.
ACTOR UNDERGOER
Arg.of 1st arg of 1st arg of 2nd arg of Arg of state
do do’ (x… pred’ (x, y) pred’ (x, y) pred’ (x)
agent effector perceiver theme patient
mover cognizer stimulus entity
s-emitter wanter attribute
consumer possessor consumed
speaker experiencer
observer emoter
attributant
Table 7. Thematic relations found in the data
SPEAKER 16 EXPERIENCER 39
SV 13 SV 35
VS 3 VS 4
OBSERVER 3 EMOTER 1
SV 2 SV 1
VS 1 VS 0
ATTRIBUTANT 3
SV 10
VS 3
6. Developmental Data
Table 12. Subject positioning and thematic role of the Subject NP (9-year-olds)
In examining these three groups of data we can begin to discern some pat-
terns and tendencies with respect to the children’s language use when semantic
factors are taken into account. First of all, we can say that in general all of the
subjects, even the youngest ones, seem to differentiate between more actor-like
roles and more undergoer-like ones and assign in each case what seems to be
the unmarked word-order for each, SV and VS respectively. In addition, we see a
decrease in SV order as one moves away from the Actor end of the continuum.
At the same time, we can see that the similarity that we had observed in Table 4
between 5 and 9-year-olds is maintained even when there is a finer break-down
according to thematic roles. Both age-groups seem to be responding in similar
ways to the thematic roles, and even though there are differences in the actual
percentages of use, the tendencies of increase or decrease seem to be going in simi-
lar directions. These two groups of children do not seem to be making any major
differentiation between the Subject NPs in the first three columns (argument of
DO, first argument of do’ and first argument of pred’). All seem to be considered
primarily as actors and thus assigned SV order. There is greater variability in the
data of the 5-year-olds; nevertheless, for both groups the percentages in favor of
SV are maintained in a range between 82% and 94%. The percentages drop off
slightly with respect to the fourth position in the continuum (2nd arg. of pred’ (x,
y)) but still SV predominates. However, it is only in the final group (Arg. of state
pred’ (x)) that we find a reversal of word-order preference and, at least for the 9-
year-olds, almost categorical VS ordering. Even though VS order increases for the
two final columns, the five year olds seem to be slightly more conservative than the
9-year-olds in their adoption of this order.
When we considered overall distribution of subject placement (Table 4), there
seemed to be a very great difference between the 3-year-olds and the older chil-
dren, with VS order being preferred overall. However, when we examine ordering
in terms of thematic roles, the differences between groups, even though they can
rosa graciela montes and alaide rodríguez corte 421
still be observed, diminish somewhat. As with the other two groups of children, the
3-year-olds show preference for SV order with more agentive subjects. However,
there is a clear-cut difference when it comes to the two final groups, which fall
more towards the role of undergoer. Here we can see VS predominating and the
children do not seem to be making a distinction between 2nd arg. of pred’ (x, y)
and Arg. of state pred’ (x). They seem to make a clear break between the three more
actor-like roles and the two more undergoer-like ones, with SV being preferred
for the first and VS for the second.
Another point to be made with respect to the data results for the 3-year-olds is
the slight reversal of expected frequencies between the second and third positions
in the continuum. One would expect greater SV order for effector roles (first ar-
gument position of do’) than for the experiencer roles in the third column (first
argument position of pred’), since they are higher ranked in the hierarchy. Instead,
we find much greater variability and decreased SV order in the effector column,
contrary to expectations. At the moment we have no explanation for this reversal.
However, the answer may lie with specific verbs found in each group. The expe-
riencer roles in the third column (first argument position of pred’) indicate that
these subjects must be animate and sentient beings with human-like characteristics
who think, feel, perceive and internalize experiences from the outside world, and it
may be these characteristics which move them up closer to agency, as subjects with
at least a potential for intentionality. On the other hand, the effector roles of the
second column do not require human or even animate Subject NPs, which may
lead to them being considered less actor-like. At the same time, some of the verbs
associated with this group of roles, as for example caer/caerse ‘to fall’, seem to pres-
ent the Subject as more undergoer-like than active. Caer/caerse is what happens
to the subject rather than what the subject does. The younger children may have
no difficulty in assigning an actor status to experiencer roles, but they may have
greater difficulty in doing so to the effector roles for the reasons just discussed.
There could also be a different explanation for the variability in the second col-
umn and the reversal of expected frequencies. These may have to do with the large
number of movement verbs that occur, many of which are realized as active ac-
complishments. The lexical decomposition for active accomplishments posits
a complex structure which combines do’ and pred’ elements:
The percentage variability that we see in Table 10 in the second column may
reflect that in cases like these the younger children may associate the subject and
its thematic role with the lower predicate (pred’) rather than with do’, thus assign-
ing it a lower place in the hierarchy and correspondingly increasing VS ordering.
In contrast, the older children, in cases like these, may be opting for the higher
predicate do’, and therefore favoring SV. At this point we can only put forward
these two possible explanations for the data, but have no way of substantiating
either of them further.
What about subjects that are perceived as having more undergoer-like char-
acteristics? When examining the two final columns, corresponding to theme sub-
jects and to patient and entity subjects, we note that although in all groups
there is an increase of VS order when it comes to these types of subjects; only for
the 3-year-olds does VS predominate.
In summary we can say that all three groups of children seem to show sensitiv-
ity to the thematic role of the subject in establishing sentence word order, and that
SV order seems to coincide with greater agentivity or conversely that VS increases
with Subjects that are perceived as more undergoer-like.
7. Conclusions
We began the study with questions as to whether there was an effect between the
type of verb and the thematic role of the Subject NP in a clause and the placement
of that NP in SV or VS order. The results show clearly that there are positive cor-
relations between thematic roles and subject-order such that particular orders are
associated with the position of the thematic role of the Subject NP along the Ac-
tor-Undergoer Hierarchy. In this respect, while SV seems to be the unmarked and
preferred word-order, VS is used for subjects lower in the hierarchy. With respect
to questions of language development we find that all children seem to be sensi-
tive to the thematic role of the subject, and even the youngest children vary appro-
priately between SV and VS depending on the status of the Subject as more or less
actor-like or undergoer-like. In fact, the youngest children seem more radical
than the older groups in assigning VS order to Subjects with thematic roles lower
along in the hierarchy, while the older children seem to apply a strategy of prefer-
ring SV order, except for patient-like subjects.
rosa graciela montes and alaide rodríguez corte 423
Finally, although this has not been a topic of this article, from our examination
of the data we agree with what has been reported in the literature that discursive
and pragmatic factors concerning the information status of the Subject NPs play a
major role in favoring SV or VS order. Nevertheless, the present study was carried
out deliberately not taking into consideration those discursive factors in order to
show that independently there are strong correlations shown with respect to verb
type and its consequent thematic roles.
References
1. Introduction
In previous analyses on the relationship between tense and mode in two Span-
ish varieties (Vázquez Laslop 2007a,b), I found a very particular behavior of poder
‘can, may’ and deber ‘ought to, must, should’ + infinitive in present perfect (in what
follows, “modal periphrases”). While in Peninsular European Spanish the prob-
ability of interpreting both verbal periphrases as epistemic or as root modes was
balanced, in Mexican Spanish the probability of root interpretations was over-
whelmingly predominant. These modal senses are as follows:
Root meanings:
a. ‘For Juan, it has been {possible/allowed} to come.’
b. ‘Juan has been able to come.’
With this term I include the entire diversity of modal meanings named in differ-
ent ways by grammatical and linguistic traditions. In RRG, for instance, illocutionary
force, evidentials, status and modality. Here, mood is understood as a purely grammati-
cal concept.
List of abbreviations: acc: accusative, asp: aspect, dat: dative, evid: evidential, f:
feminine, fg: Functional Grammar, imp: imperative, imperf: imperfect, impfv: imper-
fective, ind: indicative, inf: infinitive, lsc: layered structure of the clause, mod: modality,
n: neuter, past: simple past, perf: perfect, pl: plural, poss: possibility, pres: present, prfv:
perfective, prob: probability, prog: progressive, pstp: past participle, prt: particle, rrg:
Role and Reference Grammar, sg: singular, sta: status, tns: tense, v: verb.
425
426 a puzzle about operators: spanish modal verbs in present perfect
Inferential-Epistemic meaning:
c. ‘(Given the present circumstances, I deduce that) it is possible that Juan has
come.’
Root meanings
a. ‘For Juan it has been {obligatory/necessary} to come.’
Inferential-Epistemic meaning
b. ‘(Given the present circumstances, I deduce that) it is probable that Juan has
come.’
As observed, the paraphrases of sentences in (1) and (2) indicate not only
modal values (such as ability, possibility, necessity, permission, and obligation),
but also the scope of the modification triggered by each mode. On the one hand,
the scope which affects the relationship between the predicate and the partici-
pant referred to by the subject, corresponds to root modal senses, as in (1a), (1b)
and (2a). On the other hand, the scope associated with inferential-epistemic in-
terpretations modifies the truth-value of the propositional meaning of the clause,
as in (1c) and (2b).
In Vázquez Laslop (2007a,b), it was considered that dialectal differences among
aspectual values of the present perfect were crucial for modal interpretations. The
main conclusion was (i) that all aspectual meanings of present perfect may co-oc-
cur with any mode, and (ii) that inferential senses were restricted to the resultative
feature of some uses of present perfect, a very common aspectual value in Penin-
sular Spanish.
The following question arise from such a conclusion. In sentences with Spanish
modal periphrases in present perfect, which is the operators’ order (tense, aspect,
and root, epistemic, and inferential modes), according with the layered structure of
the clause (LSC)? In the present analysis, I begin to formulate a proposal for the
order of operators in sentences involving these modal periphrases in present per-
fect, following both, the Role and Reference Grammar (RRG) and the Functional
Grammar (FG) approaches.
maría eugenia vázquez laslop 427
2.1. Modal scope and semantic modal values of poder and deber
As regards the relationship between mode and temporality, on the one hand,
the deontic modality is commonly associated with prospectivity, since it is expect-
ed that the participant subject to the permission/obligation situation will proceed
in consequence after the performance of the act of prescription. Henceforth, very
�
As may be noted, this categorization of semantic values differs to some degree from
VV&LP: 41, who reserve modality to deontic and ability values. Instead, I follow Bybee,
Perkins & Pagliuca’s (1994), and Vázquez Laslop (2007d) distinction between root and
epistemic possibility, regarding the former as modality.
maría eugenia vázquez laslop 429
commonly (but not exclusively), deontic sentences are in non-past tenses. On the
other hand, since status indicates the speaker’s judgment realized in the speech
event, then it is always associated with present tense (cf. Palmer 2001: 33, 76) or,
more generally speaking, with time zero, as pointed out by Laca (2005: 19-20).
Specifically, in her study on the relationship between Spanish modal verbs and
tense, Laca observed that epistemic judgment is not restricted to deictic present,
but to a “zero time” which may be either the moment of speech or a “displaced
deixis,” as in the reported speech example in (3). In these situations, the modal
verb is in imperfect, pointing to the “zero time” when the epistemic judgment took
place.
As will be seen in §4.1, one of the problems to be discussed derives from the
fact that present perfect is a retrospective tense. Then, from a semantic point of
view, it is necessary to determine how it is possible for sentences in (1) and (2),
with evidential and status meanings (1c and 2b), to be interpreted in present (or
zero time), for they have a retrospective tense operator functioning as an actual
temporal codifier. Furthermore, since present perfect also functions as an aspectual
operator, the relationships generated between retrospective aspectual/tense values,
modality, status, and evidentiality have to be explained. I will concentrate the dis-
cussion on this last issue.
In the Gramática descriptiva de la lengua española, Rojo & Veiga (1999: §44.3.1.2) cate
�
gorize imperfect’s basic value in terms of temporal deixis —“copretérito,” in Bello’s terms
(1984: §§628-633)—, a tense that expresses a situation simultaneous to a moment prior to
the origin (zero time).
430 a puzzle about operators: spanish modal verbs in present perfect
�
Heine (1993) distinguishes seven stages in the grammaticalization of a verb, which go
from the possibility of accepting a clausal complement to its reduction as an affix, marking
a purely grammatical category.
�
All these properties and many others are described with detail in Fontanella de Wein-
berg (1970), Luna Traill (1980: 148-165), Gómez Torrego (1988; 1999: §51.2), and Fern-
ández de Castro (1999: 19-40), among others.
maría eugenia vázquez laslop 431
51-52; Van Valin 2005: 11). Otherwise, the relative order must be determined in
some other way. In any case, the result should be in accordance with the general
findings about the universal features of the clause, based on linguistic evidence
(VV&LP: 22; Van Valin 2005: 3).
As in the case of other Romance languages like French and Italian, where there are two
�
auxiliaries to form perfect: the copula être/essere (‘be’), respectively, for unaccusative and re-
flexive verbs, and avoir/avere (‘have’) for the rest.
432 a puzzle about operators: spanish modal verbs in present perfect
The Spanish “experiential perfect” may be defined in Comrie’s words: “[it] in-
dicates that a given situation has held at least once during some time in the past
leading up to the present” (1976: 58). Although the retrospective duration is ex-
tended up to the present, from Squartini & Bertinetto’s point of view, this value is
already perfective, for the event is “aoristic” or completed (2000: 414).
There is still another value which tends to adopt a deictic temporality rather
than just an aspectual one: the “perfect of recent past” in (7a), which contrasts with
the preterite in (7b). It may occur either in “hot news” and hodiernal contexts —as
in (7a)— or extend the currency of the referred past event to non-hodiernal con-
texts, but relevant for the present, as in Rojo & Veiga’s (1999: 2903) example in
(8). As experiential perfect, this value is also aoristic.
maría eugenia vázquez laslop 433
de la cultura occidental.
of the culture occidental
‘Greece has bequeathed to the world all the bases of Western culture.’
Peninsular and Mexican Spanish differ in the use frequency of the present
perfect aspectual values. According to the characterization put forward by Lope
Blanch (1972), Moreno de Alba (1978), and Company (1983), on the one hand,
and the empirical findings of Colombo (2003: 478), on the other, the Mexican
uses for the present perfect are mainly related to persistent and experiential values,
whereas those of Peninsular uses are mostly associated with recent past and per-
fect of result. This is schematized in Table 3.
As regards the relationship between modal operators’ scope and present perfect
aspectual values, it seems that modality interpretations may be distributed along
434 a puzzle about operators: spanish modal verbs in present perfect
4. Where is retrospectivity?
Recall that the modal status meaning is performed in present (or zero time), not
in a retrospective tense: “it is probable that Juan has come.” This is so because
The corpus already analyzed in Vázquez Laslop (2007a,b) was drawn from two sources:
the Corpus del Español Mexicano Contemporáneo (CEMC) for the Diccionario del Es-
pañol de México, a project of El Colegio de México (Lara 1996), and the Corpus de Refe
rencia del Español Actual (CREA) of the Real Academia Española. The extraction of cases
from CREA was realized during 2002 and 2003. CEMC is a corpus which has not had
additions since its construction in the 1970s. It is a stratified sample of Mexican Spanish
registers and varieties from the 1920s up to the 1970s.
maría eugenia vázquez laslop 435
por parte de funcionarios de la Junta Municipal de [sic] Centro, según indicó ayer a este
periódico el edil Eugenio Morales (0030 Hapodidoespaña2, 1193)
‘The Municipal Socialist Group in the City Council of Madrid, after studying the
Sauna Paraíso File, has detected evidence that there may have been bribery and
extortion on the part of officials belonging to the Central Municipal Board, as
was indicated yesterday by town councillor Eugenio Morales to this newspaper.’
Examples in §4 are all extracted from the corpus described in fn. 8. At the end of each
example the name of file and code-number of each item is indicated. I am in debt to Susan
Beth Kapilian, who translated all the Spanish quotations and examples here presented.
436 a puzzle about operators: spanish modal verbs in present perfect
(10) ‘Given the present circumstances, I deduce that it is {probable / possible} that p was
the case.’
The first part of this paraphrase —“Given the present circumstances, I deduce
that...”— is based on the fact that epistemic interpretations of modal verbs in
perfect are very commonly associated in context with the expression of evidence
known by the speaker, as in (9a) ha detectado evidencias ‘(it) has detected evidence’.
In fact, in previous analyses on periphrases with poder and deber in present perfect,
I concluded that the inferential meaning was directly related to the resultative
meaning in the following way: So long as present perfect (or ‘anterior to present’)
is a tense prior to a reference simultaneous to the origin (zero time) (Rojo & Veiga
1999: 288), such reference is a (set of ) fact(s) within the speech situation observed
by the speaker at the moment of utterance, regarded as evidence of the result of a
{possible/probable} completed past event (Vázquez Laslop 2007b: 211).
Evidentiality is triggered by perfect of result, while the possibility or probabil-
ity values affecting the proposition (whose content is the completed event) are
produced by the lexical meaning of the semi-auxiliary. This insight about the asso-
ciation of perfect of result with inferential evidentiality is supported by typological
views of resultatives, found in Slavic languages (Comrie 1976: 108-110; Bybee,
Perkins & Pagliuca 1994: 95-97; Lindstedt 2000), and already suggested for Span-
ish by Bermúdez (2005). Linstedt comments on South Slavic perfects expressing
evidentiality as follows: “Drawing inferences from the visible results of a non-
witnessed event is a natural extension of resultativity of C[urrent] R[elevance]”
(2000: 375). For his part, Comrie states that “the semantic similarity [...] between
perfect and inferential evidentials lies in the fact that both categories present an
event not in itself, but via results” (1976: 110).
How can the present perfect’s semantic function be represented in the LSC,
according to RRG generalizations? If the modal evaluation is being realized at
the moment of utterance: (i) tense must be present; (ii) status is modified by tense;
(iii) retrospectivity may then be situated in the aspectual operator modifying the
nucleus. This may be expressed as the general schema in (11):
propositional content as a whole, not just V, the nucleus. Recall from (10) that this
sense is expressed in the predicative clause “...that p was the case.” This possibil-
ity may be achieved with the help of Functional Grammar.
In FG, aspectuality is analyzed in different domains, corresponding to different
layers of the clause. This means that aspect is not restricted to the nucleus, since
it may function in more external layers, as the predication as a whole. Perspectival
aspect –in Dik’s words– is an operator concerned “with the way in which the State
of Affairs is viewed upon from an external point in time” (Dik 1997: 238). Its dis-
tinctions are “prospective” (‘John is going to cry’), “immediate prospective” (‘John is
about to cry’), “recent perfect” (‘John has just cried’), and “perfect” (‘John has cried’)
(Dik 1997: 239). These aspectual values are categorized in a more external di-
mension than phasal aspect and (im)perfective aspect, whose scope is the nucleus.
Thus, an alternative for the order of operators in sentences (or utterances, in
FG terms) with epistemic modal verb periphrases in perfect of result would be
as indicated in (12), considering that the inferential meaning is performed at the
moment of utterance (zero time).
In (12), the perfect’s aspectual resultative value indicates the inferential sense,
supported by context, and the past temporal value is codified by perfect as per-
spectival aspect, a predication operator.
Thus, in semantic terms, perspectival aspect, for the one hand, modifies a high-
er level than just the predicate, that is, the extended predication as a whole, a locat-
ed and qualified state of affairs (Dik 1997: 50) and, on the other hand, it does not
imply any internal change within the predication. In any case, (11) and (12) are
intended to schematize the equivalent inferential-epistemic senses of sentences
with modal verb periphrases such as (9a-b).
dialect. For instance, the perfect of recent past is more common in Peninsular than
in Mexican Spanish, as is shown in the Peninsular example in (13):
durante las pasadas navidades. En Sevilla, tres escuelas vieron como [sic] sus respec-
tivos techos caían desplomados. Por fortuna los alumnos estaban todavía de vacaciones.
(0031 Hapodidoespaña3, 1249)
‘But the tragedy could have been much greater during the past Christmas sea-
son. In Seville, three schools witnessed how their respective roofs plummeted to
the ground. ��������������������������������������������������
Fortunately, the students were still on vacation.’
Again, to test the temporality of the modal event, we may change the present
perfect modal verb in (14a) to present in (14b), and keep the predicate in infinitive
perfect:
(14a) has the root meaning: “It is true that x was able to survive.” As expressed in
this paraphrase, the root modal evaluation is within the scope of tense, which must
be retrospective, for it is part of the propositional content.
The main question, however, is whether such retrospectivity is a tense or an
aspectual category, or both. Recall from §2.1 above that root modals trigger a pro-
spective effect on the course of events, opening possible alternatives of states of
affairs. When retrospective tenses are codifying past situations, the modal event is
oriented to some anterior point —in (15) as t-1’—, followed by the effected alter-
natives of states of affairs t-1, which are likewise oriented before the speech event t0
(cf. Laca 2005: 31-32):
(15) Modal event (t-1’) < Possible states of affairs (t-1) < Speech event (t0)
Since Mama Elena died, someone was able to again perform that feat [...]’
b. ??‘�����������������������
��
Although Laca restricts counterfactuality to simple past tenses, she finds this pos-
sibility in Peninsular present perfect as well and, interestingly, she exemplifies this with an
utterance very similar to ours in example (13): Ha podido producirse un accidente terrible, pero
por fortuna no ha pasado nada (2005: 26, fn. 9). (‘A terrible accident could have occurred, but
fortunately nothing happened.’)
440 a puzzle about operators: spanish modal verbs in present perfect
(18) Moodspeechact > Moodevaluative > Moodevidential > Moodepistemic > T(Past) > T(Future) >Mood-
irrealis
> Asphabitual > T(Anterior) > Aspperfect > Aspretrospective > Aspdurative > Aspprospective /
Moodroot > Voice > Aspcelerative > Aspcompletive > Asp(semel)repetitive > AspiterativeCinque (1999: 76)
and of the other retrospective tenses, it would be necessary to design clear linguis-
tic tests for observing the behavior of aspectuality in poder and deber periphrases,
combined with their modal senses, and trying to “put between brackets” the tense
values. It sounds reasonable to adopt Laca’s proposal, since Spanish modal verbs
are still verbs. Thus, they may keep some grammatical features inherent to lexical
predicates, while also combining with their semi-auxiliary properties. One of these
features of lexical predicates would be to be within the scope of aspectual operators.
Generalizing, in RRG terms, the scope order for modal root interpretations of
Spanish modal verbs in present perfect, both for Peninsular and Mexican Spanish,
should capture (i) the particular deictic temporality of present perfect, a tense an-
terior to a reference simultaneous to the origin; (ii) such tense affecting modality,
and (iii) the specific current relevance of present perfect: as a result, as persistent
situation, as experiential, or as recent past. So long as restrictions on the order of
operators in RRG do not allow aspect to modify modality (VV&LP: 45-52, 171-
172), the kind of current relevance of perfect should be represented as affected by
the root meaning. This generalization may be represented as in (19):
(20) a. ... [π3Inferential Xi: [π2Anterior π2Inherent-Mod π2Perf ei: [π1 [...
b. ... [π3Inferential Xi: [π2Anterior π2Perf π2Inherent-Mod ei: [π1 [...
π3 = proposition operators; π2 = extended predication operators; π1 = predi-
cate perators; Mod = Modality; Perf = Perfect (Perspectival aspect)
442 a puzzle about operators: spanish modal verbs in present perfect
5. Conclusion
Role and Reference Grammar provides a rich analytic and systematic categoriza-
tion of modal senses. However, for the understanding of the interaction among
tense, aspect and modal operators within the clause, it is recommendable to cat-
egorize likewise aspectuality in a more external domain than just at the core and
at the nucleus levels, as has been proposed in Functional Grammar. In a language
like Spanish, perfect is a hybrid grammatical category, which seems to be devel-
oping from aspectual meanings (like perfect of result, of persistent situation, and
experiential) to tense values (like recent past), i. e., from internal to external se-
mantic domains. This categorial development needs to be captured in the layered
structure of the clause. Such a phenomenon has consequences in a language where
modality and status are codified by semi-auxiliaries conserving some lexical prop-
erties. Some of these consequences may be summarized as follows.
As regards the clause level, on the one hand, the aspectual perfect of result
in modal periphrases is assimilated into inferential interpretations in contexts fa-
voring evidentiality. This modal sense modifies status, the speaker’s commitment
with the truth of the proposition. On the other hand, the retrospective temporal-
ity of present perfect is within the scope of status, and functions as a perspectival
aspectual mark whose scope is the core.
At the core level, regarding aspectuality, it is possible to hypothezise that Span-
ish root modals may be modified by perfect perspectival aspect. Perfect of recent
past, combined with modality, shows similar properties of simple past in root
modals, since both retrospective forms trigger counterfactual inferences. As re-
gards tense, anterior has within its scope perfect perspectival aspect and root mod-
al core operators, and functions as a mark of deictic retrospectivity, whose scope is
the extended predication (in FG terms).
Some features of lexical predicates have yet to be explored in Spanish modal
verbs. In such a case, as with the rest of lexical predicates, it would be necessary to
define the corresponding Aktionsart(en) of poder and deber, and their consequent
semantic and grammatical properties, such as the possibility of being aspectualized.
References
Harris, M. 1982. The ‘Past simple’ and the ‘Present perfect’ in Romance. In Stud-
ies on the Romance verb, M. Harris and N. Vincent (eds.) 42-70. London:
Croom Helm.
Heine, B. 1993. Auxiliaries. Cognitive Forces and Grammaticalization, Oxford:
Oxford University Press.
Hengeveld, K. 1989. Layers and operators in Functional Grammar. Journal of
Linguistics 25: 127-157.
— 2004. Illocution, mood, and modality. In Morphology. An International
Handbook on Inflection and Word Formation, G. Booij, Ch. Lehmann, J. Mug-
dan, and S. Skopeteas (eds.), vol. 2, 1190-1201. Berlin: Mouton de Gruyter.
Laca, B. 2005. Tiempo, aspecto y la interpretación de los verbos modales en espa-
ñol. Lingüística 17: 9-43.
Lara, L. F. (dir.). 1996. Diccionario del español usual en México. México: El Cole-
gio de México.
Lindstedt, J. 2000. The perfect – aspectual, temporal and evidential. In Dahl
(ed.), 365-383.
Lope Blanch, J. M. 1972. Estudios sobre el español de México. México: Universi-
dad Nacional Autónoma de México.
Luna Traill. E. 1980. Sintaxis de los verboides en el habla culta de la Ciudad de
México. México: Universidad Nacional Autónoma de México.
Moreno de Alba, J. G. 1978. Valores de las formas verbales en el español de México.
México: Universidad Nacional Autónoma de México.
— 2006. Valores verbales de los tiempos pasados de indicativo y su evolución.
In Sintaxis histórica de la lengua española. Primera parte: La frase verbal, C.
Company Company (dir.), 3-92. México: Universidad Nacional Autónoma
de México / Fondo de Cultura Económica.
Olbertz, H. 1998. Verbal Periphrases in a Functional Grammar of Spanish. Berlin:
Mouton de Gruyter.
Palmer, F. R. 2001. Mood and Modality. 2 ed. Cambridge : Cambridge Univer-
sity Press.
Rojo, G. and Veiga, A. 1999. El tiempo verbal. Los tiempos simples. In Bosque
and Demonte (dirs.), 2867-2934.
Squartini, M. and Bertinetto, P. M. 2000. The simple and compound past in
Romance languages. In Dahl (ed.), 403-439.
Stowell, T. 2004. Tense and modals. In The Syntax of Time, J. Guéron and J.
Lecarme (eds.), 621-635. Cambridge, MA.: The MIT Press.
maría eugenia vázquez laslop 445
Van Valin, R. D. Jr. and LaPolla, R. J. 1997. Syntax. Structure, Meaning and
Function. Cambridge: Cambridge University Press.
Van Valin, R. D., Jr. 2005. Exploring the Syntax-Semantics Interface, Cambridge:
Cambridge University Press.
Vázquez Laslop, M. E. 2007a. Modalidad de deber (de) + infinitivo en ante
presente: México frente a España. In Actes du XXIV Congrès International
de Linguistique et de Philologie Romanes, D. Trotter (ed.), 591-604. Tübingen:
Niemeyer.
— 2007b. Modalidad, tiempo y aspecto de ha podido + infinitivo: México
frente a España. In La Romania en interacción: entre historia, contacto y política.
Ensayos en homenaje a Klaus Zimmermann, M. Schrader-Kniffki and L. Mor-
genthaler García (eds.), 197-213. Frankfurt am Main / Madrid: Vervuert /
Iberoamericana.
— 2007c. “Pudiera” de dicto. Nueva Revista de Filología Hispánica 50: 1-22.
— 2007d. Las modalidades de re y de dicto en la lingüística funcional contem-
poránea. In De amicitia et doctrina. Homenaje a Martha Elena Venier, L. F.
Lara, R. Y. Ortega, and M. L. Tenorio (eds.), 453-479. México: El Colegio de
México.
Vázquez Laslop, M. E. and Yee Sánchez, S. 2006. Poder + infinitivo, tiempo y
persona en el español de México. In VIII Encuentro Internacional de Lingüística
en el Noroeste, Serie Memorias, R. M. Ortiz Ciscomani (ed.), vol. 2, 369-383.
Hermosillo, Son.: Universidad de Sonora.
Epistemic adverbs
and mood alternation
Armando Mora-Bustos
Escuela Nacional de Antropología e Historia
1. Presentation
I am indebted to Luisa Becerril, Valeria Belloro, Lilián Guerrero, Laura Romero and
Veronika Somogyi for their valuable comments of this work. None of them should be held
accountable for my views or mistakes.
Abbreviations: af=affirmative, asp=aspect, apr=antepresent, aps=antepospreterit, aux=
auxiliar, cl=clitic, cp=copreterit, dat=dative, dec=declarative, des=desire, ea=epistemic
adverb, ft=future, gd=gerund, hyp=hypothetical, if=illocutionary force, in=indicative,
ing=ingressive, irr=irrealis, ls=logical structure, mod=mood, neg=negation, obl=obligation,
pc=participle, pos=possible, pr=present, prf=perfective, prb=probable, prt=preterit, ps=pos
preterit, pt=past, re=realis, sb=subjunctive, sta=status, sub=subordination, th=truthful,
tns=tense.
447
448 epistemic adverbs and mood alternation
The examples showed in (1) are evidence that the mood alternation in Spanish
is possible. Recently there have been several studies undertaken in order to give an
explanation to the above in the fields of syntax, semantics and pragmatics. In this
essay, I intend to give some answers, through an analysis based on Role and Refer-
ence Grammar (RRG), for an additional context where the alternation between
indicative and subjunctive mood is possible in the language. These sentences pres-
ent an epistemic adverb like quizás ‘maybe’, posiblemente, ‘possibly’, probablemente
‘probably’, tal vez ‘perhaps’, etc. This work is organized as follows: 2. Introduction;
3. Contexts in which epistemic adverbs appear: 3.1. Epistemic adverb and simple
or complex tense and subjunctive (simple and complex tense); 3.2. Epistemic ad-
verb and simple or complex tense and indicative (present, preterit, copreterit, fu-
ture, pospreterit) (antepresent, antepospreterit); 3.3. Simple or complex tense and
indicative and epistemic adverb. Finally, 4. Status and modal adverbs.
2. Introduction
In the descriptive studies of the semantics of adverbs (Kovacci 1999, Torner, 2005)
as well as in Spanish reference grammars (Alcina y Blecua 1994, Esbozo 2000,
armando mora-bustos 449
Bello 1984), adverbs such as quizás or quizá, tal vez, posiblemente, probablemente,
ojalá, casi, acaso and a lo mejor have been classified as adverbs of doubt, possibility
or epistemic stance. In this presentation, I analyze, out of this group, those adverbs
that allow an alternation when selecting the mood of the verb they modify, i.e.
indicative and subjunctive. The adverbs included here are: quizás or quizá, tal vez,
posiblemente and probablemente. The data comes from the Mexican Spanish variety
and have been extracted from CREA ‘Corpus de Referencia del Español Actual’.
The questions to be answered in this work are: what is the correlation between
status and modality operators and epistemic adverbs? and as it is shown in the ex-
amples (4) to (10), adverbs do have the capacity to generate mood alternation on
the verb; the question is, therefore, what is the position of these specific adverbs in
the logical structure?
In RRG, each of the layers of the clause (nucleus, core and clause) is modified
by one or more operators (time, aspect, status -negation and modality-, direction-
als, evidentials, event quantification, and illocutionary force). Operators are pro-
jected independently of clausal constituents, since they involve grammatical cat-
egories that are qualitatively different to those of predicates and their arguments.
Due to the semantic complexity of operators, Van Valin & LaPolla (1997) do not
present a substantial representation of them, but only supply their location in the
semantic representation as shown in (2). They modify the clause and its parts and
the ordering of the morphemes expressing operators with respect to the verb in-
dicates their relative scopes. Operators that are not pertinent to this explanation
have been omitted for clarity.
operators include the epistemic modality, the external negation and forms encod-
ing realis and irrealis status. Finally, “illocutionary force” encodes the distinction
between assertions, questions, orders, or expressions of a desire. This is illustrated
with the example in (3).
In (3), a peripheral adverbial ayer ‘yesterday’ is added, which interacts with op-
erators, as all peripheral extensions of this type. This is represented in the logical
structure of the sentence, where operators as well as adverbs are ordered in terms of
their relative scope. The operator that has the widest scope over the construction is
the illocutionary force operator, i.e. the communicative purpose of the speaker is to
make an assertion. Following the illocutionary force operator, the temporal adverb
ayer has scope over the matrix core and the subordinated core. The scope of the
adverb extends over the tense operator (preterit), the status operator (subjunctive
mood), the modality operator (obligation), and the aspect operator (perfective).
Detailed description of the correlation between epistemic adverbs and operators
is provided below.
Epistemic adverbs quizás or quizá, tal vez, posiblemente and probablemente may ap-
pear in the following three contexts: (i) Epistemic adverbs modify simple o com-
plex tenses. These verbs must be inflected in subjunctive. (ii) Epistemic adverbs
modify simple or complex tenses but these verbs, contrary to what is expected,
don’t code subjunctive, but indicative. (iii) Epistemic adverbs can appear in post
verbal position, so the main verb mood inflection only occurs in the indicative. To
appreciate the variability of epistemic adverbs and nuclear verb mood, the follow-
ing sections deal with particular examples.
armando mora-bustos 451
Epistemic adverbs demand subjunctive mood both in simple and complex tenses
of the verbs they modify as shown in (4) and (5). Simple and complex tense para-
digms are classified following Bello’s (1984) proposal.
Simple tenses
In (4), the subjunctive is the mood of the verb, because an epistemic adverb modi-
fies it. The adverb is in preverbal position. In all this examples one epistemic ad-
verb is modifying the main verb; the tense on the verb inflection is present.
In the same way as in (4), in (5) an epistemic adverb modifies verbs in complex
tenses, the mood on the verb is subjunctive as well, and the adverbs are in prever-
bal position.
Complex tenses
These forms are low in frequency because they constitute marked temporal forms.
As in (4) an epistemic adverb modifies the verb, but in (5a) the verb is inflected
in preterit while in (5b) and (5c) it is in present tense. In both cases the mood is
subjunctive.
452 epistemic adverbs and mood alternation
The examples in (4) and (5) illustrate what prescriptive Spanish grammars
posit as the only acceptable options. In simple, just like in complex tense, the verb
must be subjunctive (present or preterit tense) if an epistemic verb modifies it.
Simple tenses
present
preterit
(7) a. Quizás por falta de oportunidad a mí no me sucedió.
Maybe for lack of opportunity to me neg cl happen.prt.in
‘Maybe it didn’t.in happen to me because of lack of opportunities.’
copreterit
(8) a. El pastel quizá era un pastel ficticio.
The cake maybe be.cp.in a cake fictitious
‘The cake maybe was.in a fictitious one.’
future
(9) a. Quizás se hará una excepción con México.
Maybe cl do.ft.in an exception with Mexico
‘Maybe an exception will be.in made with Mexico.’
pospreterit
(10) a. Los mecanismos quizás no transferirían los recursos.
The mechanisms maybe neg transfer.ps.in the resources
‘The mechanisms may not transfer.in the resources.’
All these examples are important because they show that epistemic adverbs
may modify verbs in the indicative mood. They also show that this is possible with
different tenses. Although these constructions would not be accepted in traditional
armando mora-bustos 455
grammar studies, their frequency is high (Mora Bustos 2006). For this reason, an
explanation ought to be given for the occurrence of these kinds of combinations.
On the other hand, it is also possible to find examples, as in (11), with complex
tenses, where there is an epistemic adverb in preverbal position and it does not de-
mand subjunctive. These complex tenses, as in (11), aren’t so common in Mexican
Spanish, for this reason it is difficult to find a great variety of examples. Therefore
almost all examples are inflected in antepresent, as in (11a) to (11c). The only an-
tepospreterit occurs in (11d).
Complex tenses
(11) a. Quizá nadie ha vivido una Guerra entre naciones.
Maybe nobody aux:apr.in live.pc the war among nations
‘Maybe nobody has.in lived a war among nations.’
As it has been presented up to this point, the epistemic adverb generally occurs
preverbally, and in this case it may occur with indicative or subjunctive verbs, how-
ever if an epistemic adverb appears post verbally, as shown in (12), the nuclear verb
mood is necessarily indicative.
Adverbs such as quizás, tal vez, posiblemente and probablemente can appear in post
verbal position. Consequently, they do not have scope over the verb, which is al-
ways inflected in the indicative. This is illustrated with the examples in (12):
456 epistemic adverbs and mood alternation
Epistemic adverbs may appear immediately after the verb as in (12a) and (12b)
or right at the end of the sentence as in (12c) or between the verb and the adverb
a prepositional complement can appear, as in (12d). In (12b) and (12d) tal vez and
probablemente are between pauses. As for now, we can only state that an epistemic
adverb does not have a specific position when it appears after the verb, which must
be in the indicative.
To sum up, we may identify three syntactic contexts where epistemic adverbs
occur in the sentence; the basic schema in (13) can be further refined as in (14).
position, the verb can be in the indicative or in the subjunctive. With post-verbal
adverbs, the verb must be in the indicative. It is impossible to find a subjunctive
verb co-occurring with an epistemic adverb in post-verbal position.
In order to capture this effect in the logical structure, epistemic adverbs are lo-
cated immediately after the illocutionary force operator. Once this generalization
is made, different contexts in which epistemic adverbs appear can be accounted for
with a single logical structure template, as illustrated in (15).
(15) <IFDEC<EATALVEZ/QUIZÁS/POSIBLEMENTE<TNSPS<STARE/IRR<MODPOS/PRB<<do’
(boys, [cry’ (boys)]) >>>>>>>
Within this template all contexts are expressed in (14) and repeated in (16).
To get a general view, all of them are represented in (15). In this template, I show
epistemic adverbs in sentences and mood alternation.
According to this proposal, there is a very close semantic relation between epis-
temic modality and the realis/irrealis meanings expressed by the indicative and
subjunctive mood, respectively. Therefore, realis is associated with “truth”, as in
(17a), and irrealis with “falsity”, as in (17b).
Epistemic modality is proper of assertions and indicates the grade of commitment
conveyed by the speaker respect to the truth of the proposition. For more detail see, among
others, Lyons (1997), García Fajardo (1997) y Vázquez Laslop (2001).
armando mora-bustos 459
epistemic modality and mood is that of interdependency, that is, the speaker grad-
ually expresses his commitment toward the truthfulness of the proposition, values
which are, at the same time, lexicalized through epistemic adverbs such as quizás
or quizá, tal vez, posiblemente and probablemente, and semi lexicalized by the modal
inflection of the verb (Lyons, 1997). A very important fact to be pointed out is that
this modality type does not represent a one to one correlation with mood forms.
Instead, epistemic modality expresses at least three different meanings: truth, as
in (18a), hypothetical stance as in (18b), and possibility, as in (18c) (Van Valin
& LaPolla, 1997). These distinctions dealing with the speaker’s commitment are
not only expressed in any specific morphological form, but also there is a context
where the main and subordinate verb meanings and their respective morphologi-
cal form interact themselves for generating one specific modal sense.
their semantic nature. For example, epistemic adverbs such as quizá, tal vez, po-
siblemente, acaso, probablemente modify the core as a whole. The interrelation be-
tween status operator and epistemic adverbs is very close. The sense of doubt, pos-
sibility or probability among these adverbs corresponds to the irrealis meaning
of the subjunctive. That is, status operators and epistemic adverbs express similar
semantic notions. For example, in (19), the epistemic adverbs motivate the sub-
junctive mood on the verb. In this particular case, the sense of doubt carried by the
adverb is emphasized by the irrealis meaning of the subjunctive.
I understand by emphasis the fact that the meanings denoted in the inflected
verb form and in the lexical adverbial forms double the irrealis value. In the cases
shown in (19), the meaning of the verb inflection confirms the meaning denoted
by the adverb as default. In other words, the meaning expressed by the adverb is
absolute; consequently, the verb inflection can only stress that meaning.
On the other hand, if the adverb is on post verbal position the sentence will
denote one sense of doubt or possibility too. For this reason, epistemic adverb
position, as in (19), is the starting point to believe that epistemic adverbs are not
syntactic-semantic elements that lead the modified verb to be necessarily inflected
To determine the adverb’s scope, see Jackendoff (1972), Lakoff (1974), Fuentes (1991)
Mood alternation and adverb position lead me to suggest that these adverbs
are capable of highlighting the uncertainty or doubt modal meaning of the con-
struction as a whole. In (20), adverbs are projected or have scope over the whole
sentence. Doubt or uncertainty meanings generated by these adverbs therefore
extend over the core. In other words, the adverb meaning leaves without effect
the senses of the indicative morphological form inflected in the verb; the inflected
subjunctive form merely emphasizes this very last semantic value expressed by the
epistemic adverb in these cases.
When mentioning that doubt meaning leaves without effect the realis indica-
tive mood expressed by the matrix verb inflection, I imply that the meaning of the
modal periphery adverbial causes the loss of the realis meaning associated with
the indicative. This is due to the scope that these adverbs have within the clause
structure.
Now, I discuss the correlation between status and modal adverbs in the logical
structure. As mentioned before, the semantic features realis–irrealis, negation and
epistemic modality are contained within the status operator. An issue not com-
pletely solved by Van Valin & LaPolla (1997) is the representation of this operator
in the logical structure. I agree with the authors who place the mood operator con-
taining realis/irrealis values within the logical structure, but for Spanish, such rep-
resentation needs to be modified. As shown before, indicative-realis forms and sub-
junctive-irrealis forms can be combined with different modal senses. Realis forms
can appear along with meanings of certainty, belief and doubt in different contexts.
For that reason, I suggest that when representing the mood operator within
the logical structure, the values of epistemic modality must also be taken into ac-
armando mora-bustos 463
count. For example, the state of affairs expressed in the core of the constructions in
(21) conveys a realis sense; the main verbs are inflected in indicative mood but the
sense of the modality in each of these constructions is different, so that realis fea-
ture combines with a different modal meaning: realis-truthful, as shown in (21a),
realis-possible as in (21b) and realis-hypothetical as in (21c).
From examples in (21) and (22), we may assume that the illocutionary force,
the mood and the modality are neutralized when the main verb expresses a mean-
ing of doubt, desire, ignorance or uncertainty. This neutralization would be repre-
sented in the logical structure by the irrealis.
I would like to refer now to the logical structure in which mood alternation is
presented, such as in (23).
References
1. Introduction
Following Jolly’s work (1993) on preposition assignment, there are three types
of prepositional phrases (PPs) in Role and Reference Grammar (Van Valin and
LaPolla, 1997; Van Valin, 2005), in terms of their status within the sentence and
the relation they have with the nuclear predicate: 1) Adjunct PPs, which are intro-
duced by a predicative preposition that contributes its semantics to the argument
it licenses. They function as peripheral modifiers of the core, e.g. Mirna had lunch
in the garden; 2) PPs with non-predicative prepositions that mark an argument
directly licensed by the predicate, which function as oblique core arguments, e.g.
Oswald gave a gift to Teresa; and 3) PPs that code an argument of the verb but are
marked by a variable preposition that can contribute its semantics to the argu-
ment. They have the status of argument —adjuncts in the core— Robert put the
CD in /on / behind / under the box.
Following this general schema, on this paper I will outline a more fine-grained
characterization of PPs functions in Spanish, taking into account the semantic
and syntactic relation they have with different kinds of predicates in some particu-
lar constructions. The main goal is to try to show that Jolly’s types can be expand-
ed into sub-classes as a way of covering some more specific functions in which
the PPs can be used, at least in Spanish; I will also try to establish the inclusion of
some ‘problematic’ PPs in this language to the class of the oblique core arguments.
In particular, I will deal with the following issues: Firstly, I will present the case
of PPs in the periphery which do not modify the whole core, as they only have
scope over a part of it. Although these PPs are predicative and define the seman-
tics of the participant they introduce, because of such different scope, they have
a different logical structure (LS) from the one the standard core peripheral PPs
have. They can be considered as a sub-type of adjuncts PPs.
469
470 prepositional phrases in rrg. a case study from spanish
In relation to this kind of data, it is argued, in the first place, that the PPs in
(1a) and (1b) both fit, in general terms, in the oblique core argument category.
Secondly, it is shown that, although they seem to be similar, they fulfill different
functions and are not to be confused: They represent two subtypes of oblique core
arguments.
The organization of the work is as follows: In the second section, I will ap-
proach the adjunct PPs types; in the third section, I will deal with the argument-
sergio ibáñez cerda 471
adjunct cases, and in the fourth, I will undertake the analysis of the different kinds
of oblique core arguments in Spanish. Finally, some last conclusions are presented
in the fifth section.
Jolly (1993), Van Valin and LaPolla (1997) and Van Valin (2005) identify one type
of adjunct PP. This is exemplified in (2):
In this example, after work is a PP that codifies an element that is not seman-
tically required by the predicate; hence, it is an adjunct of adverbial category. It
belongs to the periphery and from there it modifies the core of the clause, that is,
the predicate and its arguments, as the whole event implied by baked takes place
under the temporal axis denoted by the PP. Here, the preposition after functions as
a two-argument predicate, one of them being the whole core and the other one the
NP which the preposition introduces to the clause. A standard RRG representa-
tion for (2) is the logical structure (LS) in (3):
(3) be-after’ (work, [[ do’ (John, Ø)] CAUSE [BECOME baked’ (cake)]])
Beyond this type of PPs, usually called ‘setting’ PPs, there are other kinds of
locative PPs in Spanish that can not take the clause in which they are modifiers as
an argument, because they have a more ‘local’ scope, even though they are adjuncts
in the periphery, as they are not arguments of the verb and are headed by predica-
tive prepositions. This kind of PPs is exemplified in (4):
Here, the PP sobre la tabla de madera refers only to the local placement where
the change of state of the cake takes place; more precisely, the wooden board is the
place where the event of ‘dividing the cake into pieces’ happens. Certainly, Juan is
doing something that involves the wooden board, but clearly, he is not located on
472 prepositional phrases in rrg. a case study from spanish
the board itself. Consequently, the semantic scope of the PP is not over the core
but only over a part of it. This way, we can consider PPs as the one in (4) as partial
modifiers of the core. A possible representation for (4) is shown in (5):
In constrast to the LS in (4), here the adverbial predicate be-on’ only has scope
over the sub-event of change of state, leaving the activity sub-event out. Now, be-
yond the case of a sentence like (4), the fact that adjunct PPs can modify structural
elements which are not necessarily the whole core, can be seen in a sentence with
an added instrumental complement:
In this example, not only the change of state is under the scope of the loca-
tive PP, but part of the action performed by the effector does fall inside its scope:
John’s knife manipulation and, more clearly, the contact of this instrument with
the cake, takes place on the wooden board. This fact can be represented in the
following LS:
(7) [do’ (Juan, use’ (Juan, cuchillo))] CAUSE [[be-on’ (tabla, [do’ (cuchillo, [cut’ (cuchillo,
pastel)] )])] CAUSE [BECOME cut’ (pastel)]]
Here, only one of the activity sub-events, the one that has the instrument as
an effector, is under the scope of be-on’, while the more external one, the one in
which Juan is the effector, falls out of the scope of the preposition.
To Sum up, the PPs in (4) and (6) have a different behavior from the one exem-
plified in (2). The three are structurally different, as it is not always the case that,
in their LS, the whole core of the clause they appear in is one of their arguments.
Nevertheless, all of them have in common the fact that they clearly do not codify
verbal arguments, that they are introduced by a predicative preposition and that
they are modifiers in the periphery. In this sense, they represent different cases of
adjunct PPs.
sergio ibáñez cerda 473
The discussion now moves towards the analysis and comparison of the benefi-
ciary–recipient PPs and the goal PPs. In the standard RRG analysis of English
prepositions (Van Valin and LaPolla, 1997), these two are assumed to have the
same status as argument-adjuncts in the core. In what follows, I will try to show
that in the case of their equivalents in Spanish, this is not so, and that they have a
different identity.
In section 3.1 I will undertake the analysis of the beneficiary PPs. In section
3.2 the case of the goal PPs will be presented, and in section 3.3 I will extend the
analysis to two other types of Spanish PPs.
As mentioned before, the RRG common analysis considers that the beneficiary-
recipient PPs of a sentence like Robin baked a cake for Sandy is an argument-ad-
junct. This is so, basically because the predicative preposition that heads it has a
LS which has an argument that is also an argument of the predicate that functions
as nucleus, as can be seen in the representation in (8), as proposed by Van Valin
and LaPolla (1997:383):
(8) [[do’ (Robin, Ø)] CAUSE [BECOME baked’ (cake)]] PURP [BECOME have’ (Sandy, cake)]
In this LS, the PP for Sandy is represented by the segment introduced by the
PURP operator, which basically has the purpose or finality value. This operator
has a sub-structure where there are two arguments: One of them, Sandy, the ben-
eficiary, is the participant being introduced to the sentence by the preposition;
the other one, cake, is the semantic argument shared with the LS of the verb. This
structural characteristic differentiates this kind of PPs from the adjunct ones and
that is why they are labeled argument-adjuncts in the core.
There are, however, some problems if we transport this proposal to the case of
the ‘equivalent’ PPs in Spanish, as can be seen if we look at them closer. First of all,
the beneficiary-recipient PP introduced by para does not behave as a core argu-
ment in relation to its capacity to be controller of ‘pro-drop’ arguments in coordi-
nated clauses and in non-finite final subordinate clauses, as the examples in (9)
474 prepositional phrases in rrg. a case study from spanish
show. This behavior is a standard test for identifying arguments that belong to the
core. In RRG, the omitted elements in such constructions are called pivots.1
e. Juan i envió una limosna para Pedro y María j para __ i /*j ayudar a los pobres.
‘John sent a donation for Peter and Mary to help the poor.’
In these examples, it can be seen that the recipient-argument of a verb like dar
‘to give’, an inherent participant of the act of giving, that is a core argument, can
easily control the reference of the pivot in coordinated clauses. In (9a), clearly, it is
Pedro who looks fine in the suit. Even in non-finite final clauses, as in (9b), the re-
cipient outranks the effector as the controller of the pivot. In contrast, the partici-
RRG get rid of categories like subject and direct object, notions that, accordingly to numer-
ous studies, are not universally valid. As an alternative, this theory proposes the use of the
category of privileged syntactic argument (PSA), which refers to each argument that in a
particular construction, in a particular language, has access to the majority of the most im-
portant syntactic behaviors. The two most important functions that a PSA can have is to be
a controller and to be a pivot. A controller is the syntactic element that in a given sentence
controls the reference of an omitted element in a coordinated or a subordinate clause. A
pivot is the function that bears the omitted element. It is important to consider that the
PSA functions can be split among the various arguments in the clause. In this way, while in
a given sentence an argument can control the verbal agreement, another one can control the
reference of an omitted element in a subordinate clause. Most importantly, it is the case that
in the majority of the languages, it is a requirement for a participant to be a core argument in
order to be a controller or a pivot.
sergio ibáñez cerda 475
pant introduced by the preposition para cannot control the pivot in a coordinated
clause: The example (9c) does not mean, at least in the first unmarked reading,
that Pedro looks fine in the suit. The reference of the pivot is controlled by the
undergoer un traje and this is the one that looks fine; just the same, in the example
(9d) it is Juan who likes the project and not Pedro. Finally, when a beneficiary ap-
pears in a sentence with a final clause, it is always the effector which has priority as
the controller of the pivot of that clause, as in (9e): Here, Pedro y María are more
likely to be co-referential with the undergoer los pobres of the not finite verb; it is
Juan who is helping them.
These data suggest that in a marked difference with the recipients of transfer-
ence verbs, the beneficiary introduced by para does not behave like a core element.
That this is true is confirmed by the fact that Spanish has an alternative mechanism
for making core arguments out of the beneficiaries, as they cannot be projected as
inherent participants from the LS of any kind of verbs. This mechanism is the
dative or indirect object construction, where the beneficiary appears introduced
by the preposition a and in which it can be duplicated by the clitic pronoun le:
What is interesting about these examples is that in all of them the beneficiary
functions as a controller of pivots, in both the coordinated clauses and the non-fi-
nite final clause. As opossed to the sentences in (9), it is now the case that in (10a),
at least in the first reading, it is Pedro who fits the suit well. All the same, in (10b)
it is Pedro who is preferred as the antecedent of the dative pivot of gustar. Finally,
in (10c) the PP a Pedro y a María is controlling the pivot of the final clause; in
this case, they are the ones who are helping the poor, although, cleverly, I might
say, with Juan’s money.
We can say, then, that these PPs introduced by the preposition a and duplicated
by the clitic le are core arguments: they can be controllers and, importantly, their
476 prepositional phrases in rrg. a case study from spanish
2
The idea that this kind of indirect object construction is the result of a voice process
is further developed in Ibáñez (2004) and (2008). Basically, what is argued in those works
is that the canonical way of coding the beneficiaries is by means of the preposition para.
If they appear introduced by a, they need to be reduplicated by the clitic le, as it is shown
by the agrammaticality of *Juan construyó una casa a María ‘John built a house for Mary’.
This way, the beneficiary-indirect object construction is a derived one. It supposes an ar-
gument modulation, that is, the ‘promotion’ of a peripheral argument to the core, a process
that is marked by the simultaneous presence of a and le; hence the appearence of both these
particles is determined by the argument modulation constructional scheme.
sergio ibáñez cerda 477
Now, as it has been said before, along with the recipient-beneficiary PPs, the goal
PPs of putting verbs —poner ‘to put’, colocar ‘to place’ meter ‘to put inside’, etc.— and
intransitive motion verbs —ir ‘to go’, llegar’ to arrive’ and entrar ‘to go in’, etc.— are
considered to be a case of argument-adjuncts. Nevertheless, as I will show below,
they have a different semantic and syntactic identity. A first important difference
is that goal arguments do behave as typical core arguments; that is, they can con-
trol the reference of pivots in coordinated clauses, as it is shown in (11):
In (11a) both Juan y la casa nueva could be the controller of the pivot in the
coordinated clause. Some speakers even prefer la casa nueva as the controller. The
case is even clearer with putting verbs, because, as they inherently imply a kind
of transference of a theme, the ‘focus’ of the process is in the change of place sub-
event. This way, in (11b) the effector Juan is not even a competitor for the pivot
controller function; the competition is between the theme and the goal. Examples
(11c) and (11d) show that this behavior is not permitted to peripheral locative
PPs. These data suggest that in a marked difference with the beneficiary PPs in-
troduced by para, the goal PPs of change of place verbs are core arguments. They
do not have the same status.
The syntactic differences between the goal PPs of change of place verbs and
the beneficiary PPs that appear with transitive verbs are anchored in a more im-
portant difference, a semantic one: Goal arguments are an inherent part of the
verb meaning, beneficiaries never are. There is nothing in the semantics of a verb
478 prepositional phrases in rrg. a case study from spanish
like to bake that makes us predict that the event linguistically described with that
item implies a beneficiary. These are freely added to the linguistic conceptualiza-
tion, and when they are not coded, they do not play a single role in the interpreta-
tion of a given sentence.
This again marks a big difference with the goal PPs of change of place verbs.
Beyond the fact that they can be left out of the projection without yielding un-
grammatical sentences, goals strongly tend to be coded. In a corpus study (Ibáñez
2005) of Spanish intransitive goal motion verbs like ir ‘to go’, venir ‘to come’, llegar
‘to arrive’, and entrar ‘to go in’, among others, it was found out that goals appear
with the verbs they are implied by in a range of 70% to 90% of the cases, depending
on the verb. More importantly, even when they are not coded, they still play a role
in the interpretation of the sentences they appear in; they are clearly recoverable
from the linguistic or situational context, and most of the time, they must be recov-
ered in order to allow a complete interpretation of the sentences in case. Following
this, we can say that only when an argument is not recoverable from the context it
has to be explicitly coded. In this sense, it is really this function of being an indis-
pensable element in the process of interpreting a clause that gives a certain partici-
pant its argument status, independently of its coding as an explicit complement.
Van Valin and Lapolla (1997) and Van Valin (2005) claim that the goal of put-
ting verbs is optional because it can be projected in the form of an adverb as in
John put the book down. Nevertheless, in the sense posited above, goals are always
obligatory arguments. What is optional is the codification of that goal as a PP or
as an adverb. But even in this last case, it still functions as an argument; that is why
adverbs can function as a pro-form substituting the goal PPs:
This fact shows that the correlation between adverbs and adjuncts does not
have to be a perfect match. It is true that linguistic categories tent to specialize for
sergio ibáñez cerda 479
certain uses: Names are mostly used as arguments, verbs as predicates and adjec-
tives, and adverbs as modifiers. But in all languages there are plenty of mismatch-
es. For example, it is not rare to find names (or NPs) functioning as adjuncts, e.g.,
Juan nadó el domingo ‘John swam on Sunday’. All the same, adverbs are mostly
used for coding the circumstances that surround or that modify the states of affairs
denoted by predicates, but this is not always the case. The examples in (12) show
that the PPs can indeed be substituted by adverbs. More importantly, one should
be aware of the fact that even when the sustitution is not formally realized, a PP is
always potentially replaceable by an adverb. Basically, in categorical and paradig-
matical terms, they are the same.
Goal arguments, then, at least in Spanish, share some important semantic and
syntactic characteristics with the recipient arguments of verbs like dar ‘to give’: a)
They are inherently implied by the verbs they appear with; b) they are obligatory
and c) they can be controllers of pivots. Based on this, we can say that they have
full status as oblique core arguments.
This argumentation is valid for the cases where the goal arguments are coded
by means of a PP introduced by the preposition a ‘to’, in the case of intransitive
motion verbs, and by the preposition en ‘in/on’, in the case of change of place verbs.
These prepositions are the canonical ones with these verb classes. In this respect,
the corpus study that is presented in Ibáñez (2005) shows that the goal PPs of
intransitive motion verbs tend to be coded with the preposition a in over 90% of
the cases. A similar study in process, preliminarily shows a same pattern for the
PPs of the change of place verbs: They strongly tend to be introduced by en in the
corpus examples. It is possible to say that these prepositions are systematically as-
signed, at least in the unmarked cases, within the following structural contexts: a
appears when the LS of a predicate has a BECOME pred’ (z, y), which is the case
of motion verbs, and en shows up in the structural environment characterized by
the present of INGR pred’ (z, y), which is the case of ‘putting verbs’. The basic dif-
ference between these two is that the motion verbs as ir ‘to go’ imply a change of
place with a displacement or a movement on focus or as part of the semantics; that
is why they are durative in aksiontsart terms and are characterized by the presence
in their LS of the BECOME operator. On the other hand, a verb like poner ‘to put’
has an inherent punctual aspect. It does not imply a displacement and that is why
its LS has the INGR operator in it.
Now, the goal PPs of these two types of verbs can alternatively be projected
with other prepositions beyond the canonicals a and en. These alternative preposi-
480 prepositional phrases in rrg. a case study from spanish
tions such as hacia ‘towards’, dentro ‘inside’, detrás ‘behind’, debajo de ‘under’, etc.,
are predicative, since they have an inherent meaning which plays a role in deter-
mining the referential meaning of the goal argument in case, as the examples in
(13) show:
We can see in these examples that prepositions such as dentro ‘inside’ and detrás
‘behind’ specify the goal referent with more detail than the preposition en. They
add their own semantics and it is in this sense that they are predicative. The ex-
ample in (13c) is a little different. The preposition atrás ‘behind’ does not specify
the same meaning of en; when something is behind an object, it is not on/in it.
Rather, a preposition like atrás establishes the referent of its object as a point of
reference that serves to construct the referent of the goal: ‘Behind the box’ denotes
a place that is not the box, but a place that can be identified with the box as a point
of reference.
Nevertheless, although this kind of prepositions are clearly predicative, the pos-
sibility to use them to introduce the goal arguments of the change of place verbs is
not entirely free, as it is the case of the locative PPs that have the status of adjuncts.
Poner ‘to put’, the predicate that serves as the nucleus in the sentences exempli-
fied in (13), is the hyperonime in the domain of change of place verbs. As such,
it has a very general and abstract locative meaning; it does not internally specify
that much about its goal and that is why it can be used with goals introduced by
almost any locative preposition. But this is not the case of other verbs that in terms
of their own inherent semantics can prevent the use of certain prepositions to code
their goals, as the examples of (14) show:
What we can see here is that besides the fact that the preposition in case can
add a portion of meaning to the argument, such preposition is determined by the
predicate it appears with, and hence, it is a function of that predicate.
Importantly, these PPs with non-canonical prepositions seem to behave as core
arguments: They still can control pivots in coordinated clauses:
(15) a. Paseando por el bosque, Juan i llegó hasta a la vieja cabañaj y aún i / j se veía bien.
‘Walking through the forest, John got to the old cabin and (it) still looked fine.’
The first more accessible reading for (15a) is the one in which it is the cabin
that looked fine. Much the same, in (15b) it is the fish bowl that looks good.
Therefore, these data show that in the case of the goal arguments introduced by
non-canonical prepositions, we do have a case of a predicative PP which functions
as a core argument. That is what is properly called an argument-adjunct in the
core in RRG. This situation clearly contrasts with the case of the beneficiary PPs.
These last ones are argument-adjuncts in the periphery.
482 prepositional phrases in rrg. a case study from spanish
Other cases that can render the distinction between argument-adjuncts in the pe-
riphery (as opposed to adjuncts) and argument-adjuncts in the core (as opposed
to oblique core arguments) productive are the PPs in the passive construction and
the PPs in what is usually called the anticausative or inchoative construction. Both
are exemplified in (16):
(16a) and (16c) are canonical transitive sentences, while (16b) and (16d) are
the corresponding passive and anticausative versions of the formers, respectively.
On one hand, as one can see, the actor of passive construction, an agent-effector,
is coded as a PP introduced by the preposition por. On the other hand, the actor
of the anticausative construction, a cause-effector, is coded as a PP introduced by
con ‘with’. Following the standard treatment in RRG for this kind of phenomena,
we can say that the assignment of these prepositions is ruled by constructional
schemas that specify the characteristics of each one of these voice modulations.
In RRG terms, both are PSA and argument modulations,3 because both sup-
pose a marked assignment of the PSA function to the undergoer and a marked
coding of the actor as a PP. The difference between the two, for our purposes, is
that beyond losing its PSA status and being marked as an oblique, the actor in
the passive voice can still function as the controller of a pivot in a non-finite final
subordinate clause, as can be seen in the examples (17a) and (17b) below, while
3
In RRG, a PSA modulation is a morpho-syntactic process that has as a result a change
of the PSA. This is the case of the passive voice, which implies that the actor argument los-
sergio ibáñez cerda 483
(17) a. La ciudadi fue atacada por el violento ejércitoj para __* i / j saquearla
‘The city was attacked by the violent army in order to sack it.’
es its syntactic privileges as the controller of the verbal agreement and as the bearer of the
nominative case in favor of the undergoer. In contrast, an argument modulation is a process
that implies a change in the morpho-syntactic properties of one or more of the arguments,
but without taking out the PSA role from the argument which originally bears it.
484 prepositional phrases in rrg. a case study from spanish
(18) a. A las tres de la tarde llegaron los trabajadoresi. A las 5, la puertaj fue desmontada
para__ i /* j pintarla.
‘At three in the afternoon, the workers arrived. At 5, the door was taken apart
in order to paint it.’
b. la puerta se abrió sola.
‘The door opened by itself.’
As it has been noted in the literature before, when the actor is not coded, the
anticausative construction can have a spontaneous meaning, as in (18b). In this
version of the construction, the cause has been completely removed, not only in
syntactic terms, but also in semantic ones, and does not play any function at all.
To sum up, on one hand, passive actors and goal PPs, as argument-adjuncts in
the core, have in common the fact that they are inherently required by the verb se-
mantics and that they can function as controllers of pivots; they differ in the sense
that the preposition which they are introduced by can be predicative (non-canoni-
cal prepositions of goals) or non-predicative (passive actors). On the other hand,
the cause PP of the inchoative construction and the beneficiary-recipient intro-
duced by the preposition para have in common the fact that they do not function
as pivot controllers and that they do not have the clause as one of their arguments;
They are in the periphery but they are not completely adjuncts.
The category of oblique core argument in Spanish subsumes at least three differ-
ent classes of PPs. The first type is the case of the recipient argument of transfer-
ence verbs like dar ‘to give’ and ofrecer ‘to offer’, the addressee of saying verbs like
decir ‘to say’ and contar ‘to tell’, and of the source PPs of verbs like sacar ‘to take out’
and quitar ‘to remove’. As Spanish is a language that does not allow more than two
direct core arguments, the third arguments of three-place predicates are always
coded as PPs; hence, they are oblique core arguments. As shown in section 3.2, this
is also the case of the goal arguments of change of place verbs like poner ‘to put’.
A second class of oblique core arguments is the one that groups together the
PPs that are introduced by the preposition con ‘with’. In a very similar fashion as it
is in English, these PPs code, in general terms, arguments that are competitors for
sergio ibáñez cerda 485
the macro-role assignment but are not selected for it ( Jolly, 1993; Van Valin and
LaPolla, 1997).
These two types of oblique core arguments have in common the fact that the
preposition they are introduced by is systematically assigned and can be predicted
from specific structural environments: Recipient and addressee PPs get their prep-
osition a from the context characterized by the presence of a BECOME pred’ (z,
y); sources get de from the presence of a BECOME NOT pred’ (z, y) in the LS;
and, as mentioned before, goals of change of place verbs get en from an INGR
pred’ (z, y) context.
A third kind of oblique core arguments in Spanish corresponds to PPs usually
grouped together under the category of ‘suplemento’ in the Hispanic linguistic tra-
dition (cf. Alarcos 1968, 1994). These are exemplified in (19):
(20) a. Atreverse ‘to dare’, oler ‘to smell like’, recurrir, ‘to resort to’,
renunciar, ‘to renounce’, aspirar ‘to aspire’, atenerse ‘to rely on’.
b. Carecer, ‘to lack’, abstenerse ‘to abstain’, abusar ‘to abuse’, constar,
‘to consist of ’, desistir ‘to desist’, desconfiar ‘to distrust’.
Among the items on these lists, there are activity predicates and state verbs. It is
possibly true that the verbs that are used with the preposition a tend to be activi-
ties and the verbs that are used with de usually correspond to state predicates, but
this is not an exhaustive condition. Besides this, it is commonly assumed that these
verbs are learned along with their preposition in the acquisition process. In some
way the preposition seems to be part of the verb. That is why these items are com-
monly referred to in the Spanish grammars as ‘verbs that govern their preposition’.
So, although they are non-predicative, the assignment of these prepositions seems
to be more of an idiosyncratic lexical phenomenon than something systematically
ruled. It is likely that these prepositions have to be listed in the lexical entry.
Another factor that points out the idiosyncratic identity of this type of PP
marking is that a and de, although they are the most commonly used, are not the
only prepositions that appear with this kind of verbs. Some of these verbs can have
their second argument coded by por as abogar ‘to advocate’, by con, as contar ‘to
count on’, and by en, as consistir ‘to consist’.
To sum up, the PPs in the examples of (19) above, code semantic arguments
that are introduced by non-predicative prepositions. They are, then, oblique core
arguments, but they are a special case, different from the two other types men-
tioned above, because their preposition is not assigned by a systematic rule; it is in
some way lexically determined.
Now, although in a general level all the PPs in (19) have the same oblique
core argument status, and although they have always been grouped together in
the Hispanic linguistic tradition, it seems that, in a closer look, they belong to two
different sub-types of oblique arguments. Let us look at these examples again, but
now with the added counterparts that are shown in (21):
This way, it seems that the alternation in case is a voice phenomenon that does
not imply a PSA modulation, but only an argument modulation. In this sense, we
can say that the assignment of the preposition takes place through the application
of a voice constructional scheme during the linking. The problem with this way of
solving this situation is that, even though the preposition de is the one used with
the majority of the verbs that appear in this construction, other prepositions like
en and con can also be used with certain verbs —cumplir con ‘to comply’, creer en
‘believe in’—. This suggests that the phenomenon is quite idiosyncratic and that
the assignment of the preposition is determined by something more than the con-
structional scheme. This situation calls for a more detailed analysis, something
that goes beyond the scope of this work.
To Sum up, then, the PPs that appear within this voice construction type code
a semantic argument in the core and are introduced by a non-predicative preposi-
tion assigned through the application of a constructional scheme in the linking. In
this respect these PPs are quite different from those in the (a) and (b) examples
of (20) and (22). Nevertheless, they are all oblique core arguments in RRG terms,
just as the recipient, the addressee, the goal and the source PPs of three-place
predicates are.
5. Conclusions
This work constitutes a first attempt to show that, in general terms, the RRG no-
tions of predicative adjunct PP, argument-adjunct PP and oblique core argument
PP are useful for covering a broad range of PP types in Spanish. Nevertheless, it
also tries to show that these notions can be, in the same RRG terms, ‘expanded’ in
subcategories, in order to capture more specific differences among the PPs in this
language. In particular, it has been proposed that: a) There can be PP adjuncts in
the periphery modifying other structural levels than the core; b) the argument-ad-
junct category, one that already covers frontier cases, can be split into argument-
adjuncts in the core and argument-adjuncts in the periphery in order to cover
some PPs in Spanish that do not fit the argument-adjunct category as it stands,
and c) there are different kinds of PPs that belong to the same general oblique
core argument category.
sergio ibáñez cerda 489
References
1. Introduction
Most current studies on Spanish datives are primarily concerned with the syntac-
tic and semantic aspects of the constructions in which these arguments may inter-
vene, but much less importance has been given to the pragmatic function of the
grammatical alternatives by which datives are encoded. ��������������������������
This article examines the
three most frequent ones; namely, the cases where (i) the dative is encoded exclu-
sively by a lexical phrase; (ii) the dative is encoded exclusively by a clitic; and (iii)
the dative is encoded via a so-called “clitic doubling” construction.
In which natural contexts does each formal alternant occur? What does it serve
to express? How are the alternants functionally related? Are they related with the
encoding alternatives available for accusative arguments? It is these questions
which this paper seeks to address. Most of the data will be drawn from a corpus
of oral interactions among native speakers of the Buenos Aires dialect, to which I
will refer as the “Buenos Aires” corpus (Barrenechea 1987), as well as from previ-
ous studies based on Mexican Spanish. The analysis will be formalized within the
model of Role and Reference Grammar, as presented in Van Valin (2005) and Van
Valin and LaPolla (1997).
Let us start by defining so-called “dative doubling” constructions. As under-
stood in this paper, these constructions are those in which a lexical dative argu-
ment in canonical position co-occurs with a dative clitic, e.g. le, les or its allomorph
se, as in the following examples.
491
492 spanish datives
The dative clitic inflects for number but not for gender, as illustrated in (1a)
and (1b). The allomorph se occurs when the third person dative and accusative
clitics co-occur. This form is invariable for both number and gender (1c). On the
other hand, lexical datives are always marked by a.
Spanish datives are often ascribed to one of two basic groups. Following De-
monte’s (1994) classification, the first one involves those datives occurring with
verbs of transfer, such as dar ‘give’ or donar ‘donate’, whereas the second concerns
datives associated with verbs of creation such as hacer ‘make’ or cocinar ‘cook’. This
distinction is relevant to our purposes because even though both groups allow clit-
ic doubling constructions, only verbs of the first type may be associated with a
“recipient” encoded exclusively as a lexical phrase (Strozer 1976, Demonte 1994,
1995, among others), as illustrated in (2b).
For verbs which do not belong to this group, on the other hand, the alternative
to a lexical dative is a prepositionally marked adjunct:
Based on cases like (3a), it has been proposed that the Spanish dative clitic
should be analyzed as an applicative morpheme, licensing the occurrence of a third
In distinguishing different types of datives, some classifications rely on the inherent se-
mantics of the predicate (e.g. Demonte 1994), and others on its syntactic transitivity (e.g. Or-
dóñez 1999). For a survey of different approaches and the issues involved, see Company (2006).
valeria a. belloro 493
participant with argument status (Cuervo 2003, Ibáñez 2003). This is consistent
with the fact that, with verbs like cocinar, the occurrence of a lexical dative in fact
requires the co-occurrence of the clitic (4a), which is in turn incompatible with a
coreferential interpretation of a prepositional phrase (4b).
One of the challenges this hypothesis faces is explaining what the function of
the dative clitic is when it is associated with verbs like dar, since in these cases the
third lexical argument seems to be able to occur independently of the presence
of the clitic (cf. (2) above). Current analyses propose that the a marking the third
participant in those cases should be analyzed as a preposition akin to the one in
(3b). Hence, these sentences would not involve a nominal dative argument, but
a prepositionally marked oblique (Demonte 1995, Cuervo 2003; but see Strozer
1976, Suñer 1988, Campos 1999).
On the other hand, in analyzes based, for instance, on the Principles and Pa-
rameters framework, it is normally assumed that in the instances where there is a
dative clitic but no coreferential phrase, the argument is realized by a phonologi-
cally silent category (i.e. pro).
Note that, together, these two assumptions justify the common claim that da-
tive doubling is always obligatory (Masullo 1992, Demonte 1995, Cuervo 2003,
among others). This is so because, under these approaches, a structure such as (5a)
constitutes a case of dative doubling on a par with (5b), whereas examples as (6)
are excluded as alternants since they are not considered to involve a dative-marked
NP but a prepositional phrase (cf. e.g. Cuervo 2007:587).
For simplicity, both “doubled” and “not doubled” a-marked lexical phrases will be re-
ferred to as NPs, in order to distinguish them from phrases marked by para or other more
contentful prepositions, which cannot be clitic doubled.
valeria a. belloro 495
called “linking algorithms”, which are meant to represent the production (i.e. “se-
mantics-to-syntax”) and comprehension (i.e. “syntax-to-semantics”) aspects of
language use. Besides semantic and morphosyntactic categories, the linking algo-
rithms make reference to discourse-pragmatics ones, and it is these which I will
focus on here.
Following Lambrecht (1994), RRG recognizes two aspects of discourse-prag-
matics as influencing sentence structure. The first concerns the pragmatic rela-
tions established among referents in terms of their communicative dynamism, and
is expressed in a “focus structure” which segments the morphosyntactic string into
topical and focal components. An important distinction in the theory is that be-
tween “potential focus domain” (PFD) and “actual focus domain” (AFD). The first
refers to the syntactic domain where focus may fall, whereas the latter targets the
actual focal elements in a specific utterance. This distinction makes possible to
capture, for instance, the motivation for word-order differences that may be ob-
served between a language such as Spanish, in which the PDF is assumed to co-
incide with the whole clause and, say, Italian, where preverbal elements inside the
core are banned from bearing focus. Compare the structures in Figure 1.
The examples in Figure 1 show that in Spanish it is possible to adapt the focus
structure to an unmarked word order, with the result that the first element in the
core may in fact be topical (a) or focal (a’) (Zubizarreta 1999: 4225). This is not an
option in Italian, where focal status cannot be assigned to the core-initial position.
Thus, in Italian it is the syntax (word order) which has to adapt to the focus struc-
ture. The difference between the two languages can be accounted for in terms of
the potential focus domain in each language, as the comparison of the representa-
tions illustrate.
Besides the pragmatic relations manifested by focus structure, the second in-
formation-structure aspect which RRG recognizes to influence morphosyntax is
the pragmatic properties of the discourse referents at a given point in the com-
municative exchange. By “pragmatic properties” is meant the degree of easiness
or difficulty that the addressee may have in accessing or building a representation
of the referent the speaker denotes. This is of course dependent not only on the
addressee’s knowledge, but also on their attentional state. Entities which are as-
sumed to be known by the addressee are considered “identifiable”. Those which
are further in the focus of the addressee’s consciousness are “active”. “Accessible”
referents are those which are peripherally activated (e.g., via their prior mention
a few clauses back in the text), and “inactive” referents are those which are not fo-
496 spanish datives
cally nor peripherally active at the current point in the interaction. Building on
the works by Prince (1981), Chafe (1987) and Lambrecht (1994), among others,
RRG recognizes the five cognitive states illustrated on the third row of Figure 2
(from Van Valin and LaPolla 1997: 201).
The relationship between pragmatic function and cognitive states, in turn, can
be expressed by a scale of topic acceptability. This scale (from Lambrecht 1994:
165) indicates that active referents are the most acceptable topics, with more
marked choices as the referents’ activation decreases.
Active > Accessible > Inactive > Brand-new anchored > Brand-new unanchored
Finally, the emphasis may be put more directly on the relationship between
form of encoding and cognitive state. An illustration of this approach is presented
in Figure 5 (from Gundel et al. 1993: 275), which captures the relative accessibility
associated with different (pro)nominal forms:
The interaction between these different scales captures the expected correla-
tions between attenuated forms (e.g. zeroes, bound pronouns), sentence topics (i.e.
constituents denoting entities about which some new information is asserted), and
active referents (i.e. entities which the speaker assumes are the current focus of at-
tention for the hearer). Likewise, there is an expected correlation between more
complex, semantically richer forms (e.g. modified nouns), sentence focus, and ref-
erents which are not assumed to be currently active in the mind of the addressee.
It is however important to distinguish between the three factors, since the antici-
pated correlations do not always hold. For instance, it is not necessary for a focal
element to be inactive or unidentifiable (Lambrecht 1994). In the reply in (7), the
NP Eve is focal, as it provides the correct value for the variable in the presupposi-
tion speaker will invite x. Yet, it cannot be claimed to be inactive, since it was just
mentioned in the preceding context:
The mapping between these logical structures and the final morphosyntactic
string, as determined by the linking algorithms, will be illustrated with Spanish
examples in the final section of this paper.
It was mentioned in the introduction that in Spanish there are two groups of verbs
which differ on the morphosyntactic properties of their (potentially) associated
dative arguments: One group, with verbs like dar ‘give’, which allow datives to oc-
cur in “clitic-only”, “clitic doubling” or “NP-only” structures; and the other, with
verbs like cocinar ‘cook’, allowing only “clitic-only” and “clitic doubling”, but ban-
ning the “NP-only” alternative.
The question then arises of what other verbs belong to each type. Intuitions
about the inherent semantics of the predicate do not provide conclusive results,
as the same verb may be adscribed to different classes based on slightly different
criteria. An additional difficulty comes from the fact that different verbs of trans-
fer included in the dar-group exhibit different frequencies of dative encoding of
Abbreviations: acs= accessible, act= active, ina= inactive, m=masculine, pl= plural,
sg= singular.
500 spanish datives
any type. A search on the ADESSE database shows that dar ‘give’ takes a dative
argument in 98.8% of the clauses where it occurs (1328/1344), whereas for ofrecer
‘offer’ this percentage falls to 56.5% (140/248); and for vender ‘sell’ the encoding of
a dative only occurs in 23.1% of the cases (27/117).
Thus, in order to track possible occurrences of datives encoded in exclusive
nominal form in the Buenos Aires dialect (the main variety studied in the research
from which this work is part), I selected the 12 verbs of transfer of knowledge,
possession and information that appeared as the most prototypical, based on the
relative frequency with which a dative argument occurred with them, as well as the
overall number of tokens available. The resulting data are presented in Table 1.
encoding
Alternancias de Diátesis y Esquemas Sintáctico-Semánticos del Español. Developed
at the Universidad de Vigo. http://webs.uvigo.es/adesse/enlaces.html. Accessed June 16,
2007.
The total numbers reflect the Buenos Aires corpus captured in ADESSE. When a
verb returned no instance of dative encoding in this dialect, I replaced it with the next best
alternative.
valeria a. belloro 501
This table shows that, in fact, the most frequent type of dative encoding occur-
ring with transfer verbs in the Buenos Aires corpus is by means of a clitic alone
(154/180=85.6%). Whereas there is some incidence of clitic doubling structures
(23/180=12.8%), NP-exclusive encoding is extremely low (3/180=1.7%). These
frequencies are similar to the ones reported in Weissenrieder (1995), based on
analysis of the novel El beso de la mujer araña, by the Argentinean author Manuel
Puig. The relevant data appears in Table 2 (adapted from Weissenrieder 1995: 173):
An additional search for occurrences of the dative clitic le in the Buenos Aires
corpus confirmed the preeminence of clitic-only structures in oral interactions.
From a total of 780 clauses, this time combining verbs of the two groups, in 76%
of the instances (593/780) the dative argument is minimally realized with a clitic
form.
The prevalence of clitic-only is far from surprising. In effect, there is extensive
evidence that the dative case-role is associated entities which are topical, and have
high degrees of individuation and discourse saliency (Greenberg 1974, Givón
1984, 2001), all factors which justify attenuated forms of encoding. Typical ex-
amples of this class involve instances where the referent denoted by the clitic has
been mentioned in the preceding clause, as in the sentences presented in (9) (the
antecedents appear underlined).
b. Ya he dicho muchas veces que la originalidad no tenía el valor en aquel entonces que se le
atribuye hoy. (hc:xx)
‘I have said many times that back then originality didn’t have the value that is at-
tributed to it today’
502 spanish datives
After clitic-only, the next most frequent type is clitic-doubling. Typical exam-
ples of clitic doubling occur when the target participant was mentioned farther
away in the discourse context, or when it is identifiable but discourse-new. Con-
sider the following examples:
(10) a. ¿Pedro Páramo? Eh... escuchame, Pedro Páramo, mirá--- yo lo leí este año cuando fui a la
facultad. Eh... es la historia de un tipo--- cuya madre al momento de morir--- le dice que
vaya--- a un pueblo donde vive--- Pedro Páramo. Pedro Páramo es su padre, él es hijo de
Pedro Páramo. Entonces el tipo le cierra los ojos a su madre--- y va a ese pueblo. (hc:xxx)
‘Pedro Páramo? Eh… listen to me, Pedro Páramo, look--- I read it this year when I
was in college. Eh… it is the story of a guy--- whose mother at the time of death-
-- tells him to go--- to a town where ---Pedro Páramo lives. Pedro Páramo is his
father, he is Pedro Páramo’s son. Then the guy closes his mother’s eyes (lit. ‘he
closes the eyes to his mother’) ---and goes to that town.’
c. Ayer a las seis de la tarde salí a comprar ....eh... eh... unas cosas ... eh... eh... u... unas
cosas a Alvarito. (hc:xxxii)
‘Yesterday at six in the afternoon I went out to buy…eh…eh… some things…
eh…eh… s… some things for Alvarito.’
Excluding the anomalous utterance in (11c), these examples suggest that NP-
exclusive encoding correlates with non referential entities. This trend is also ob-
served by Ibáñez (2008), who reaches a similar conclusion based on the Mexican
dialect, associating NP-exclusive with the denotation to “non referential generic
entities”.
Data presented by Maldonado (2002: 18), also from Mexican Spanish, can be
interpreted in analogous fashion. Analyzing the phenomenon within the frame-
work of Cognitive Grammar, Maldonado argues that the omission of the clitic
marks a weaker conceptual linkage between discourse-participants. Based on ex-
amples from a corpus of newspaper articles, he notes that this weaker linkage finds
expression in third-person plural-impersonal constructions (12a) and in reference
to generic participants such as institutions, groups, or masses (12b); contexts in
which the clitic can be left out. (The translations have been slightly modified from
their original rendition).
Although Ibáñez and Maldonado do not provide data about the relative fre-
quency of NP-exclusive encoding in the Mexican dialect, results presented by Bo-
gard (1992) indicate that it is also very low, accounting, in his corpus, for less than
8% of the cases (38/491).
504 spanish datives
Note that the correspondence between the two types of markedness relations is
not a given. It does not apply, for instance, in the case of accusatives. Consider the
following Figure:
“Direct objects” are typically associated with new referents and expressed as
nominals. Thus, the series starts from the opposite end, with “least active” and
“NP-only” as the unmarked forms. If the expected correspondence obtained, we
would find the next step up in the cognitive continuum to be expressed by the
next most frequent encoding type. This is, however, not the case, as “accessible”
(i.e. “less active”) referents correlate with clitic-doubling constructions (Belloro
2007), which is formally the most marked type. This “crossing” is partly due to the
interaction between activation and focus structure. Lexical datives and accusatives
in canonical position are normally focal, and focal constituents normally denote
least active referents. In the case of dative doubling, the association between a fo-
cal NP and a least active referent fits nicely with the interpretation that “dative
doubling” constructions occur when the target referent is not as active as expected,
506 spanish datives
and a lexical phrase has to be added to the structure to help establish reference. On
the other hand, in accusative doubling constructions the expected lexical phrase
in focal position encodes a referent that is nevertheless relatively active (specifi-
cally, “accessible”), and this contrary-to-expectation fact is marked by adding to
the structure the accusative clitic. This violation of the default expectations, in ad-
dition to the availability of alternative structures for expressing potentially analo-
gous pragmatic meanings (e.g. topicalization), as well as the long history of pre-
scriptive stigmatization that holds over accusative doubling constructions, are all
factors which conspire for making these structures relatively infrequent in actual
corpora.
The situation with datives, as we saw, is more straightforward: the most fre-
quent alternant is used with the most active referents, the less frequent alternant
with the less active referent, and the least frequent alternant with the least active
referent. Let us provisionally adopt RRG’s three-way distinction between “active”,
“accessible” and “inactive” as corresponding to the three activation levels encoded
by the dative alternants analyzed here. The next question is how dative doubling
constructions and their allosentences should be represented, and how each encod-
ing type could be derived. The next section, therefore, advances a Role and Refer-
ence Grammar analysis of the constructions at hand.
In section §2 it was mentioned that sentence structure may be affected by two in-
formation-structure components; one concerning the cognitive state of discourse
referents and the other dependent on the pragmatic relations (i.e. topic and focus)
established by the phrases used to denote those referents.
We mentioned that even though there are expected correlations between fo-
cus structure and cognitive states, these are not categorical. This was illustrated
with an example of an accessible referent encoded within the focal domain (7).
A similar mismatch may apply in Spanish. It is clear that in clitic-only structures
the dative referent must be active and must belong to the pragmatic presupposi-
tion (i.e. it must be topical), so that the expectations associated with focus struc-
ture and activation status coincide. But this does not need to be so when there is
a lexical phrase occupying the unmarked focus position, as it is the case with clitic
doubling and NP-only structures. In §3 it was argued that the pragmatic distinc-
valeria a. belloro 507
In both the clitic doubling structure in (13a) and the NP-only structure in (13b)
the dative phrase belongs to the focal constituent: in the first case, along with the
rest of the predicate, as this is a “predicate focus” construction. The second exam-
ple involves “sentence focus”, since there is no presupposition. (For definitions and
cross-linguistic examples of the different focus types recognized in this theory
see Lambrecht 1994 and Van Valin 2005; for some Spanish examples see Belloro
2007). Since in both clitic doubling and NP-only structures the dative phrase is
part of the focus domain, a pragmatic analysis of dative alternants in terms of fo-
cus structure can only account for part of the phenomenon, as it cannot be used to
distinguish between clitic doubling and NP-only constructions.
Yet a different partial classification arises with approaches based on affected-
ness, where the division would be established between NP-only (�����������������
[-���������������
affected]) and
non-NP-only alternants (i.e. clitic-only and clitic-doubling, both [+affected]). The
main distinctions captured by a pragmatic approach based on the focus structure
and a semantic approach based on affectedness can be represented as in Figure 8.
Note that most current studies of DO-doubling also take a binary approach, propos-
ing, for instance, that it can only occur if the “doubled referent”, so to speak, is topical (as
opposed to focal), specific (as opposed to non-specific), or discourse-old (as opposed to
discourse-new). Again, whether these features may be invoked to distinguish, with different
degrees of success, between NP-only and clitic doubling structures, neither of them can be
used to motivate the functional distinction between clitic-only and clitic-doubling since,
under this approach, the doubled phrase has the same semantic/pragmatic features associ-
ated with “zeroes”. In other words, clitic-only structures, regardless of whether one consid-
ers the clitic or a pro as the manifestation of the argument, are expected to associate with
508 spanish datives
topical, specific, discourse-old participants just as much as clitic doubled ones. An approach
in terms of cognitive states, on the other hand, allows us to distinguish between the three al-
ternatives available in the grammar. Moreover, it also makes it possible to generalize over the
functional motivations which affect the encoding of both dative and accusative arguments,
since in both cases doubling constructions mark the deviation of the target referent with re-
spect to the activation level typically associated with the case-role chosen to encode it: more
active than expected for accusatives, less active than expected for datives (Belloro 2007).
valeria a. belloro 509
a (simplified) version of the linking algorithms that govern the mapping from
semantics to syntax in Spanish, and which incorporate the pragmatic information
determining the appropriate encoding for the dative argument (for the complete
version and for the syntax-to-semantics algorithm, see Belloro 2007).
We can now illustrate how these linking rules work, based on one of the ex-
amples presented above. Let us select a simplified version of the sentence in (12a),
along with its potential “allosentences”, as below:
We need to create an appropriate logical structure for each sentence. All in-
volve a predicate of transfer; in RRG terms, a causative accomplishment, of the
type do’ (x, Ø) cause [become have’ (y, z)]. We argued that each of the sentences
in (15) differ on the relative activation of the dative argument, and we said that in
RRG this information is indexed in each argument position. Therefore each logi-
cal structure will also differ on how the position for this argument is filled. The
resulting structures are presented in (16).
The structures in (16) count as the output of the first step of the semantic-to-
syntax linking. The second step is to determine the actor-undergoer assignment.
Following RRG’s AU-hierarchy, the first argument of do’ (x, y) is selected as the
actor macrorole, and the second argument of become have’ (y, z) as the undergoer,
leaving the remaining argument as a non-macrorole. The third step in the linking
involves the selection of the PSA and the assignment of case. PSA status falls on
the first argument of do’ (x, y), which is assigned nominative case. There is no lexi-
cal PSA, and thus the case features are assigned exclusively to the pronominal fea-
tures. Following the appropriate case assignment rules, the highest ranking core
Considering the actor participant of so-called “third person impersonal constructions”
as “active” is an oversimplification, since it cannot be identified. One of the characteristics
of the construction is precisely that encodes it as if active, and thus I will code it thus here,
leaving for further studies a more insightful account of its cognitive status. The “direct ob-
ject” realized as the indefinite un asueto is inactive in all three cases, which accounts for the
fact that it will be encoded exclusively as an NP (i.e. not in a DO-doubling construction).
valeria a. belloro 511
macrorole (i.e. the actor) receives nominative case, the other macrorole receives
accusative, and the non-macrorole receives dative.
The fourth step in the linking entails the selection of the appropriate syntac-
tic templates. The three sentences contain three specified argument positions in
their semantic representation. In (16a) and (16b), however, one of these argument
positions is filled exclusively by feature bundles, and thus for these two structures
we must select core templates with only two syntactic slots (Belloro 2004, 2007;
González Vergara 2006). The structure in (16c) contains two argument positions
filled exclusively by feature bundles, and therefore for this structure we select a
core template with just one syntactic slot.
The final step in the linking implies the assignment of the elements in each
of the argument positions to the appropriate slots in the syntactic template: the
pronominal features to the AGX, and the nominals to the NP nodes. The dia-
grams that illustrate the result of the linking for the three structures are pre-
sented below.
5. Conclusions
In this paper I examined the information structure of the three most frequent
grammatical alternatives available for encoding dative arguments in Spanish, based
on examples from interactional corpora. It was suggested that these three alterna-
tives can be conceived of as a small system, with each member representing a dif-
ferent point on a markedness hierarchy based on correlations between frequency
of occurrence and the prototypical cognitive state associated with the dative case.
Specifically, it was argued that clitic-only, clitic doubling and NP-only structures
correlate with most active, less active and least active participants, respectively.
It was shown that the kind of approach to discourse-pragmatics incorporated
in the model of Role and Reference Grammar can consistently capture the func-
tional differences unveiled by the empirical data.
There are two issues which were not dealt with in this study, and that deserve
careful attention. The first concerns the comparison of the pragmatic status of
lexical phrases marked by a versus those marked by (other) prepositions (e.g. para).
The second concerns the analysis of lexical datives which occur by default in topi-
cal positions, such as those associated with psych-verbs. The role of accessibility
scales in determining the conditions of use of these two extra types, and the extent
to which affectedness, focus structure or other factors to be uncovered need to be
invoked, are issues that I have to leave here open, in the hopes of addressing them
adequately in future research.
References
Ariel, M. 1990. Accessing noun-phrase antecedents. London and New York: Rout-
ledge.
514 spanish datives