Interaction of verbal categories: resolution of infelicitous grammeme combinations*

ANDREJ MALCHUKOV

Abstract In this paper I propose, building on an earlier work by Xrakovskij (1996) and Aikhenvald and Dixon (1998), a general approach to the study of grammeme interaction, focusing on functionally infelicitous grammeme combinations (such as perfective presents), and demonstrate its applicability to the domain of verbal categories. It is argued that the outcome of grammeme conflict depends on the general principles of markedness and semantic scope. I further consider other factors, both functional and structural, which constrain grammeme combinability, and which are regarded as competing motivations in the domain of grammeme interaction. Finally, I demonstrate how the various factors can be integrated into a single model, which crucially relies on the concept of local markedness and markedness hierarchies. 1. Introduction: Typology of syntagmatic dependencies between grammatical categories Although almost any grammatical description of a language contains information about interaction between grammatical categories, there are still few typological studies that focus on this issue (see, however, 5.1 below for further discussion and qualification).1 To my knowledge, the only two studies which explicitly address this issue from a typological point of view are Xrakovskij (1996) and Aikhenvald and Dixon (1998).2 Xrakovskij (1996) (cf. also Xrakovskij 1990) represents a pioneering case study of interaction of verbal categories, focusing on the interaction of mood (in particular, imperative, as opposed to indicative) with tense, aspect, voice and person agreement. He concludes that imperative frequently induces changes in the grammemes belonging to other categories, leading to the loss of a category altogether (e.g., tense), of some grammeme of a category (e.g., the passive is normally lacking an imperative), or of some function of a grammeme (as in the case of reinterpretation
Linguistics 491 (2011), 229282 DOI 10.1515/LING.2011.006 00243949/11/00490229 Walter de Gruyter

230

A. Malchukov

of aspectual values in the imperative), or else leading to a change in its formal expression (cf. the use of special forms of person agreement in imperative as compared to indicative). Further, Xrakovskij makes the important point that results of grammeme interaction may be asymmetrical: more often than not, (only) one of the grammemes changes its meaning when combined with another grammeme (the one undergoing the semantic shift is called recessive, the one inducing the change is called dominant). With regard to imperative, his conclusion is that imperative normally acts as a dominant category with respect to other categories. The paper by Aikhenvald and Dixon (1998) is broader in scope, as it studies mutual dependencies between various grammatical categories, both verbal and nominal, in a wide range of languages. One of the most general results of this study is to show that interpretation/availability of nominal categories is more often determined by verbal categories than the other way round (for example, case marking of arguments may depend on choices in the TAM system). Some other unilateral dependencies have been noted as well; for example, negation is found to be more likely to influence availability/realization of other categories, but is hardly affected by other categories itself. However, many other dependencies have been found to be bidirectional; for example either a choice of number or a choice of case system can impose restrictions on members of the other category. Yet, it seems that further crosslinguistic generalizations can be established in this field once a more fine-grained approach is adopted to category interaction. That is, it is important to distinguish between three distinct albeit related phenomena in the domain of interaction between grammatical categories, which are treated indiscriminately in Aikhenvald and Dixon (1998): (i) The choice of grammeme X of category x excludes category y (e.g., in imperatives/subjunctives tense distinctions are normally missing); (ii) The choice of grammeme X of category x excludes grammeme Y of category y (for example, perfective aspect in many languages is incompatible with the present tense); (iii) The choice of grammeme X of category x leads to formal lack of distinction between grammemes Y1 and Y2 of category y (e.g., in many languages there is a larger number of distinct case forms in singular than in plural). Admittedly, this classification may be less clear-cut in certain cases; for example, if the number of grammemes is restricted to two, the first case is indistinguishable from the third. On the other hand, if a grammeme X may exclude (possibly, for different reasons) all grammemes of y, then the second case converges with the first one. Yet these cases should be kept distinct, since the motivation behind these types of category interaction may be different. In particular, the third case, dealing with neutralization/syncretism, has usually been

Interaction of verbal categories 231 related to markedness: combinations of unmarked members of categories are known to be less restricted, as compared to the marked one (see 5.1). This is different from the case of the second type, which is inasmuch as it is crosslinguistically recurrent due to functional incompatibility:3 here we are dealing with absolute restrictions on certain combinations rather than with relative preferences. The first case is probably the most complex, since exclusion of different members of a certain grammatical category may be due to both factors (see 6.2 below for an account of gender/number dependencies, or tense/ mood dependencies). The present paper continues the typological research into category interaction initiated by Xrakovskij (1996) and Aikhenvald and Dixon (1998), as well as earlier studies under the rubric of markedness (see Plank and Schellinger [1997] for a critical discussion). Here I shall mostly concentrate on the cases of the second kind (called hereafter infelicitous combinations, occasionally abbreviated as IC), where grammemes X and Y are functionally incompatible and therefore a combination of these values is systematically excluded. In Section 2 I exemplify different types of infelicitous combinations as well as outcomes of grammeme conflicts. Section 3 provides an overview of types of infelicitous combinations in the domain of verbal categories. Section 4 offers a general discussion of factors which determine the outcome in infelicitous combinations (resolution rules). Section 5 broadens the discussion to include apart from semantic (in)compatibility other factors constraining grammeme interaction, both functional, such as relevance and economy, and also structural and diachronic. Section 6 discusses interaction, competition and conspiracies between different factors and attempts to integrate different factors into a general picture, taking the form of markedness hierarchies. Finally, Section 7 summarizes the main results of the study and poses questions for further research. The main goal of this paper is to outline a general approach to the study of the interaction of grammatical categories in reliance to the previous literature, so some parts (in particular, Sections 5 and 6) are intentionally programmatic. A full-scale typological study of interaction of individual categories (along the lines of Plank and Schellingers (1997) work on gender-number interaction) will not be attempted in this general article, but is left for future research. 2. Types of infelicitous combinations and their outcomes: some illustrations Note that since infelicitous combinations are infelicitous for functional reasons, effects of grammeme conflicts will be observed independently of the mode of expression of the respective categories. Yet the outcome of a grammeme conflict may be different depending on the structural properties of the

232

A. Malchukov

language. If the respective values are expressed cumulatively, as is typically the case in a fusional language, one should expect that the infelicitous combination will not be expressed at all, which will result in a paradigm gap. For example, in Romance languages the distinction between perfective and imperfective (aorist/imperfect) is restricted to past tense and is not found in the present (see below for further discussion). If categories are expressed independently, as is often the case in agglutinative languages, the outcome may be more diverse. In the latter case we can imagine three primary techniques for conflict resolution4 of infelicitous combinations: (i) the infelicitous combination is not available at all, due to the mutual incompatibility of the categories in question; (ii) the infelicitous combination is available, but involves a change of meaning of one of the grammemes (the recessive grammeme, in Xrakovskijs [1996] terms); (iii) the infelicitous combination is available, but involves a change of meaning in both grammemes. These three cases may be symbolically represented here as: (i) X * Y; (ii) X Y; (iii) X Y. In the first case (X * Y) the resolution rules are similar to what we observed in cases of cumulative expression: semantic incompatibility leads to a gap in a paradigm. Consider, for example, the interaction between modal categories in Korean, as described in Sohn (1994). In Korean, the categories of (epistemic) mood and illocutionary force, which crosslinguistically are most often expressed cumulatively, constitute independent categories. Yet not all theoretically possible combinations of moods (indicative, retrospective, requestive and suppositive) and illocutionary force markers (declarative, interrogative, imperative and propositive) are found. While declaratives and interrogatives combine with indicative and retrospective (i.e., experiential) moods (see (1)(3)), imperatives and propositives (the latter expressing the lets do V meaning) share the requestive mood (see (4)(5)): Korean (1) Ka-n-ta / ka-te-ta. go-IND-DC/ go-RETR-DC S/he goes/went (I noticed). (2) Mek-ess-n-unya eat-PST-IND-INT.PLN Did (s/he) eat? (3) W-ass-te-la. come-PST-RETR-DC S/he came (I noticed). (4) Po-si-p-si-o! see-SH-AH-REQ-IMP.DEF Please, look!

Interaction of verbal categories 233 (5) Wuli ilccik ttena-sip-sita! We early leave-SH-AH-REQ-PROP Lets leave early! (Sohn 1994: 338, 339, 342, 40, 45) Notably, other theoretically conceivable combinations (declaratives and interrogatives with requestive mood, or imperatives and propositives with indicative and retrospective moods) are not found (Sohn 1994). Such patterns, where only natural combinations of grammemes are available, while less natural combinations are avoided, will be explained as instantiations of local markedness in 5.2. below. The second case, where one grammeme (the recessive one) undergoes a semantic shift when combined with another grammeme (the dominant grammeme), is illustrated here with data from the Tungusic languages Even and Evenki. Even, like the genetically related Evenki, has a special habitual marker grA- which is normally used in the past tense. When combined with the nonfuture (aorist) marker it induces a past tense interpretation. Compare the base form in (6) which, when derived from atelics, has a present tense reference, and the habitual form in (7), referring to the past: Even (6) Etiken nulge-n. old.man nomadize-AOR.3SG Old man nomadizes. (7) Etiken nulge-gre-n. old.man nomadize-HAB-AOR.3SG Old man used to nomadize. In Evenki, the combination of the past habitual aspect and future tense is, as expected, excluded (Nedjalkov 1992). In Even, however, such a combination is possible, but the meaning of the aspectual marker is reinterpreted to mean as before in the context of tense/mood forms referring to the future: Even (8) Nulge-gre-di-n. nomadize-HAB-FUT-3SG (He) will nomadize as before. (9) Nulge-gre-li! nomadize-HAB-IMP.2SG Nomadize as before! The third case (X Y), where both morphemes undergo a semantic shift, will be illustrated by data from the Mesoamerican language Totonac, where the combination of the first person singular prefix with the 1st person plural suffix

234

A. Malchukov

yields a 1st person plural exclusive interpretation (Beck 2004: 29). Compare (10b) which normally has the inclusive meaning (we with you), but in the presence of the 1st singular prefix receives the exclusive reading (we without you): Totonac (10) a. Ik-tks-a. 1SG.S-hit-IMFV I hit (him). b. Tks-ya:-w. hit-IMPFV-1PL.S We (incl.) hit (him). c. Ik-tks-ya:-w. 1sgS-hit-IMFV-1PL.S We (excl.) hit (him). (Beck 2004: 29) Another, more common, case of a symmetric grammeme interaction concerns the situation where two ambiguous grammemes undergo mutual adjustment when occurring in combination, allowing only certain values to combine. This case, however, will not necessarily qualify as an infelicitous combination, since the meaning of the two grammemes is not restricted to the values which are semantically incompatible. To conclude: if two grammemes are functionally incompatible, the infelicitous combination will either not be available at all, or else cause a shift in meaning on the part of one or both of the grammemes. In Section 4 I shall return to the discussion of semantic aspects of reinterpretation within infelicitous combinations. In what follows, I shall focus primarily on the second case of asymmetric infelicitous combinations, which involves a dominant and a recessive grammeme. It should be noted, however, that the distinction between asymmetric dependency and mutual dependency is often not clear (as in the case of imperative past used in the obligative function, discussed in 3.3). Cases of the first type (exclusion) will be also relevant for the following discussion, inasmuch as recurrent crosslinguistic incompatibility is a hallmark of an infelicitous combination and thus can be used as a diagnostic for an infelicitous combination. 3. Infelicitous combinations in the domain of verbal categories Here I shall consider several infelicitous grammeme combinations and the outcome of their interaction, focusing as with much of the other work within

Interaction of verbal categories 235 the St. Petersburg Typology Group tradition on the domain of verbal categories. I start with a discussion of interaction of grammemes in the aspectotemporal domain, which is probably the best studied in the literature. Next I provide an example of interaction of mood with other categories, elaborating on the work by Xrakovskij (1996), and finally, address the interaction of valency-related categories. The focus will be on interaction between grammatical categories, although occasionally (as in the case of valency-related interaction, but also actionality) the discussion will be extended beyond grammatical categories per se to include lexical categories as well. 3.1. Perfective presents One of the best studied cases of infelicitous combinations in the aspectotemporal domain is the semantic incompatibility of perfective aspect with the present tense.5 As repeatedly noted in the literature (Comrie 1976; Bybee et al. 1994: 83; Bache 1995), this grammeme combination is systematically avoided, due to the semantic incompatibility of the perfective aspect, which imposes a bounded, closed view of the situation, and the (central) meaning of the present tense, locating an event at the moment of speech.6 The effects of this semantic incompatibility can be observed irrespective of the mode of expression of tense/aspect categories. Thus, in flectional languages, where aspectotemporal values are expressed cumulatively, this feature combination is not found; compare the situation in the Romance languages, where the perfective has past time reference only (Comrie 1976). Where categories of tense and aspect are expressed independently, as in Slavic languages, this grammeme combination is available, but involves reinterpretation of one or the other of the two grammemes. As is well known, Slavic languages differ in the way that perfective presents are interpreted (see, e.g., Bondarko 1971: 250253; Comrie 1976: 6671; Breu 1994).7 In South Slavic languages, such as Bulgarian, the default meaning of a perfective present is generic present (narrative or habitual); see Comrie (1976: 6671; Breu 1994): Bulgarian (11) Spoglednat se, pousmixnat, devojki . . . glance.PFV.PRES.3PL REFL smile.PFV.PRES.3PL girls The girls (used to) look at one another, smile at one another . . . (Comrie 1976: 69) By contrast, in East Slavic languages, such as Russian, a perfective present is normally interpreted as future:

236

A. Malchukov

Russian (12) a. On idt. He go.IMFV.PRES.3SG He goes. b. On pri-dt. he PFV-go.PRES.3SG He will come. Only in certain contexts (such as in the presence of the habitual particle byvalo or modal negation nikak ne (see Bondarko 1993 for other contexts) can the present perfective forms have a generic interpretation: Russian (13) On byvalo pri-dt, skaet. He PTCL PFV-go.PRES.3SG say.PFV.PRES.3SG He used to come and say. As the tense form in (12)(13) is used to refer to the present when derived from imperfective verbs, and to the future or generic present when combined with perfectives, the meaning of this form in Russian is traditionally defined as nonpast or as ambiguous between present and future (see Bondarko 1971). Yet formally, we are dealing here with a present form, which is completely parallel to the present forms in South Slavic. Note also that while perfective presents can refer to the (generic) present,8 imperfective presents cannot (cases of temporal transposition aside) refer to the future. Also in diachronic perspective it is clear that the present was originally the basic meaning of this form (Bondarko 1971: 51; Comrie 1976); hence the rise of the future meaning can be attributed to the clash with the perfective value within an emerging aspectual opposition. Breu (1994) also attributes the difference between South Slavic and East Slavic languages to the fact that in South Slavic the (present) tense is dominant with respect to the aspect, while in East Slavic the aspectual meaning (perfective) is dominant with respect to the temporal one. In our terms, the difference between Slavic languages relates to the fact that in Bulgarian the (present) tense is a dominant category and the (perfective) aspect is a recessive category, while in Russian the aspectual grammeme (perfective) is dominant while the temporal one is recessive. Since the infelicity of the perfective present combination is semantically motivated, IC effects will be observed independently of the concrete mode of expression of tense and aspect categories in the particular language. In this context it is instructive to compare Russian to Finnish, as the outcome of the grammeme conflict is similar in these two languages, even though Finnish differs radically from Russian in the expression of aspectual distinctions. As is

Interaction of verbal categories 237 well known, Finnish lacks verbal aspect, but an aspectual distinction can be rendered through a case alternation on the object (therefore one sometimes speaks of aspectual case in Finnish). The accusative marking of the object as in (14) is used to express perfective meaning, while the use of the partitive case as in (15) is used to express imperfective meaning: Finnish (14) Outi luki kirjan. Outi read.PAST.3SG book.ACC Outi read a book. (15) Outi luki kirjaa. Outi read.PAST.3SG book.PART Outi was reading a book. (Sulkala and Karjalainen 1992: 306, 308) Notably, if a verb is in the (unmarked) present, rather than in the imperfect (past) as in the above examples, the use of the perfective construction (with the object in the accusative) yields the future meaning, similarly to what we observed for Russian: Finnish (16) Outi lukee kirjaa. Outi read.PRES.3SG book.PART Outi reads/is reading a book. (17) Outi lukee kirjan. Outi read.PRES.3SG book.ACC Outi will read a book. (Sulkala and Karjalainen 1992: 308, 306) Thus, resolution of the perfective present combination in Finnish proceeds similarly to Russian, even though aspect is expressed by a syntactic construction in Finnish rather than an inflectional category as in Russian: in both cases tense is a recessive category, and aspect is dominant. The same effects can also be observed in languages which have a category of aspect but lack a category of tense altogether. In these languages, a category with the perfective meaning cannot be interpreted as referring to present even in strong contexts. Maltese Arabic is instructive in this respect. Maltese lacks tense, but uses aspectual (perfective/imperfective) forms to render tense distinctions. Notably, the perfective normally refers to the past, in strong contexts it can refer to the future, but it never refers to the present (Borg and AzzopardiAlexander 1997: 234). In Lango, too, which distinguishes between perfective, habitual and progressive aspects, the perfective aspect may refer to either past or future, but not to the present (Noonan 1992: 138).

238

A. Malchukov

Similar restrictions may be observed even if one or both grammemes in the perfective present combination are unmarked. Consider an example from Limbu (Tibeto-Burman), which makes a distinction between unmarked nonpast and marked past (preterite) in the temporal system and between unmarked perfective and marked imperfective in the aspectual system; all combinations of tenses and aspects are possible. Notably, the present (non-preterite) perfective, a combination of the two unmarked categories, is ambiguous between the future and general present (van Driem 1987: 115): Limbu (18) Hn k-dzok--? What you-do(-NPRET-PFV) What are you doing (generally)? or What are you going to do? (19) Hn k-dzok--pa? What you-do(-NPRET-IPFV) What are you doing (now)? (van Driem 1987: 115) The examples above show that when a grammeme is (structurally) unmarked it is open to more interpretations, and is more likely to be a recessive category. We shall return to the role of markedness for the outcome of grammeme conflicts in 4.1 below. 3.2. Perfective statives and actionality clashes

As is well known in aspectology, not all actional classes of verbs (that is, Vendlers verb classes of Accomplishments, Achievements, Activities and States) are equally felicitous with all aspects (in Smiths terms [1991] there are certain restrictions on combinations of situation aspect and viewpoint aspect). For example, in some languages like Abkhaz (Hewitt 1979) or Totonac (Beck 2004) perfective aspect is available for dynamic events, but unavailable for statives (cf. Bybee et al. 1994: 74). On the other hand, Achievements, which are punctual, may be incompatible with progressive/imperfective values (as is normally the case in English). In other languages such combinations may be possible, but this coercion leads to an actional reinterpretation. Breu (1994) proposed the following rules of actional reinterpretation on the basis of data from European languages (here illustrated with Russian and Italian): a) Atelic verbs (states and activities), if occurring in perfective aspect, receive an inchoative (cf. Rus. znat/uznat, It. sapeva/seppe know/ learn) or a delimitative interpretation (cf. Rus. rabotat/porabotat, It. lavorava/lavor work/work for a while);9

Interaction of verbal categories 239 b) Achievements (find, etc.), if occurring in imperfective aspect, receive an iterative interpretation (Rus. naxodit/najti, It. trovava/trav find repeatedly/find). These rules have been formulated on the basis of few European languages, although in more recent work the data base has been extended (see, e.g., Smith 1991; Johanson 2000; Breu 2000; Tatevosov 2002). For reasons of space I do not venture to delve into the vast topic of aspect/actionality interaction, which has been extensively studied both in the typological (for recent overviews see Sasse 2002 and Tatevosov 2002) and the semantic literature (see Smith 1997; de Swart 1998; Michaelis 2003; Michaelis 2004). Here I shall confine myself to illustrating a few cases of actionality/aspect interaction in order to demonstrate the relative independence of the outcome of actionality clashes from the concrete mode of expression of individual categories, similarly to what we observed above for perfective presents. In fact, similar IC effects have been observed for the interaction of the actional class with temporal adverbs. Thus, as repeatedly noted in the literature, punctual adverbs may induce an inchoative interpretation when combined with statives (cf. Suddenly he knew the answer), while duration adverbs can coerce iterative-habitual interpretation of achievements (cf. For months the train arrived late); see Smith 1997; de Swart 1998 for a discussion of the semantic processes involved here. Similar effects can also be observed with regard to the interaction of actional classes with the meaning of nonfinite verb forms having an inherent aspectual value, although this case has been less studied in the literature. This may be illustrated for Even, which has about a dozen converbial forms, some of which are aspectually neutral (e.g., conditional converbs), while others have an inherent imperfective or perfective value (Benzing 1955; Malchukov 1995). The group of imperfective converbs includes the simultaneous converb in nikAn, while the group of perfective converbs includes the anterior converb in mnin, which has a successive value. As expected, the anterior converb normally combines with telic verbs to indicate immediate anteriority, while the simultaneous converb combines with atelic verbs to indicate (inclusive) simultaneity: Even (20) Eme-mnin, gn-ni. come-CONV say-AOR.3SG As soon as he came, he said. (21) Nulge-niken, it-te-n. nomadize-CONV see-AOR-3SG When he was nomadizing, he saw. Non-harmonic pairings of actionality class and converb form are not excluded, yet such grammeme combinations undergo reinterpretation. Thus,

240

A. Malchukov

statives (like haa- know) when taking the anterior converb form take on an inchoative meaning: Even (22) Haa-mnin, gn-ni. know-CONV say-AOR.3SG As soon as he learned, he said. On the other hand, achievements (like maa- kill) can take the simultaneous converb form only in iterative or distributive context; note that the singular object gets a distributive interpretation in the following example: Even (23) Oro-m maa-nikan, nulge-n. Reindeer kill-SIM.CONV nomadize-AOR.3SG He nomadizes, killing reindeer. In other languages, however, such non-harmonic combinations are disallowed. Thus, in Modern Russian simultaneous converbs can normally be built only from imperfective verbs (from present stems, in traditional terminology), while anterior converbs can be derived only from perfective verbs (from past stems).10 This is demonstrated for the aspectual pair delat/sdelat do (imperfective/perfective), where the perfective form is derived through the use of the prefixal perfectivizer s-: Russian (24) Delaj-a (*s-delaj-a) do.IMFV-SIM.CONV (PFV-do.IMFV-SIM.CONV) Kola zasnul. K. fell_asleep While doing the homework, Kolja fell asleep. (25) S-dela-v (*dela-v) uroki, PFV-do-ANT.CONV (do-ANT.CONV) homework zasnul. fell_asleep After doing the homework, Kolja fell asleep. uroki, homework,

Kola K.

Interestingly, in Old Russian (as used in the 1517th century) all combinations of converbs and actionality classes were possible, but later the infelicitous combinations simultaneous/perfective and anterior/imperfective fell into disuse (Kuzmina and Nemenko 1982: 342346). As a final illustration, consider the interaction between inflectional and derivational aspect (Aktionsart) in South Slavic, discussed by Comrie (1976) and Breu (1994). Apart from the derivational aspect distinction between perfective and imperfective verbs, as attested in other Slavic languages, Bulgarian has

Interaction of verbal categories 241 preserved an older inflectional distinction between aorist, with an inherently perfective value, and imperfect, with an inherently imperfective value. As expected, most common are the harmonic combinations of derivational and inflectional aspects; that is, perfective verbs most often appear in the aorist, and imperfective in the imperfect. However, non-harmonic combinations are found as well (Breu 1994: 38). As with the interaction between actionality and aspect (not surprisingly, as Aktionsart can be regarded as a derived actional class), the meaning shifts are as expected: imperfective verbs in the aorist receive delimitative interpretation (e.g., vsluva [listen.IMFV.AOR.3SG] listen (for some time)), while perfective verbs in the imperfect receive an iterative reading (krivnee [turn.PFV.IMP.3SG] (repeatedly) turn). To conclude: the effects of reinterpretation are similar in various cases of interaction: forms with the perfective value, when applied to a base with an inherent (actional class) or derived (Aktionsart) static/atelic value, receive a delimitative or inchoative reading, while forms with the perfective value, when applied to punctual verbs (basic or derived), receive an iterative interpretation. One important generalization that has emerged from the study of actionality mismatches is that it is usually the more inflectional category that overrides the value of a more derivational or lexical category. For example, regarding the interaction of derivational and inflectional aspect, Breu observes that in all such cases the resultant aspectual value of the verb form as well as its discourse function ultimately depends on the inflectional aspect; that is, in our terms inflectional aspect is a dominant category while derivational aspect is a recessive category. In the semantic literature aspectual shifts have been analyzed in terms of aspectual coercion: the basic actional value is readjusted to satisfy the selectional restrictions of an aspectual operator or an adverb (de Swart 1998). We will return to the semantic aspects of interaction in infelicitous combinations in Section 4 below. 3.3. Past imperatives As the imperative inherently has future time reference, it is normally incompatible with the expression of past tense; even in those languages where expression of tense is independent of mood, such forms are usually absent (Xrakovskij 1992, 2001). If such forms are available at all, they tend to be reinterpreted. Consider an example from Syrian Arabic, where (a periphrastic form of) the imperative past is reinterpreted as past obligation: Syrian Arabic (26) Knt kl lamma knt fl-bt. you.were eat.IMP when you.were in.the.house

242

A. Malchukov You should have eaten when you were at home. (Palmer 1986: 112 citing Cowell 1964: 36)

In Alamblak (Papuan) imperative forms proper (restricted to 2nd person) are not compatible with the past tense markers. Hortative forms (1st, 3rd person imperatives), on the other hand, may combine with the past (in combination with the irrealis marker), but this combination is reinterpreted as counterfactual obligation: Alamblak (27) A-ya-r-t. HORT-eat-3SM-3SF Let us eat it. (28) Ynr a-yakrmay-r-m-r. boy HORT-run.away-IRR-R.Past-3SM The boy should have run away. (Bruce 1984: 137) A somewhat similar situation is found in Finnish, although it concerns the combinability of the imperative with the perfect rather than the simple past. As noted in Sulkala and Karjalainen (1992: 303), imperative perfect forms in Finnish are very rare and when used are interpreted not as a command, but as an admission or a statement. Similarly in Oriya (Neukom and Patnaik 2003: 201), use of the imperative perfect is considered to be more polite, although a (more compositional) delimitative sense (do for a while) is also available. As these examples illustrate, the meaning of the imperative undergoes a semantic shift in such combinations, although the cases from Alamblak and Finnish could be also interpreted as cases of a symmetrical (X Y) interaction in an infelicitous combination. In any case, contrary to what is proposed in Xrakovskij (1996), these examples show that an imperative can be not only a dominant but also a recessive category. 3.4. First person (singular) imperatives

As repeatedly observed in the literature (Birulin and Xrakovskij 1992: 30; van der Auwera et al. 2004), in many languages the first person singular imperative forms are lacking. Thus, in Finnish this is the only form which is absent in the imperative paradigm. A specialized form for the first person imperative is also missing in Turkic languages (Xrakovskij 1996: 39). This restriction is in fact part of a more general markedness pattern, where second person imperative forms are the least marked (and universally available) forms, while first person singular imperative forms (as well as first person exclusive plural, as opposed

Interaction of verbal categories 243 to first person inclusive plural forms) are the most marked (see also 6.3 below for discussion). Generally, one can follow Xrakovskij in his conclusion that an imperative usually acts as a dominant category with respect to a category of person; note that a shift from indicative to imperative in many languages severely restricts the range of possible person grammemes (in many languages only 2nd person imperative forms exist), and can also influence the formal expression of agreement (in many languages the form of imperative agreement markers differs from the indicative; see Xrakovskij [1992, 2001] for discussion and exemplification). However, if a language extends the range of imperative forms to include 1st person forms, the result of grammeme interaction within this Infelicitous Combination can be rather variable. In Malchukov (2001: 172) this was demonstrated for Even (Tungusic), where in combinations of 1st person forms with the second (distant) imperative in dA- either grammeme may be dominant. On the one hand, in combination with the 1st person singular, forms in dAhave developed the future tense function. This reinterpretation is clear from the fact that form in dA-ku may be used in questions: Even (29) Hr-de-j! go-IMP-REF.SG Go (later)! (30) Hr-de-ku? Inge, hr-li-e! go-IMP-1SG OK, go-IMP-PTCL May I go? OK, go! On the other hand, the combination of the distant imperative with the 1st person plural exclusive marker un, which in other moods indicates a set of persons excluding the hearer (cf. Hr-ri-vun We left [without you] vs. Hr-ri-t We left [with you]), results in reinterpretation of the person marker. The latter has developed an inclusive reading, which is semantically more compatible with the function of the imperative speech act: Even (31) Hr-de-kun! go-IMP-1PL.EXCL Lets go! Thus, in the former case the person grammeme is dominant and the (imperative) mood is recessive, while in the latter case the imperative is dominant and the person is recessive. A similar case of reinterpretation of person markers in the imperative mood has been reported for Rapanui (du Feu 1996), although here the effects of reinterpretation are less drastic. In Rapanui, imperative is marked by special momentaneous particles ka for the 2nd person vs. ki for

244

A. Malchukov

the 1st and 3rd person. As noted by du Feu (1996: 37, 158), these particles are also used as the definite future, but in that case the formal differentiation between 2nd and non-2nd person which is natural in imperatives but unnatural for other moods is not as strictly adhered to as in the imperative use. 3.5. Passive imperatives

Inasmuch as passive forms refer to non-controllable states, while the imperative implies that accomplishment of an event is under the agents control, the combination of the two grammemes is semantically infelicitous. In most languages passive imperative forms are therefore absent or marginal (Xrakovskij 1992, 2001; Gusev 2005). If such combinations are available at all, either grammeme may get reinterpreted. Thus, passive can be reinterpreted as a reflexive causative, as has been observed by (Xrakovskij 1996) for English. Xrakovskij illustrates this point with the following example from (Davies 1986: 15): Be checked by a doctor! (= let a doctor check you). Alternatively, the imperative may receive an optative interpretation, expressing not a command but a wish, which does not need to be under the agents control. Thus, in Even the following medio-passive form expresses a malevolent wish (curse): Even (32) Nikata-di ubas tini-v-li! dead.tree-DAT throughout pierce-PASS-IMP May you be pierced by a dead tree! (Malchukov 2001) Note that some other languages use a distinct optative construction in such cases; cf. the passive imperative in Mansi, involving the optative particle wos may in combination with the passive indicative form: Mansi/Vogul (33) Nang wos tota-we-n! you may take-PASS-2SG May you be taken! (Riese 2001: 52) The interpretation of the passive imperative combination as intrinsically infelicitous is apparently contradicted by the fact that some languages like Indonesian and Maori, mentioned in this connection by Xrakovskij (1996: 35) do not show such a restriction. Especially troubling is the evidence from languages such as Maori, where imperatives are possible only in the passive form. Cf.:

Interaction of verbal categories 245 Maori (34) Huri.hia too ika! turn.PASS your fish Turn your fish over! (Bauer et al. 1993: 400) As Bauer et al. (1993: 400) note, these are true imperatives, and the corresponding active forms are impossible with imperatives. The explanation for this restriction, which runs directly counter to the crosslinguistic pattern, lies in the correlation between passive and perfective forms observed in Maori. As noted by Bauer et al. (1993: 400) (referring further to Clark [1973] and Chung [1978]11), in Maori perfectives almost always appear in the passive form (this also accounts for the fact why passives predominate in past tense contexts). On the other hand, imperatives seem universally to align with perfectives rather than imperfectives, as imperatives normally constitute an instruction for some action to be accomplished, without focusing on the internal structure of the event.12 (See also below for a reinterpretation of aspect in imperative forms in Russian.) These two correlations taken together will explain why some languages like Maori make use of passive imperatives. Although general markedness cannot explain this restriction (as passive is usually taken to be the marked grammeme in the voice opposition), one can invoke the notion of local markedness as an explanation (see Tiersma 1982; Croft 1990 for a distinction between general and local markedness). Local markedness pertains to markedness (naturalness) of certain grammeme combinations rather than of grammemes per se in the above case to naturalness of the perfect passive combination, which seems to be universally unmarked (see Section 6.1 below for the Russian data). On the other hand, general markedness can help to explain why many other Austronesian languages (cf. e.g., Himmelmann 1999: 24 on Toratan) make liberal use of passive imperatives, as here passive cannot be regarded as a marked category (cf. Xrakovskij 1996: 35; Birulin and Xrakovskij 1992: 33).13 We shall return to the issue of markedness (both general and local markedness) in Section 5.1 below. 3.6. First person indirect evidential forms In those languages which distinguish between verbal forms indicating direct vs. indirect evidence, indirective (i.e., non-witnessed or non-firsthand) forms are often missing in the 1st person, or else are reinterpreted (Willet 1988; Kozintseva 1994; Johanson and Utas 2000; Curnow 2003; Aikhenvald 200414). For example, in Hunzib the evidential perfect (a periphrastic form built on the gerund form) is not used in the 1st person (van den Berg 1995: 103). The same

246

A. Malchukov

holds for Komi-Zyrjan (Kozintseva [1994: 100]). More examples of languages where 1st person forms are lacking with non-firsthand evidentials are given in Aikhenvald (2004: 232). A somewhat similar situation obtains in Nenets (Samoyedic), where the non-firsthand non-visual evidential (traditionally called the auditive mood) is mostly used in the 3rd person, rare in the 2nd, and absent in the 1st person (cf. Tereenko 1947: 224; cf. Lublinskaja and Malchukov 2007): Nenets (35) Nabi ne ma-mon-da. other woman say-AUD-3SG (It was heard that) the other woman said. (Tereenko 1947: 225) If unwitnessed evidential forms can be used with the 1st person at all, they are often interpreted as mirative, to register the fact of new knowledge on the speakers part (it turns out that P; the term mirative is due to DeLancey 1997). This has been noted for the evidential forms in Even in Malchukov (2000: 443447). Even, like many other Altaic languages (see Johanson and Utas 2000), uses evidential perfect forms to refer to unwitnessed events; when used in the 1st person these forms regularly acquire a mirative meaning: Even (36) Oma-a. forget-PF.PART (S/He) has forgotten (it). (37) Oma-a bi-he-m. forget-PF.PART be-PRES-1SG It turns out that I have forgotten (it). (Malchukov 2000: 445) A similar interpretation is found with the first person evidential perfect forms in Tatar (Majsak and Tatevosov 2000: 72): Tatar (38) Min balta al-gan-man. I axe take-PF-1SG It turned out that I have taken the axe. A somewhat different case is attested for Archi (Daghestanian), although here the indirect evidential in the 1st person is used to convey the new knowledge on the part of the hearer, rather than the speaker (Kozintseva [1994: 100]). More examples of the mirative and related uses of 1st person non-firsthand forms are discussed by Aikhenvald (2004: 219231). There can be further semantic extensions from the mirative as well, in particular leading to an

Interaction of verbal categories 247 involuntary-action interpretation of the 1st person unwitnessed forms (see Curnow 2003). 3.7. Singular forms of reciprocals and sociatives It is well known that reciprocal as well as sociative15 forms (the latter denoting joint actions) are normally used with plural subjects. One systematic exception to this generalization is provided by languages which permit discontinuous reciprocal and sociative constructions, where the grammatical subject is singular and the co-agent is expressed by a comitative phrase (John quarrels with Mary). It should be noted, however, that such constructions are normally available for either lexical reciprocals (such as quarrel above) or morphologically light reciprocals, which are more grammaticalized, reduced in form, and show more polysemy (for the distinction between heavy and light reciprocals, see Kemmer 1993). Conversely, they are normally excluded with heavy reciprocal markers such as reciprocal pronouns, which are less grammaticalized and normally unambiguous (cf. *John quarrels with each other with Mary). And even for light reciprocals such combinations can be questionable; cf. the following German example (from Nedjalkov 1991: 298) where the comitative construction is excluded with the heavy reciprocal einander, but is hardly acceptable with the light reciprocal/reflexive sich either (see also Plank 2006: 257266 for a detailed analysis of these patterns): German (Nedjalkov 1991: 298) (39) a. Sie kten sich. b. ??Er kte sich mit ihr. c. *Er kte einander mit ihr. They kissed each other. The same conclusion concerning the semantic infelicity of singular forms of reciprocals and sociatives can be obtained by looking at cases of reinterpretation of these combinations. Thus, Nedjalkov (1991: 299; Nedjalkov 2007b: 277278) notes that reciprocal-sociative forms of verbs in Turkic and Mongolian languages express the assistive reading (help to V) when used in the singular. This is illustrated below for Yakut (Siberian Turkic); cf. (40a), where the verb expresses the sociative meaning when used in the plural, and (40b), where the verb expresses the assistive meaning when used in the singular: Yakut (40) a. Kini.ler ot tiej-is-ti.ler. they hay carry-REC-PRES.3PL They carry hay together.

248

A. Malchukov b. Kini aga-ty-gar ot tiej-s-er. s/he father-his-DAT hay carry-REC-PRES.3SG He helps his father to carry hay. (A.A. Petrov p.c)

erbak (1981: 114) provides further examples of the shift of reciprocalsociative forms to assistive meaning from other Turkic languages (Bashkir, Kazakh, Tatar, Turkmen). On the other hand, in some Turkic languages reciprocal singular forms have been reanalyzed as plurals, as illustrated below for Kirgiz: Kirgiz (41) az-.at. write-SOC.PRES.3SG They write. (erbak 1981: 114) Thus in some Turkic languages allowing for reciprocal/sociative singular forms, the (singular) number category leads to reinterpretation of the reciprocal (that is, the voice grammeme is recessive), while in the case of Kirghiz the reinterpretation of reciprocal singular forms as plurals can be viewed as symmetrical. 3.8. Transitive verbs with intransitive conjugation

In many languages which distinguish between intransitive conjugation (crossreferencing the subject alone) and transitive conjugation (cross-referencing both subject and object), the former conjugation is predictably restricted to intransitive verbs and the latter to transitive verbs. Those languages which do permit the intransitive conjugation with transitive verbs may use it to indicate reduced semantic transitivity (in the sense of Hopper and Thompson 1980), in particular, that an object is indefinite or non-specific (as in many Uralic languages; cf. Moravcsik 1978). In some other languages, where the transitive conjugation is not restricted to constructions with definite objects, the use of the intransitive conjugation may have a more radical detransitivizing effect. Thus, in (West Greenlandic) Eskimo the change from transitive (polypersonal) to intransitive (monopersonal) agreement results in a reflexive or anticausative interpretation: Eskimo (42) Napi-vaa. break-IND.3->3 He broke it.

Interaction of verbal categories 249 (43) Napi-vuq. break-IND.3SG It broke. (Fortescue 1984: 85) A similar situation obtains in Basque, although here the choice between transitive and intransitive conjugation is additionally encoded through the use of different auxiliaries. Thus, as noted by Saltarelli (1988: 6465), some transitive verbs may take the intransitive auxiliary16, thereby becoming intransitive (anticausative or pseudopassive): Basque (44) Jon-ek hil d-u. John-ERG.SG kill(PF) 3sA-(PRS)-AUX2-(3ERG.SG) John has killed (him). (45) Jon hil d-a. John-ABS.SG kill(PF) 3sABS.SG-(PRS)-AUX1 John has died. (Saltarelli 1988: 6465) Admittedly some of these cases, including the verb hil in (44)(45) above, can also be analyzed as labile verbs, disambiguated by the auxiliary choice (cf. Saltarelli 1988: 64). However, in other cases, it is clear that the transitive use is the basic one, and the intransitive (pseudopassive) interpretation is coerced through the use of the intransitive auxiliary: Basque (46) Zur-i askata-sun handi-a eman you-DAT freedom big-ABS.SG give za-izu. 3ABS-PRS-AUX1-2DAT.SG You have been given much freedom. (Saltarelli 1988: 219) Note that in the latter case the reinterpretation is partial, as the verb may be argued to remain semantically transitive (the agent is present in the argument structure, even if suppressed in the syntax), unlike what we observed for the verb hil kill; die above, where the agent is absent in the intransitive use at the semantic level as well. Similar phenomena have been noted for other languages as well. For example, in Alamblak, virtually every verb can form a pseudo-passive by replacing the transitive agreement by an intransitive one (Bruce 1984: 195). An

250

A. Malchukov

opposite case, where a basically intransitive verb is coerced into a transitive interpretation through the use of a transitive conjugation, is also attested. Thus, according to Haas (1941: 46), in Tunica any intransitive verb can be used transitively when taking a transitive inflection. In all the cases discussed the use of agreement induces a shift in verbal valency; that is, the verbal valency is a recessive category here, while agreement is a dominant one. Admittedly, this case is somewhat different from those discussed above, since the recessive category is lexical rather than morphological. 4. Resolution of infelicitous combinations Here we shall discuss factors which determine the outcome of conflict in infelicitous combinations. In what follows we disregard cases where the grammemes are expressed cumulatively: as stated above, for such cases the infelicitous combination of the respective values will simply be missing from the paradigm. If the respective grammemes belong to different categories, such combinations will be possible, as far as structural factors are concerned, but are disfavored for functional reasons. As stated above semantic incompatibility often leads to reinterpretation of either (or both) of the two grammemes. An interesting question is what factors determine which grammeme wins out in infelicitous combinations, that is, which grammeme is dominant and which is recessive. The two general semantic factors to be discussed here are (functional) markedness and scope. 4.1. Functional markedness

The notion of markedness is known to be notoriously complex, as it can pertain to formal markedness (overt vs. zero coding), textual markedness (frequency), distributional markedness (restricted/unrestricted distribution) and functional markedness (specificity vs. generality of meaning), among others (see Croft [1990]; Plank and Schellinger [1997]; Aikhenvald and Dixon [1998]; Haspelmath [2006]; see also 5.2. below for further discussion). In the present context we are mostly concerned with functional markedness, pertaining to specificity of meaning or following a traditional interpretation going back to Trubetskoj (1939) and Jakobson (1957) to the distinction between forms specified for a particular value and forms having a neutral value within a privative opposition. Thus, it is arguably the case that the unmarked grammeme (the form which is more general in meaning or ambiguous) is more likely to be recessive, while the marked member (with a more specific meaning) is more likely to be dominant.17 Similar suggestions, which relate results of interaction

Interaction of verbal categories 251 to markedness of the respective categories, have been made by Bache (1995) with respect to tense/aspect interaction (see the quote below), and by Apresjan (2004) with respect to interaction between aspect and voice. The role of this factor can be best demonstrated for the case of present perfectives discussed above. This combination provides a good testing ground, as languages show variation with regard to which member of the aspectual paradigm (perfective or imperfective) is unmarked. Thus, it can be shown that perfective aspect is more likely to cause reinterpretation of tense in the present perfective when it is the marked member in the aspectual opposition. This can account for the fact that in Russian perfective presents are reinterpreted as future, as perfective is traditionally considered to be the marked aspect and imperfective unmarked (Jakobson 1957; Bondarko 1971; Meluk 1998: 28).18 On the other hand, unmarked perfectives (in particular, those lacking overt marking)19 are more likely to be recessive when combined with the present. Itelmen (Paleosiberian) is instructive in this respect. Itelmen has perfective as an unmarked member in the aspectual paradigm (imperfective is marked by the suffix kz-), while in the tense paradigm the past is unmarked and the present is a marked category (with the s- marker) (Volodin 1976). Notably, a combination of the unmarked perfective and the marked present performs the function of a present: Itelmen (47) T-entxla--s-kicen. 1SG-lead-ASP-PRS-3SG.O I lead him. (Volodin 1976: 248) In accordance with the markedness pattern, the marked category is dominant here, while the unmarked one is recessive. Of course, the same point can be made with respect to English, if we follow Smith (1991, 1997) in her assumption that the aspectual paradigm in English includes a (marked) imperfective (progressive) as opposed to an unmarked perfective (indefinite) forms. As expected, present perfective (present indefinite) forms have a generic present interpretation when derived from telic verbs. A similar proposal relating the outcome of grammeme conflict to markedness in the tense and aspect system was made by Bache (1995: 190) in his discussion of perfective presents in English and Russian.20 Finally it can be observed that, when both present and perfective are unmarked, as in case of Limbu in (18), either grammeme may be dominant or recessive (hence the ambiguous interpretation of (18)). Another illustration of the role of markedness can be provided from the domain of valency-related categories. In those languages where transitive agreement is restricted in use (to constructions with definite object as in the case of

252

A. Malchukov

many Uralic languages), its non-occurrence with transitive verbs does not coerce the intransitive interpretation. However, in those languages where both intransitive and transitive conjugations can be regarded as equally marked, the use of intransitive conjugation with basic transitives can produce a detransitivizing effect (as we observed for Basque and Eskimo, discussed in 3.8). However, markedness is not sufficient to account for all cases of category interaction. To begin with, in many of the cases discussed above both of the interacting categories seem to be marked, as in the case of the imperative (which is traditionally regarded as a marked mood in contrast to the unmarked indicative) in combination with other equally marked categories (the 1st person, the passive, the past tense). Further, if the interpretation of the perfective present combination is due solely to markedness, it is not clear how this explanation applies to Bulgarian, where the outcome of the grammeme conflict is quite different. Note that there is no way to argue that in Bulgarian the perfective is less marked; in fact, if anything, the perfective aspect in Bulgarian is semantically richer (more specific), being more lexical and less grammaticalized than aspect in East Slavic (Russian). This, however, provides a clue as to why the aspectual grammeme is recessive in Bulgarian. As we have seen in 3.2 above, in case of a conflict between the meaning of an inflectional and a lexical (or derivational) category, the conflict is usually resolved in favor of the inflectional category. In the next section this observation will be related to the general principle of scope. 4.2. Scope

Another relevant factor that has a bearing on the resolution of infelicitous combinations, which emerges the study of interaction of grammemes within the aspecto-temporal domain, even though this principle is rarely formulated in general terms in the typological literature, is semantic scope (cf. Rice 2000).21 It appears that, other things being equal, a grammeme with a wider scope is more likely to be dominant. Thus, according to Breu (1994), in Bulgarian inflectional aspecto-temporal forms override the values of the derivationally expressed aspect in non-harmonic combinations, such as imperfective aorist (see 3.2 above). The same holds for cases where recategorization in the actionality class of a verb is due to aspect as in the cases discussed above where statives acquire an inchoative interpretation when used in perfective aspect. A similar reinterpretation may arise from an infelicitous combination with an adverbial, as noted by Smith (1991), among others. Smith (1991) introduces a special rule (dubbed external override) which states that in case of a conflict between aspectual features of a verb and an adverb, the latter wins out. For example, the verb know is basically a stative, but when combined with punc-

Interaction of verbal categories 253 tual time adverbials like suddenly, it is reinterpreted as an achievement (Suddenly I knew the answer). A more general version of the Override Principle to the effect that the meaning of the lexical item adjusts to the meaning of the structure in which it is embedded has been proposed by Michaelis (2003: 268). In all these cases the results of interaction seem to be related to semantic compositionality: an application of the category with a broader scope may be blocked by the semantics of the base, but if not, then the feature of the external category can override the feature of the base. In the semantic literature this phenomenon is known as coercion: an argument is reinterpreted to meet the selectional restrictions of its functor.22 For cases of aspectual coercion, the aspectual operator (or a temporal adverb) is said to shift the interpretation of the actional class in cases where the latter does not satisfy the selectional restrictions of the former (see de Swart 1998 for discussion). In de Swarts theory of aspectual coercion, conflicts between aspectual operators or scoping adverbs and the lexical class of the verb are resolved by introducing (coercing) implicit aspectual operators readjusting the values of the actional class. For example, perfective aspect normally applies to dynamic events; hence, when applied to a state, it coerces the state into a process by interpolation of an implicit (inchoative) operator. Similarly, since progressives normally apply to durative events, when applied to achievements they coerce an iterative interpretation (The light is flashing). A similar analysis in terms of coercion can be offered for cases where perfective presents receive a habitual interpretation, on the assumption that presents intrinsically select for unbounded (imperfective) events; application to a bounded (perfective) event will therefore coerce the habitual operator. As Michaelis puts it (Michaelis 2004: 60): present constructions are intrinsically state selectors. An explanation in terms of scope/coercion is most appropriate for cases in which there are clear scope relations between categories (as in case of interaction between inflectional and lexical categories, such as inflectional aspect and actionality, or intransitive conjugation and verbal valency). Conversely, it is less applicable to cases in which the scope relations are less clear (e.g., in the case of imperative mood and person discussed above). It remains to be seen whether all cases of resolution in infelicitous combinations can be accounted for in terms of scope and markedness (some cases will remain controversial, as in many cases the scope relations or the markedness status or both may not be obvious). For the time being it can be stated that both markedness and scope seem to be important (and potentially conflicting) motivations in the field of grammeme interaction. It is telling that the imperative, which has been shown by Xrakovskij (1996) to be usually a dominant category, is both the marked category in the mood paradigm and the category with a wide (propositional) scope.

254 4.3.

A. Malchukov Conclusions: mechanisms of resolution of infelicitous combinations

On the basis of the previous discussion, we can tentatively suggest that there are at least three different mechanisms involved in the resolution of infelicitous combinations: (a) blocking: the internal category (the category with a narrow scope, the argument in an argument-functor pair) blocks application of a wide-scope category having an incompatible value (for example, statives resist combinability with a perfective gram); (b) coercion: an external category (the one with a wider scope, the functor in an argument-functor pair) coerces a shifted interpretation of its argument (for example, perfectives coerce states into inchoative processes); (c) adaptation: a grammeme narrows its use to be compatible with the meaning of another grammeme (for example, the imperative receives an optative interpretation when applied to the passive). These mechanisms differ with respect to the principles of markedness and scope discussed above. Thus, both blocking and coercion pertain primarily to infelicitous combinations involving two marked categories. Adaptation, on the other hand, pertains primarily to an unmarked grammeme, as the unmarked category is open to more interpretations (i.e., is vague or ambiguous) as compared to the marked one, and therefore adapts to the context more easily. Adaptation differs from the other two properties in terms of scope as well. As is clear from the above, both blocking and coercion are unidirectional processes: the former operates outwards (blocking application of the functor category), the latter inwards (affecting the meaning of the argument). Adaptation (or selection23), by contrast, is a bidirectional process, which may involve functional readjustment on the part of either external or internal categories (or both), in order to produce the most natural interpretation given the range of meanings of the respective grammemes. Sometimes it may be difficult to distinguish cases of coercion and adaptation/ selection. Indeed, some cases can be interpreted in both ways. Even classic examples of aspectual coercion discussed in the literature (Smith 1997; de Swart 1998; Michaelis 2003; 2004) may allow for an alternative interpretation. In fact, some cases allegedly involving coercion, as when statives receive an inchoative interpretation in perfective contexts, have been treated in aspectology in terms of aspectual ambiguity on the part of verbs referred to as inchoative-stative or initio-transformatives (Sasse 2002; Johanson 1998). Such an interpretation is more plausible in cases where this aspectual alternation is lexically restricted to certain verbs, as in the case of initio-transformatives in Turkic languages discussed by Johanson (1996). In other cases where no such restrictions are found (as in the case of reinterpretation triggered by temporal adverbs) a more general interpretation in terms of coercion may be preferable. Generally, the analysis in terms of selection/adaptation is preferable for cases where the primary and the shifted meaning do not show a clear asym-

Interaction of verbal categories 255 metry, while coercion is a better explanation for cases where the shifted meaning is marginal and restricted to particular infelicitous combinations. Thus, it is perfectly possible to account for the future interpretation of perfective presents in Russian in terms of adaptation: the secondary future meaning is selected in preference to the primary present-tense meaning because it is more compatible with the context of the perfective aspect. In other cases, the adaptation scenario seems less applicable. Thus, it will be misleading to treat, say, the 1st person plural form kun in Even (see (30)(31) above) as ambiguous between exclusive and inclusive interpretation (or as an underspecified 1st person plural form), as the inclusive interpretation is found only in combination with the imperative, where the exclusive interpretation would be infelicitous. Ultimately, the problem concerning interpretation of particular cases in terms of coercion or adaptation may need to be solved on a language-particular basis. However, there is also a theoretical issue involved, if we assume, as above, that coercion operates only inwards (affecting interpretation of the internal category). This means that all shifted interpretations of an external category will need to be analyzed as adaptations, choosing the interpretation of the external category to fit the meaning of the base. On these assumptions, all those cases in which the external category is recessive should be due to markedness: the external category, which is open to several interpretations, is interpreted in a way which is more harmonic with respect to the meaning of the other grammeme. Clearly further research is needed to determine whether such an analysis is universally valid. For the time being we shall use the term shifted interpretation (on the part of a recessive grammeme) in a broad sense which is noncommittal as to the process involved, be it coercion (semantic augmentation triggered by an external category) or a less drastic semantic modification involving adaptation (shifting to a secondary meaning which is more harmonic in a particular combination). 5. Factors underlying grammeme (in)compatibility Above we have considered some factors underlying grammeme (in)compatibility, focusing on cases in which compatibility is semantically motivated. However, semantics is clearly not the only factor constraining syntagmatic combinability between categories. Such factors can be divided into function-related, form-related, and more complex cases which involve both function and form. 5.1. Functional factors

Apart from semantic (in)compatibility, the following factors are considered here as function-related: relevance, economy/redundancy, and functional markedness.

256

A. Malchukov

5.1.1. Semantic incompatibility. The role of this factor has been discussed and illustrated above, and can be captured in the form of the following constraint: Generalization 1. Semantically infelicitous grammeme combinations are avoided. As a consequence of this constraint, semantically infelicitous combinations will either be nonexistent, or if the respective grammemes (their exponents) do occur in combination, the combination will be reinterpreted. 5.1.2. here: Relevance. Another functional factor will be dubbed relevance

Generalization 2. Regularity of co-occurrence between the members of grammatical categories reflects the degree of their mutual relevance. The role of this factor can be illustrated from the domain of tense-aspect interaction. A well-known crosslinguistic generalization due to Comrie (1976) states that aspectual distinctions are more often observed in the domain of past tenses. In particular, Comrie (1976: 7172) cites examples from Romance languages where the aspectual distinction between perfective and imperfective is restricted to the past (aorist vs. imperfect). Other examples are not difficult to come by. For example, in Hixkaryana the tense/aspect system includes, apart from the nonpast in yaha, three different forms referring to the past: immediate past in no, recent past completive in yako and recent past continuative in yakonano. In Mangarayi, the aspectual distinction (punctual/continuous) is also restricted to the past. Comries explanation for this asymmetry invokes the notion of relevance: aspectual qualification is less relevant for actions that have not (yet) occurred (Comrie 1976). On a more general understanding, this notion of relevance is also reminiscent of Joan Bybees (1985) relevance, yet the latter notion pertains to the relevance of the category expressed by the bound morpheme to the stem, while we are rather interested in semantic relevance of one (verbal) category to another. Perhaps more importantly, Bybees (1985) relevance reflects the degree of semantic impact (i.e., to what extent the meaning of the verbal lexeme is affected), while the notion of relevance as used here rather reflects functional usefulness of a particular distinction to a cross-cutting category. Note that explanation in terms of relevance is more general than explanation in terms of semantic compatibility. Indeed, while restrictions on the use of aspects with the present tense can be explained in terms of (in)compatibility, this explanation does not carry over to the future tense, which is conceptually com-

Interaction of verbal categories 257 patible with both aspects. Yet we can still follow Comrie in his conclusion that aspectual qualification is less relevant for future non-factive events than for past actions, for which it is often important whether an action has been completed or not. This explanation carries over to the frequent lack/neutralization of tense and aspect distinctions in negative clauses and irrealis moods, as documented by Aikhenvald and Dixon (1998). Similarly, tense distinctions are frequently found only in the indicative mood. By way of illustration, consider the case of Nkore-Kiga, where seven different tenses are distinguished in the indicative, while in the subjunctive the tense paradigm is reduced to two, and in the imperative is lost altogether (Taylor 1985: 154). In fact, mutual incompatibility of tense with non-indicative (irrealis) moods, observed in many languages, has led some authors to suggest that tense and mood should be subsumed under one category (Xrakovskij and Volodin 1979). The same observation holds for aspectual distinctions, although the restrictions are less regular here. This case deserves special attention since it is less well documented in the literature (interaction of mood with other categories is not addressed in Aikhenvald and Dixon 1998). Thus, in Koromfe (Rennison 1997), the indicative has a four way tense/aspect opposition (between aorist, past, durative and progressive), while in the imperative only an unmarked/durative opposition survives. In Sanskrit, imperative and optative forms derive from present/ imperfective stems (Kulikov 2001, and p.c.). Originally, irrealis moods could be also formed from stems of the aorist and perfect, but later the latter forms disappeared, leading to neutralization of the aspectual distinction. In Basque, the subjunctive mood does not distinguish aspects either (Saltarelli 1988: 230). And in Tsakhur (Daghestanian), the aspectual opposition is obligatory in realis (referential) moods, is optional in hypothetical moods and is absent in counterfactual moods (Majsak and Tatevosov 1998). Significantly, Majsak and Tatevosov also invoke the notion of relevance to account for this gradual reduction of aspectual distinctions in Tsakhur. Partial neutralization of aspectual distinctions in the imperative form has also been noted for Russian (Xrakovskij 1996). Although both perfective and imperfective forms are available for the imperative in Russian, the imperfective imperative forms need not have a progressive meaning (Continue V-ing!), which is a primary function of imperfective indicative forms. Thus, Delaj uroki Do (imperfective) homework does not necessarily imply that the addressee is already in the process of doing homework (continue doing), unlike imperfective indicative forms; cf. Delaet uroki (S/he) is doing (imperfective) homework.24 This may be taken as an indication that universally the imperative aligns with the perfective rather than the imperfective, as it focuses on the accomplishment of a command rather than on the process itself (recall the discussion of passive imperatives in Maori in (33)). See van der Auwera et al. (2009) for more discussion of the perfective bias of imperatives.

258

A. Malchukov

5.1.3. Redundancy. Above we have noted that imperatives normally do not combine with the past, for obvious semantic reasons. This does not explain, however, why imperative forms do not combine with the future, as their meanings are perfectly compatible: in fact, the meaning of the imperative implies future time reference. But this provides a clue to an explanation of their incompatibility: expression of future tense is redundant in combination with the imperative, hence its overt encoding is dispensable. Other cases of grammeme incompatibility due to redundancy are not hard to find. Thus, in Nambiquara (Brazil) there are, apart from the familiar restrictions on the use of indirect evidential forms (inferential-evidential and reportative) with the 1st person, restrictions on the use of the direct evidential forms (referring to the speakers observation) in the 2nd person: thus, there are no 2nd person forms meaning You V-ed (I saw it) parallel to the grammatical 3rd person direct evidential forms S/He V-ed (I saw it) (Palmer 1986: 221222). As Palmer notes, this must be due to redundancy. More generally, redundancy can account for the crosslinguistic tendency for cumulative expression in the domain of tense/aspect/mood, given many semantic interconnections in this domain (e.g., between future and irrealis, past and perfective, and present and imperfective). A similar observation holds for mood/person interaction in imperatives: cumulative expression of 2nd person imperative forms is definitely favored as compared to independent expression of these categories, especially in languages with a restricted imperative paradigm (Xrakovskij 1992). Similar economy effects are attested in the domain of nominal categories as well. Thus, as shown by Haspelmath 1999, cases of over-determination involving use of both possessor and (definite) determiner in the same construction are crosslinguistically avoided. This restriction, which is familiar from English (cf. * the my house), is also found in a number of other languages, including Armenian (Kozintseva 1995), Abkhaz and Lango (see Haspelmath [1999] for further exemplification). Haspelmath explains this restriction in terms of economy: the expression of definiteness is redundant in construction with the possessor, since the latter is itself used to identify (anchor) the reference of the possessed item. Similarly, in many languages having a grammaticalized article, its use is normally avoided with inherently definite nominals (e.g., proper names), although some other languages like Greek may permit this (see also Plank 2003 for further discussion of patterns of overdetermination in the noun phrase). Thus, the economy constraint on grammeme combinability can be formulated as follows: Generalization 3. Overt expression of a semantically redundant grammeme is avoided.

Interaction of verbal categories 259 Note that since economy constraints, unlike constraints due to semantic incompatibility, refer both to form and function, they are only operative in cases where both categories have an overt expression. 5.1.4. Markedness. One general factor frequently invoked in discussions of incompatibility between grammatical categories is markedness (Greenberg [1966]; Croft [1990]; Aikhenvald and Dixon [1998]; Meluk [1998]; Croft [2003]; but see Plank and Schellinger [1997] and Haspelmath [2006] for a critical discussion of this notion). For example, Meluk (1998: 26) notes that a combination of two marked categories is avoided citing a spectacular case of Koryak (Paleosiberian), where case forms are distinguished only in the unmarked (singular) number, while numbers are distinguished in the unmarked (absolutive) case. Some other authors even build this generalization into the definition of markedness (in this case one speaks of distributional markedness). Thus, both Greenberg (1966) and Croft (1990: 157; Croft 2003: 9597) have observed that the number of cross-cutting inflectional distinctions of the unmarked gram is larger as compared to the marked one. Croft (2003: 97) refers to inflectional potential and distributional potential as manifestations of behavioral potential, which he considers one of the diagnostics for (typological) markedness. He provides the following definition of inflectional potential, If the marked value has a certain number of formal distinctions in an inflectional paradigm, then the unmarked value will have at least as many distinctions in the same paradigm (Croft 2003: 97). Aikhenvald and Dixon (1998) come to the same conclusion in their study of dependencies between grammatical categories. An instructive example of markedness effects from the domain of tense/ aspect interaction comes from Alamblak (Papuan). In Alamblak, as in Limbu (see (18)), the perfective grammeme is the unmarked member in the aspect category, and the present grammeme is the unmarked member in the tense category. Again as in Limbu, Alamblak permits some verbs to appear in the unmarked present perfective form with a generic present meaning: Alamblak (48) Hoit--w-r. sleep-PRES-IMFV-3MS.SG He is sleeping. (49) Hoit-- r. sleep-PRES-PFV-3SG He sleeps. (Bruce 1984: 135) Interestingly, the combination of the marked categories immediate past (-yand its allomorphs) and imperfective (-w) is excluded, although this can hardly be attributed to semantic incompatibility.

260

A. Malchukov

Rich data illustrating the role of markedness from the domain of both nominal and verbal categories has been provided by Aikhenvald and Dixon (1998), who show in particular that the use of a marked category frequently restricts the choice of grammemes in another grammatical category. For example, gender/ person distinctions available in the singular (the unmarked member of the number category) may be lacking in the plural (cf. Greenberg [1966]; but see Plank and Schellinger [1997] for a cautionary note25). Similarly, in negative clauses TAM distinctions are frequently neutralized or absent; and in irrealis mood tense/aspect distinctions are often lacking. Some of these cases can also be attributed to other factors; thus loss of TAM distinctions with negatives, like loss of tense/aspect distinctions in irrealis forms, can also be explained by relevance (see above). Some other markedness effects, however, in particular those observed for cases of interaction between voice and TAM categories (not discussed by Aikhenvald and Dixon [1998]), cannot be explained in such way. Thus, it is not obvious whether TAM distinctions should be less relevant in the (marked) passive voice; still, in many languages TAM distinctions are frequently neutralized in the passive. For instance, in Wari (Chapakuran; Amazonia), there is only one passive marker (inflectional particle), which does not differentiate TAM, in contrast to the active TAM markers (Everett and Kern 1997). In Sanskrit, a morphological passive (as distinct from middle inflection) is normally built only from present/imperfective stems (Kulikov 2001 and p.c.): Sanskrit (50) a. KR-No-ti. make.PRS.ST-3SG He makes. b. KR-Nu-te. make.PRS.ST-3SG.MID He makes for himself. c. Kri-ya-te. make.PRS.ST-PASS-3SG.MID He is made. In Russian, there are restrictions on the combination of both synthetic (reflexive) and periphrastic passive with aspects and further restrictions on what specific meanings of aspect are available in passive forms (Poupynin 1999; Apresjan 2004). Thus we concur with other authors in accepting the following generalization: Generalization 4. Combinations of marked grammemes are more restricted (i.e., less frequent crosslinguistically) than combinations of unmarked ones. Although it is not difficult to find support for this generalization, it is more difficult to pinpoint the contribution of the different aspects of markedness to the

Interaction of verbal categories 261 attested restrictions. Indeed, as already mentioned (cf. 4.1 above), markedness itself is a complex notion, as marked terms are identified by different criteria (see Croft 1990; Battistella 1990; Plank and Schellinger 1997; Haspelmath 2006). Thus it is common to distinguish between functional markedness, formal markedness, textual markedness, and distributional markedness. The unmarked member is assumed to be more general in meaning as compared to the marked one, to have less complex overt marking (in particular, to be zeromarked), to be more frequent in texts and less restricted in its use. Distributional markedness is directly related to the issue of category interaction (the unmarked category appears in more morphological contexts), hence invoking it here as an explanatory factor is circular. Textual markedness (frequency) may indeed be the driving force behind the correlations between different markedness phenomena (cf. Haspelmath 2006), but its role in category interaction seems to be indirect. On the other hand, both functional and formal markedness seem to contribute to restrictions on grammeme co-occurrence. Thus, it is conceivable that a grammeme with a more specific meaning will incur more restrictions, as it is more likely to be semantically incompatible or redundant in certain grammeme combinations. On the other hand, there may be formal restrictions on combinations of morphemes (not necessarily found with zero morphemes), which will be discussed in the next section. A second qualification concerning the role of markedness for grammeme interaction pertains to the phenomenon of local (un)markedness. In fact, general markedness is often unable to explain co-occurrence restrictions between different grammemes, as a particular combination may be marked (i.e., less natural) for some members of a grammatical category but unmarked (natural) for others (see Tiersma 1982; Croft 1990: 144146 on local markedness). The interaction of imperative mood with the category of person, discussed in 3.4 above, is instructive in this respect. Indeed, the 1st person is arguably the most marked category with respect to the imperative, while the 2nd person is the least marked (cf., e.g., Croft 1990: 149). Similar patterns of local markedness can be found in the interaction of other mood (illocutionary force) markers with person as well. Kwaza (isolate, Amazonia) is instructive in this respect. In this language, mood markers occur only with certain person forms and are in complementary distribution: imperative forms are found only in the 2nd person, hortative forms (distinct from imperative) in the 3rd person, while special volitional/intentional forms are used only for the 1st person (van der Voort 2004). Analogous relations of local markedness can be observed with other categories. For example, indirect evidential forms are clearly marked with respect to the 1st person, but can be regarded as unmarked in relation to the 3rd person. Below, the relations of local markedness which obtain between particular grammemes will be captured in the form of markedness hierarchies (see Section 6).

262 5.2.

A. Malchukov Structural factors

It is generally acknowledged that interaction of categories, insofar as it is crosslinguistically consistent, usually has a functional motivation, such as relevance or semantic compatibility. This is also evident from the fact that crosslinguistically recurrent types of category interaction pertain to categories that share certain semantic components. Therefore one far more often finds interaction between, say, mood and tense, or tense and aspect which pertain to a shared semantic domain, while some other types of interaction are attested less regularly, e.g., between person and tense, or mood and transitivity (Xrakovskij 1996; cf. Chung and Timberlake 1985: 206; Palmer 1986: 209). This, however, does not preclude the existence, in individual languages, of interdependencies between grammemes for structural reasons. Although most such structural factors are language-specific and hardly generalizable, the following generalization seems to hold:26 Generalization 5. Combinability of morphological categories is crosslinguistically more restricted than combinability of periphrastic categories. For example, in Basque there are generally no restrictions on combinations of tenses (present/past) and aspects, as the former are expressed on the participle and the latter on the auxiliary verb. But the minor class of synthetic verbs, which directly inflect for tense and take no auxiliary, lacks aspectual distinctions (Saltarelli 1988). In Maltese (Arabic) there are no restrictions on combinations of voices (expressed by suppletive forms) and aspects, which are expressed by a combination of finite verbs with auxiliaries or particles (Borg and Azzopardi-Alexander 1997). In Turkish, the four basic tense/aspect forms, perfective past (-dI), aorist (-Ar), past (-mI), and progressive, are not combinable within one word form, but such combinations are possible in periphrastic forms (Kornfilt 1997). This does not mean, of course, that there is any general constraint on combination of morphological categories, although most fusional languages do impose restrictions of some sort in that domain. In agglutinative languages, on the other hand, there may be very few such restrictions. An extreme case in this respect is (West Greenlandic) Eskimo, which in spite of its rich (agglutinative) morphology apparently has no restrictions on combinations of numerous verbal (aspectual, tense and voice) categories (Fortescue 1984). One of the few restrictions noted by Fortescue (1984: 287), which is clearly due to semantic factors, is the incompatibility of the imperative with the perfect marker -rir- already. (Recall the restricted use of perfect imperatives in Finnish). However, in some other languages, morpheme combinations are highly restricted within a single word form. Thus, in Ika (Chibchan) a constraint is

Interaction of verbal categories 263 found which can be labeled a one suffix per word constraint. Normally only one suffix slot is available for expression of verbal categories; if several categories need to be expressed (for example, negation and aspect, as in the following example), auxiliary support is needed (Frank 1985): Ika (51) Nik-u nan-an work-NEG AUX-IMFV I will not work. (Frank 1985: 64) na-ngua. 1SG-will

Note that if one takes mutual incompatibility as a diagnostic for assigning grammemes to the same category, it can be claimed that all verbal grammemes belong to the same grammatical category in Ika. This brings us to the second general structural principle regulating compatibility between grammemes, which is built into the definition of the grammatical category: Generalization 6. Members of the same grammatical category (of the same paradigm) in individual languages are mutually exclusive by definition. As is also clear from the formulation of this constraint, Generalization 6 has a somewhat different status, as compared to Generalization 15. First, it is formulated in absolute terms: members of the same paradigm (or, maybe better, incumbents of the same structural slot) are mutually exclusive per definition. Second, this constraint (as well as the notion of a paradigm from which it is derivative) is language-particular, while other generalizations are crosslinguistic (see Haspelmath 2007 on the frequent confusion of language-particular and crosslinguistic categories and generalizations). Now, a paradigm invoked by Generalization 6 is a product of a language-specific diachronic developments involving successive grammaticalization of individuals forms. Such processes are regulated by general principles (as reflected in Generalizations 15), but importantly cannot be reduced to/explained by a single principle (such as semantic compatibility). Of course, as long as paradigms are semantically coherent, incompatibility of members of a grammatical category is expected and may be argued to be semantically motivated. For example, incompatibility of past and future within the same form may be explained by their semantic incompatibility. However, contrary to authors who stress the semantic basis of organization of grammatical categories and even build in a semantic/logical incompatibility of grammemes belonging to the same category into the definition of grammatical category (cf. e.g., Meluk 1997: 247248), it should be noted that, crosslinguistically, grammatical categories are quite often semantically incoherent (cf. Plungian 2000: 115 for a similar view). Some Amerindian polysynthetic languages with templatic morphology provide a good illustration

264

A. Malchukov

in this regard. For example, in Koasati (Muskogean), which has 9 prefixal and 15 suffixal slots, the distribution of particular grammemes across the slots is largely idiosyncratic (Kimball 1991). Thus, grammemes of the same functional class are often distributed into different slots, while grammemes with radically different functions can be hosted in the same slot and be mutually exclusive. For example, one of the suffixal slots (slot +5 in Kimballs notation) hosts suffixes meaning can, regularly and reluctantly, while epistemic meaning may be rendered through suffixes appearing in three different slots (slots +7, +8 and +9 in Kimballs notation). Further, the tense category is expressed by 2 discontinuous suffix orders, as well as by stem modification. Still more instructive in this respect is the case of Kwamera (Lindstrom and Lynch 1994), which assigns future and non-future (past/present) categories to different slots. Unexpectedly, the combination of future and nonfuture grammemes is possible, but is reinterpreted as an immediate future. This brings out an important point: while grammemes from the same slot are always structurally incompatible (by definition), the reverse rule does not hold: grammemes of different categories which should be structurally compatible may be avoided or reinterpreted due to semantic reasons. Here, then, we see a compromise between functional and structural factors. Most often, however, structural factors are too language-specific to be generalizable and are better explained in diachronic rather than synchronic terms. For example, nothing a priori can explain why the emphatic mood (Avrorins 1962 utverditeljnoe naklonenie lit. assertive mood) in Nanai (Tungusic) cannot be negated. Note that this cannot be due to markedness alone, as other moods both indicative and non-indicative do not show any restrictions on combinability with negation. The reason however, is simple: the emphatic mood forms derive historically from the combination of a negative verb form and an emphatic particle. Compare the assertive mood in (52) and the negative form of the indicative mood in (53): Nanai (52) Debo-a.ca-i=ka! work-PST.AFF-1SG=PTCL I did work! (or more literally Didnt I work?!) (53) Debo-a.ca-i. work-PST.NEG-1SG I didnt work. (Avrorin 1962: 186) Thus the restriction on the use of negation in the emphatic mood can be attributed, in historical perspective, to the general constraint against doubling of negation in a verb form.

Interaction of verbal categories 265 5.3. Diachronic factors As noted above, structural factors are often best accounted for in diachronic terms. Some of the examples of category interaction discussed in the literature indeed appeal to diachronic factors. Thus, the cumulative expression of gender/ number/person in verbal agreement, which is very common crosslinguistically, is determined not only by the mutual relevance and/or semantic overlap between these categories, but also by the fact that agreement commonly develops from encliticized pronouns, where these categories are inherent. Another example, discussed by Aikhenvald and Dixon (1998), is the interdependence between the verbal category of tense and the nominal categories of person and gender. Thus in Russian past-tense forms distinguish gender/number, while present-tense forms distinguish person. Cf.: Russian (54) a. stojal, stojal-a, stojal-o, stand.PAST.M.SG stand.PAST-F.SG stand.PAST-N.SG stojal-i stand.PAST-PL he/she/it/they stood b. stoj-u, sto-i, sto-it, . . . stand.PRES-SG stand.PRES-2SG stand.PRES-3SG (I) stand; (you) stand; (s/he) stands This is not an isolated phenomenon; a similar asymmetry obtains in other languages as well. Thus, in Hunzib (Daghestanian) person agreement (expressed by suffixes, in contrast to the class agreement prefixes) is found only in the present: Hunzib (55) a. x-o I/you write b. x S/he writes (van den Berg 1995: 76) In this case, too, the correlation between tense and person categories has a historic explanation, as Aikhenvald and Dixon (1998) do not fail to observe. In my view, the predisposition of past tense for gender agreement,27 can be explained in terms of three diachronic processes (well known from the grammaticalization literature), which are stated here in retrospect. Past tense forms frequently develop from perfects and resultatives (Bybee et al. 1994)

266

A. Malchukov

Resultatives and perfects frequently develop from nonfinite participial forms (cf. Nedjalkov 1988) Attributive nonfinite forms have a predisposition for developing gender agreement, while predicative (finite) forms have a predisposition for developing person agreement (note that person agreement is irrelevant for attributive forms, since the head of an attributive NP is usually 3rd person; Lehmann 1982). Thus three different diachronic processes conspire here to produce an asymmetry in combinability of different tense grammemes with gender and person markers. Similarly, a diachronic explanation can be invoked to explain voice neutralization in the evidential narrative present in Vogul (Mansi), exemplified in (56): Vogul (56) Tot-ne-m. bring-EV-1SG I supposedly brought/was brought. (Riese 2001: 53; Rombandeyeva 1973: 39) Although markedness can also be implicated here, the restriction on indirect evidential passive forms in Vogul seems to be due largely to the participial/ nonfinite origin of the -ne- form. Since this form was originally a present participle marker, the lack of the passive voice is unsurprising, given that nonfinite forms frequently lack some of the verbal trappings found on finite verbs. A diachronic explanation will likewise be offered in 6.1 below for cases where transitive perfects are restricted to the passive voice, which runs against the markedness pattern. 6. Discussion: interaction between different factors In this section I shall discuss some general issues pertaining to the interaction of different factors in constraining grammeme combinability, and suggest a way to integrate infelicitous combinations into a more general markedness pattern. 6.1. Interplay between different factors

As noted above, several different factors can contribute to grammeme (in)compatibility. Some of these factors reinforce rather than conflict with each other.

Interaction of verbal categories 267 Thus, semantic (in)compatibility is arguably related to relevance, inasmuch as infelicitous combinations can be viewed as cases where one grammeme is completely irrelevant to another. Both of these factors can be further related to markedness through the notion of local (un)markedness, understood as a natural correlation between values of different grammatical categories (Tiersma 1982; Croft 1990: 135). Some other factors may be but do not need to be in conflict. Thus, functional factors of local markedness, relevance and compatibility may be in conflict with economy, as in the case of redundant expression of categories which are by definition semantically compatible. Note, however, that economy constraints are operative only in cases when both categories have marked values and are overtly expressed, and even here a certain redundancy may be tolerated (see the Korean example on mood combinations in Section 2 above). The usual strategy to comply with both markedness and economy is to use cumulative expression for the most natural (unmarked) combination of values. This explains, for example, why second person imperatives are commonly expressed by cumulative markers, rather than by two separate grammemes. In some cases there may be a conflict between functional and structural factors. Generally it seems that functional factors pertaining to grammeme incompatibility are usually complied with. As we have seen above, certain functionally infelicitous combinations are consistently avoided, while more natural combinations of grammemes from the same categories are permitted. In other cases, however, both functional factors and markedness may be violated due to structural and/or diachronic factors. Thus, consider the case of Irish, where the passive is found only in progressive aspect: Irish (57) T Mire ag ol tae. is Mary at drink(ing) tea.GEN Mary is drinking tea. (58) T an tae ol ag Mire. is the tea PTCL drink(ing) at Mary The tea is being drunk by Mary. ( Siadhail 1989: 42) This is a violation of a markedness pattern which states that active, by virtue of being the unmarked voice, should be more receptive to co-occurrence with aspects. The Irish restriction has a straightforward structural explanation, however. In Irish only habitual Aspect is expressed inflectionally, while other aspects (including progressive) are periphrastic, involving a verbal noun (see (58) above). Since passive is also periphrastic, this results in a pattern where passive is available only in a marked periphrastic aspect, there being no simple (inflectional) passive forms corresponding to nonperiphrastic aspects.

268

A. Malchukov

One example where a markedness violation is due to diachronic factors involves constraints on the use of perfect forms in different voices, as discussed by Comrie (1976: 8486). Comrie notes that in some languages (overt) expression of the perfect is possible only in the passive voice and not in the active. He proceeds (1976: 84), In Russian, for instance, a sentence like konjak vypit means explicitly the brandy has been drunk, and not the brandy was drunk, which would be konjak byl vypit; but this distinction cannot be made in the active voice, where on vypil konjak corresponds to both he drank the brandy and he has drunk the brandy . Comrie, in part following Kuryowicz, explains this restriction in diachronic terms: transitive perfects derive from transitive stative (resultative) forms, which normally have a passive diathesis (cf. He is gone, but The stick is broken; on the correlation between objective resultative and passive, see Nedjalkov 1988). Thus the origin of the markedness violation may be related to the semantic evolution of individual grammemes.28 Note, however, that the latter example constitutes a markedness violation only if markedness is understood in the general sense (passive taken as a marked category as compared to active). It can be argued, however, that this pattern is consistent with the logic of local markedness, given that the correlation between the categories of perfect and passive is seen as natural or unmarked.29 (Note in this connection that this combination is also formally unmarked in Russian, as the perfect passive in the examples cited by Comrie lacks the auxiliary byl was, present in the past passive construction.) As we have seen in Section 3.5 above, languages such as Maori provide further evidence for regarding a correlation between passive and perfect(ive) as unmarked (see also Comrie 1981 for further discussion). Yet, on the whole there is little doubt that diachronic developments may interfere with distributional patterns as predicted by (general) markedness. For example, for the domain of nominal categories Plank and Schellinger (1997), reported cases where the pattern of gender/number neutralization runs counter to the general markedness pattern inasmuch as some languages may have forms with gender distinctions restricted to marked numbers. The authors provide a diachronic explanation for some of these cases, attributing such patterns to late grammaticalization of marked members of the number paradigm (especially duals) from numerals or nouns differentiated for gender (Plank and Schellinger 1997: 6973). 6.2. Multiple motivations Above we have discussed cases where several factors pertaining to category interaction are in conflict. In some other cases, however, different factors con-

Interaction of verbal categories 269 spire in producing the same outcome; in other words, the result of grammeme interaction can be attributed to several factors at once. This is already evident from the fact that markedness, which is frequently appealed to as a conditioning factor, is itself a cover term for several factors. Some examples of multiple motivations which conspire in producing specific patterns of dependencies between grammatical categories are given below (the first two cases, pertaining to dependencies between nominal categories, are based on Aikhenvald and Dixon 1998): Gender neutralization in the 1st and 2nd person, frequent in pronouns as well as in agreement forms, is due to both markedness and economy (that is, functional dispensability of gender specification for the speech act participants). Gender neutralization in plural pronouns is motivated, apart from markedness, by semantic compatibility, given that plural pronouns can refer to mixed genders groups.30 Number neutralization in oblique cases, as attested in Chukotkan languages (Meluk 1998), is not only due to markedness, but also to the fact that absolutive arguments are topics in these ergative languages, and (number) specification is more relevant for more topical participants. Tense neutralization in imperatives: in the past (and present) the expression of tense is disfavored due to semantic incompatibility; in the future it is disfavored due to redundancy. Lack of subject agreement on imperatives is motivated not only by markedness but also by economy considerations (that is, functional dispensability of 2nd person marking in languages with a restricted imperative paradigm). Voice neutralization in evidential forms (as exemplified for Vogul in (56) above) is motivated not only by markedness, but also diachronically (many evidential forms have a nonfinite origin). Asymmetries in combinability involving different tense grammemes with respect to gender and person agreement was attributed to a conspiracy of three different diachronic factors (see 5.3 above). Note that different factors may either conspire in (dis)favoring particular grammeme combinations (as in the last example), or they can also disfavor combinations involving several grammemes of a cross-cutting category (as in the case of restrictions on combinability of the imperative with various tense grammemes). The latter may result in a type of dependency (the 1st case of category interaction discussed in Section 1) where the presence of grammeme X of category x excludes category y (but different grammemes of category y are excluded for different reasons).

270 6.3.

A. Malchukov Infelicitous combinations and markedness: a reconciliation

In the above discussion we have regarded semantic incompatibility and (functional) markedness as distinct phenomena. However, it was also noted that these functional factors can be reconciled with each other as well as with relevance, as all three are related to the notion of local markedness. The phenomenon of local markedness is particularly relevant for the study of the interaction between verbal categories, as it pertains to markedness of certain grammeme combinations, rather than to markedness of grammemes per se in absolute terms. In the literature, local markedness is also known under the name of markedness reversal (Croft 1990). However, given that some categories involve more than one member, patterns of local markedness are better viewed as markedness hierarchies, reflecting the relative naturalness of certain grammeme combinations. This is consistent with Crofts (1990: 150) observation that many markedness reversal patterns turn out on closer inspection to be multivalued hierarchies. These hierarchies extend from most natural (unmarked) combinations, where grammemes are both compatible and highly relevant to each others content, at the one end, to combinations which are functionally incompatible and hence irrelevant, at the other end. In between we find combinations of categories which, although functionally compatible, are less relevant to one another. On this account the infelicitous combinations discussed in this paper represent the most marked combination of grammemes on the markedness hierarchies. By way of illustration, I present below three hierarchies pertaining to interaction of tense and aspect (Figure 1), interaction of person and (imperative) mood (Figure 2), and interaction of person and (indirect) evidentiality (Figure 3). The hierarchy in Figure 2 is adopted from the work by van der Auwera and his associates (van der Auwera et al. 2004; Gusev 2005; Schalley 2008). The two other hierarchies have not been proposed in the earlier literature to the best of my knowledge; they are set up here in reliance to the previous discussion as well as to the relevant literature. As observed by Comrie (1976: 73) and Dahl (1985: 80), the perfective grammeme (and, more generally, aspectual distinctions involving perfective as a marked member) is not equally compatible with different tenses: it is more often found in the past, less often in the future, and is usually lacking in the

Figure 1.

Tense Hierarchy for perfective aspect

Interaction of verbal categories 271

Figure 2. Person hierarchy for imperatives (van der Auwera et al. 2004; Gusev 2005; Schalley 2008)

Figure 3. Person hierarchy for evidentials

present, or else is reinterpreted. As noted above, in Romance languages the aspectual opposition obtains only in the past, while in Greek it is found in both past and future, but not in the present. In the Slavic languages it is extended to the present as well but the present perfective combination is reinterpreted (recall the discussion in 3.1). Evidence for all parts of the hierarchy can also be found together in one single language. Thus, in ChiBemba (Bantu), the aspectual distinctions (between perfective, imperfective and perfect) found in the past are somewhat reduced in the future (future perfect is lacking), and are completely neutralized in the present, which exists only in the imperfective (Chung and Timberlake 1985: 227228 citing Givn 1972). This pattern is expected, given that aspectual distinctions are most relevant for the past, less relevant for the future, and irrelevant for the present, as the present perfective combination is semantically infelicitous. For imperfective aspect this hierarchy is partially reversed, as imperfective naturally correlates with the present,31 but is partially retained. As noted by Dahl (1985: 110) past imperfective forms are more frequent crosslinguistically than future imperfective forms; this is clearly due to relevance. As noted in 3.4 above, there is a particular markedness pattern concerning the combinations of imperative and person (see Xrakovskij and Volodin 1986; van der Auwera et al. 2004; cf. also Croft 1990: 149). In van der Auwera et al. (2004) this markedness pattern is captured in the form of a semantic map, predicting the availability of dedicated imperative forms for particular persons. A simplified form of this map (disregarding the number dimension, among others) is represented in Figure 2 above (see Gusev [2005] and Schalley [2008] for similar proposals): it shows that 2nd person forms are most prone to being encoded by dedicated imperative forms (if a language has dedicated imperative forms at all), while 1st person singular and 1st person plural exclusive are

272

A. Malchukov

least likely to be found in an imperative paradigm. Thus, for example, in Armenian the imperative is restricted to the 2nd person, in Eskimo (West Greenlandic) it is restricted to the 2nd and 1st plural inclusive, in Finnish it is found in all persons apart from 1st person singular, and in Lingala it extends to all person categories (see van der Auwera et al. 2004; Gusev 2005; and Schalley 2008 for further discussion and exemplification). However, in the latter case, the infelicitous 1st person (singular or exclusive plural) imperative combination is likely to be reinterpreted, as shown above for Even. Again, other mood forms (illocutionary force markers) interact with person in a different way. Recall that in Kwaza, intentional/volitional forms are restricted to the 1st person, which suggests that 1st person volitional is the unmarked combination. It has been repeatedly noted in the literature that the use of indirect evidential (unwitnessed) forms is not equally felicitous with different persons (see Willet 1988; Kozintseva 1994; Johanson and Utas 2000; Aikhenvald 2004). Building on the earlier literature (in particular, on Aikhenvalds comprehensive study), one can set up the hierarchy in Figure 3 above, which restricts the use of indirect evidential forms in different persons. In accordance with the hierarchy, some languages effectively restrict indirect evidential forms to the 3rd person (Aikhenvald 2004: 232 cites Koasati, Kham, and Georgian as examples), some others to 3rd and 2nd person (e.g., Hunzib and Komi mentioned in 3.6; for other languages see Aikhenvald 2004: 232), while still others allow these forms for all persons. Yet, as noted in 3.6 above, the 1st person unwitnessed combination is often reinterpreted (see Curnow 2003 and Aikhenvald 2004 for further discussion and exemplification of 1st person effects in evidentials). The combined effects of all the subparts of this hierarchy can also be observed together in one single language: as noted above, indirect (non-visual) evidential forms in Nenets are frequent in the 3rd person, infrequent in the 2nd, and absent in the 1st person. Conversely, for the witnessed forms the unmarked combination is predictably the one involving the 1st person, although overt expression of both grammemes will be disfavored due to redundancy. Thus, in all cases functional principles underlying relations of local markedness, such as relevance and semantic compatibility, jointly shape the markedness hierarchies. A final qualification concerning the role of economy is in order here. As was noted in 5.1, certain grammeme combinations which are least marked and most natural may be avoided due to redundancy. However, this restriction pertains only to cases in which the respective categories are overt and independently expressed. In any case the combination of the respective values (not forms!) will always be available, but preferably encoded by a cumulative form in order to comply with economy. It is possible, as suggested in the functional-typological literature (Greenberg 1966; Croft 1990; Haspelmath 2006), that frequency is ultimately the driving force behind markedness relations and more generally behind markedness as a multidimen-

Interaction of verbal categories 273 sional correlation. Frequency can also explain the correlation between (local) markedness and economy: the most frequent forms are known to opt for the shortest form, hence the attested economy effects (see Haspelmath 2006 for further discussion). 7. Conclusion In this paper I have proposed a general approach to the study of infelicitous morpheme combinations and demonstrated its applicability to the domain of verbal categories. I also discussed the factors that determine the outcome of grammeme conflict. Resolution of infelicitous combinations has been shown to depend both on the particular structural properties of the given language and on the general principles of markedness and scope. All other things being equal the category which is (functionally) unmarked and which has narrow scope is more likely to be recessive and undergo the meaning shift. Broadening the discussion, I addressed other factors which influence combinability between grammemes both functional factors, such as relevance and economy, and structural and diachronic factors. Sometimes different factors conspire to produce the same pattern, while in other cases they are in conflict and thus can be regarded as competing motivations in the domain of grammeme interaction. Finally, I showed how different factors can be integrated into one model, which crucially relies on the concept of local markedness and markedness hierarchies. On this account, infelicitous combinations represent the most marked combination of values on the markedness hierarchies. In conclusion, I should mention some open questions which cannot be addressed in this exploratory paper and are left for future research. Thus, results of category interaction should be studied more systematically on different levels of grammar at the level of grammatical categories (as in Aikhenvald and Dixon [1998]), at the level of individual grammemes (as in the present paper), but also at the sublevels of forms and functions of individual grammemes (see Xrakovskij 2003). The discussion, mostly confined here to verbal categories, should be extended to include nominal categories (see, e.g., Plank and Schellinger 1997 on gender/number interaction),32 as well as to cases of interaction between grammatical and lexical categories (see, e.g., Xrakovskij et al. [2010] on interaction of lexical and grammatical aspect from a perspective similar to the one adopted here). Apart from direct dependencies, indirect dependencies between categories should be studied more systematically (the bestknown example of the latter is interdependency between case and person categories as captured by Silversteins generalization concerning the distribution of accusative and ergative patterns across languages).33 Also, the theory of grammeme interaction outlined here should be embedded into the general

274

A. Malchukov

theory of (resolution) of grammatical conflicts, as suggested by Moravcsik (forthcoming)34. Last but not least, an articulated semantic theory should be proposed to account for reinterpretation of interacting categories (see Malchukov 2009 for a proposal how to model meaning shifts in Optimality Theory). Yet, I believe that the present article has outlined a general research program for the study of interaction of grammatical categories, which will be hopefully followed up by in-depth typological studies of individual cases of category interaction. Received 9 January 2009 Revised version received 16 April 2010 Appendix. Abbreviations
A agent, ABS marker of agreement with the absolutive argument, ACC accusative, AFF affirmative, AH addressee honorific, ANT anterior (converb), AOR aorist, ASP aspect, AUD auditive (mood), AUX auxiliary, CONV converb, DAT dative, DC declarative (mood), DEF deferential, ERG ergative, EV evidential, EXCL exclusive, F feminine, FUT future, GEN genitive, HAB habitual, HORT hortative (mood), IMP imperative; IMFV imperfective, INC inclusive, IND indicative, INT interrogative, IRR irrealis, LOC locative, M masculine, MID middle (voice), NEG negation, NPRET non-preterite, O object, P patient, PART partitive, PASS passive, PF perfect, PART participle, PASS passive, PL plural, PLN plain (style), POS possessive, PR(E)S present, PROP propositive (mood), PTCL particle, REC reciprocal, R.Past realis past, REF reflexive, RETR retrospective (mood) , S subject, SG singular, SH subject honorific, SIM simultaneous (converb), SOC sociative; TOP topic, USIT usitative, 3->3 cumulative agreement marker of the 3rd person A acting on 3rd person O.

Max Planck Institute for Evolutionary Anthropology, Leipzig Institute for Linguistic Studies, St. Petersburg

Notes
* The paper originated from a project on interaction of verbal categories of the St. Petersburg typology group. I am particularly indebted to V. S. Xrakovskij for his advice and inspiration. The versions of this paper have been presented at the workshop on Interaction of verbal categories (St. Petersburg, November 2003), at the TAMTAM-workshop (Nijmegen, October 2006), and at the 7th ALT meeting (Paris, November 2008). I am grateful to the audiences for useful discussion. I am also indebted to Juliette Blevins, Jim Blevins, Denis Creissels, Misha Daniel, Orin Gensler, Martin Haspelmath, Lars Johanson, Leonid Kulikov, Edith Moravcsik, Toshiyuki Sadanobu, and Johan van der Auwera for the helpful comments on the earlier draft of this paper. I also benefited from discussion with Helen de Hoop, Henritte de Swart, Henk Zeevat and Joost Zwarts of some of the semantic issues pertaining to the domain of category interaction. Finally, the insightful comments of the anonymous reviewers are highly appreciated. The usual disclaimers apply. Correspondence address: Max Planck Insti-

Interaction of verbal categories 275

tute for Evolutionary Anthropology, Deutscher Platz 6, 04103 Leipzig, Germany. E-mail: Andrej_Malchukov@eva.mpg.de. Note that here we will only be concerned with cases of syntagmatic interaction of grammemes belonging to different categories (as opposed to paradigmatic interaction of grammemes belonging to the same category), which is a traditional topic of investigation in linguistic including typological studies. It should be noted that certain issues pertaining to syntagmatic interaction have been addressed under the rubric of markedness (see, e.g., Plank and Schellinger 1997 and other work cited in Section 5.1), yet these studies only partially cover the same field as they focus on restrictions (and patterns of syncretism) rather than on (re)interpretation. Equally few are monographic studies which specifically address category interaction in individual languages or cross-linguistically: in this connection a study by Poupynin (1999) on tense/aspect interaction in Russian, and a typological study by de Haan (1997) on interaction of modality and negation should be mentioned. Some cases discussed in this paper are better interpreted as being pragmatically infelicitous, rather than semantically incompatible. The term conflict resolution has been used in the literature in a somewhat different sense. In particular, in the literature on agreement (e.g., Corbett 2006), it is conventional to speak of resolution of gender/number conflicts when multiple agreement triggers impose conflicting requirements on an agreement target. For an illuminating general discussion of (kinds of) conflicts in syntax and their resolution see Moravcsik, forthcoming. Note that in some grammatical descriptions the term present perfective turns out on closer inspection to be a misnomer. For example, the present perfective forms reported for Konkani (Almeida 1985) and Yimas (Foley 1986) should be more appropriately qualified as present perfect forms. Thus, the present perfective in Yimas is used to refer to events which have already occurred and had been completed during today, including last night (Foley 1986: 244). Note that, unlike present perfectives, present perfects are not considered here as instantiating an infelicitous combination. We abstract away from those marginal contexts (such as performative use) where the semantic incompatibility of the present tense with the perfective aspect is suspended (see Note 8 below). There is a vast literature on the interaction of tense and aspect in Slavic; my account is closest to Breu 1994 (and from a more general perspective to other interactional approaches to the aspecto-temporal domain, such as Bache 1995, Smith 1997, Johanson 2000). And even to actual present when the incompatibility between present and perfective is suspended in special contexts, such as in performative use; Poprou vyjti [ask.PFV.1SG leave] I ask you to leave. Such performative uses of perfective presents are rather marginal in Russian, but are more widespread in some other Slavic languages, such as Slovene (Ljudmila Popovic p.c.). Breu (1994) suggests that (permanent) states (such as know) always get the inchoative interpretation in these contexts, while activities (such as play) receive the delimitative interpretation, and further attributes this asymmetry to the fact that activities, being temporary, imply the existence of both an initial and a final point, while states, being permanent, imply only an initial point. For some exceptions to this rule, representing residues of earlier stages of the languages development, see vedova (ed.) 1970: 422423; Knjazev (1998). Clark (1973) and Chung (1978) differ somewhat in their formulation of this correlation: Clark refers to perfectivity, while Chung refers to the objects affectedness as a triggering factor for the use of the passive. However, the relation between perfectivity and the objects affectedness is well established in typological literature (see Moravcsik 1978; Hopper and Thompson 1980; Tsunoda 1981; Malchukov 2006).

1.

2.

3. 4.

5.

6. 7. 8.

9.

10. 11.

276
12.

A. Malchukov
Note that in some languages like Paanci (Skinner 1979: 48) imperatives are restricted to the perfective aspect. We would predict that in all cases where neutralization goes in the opposite direction, that is, where imperatives are found only in the imperfective, the imperfective is the unmarked default aspect. This partially explains why imperfective imperatives are widely used in Russian, as the imperfective is generally held to be the unmarked member of the aspectual paradigm. See 4.1 and 5.2 below for discussion of markedness; also see van der Auwera et al. (2009) for further discussion of the perfective bias of imperatives. It should be noted that for many (western) Austronesian languages, the corresponding forms are qualified as patient focus forms rather than passives. Still the explanation provided above holds: the patient-focus form is NOT a marked member of a voice paradigm. Indeed, for different western Austronesian languages patient-focus forms have been argued to be either unmarked (which sometimes is interpreted as evidence for ergative alignment), or both a patient-focus and an agent-focus form are considered to be equally (un)marked (see Himmelmann 2002 for discussion and references). According to Aikhenvald (2004), these restrictions pertain in the first instance to languages with a small evidential system (and presumably where the non-firsthand forms are the marked member in the evidential paradigm). The terminology in this section follows Nedjalkov (2007a). In the glosses, AUX1 denotes the intransitive auxiliary, and AUX2 the transitive auxiliary. It is interesting to note that another famous structuralist, Louis Hjelmslev used the term dominant category in a sense somewhat similar to its use by Xrakovskij and also by the present author. Hjelmslev characterized a category as dominant with respect to another category if it is less prone to be affected by the syncretism than the category it dominates. See Plank and Schellinger (1997: 5658) for discussion of Hjelmslevs approach to markedness. Similarly in Finnish, which shows the same temporal shift, the accusative case (which triggers perfective interpretation when used on objects) is usually regarded as a marked case, and the partitive case (which triggers imperfective interpretation when used on objects) as an unmarked case for objects (Vainikka 1989). Although there may be mismatches between formal and functional markedness, we use here formal (un)markedness as a diagnostic for functional (un)markedness, on iconicity assumption that the two usually go together. See van der Auwera et al. (2009) for a similar approach. Cf. In Russian, the use of the present perfective form for future rather than present temporality supports the status of the perfective as the marked member of the Russian aspect opposition. In English, the occasional use of the simple (perfective) present for the expression of present temporality (e.g., present stative or habitual meaning) indicates that the simple form is the unmarked member of the English aspect opposition (Bache 1995: 190). A general definition of scope provided by Rice (2000: 24) appeals to semantic compositionality: given three items X, Y and Z, Z is said to have a scope over X and Y, if the semantics of Z is added to that of X and Y as a unit. In general terms, coercion is understood as an inferential process through which operatorargument conflicts are resolved in favor of an operator, which may reinterpret its argument in order to satisfy its selectional restrictions (Jackendoff 1990; cf. Pustejovsky 1995 on the rules of semantic coercion forcing reinterpretation of arguments to comply with the selectional restrictions of a verb). Daniel (2003) calls this phenomenon selection, illustrating it with a familiar case of ergative/instrumental polysemy: the same case marker is used in many languages (e.g., many Australian languages) as instrumental case when combined with inanimates, and as ergative case when combined with animates. (See also Aristar [1997] for a general discussion of the role of animacy in case-marking.) I prefer the term adaptation for cases where in an infe-

13.

14. 15. 16. 17.

18.

19.

20.

21. 22.

23.

Interaction of verbal categories 277

licitous combination a secondary meaning is selected (in preference to the default one) to meet the requirements of another grammeme. The neutralization is only partial, though, as residual aspectual distinctions are preserved in the imperative. Thus, in cases where the action is continued, only the imperfective imperative form can be used. Similarly, the imperfective form is preferred if the action is imminent or the preparatory phases are completed; thus Zaxodi Come in (imperfective)! is used in preference to Zajdi Come in (perfective)! when the addressee is already standing in the doorway (Birjulin and Xrakovskij 1992: 33; Padueva 1996: 6669). In a detailed crosslinguistic study, Plank and Schellinger (1997) note a number of counterexamples to this generalization, which is originally due to Greenberg (1966). However, as also acknowledged by the authors, most of such problematic cases violate Greenbergs generalization only on a stricter interpretation, which pertains to individual forms rather than to languages in general. This can also explain the fact, mentioned by Aikhenvald and Dixon 1998 (in Note 1), that although the authors assiduously searched the grammars of isolating languages, few dependencies between grammatical categories were found in these languages. Counterexamples to this generalization can likewise be accounted for diachronically. Thus Aikhenvald and Dixon (1998) note that in Hebrew present-tense forms distinguish gender forms more consistently than other tense forms, and mention in this connection that these forms are of participial origin. As noted by an anonymous reviewer, this explanation applies to Russian constructions synchronically as well, since forms like vypit have basically a stative interpretation. Similarly, Trask (1979) invokes (local) markedness to explain a correlation not only between perfect(ive) and passive, but also between perfect(ive) and ergative. See, however, Note 25 above for a qualification. Meluk (1998: 106) notes, for example, that in Uzbek the imperfective/progressive is restricted to the present tense. More examples of progressives restricted to present tense can be found in Dahl (1985: 94). Plank and Schellinger (1997) provide a comprehensive crosslinguistic study of interaction between gender and number, with a focus on patterns of syncretism rather than on reinterpretation. Some interesting examples of reinterpretation of infelicitous combinations in the nominal domain have been discussed by Daniel (2003), who notes in particular that plural forms of a noun with a unique referent (e.g., kin terms or proper nouns), if available at all, are often reinterpreted as associative plurals (e.g., in Turkic languages). For an interesting discussion of some of these indirect dependencies (in particular, those which concern the trade-off and complementarity between case and agreement in encoding argument relations) see Frajzyngier and Shay (2003). Moravcsiks (forthcoming) approach to conflicts in grammar is more inclusive insofar as she also considers cases of paradigmatic conflicts (in particular, deviations from isomorphic relation between form and function).

24.

25.

26. 27.

28. 29. 30. 31. 32.

33. 34.

References
Aikhenvald, Alexandra Y. 2004. Evidentiality. Oxford: Oxford University Press. Aikhenvald, Alexandra Y. & Robert Malcolm Ward Dixon. 1998. Dependencies between grammatical systems. Language 74. 5680. Almeida, Matthew. 1985. A description of Konkani. Washington DC: Georgetown University dissertation.

278

A. Malchukov

Apresjan, Jurij D. 2004. Principy organizacii centra i periferii v leksike i grammatike [Principles of organization of center and periphery in grammar and lexicon]. In Alexander P. Volodin (ed.), Tipologieskoe obosnovanie v grammatike [Typological explanation in grammar], 2036. Moscow: Znak. Aristar, Anthony R. 1997. Marking and hierarchy types, and the grammaticalization of casemarkers. Studies in Language 21. 313368. van der Auwera, Johan, Nina Dobrushina & Valentin Goussev. 2004. A semantic map for Imperatives-Hortatives. In Dominique Willems, Bart Defrancq, Timothy Colleman, Dirk Noel (eds.), Contrastive analysis in language: identifying linguistic units in comparison, 4469. New York: Palgrave Macmillan. van der Auwera, Johan, Andrej Malchukov & Ewa Schalley. 2009. Thoughts on (im)perfective imperatives. In Johannes Helmbrecht, Yoko Nishina, Yong-Min Shin, Stavros Skopeteas, Elisabeth Verhoeven (eds.), Form and function in language research: Festschrift for Christian Lehmann, 93107. Berlin & New York: Mouton de Gruyter. Avrorin, Valentin A. 1962. Grammatika nanajskogo jazyka [Nanai grammar]. Leningrad: Nauka. Bache, Carl. 1995. The study of aspect, tense and action: towards a theory of the semantics of grammatical categories. Frankfurt: Lang. Battistella, Edwin L. 1996. The logic of markedness. New York: Oxford University Press. Bauer, Winnifred, with William Parker and Te Kareongawai Evans. 1993. Maori. London: Routledge. Beck, David. 2004. Upper Necaxa Totanac. Munich: Lincom. Benzing, Johannes. 1955. Lamutische Grammatik. Wiesbaden: Steiner. van den Berg, Helma. 1995. A grammar of Hunzib. Munich: Lincom. Birulin, Leonid A. & Viktor S. Xrakovskij 1992. Povelitelnye predloenija: problemy teorii [Imperative constructions: theoretical framework]. In Viktor S. Xrakovskij (ed.), Tipologija imperativnyx konstrukcij [Typology of imperative constructions], 550. St. Petersburg: Nauka. Bondarko, Alexander V. 1971. Vid i vremja russkogo glagola [Tense and aspect of the Russian verb]. Leningrad: Nauka. Borg, Albert J. & Marie Azzopardi-Alexander. 1997. Maltese. London: Routledge. Breu, Walter. 1994. Interactions between lexical, temporal and aspectual meanings. Studies in Language 18(1). 2344. Breu, Walter (ed.). 2000. Probleme der Interaktion von Lexik und Aspekt. Tbingen: Niemeyer. Bruce, Leslie P. 1984. The Alamblak language of Papua New Guinea (East Sepik). Canberra: Australian National University. Bybee, Joan. 1985. Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins. Bybee, Joan, Revere Perkins & William Pagliuca. 1994. The evolution of grammar: tense, aspect, and modality in the languages of the world. Chicago: University of Chicago Press. Chung, Sandra. 1978. Case marking and grammatical relations in Polynesian. Austin & London: University of Texas Press. Chung, Sandra & Alan Timberlake. 1985. Tense, aspect and mood. In Timothy Shopen (ed.), Language typology and syntactic description: Vol. 3: Grammatical categories and the lexicon, 202207. Cambridge: Cambridge University Press. Clark, Ross. 1973. Passive and surface subject in Maori. Paper presented at the LSA meeting, San Diego, 1973. Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. Comrie, Bernard. 1981. Aspect and voice: Some reflections on perfect and passive. In Philip J. Tedeschi & Annie Zaenen (eds.), Tense and aspect (Syntax and Semantics 14.), 6578. New York: Academic Press. Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press.

Interaction of verbal categories 279

Cowell, Mark W. 1964. A reference grammar of Syrian Arabic. Washington DC: Georgetown University Press. Croft, William. 1990. Typology and universals. Cambridge: Cambridge University Press. Croft, William. 2003. Typology and universals, 2nd edn. Cambridge: Cambridge University Press. Curnow, Timothy J. 2003. Non-volitionality expressed through evidentials. Studies in Language 27. 3960. Dahl, sten. 1985. Tense and aspect systems. Oxford: Basil Blackwell. Daniel, Michael A. 2003. Dva primera vzaimodejstvija grammatieskix znaenij vnutri slovoform [Two cases of category interaction within a word form]. In Viktor S. Xrakovskij (ed.), Grammatieskie kategorii, ierarxii, svazi, vzaimodejstvie. [Grammatical categories: hierarchies, dependencies, and interactions. Proceedings of the St. Petersburg Conference], 5256. St. Petersburg: Nauka. Davies, Eirlys. 1986. The English imperative. London: Croom Helm. DeLancey, Scott. 1997. Mirativity: the grammatical marking of unexpected information. Linguistic Typology 1. 3352. Derbyshire, Desmond C. 1979. Hixkaryana. Amsterdam: North-Holland. van Driem, George. 1987. Grammar of Limbu. Berlin & New York: Mouton de Gruyter. Everett, Daniel & Barbara Kern. 1997. Wari: The Pacaas Novos language of western Brazil. London: Routledge. du Feu, Veronica. 1996. Rapanui. London: Routledge. Foley, William A. 1986. The Papuan languages of New Guinea. Cambridge: Cambridge University Press. Fortescue, Michael. 1984. West Greenlandic. London: Croom Helm. Frajzyngier, Zygmunt & Erin Shay. 2003. Explaining language structure through system interaction. Philadelphia: John Benjamins. Frank, Paul. 1985. A grammar of Ika. Philadelphia: University of Pennsylvania dissertation Givn, Talmy. 1972. Studies in ChiBemba and Bantu grammar (Studies in African Linguistics Supplement 3) Greenberg, Joseph. 1966. Language universals with special reference to feature hierarchies. The Hague: Mouton. Gusev, Valentin Ju. 2005. Tipologija specializirovannyx glagolnyx form imperativa [Typology of specialized verbal imperative forms]. Moscow: Moscow University. de Haan, Ferdinand. 1997. The interaction of modality and negation: Aa typological study. New York: Garland. Haas, Mary R. 1941. Tunica. Extract from Franz Boas (ed.), Handbook of American Indian Languages IV, 9143. New York: J. J. Augustin. Haspelmath, Martin. 1999. Explaining article-possessor complementarity: economic motivation in noun phrase syntax. Language 75(2). 227243. Haspelmath, Martin. 2006. Against markedness (and what to replace it with). Journal of Linguistics. 42(1). 2570. Hewitt, Brian G. 1979. Abkhaz. Amsterdam: North-Holland. Himmelmann, Nikolaus. 1999. Toratn. Munich: Lincom. Himmelmann, Nikolaus. 2002. Voice in western Austronesian: an update. In Fay Wouk & Malcolm Ross (eds.), The history and typology of Western Austronesian voice systems, 716. Canberra: Pacific Linguistics. Hopper, Paul J. & Sara A. Thompson. 1980. Transitivity in grammar and discourse. Language 56. 251299. Jackendoff, Ray. 1990. Semantic structures. Cambridge, MA: MIT Press. Jakobson, Roman. 1957. Shifters, verbal categories and the Russian verb. In Selected writings, vol. 2. Word and language, 130147. The Hague: Mouton.

280

A. Malchukov

Johanson, Lars. 1996. Terminality operators and their hierarchical status. In Betty Devriendt, Lois Goossens & Johan van der Auwera (eds.), Complex structures: A functionalist perspective, 229259. Berlin & New York: Mouton de Gruyter. Johanson, L. 1996. Terminality operators and their hierarchical status. In: Devriendt, B. et al. (eds.) Complex structures: a functionalist perspective, 229259. Berlin: Mouton de Gruyter. Johanson, Lars. 2000. Viewpoint operators in European languages. In sten Dahl (ed.), Tense and aspect in the languages of Europe, 27187. Berlin & New York: Mouton de Gruyter. Johanson, Lars & Bo Utas (eds.). 2000. Evidentials. Turkic, Iranian and neighboring languages. Berlin & New York: Mouton de Gruyter. Kemmer, Suzanne. 1993. The middle voice. Amsterdam & Philadelphia: John Benjamins. Kimball, Geoffrey D. 1991. Koasati grammar. Lincoln, NE: University of Nebraska Press. Knjazev, Jurij P. 1998. Parametry dla tipologii vida i russkij vid. [Parameters for the typology of aspect]. In Marina Ju. ertkova (ed.), Tipologija vida [Typology of aspect], 193207. Moscow: Studia Philologica. Kornfilt, Jaklin. 1997. Turkish. London: Routledge. Kozintseva, Natalia A. 1994. Kategorija evidencialnosti. Problemy tipologieskogo analiza [Evidentiality: problems of typological analysis]. Voprosy Jazykoznanija 3. 92105. Kozintseva, Natalia A. 1995. Modern Eastern Armenian. Mnchen/Newcastle: Lincom Europa. Kulikov, Leonid. 2001. The Vedic ya- presents. Leiden: University of Leiden dissertation. Kuzmina I. B. & E. V. Nemenko 1982. Istorija priastij. [History of participial forms]. In R. I. Avanesov & V. V. Ivanov (eds.), Istorieskaja grammatika russkogo jazyka. Morfologija. Glagol. [Historical grammar of Russian. Verbal morphology], 280411. Moscow: Nauka. Lehmann, Christian. 1982. Universal and typological aspects of agreement. In Hansjakob Seiler & Franz J. Stochowiak (eds.) Apprehension. Das sprachliche Erfassen von Gegenstnden. Teil II: Die Techniken und ihr Zusammenhang in Einzelsprachen, 201267. Tbingen: Narr. Lindstrm, Lamont & John Lynch. 1994. Kwamera. Munich: Lincom. Lublinskaja, Marina D. & Andrej L. Malchukov. 2007. Evidencialnost v nenetskom jazyke. [Evidentiality in Nenets]. In Viktor S. Xrakovskij (ed.), Evidencialnost v jazykax Evropy i Azii [Evidentiality in languages of Europe and Asia], 445468. Gedenkschrift for N. A. Kozintseva. St. Petersburg: Nauka. Majsak, Timur A. & Sergej G. Tatevosov. 1998. Vid i modalnostj: sposoby vzaimodejstvija (na materiale caxurskogo jazyka) [Aspect and modality: modes of interaction: Tsakhur evidence]. In Marina Ju. ertkova (ed.), Tipologija vida. [Typology of aspect], Moscow: Studia Philologica. Majsak, Timur A. & Sergej G. Tatevosov. 2000. Prostranstvo govoraego v kategorijax grammatiki [Egocentricity and grammatical categories]. Voprosy Jazykoznanija 5: 6881. Malchukov, Andrej L. 1995. Even. Munich: Lincom. Malchukov, Andrej L. 2000. Perfect, evidentiality and related categories in Tungusic languages. In: L. Johanson & B. Utas (eds.), Evidentials: Turkic, Iranian and neighbouring languages, 441471. Berlin: Mouton de Gruyter. Malchukov, Andrej. 2001. Imperative constructions in Even. In Viktor S. Xrakovskij (ed.), Typology of imperative constructions, 159180. Munich: Lincom. Malchukov, Andrej. 2006. Transitivity parameters and transitivity alternations: constraining covariation. In Leonid Kulikov, Andrej Malchukov & Peter de Swart (eds.), Studies on case, valency and transitivity, 329359. Amsterdam & Philadelphia: John Benjamins. Malchukov, Andrej. 2009. Incompatible categories: resolving the present perfective paradox. In Lotte Hogeweg, Helen de Hoop & Andrej Malchukov (eds.), Cross-linguistic Semantics of Tense, Aspect and Modality, 1333. Amsterdam & Philadelphia: John Benjamins. Meluk, Igor A. 1997. Kurs obej morfologii. Tom I. Vvedenie. ast pervaja: Slovo. Moskva: Jazyki russkoj kultury. 1997. Meluk, Igor A. 1998. Kurs obej morfologii [A course in general morphology], vol. 2. Moscow: Jazyki russkoj kultury.

Interaction of verbal categories 281

Michaelis, Laura A. 2003. Headless constructions and coercion by construction. In Elaine J. Francis & Laura A. Michaelis (eds.), Mismatch: Form-Function Incongruity and the Architecture of Grammar, 259311. Stanford: CSLI. Michaelis, Laura A. 2004. Type shifting in Construction Grammar: an integrated approach to aspectual coercion. Cognitive Linguistics 15. 167. Moravcsik, Edith. 1978. On the case marking of objects. In Joseph Greenberg (ed.), Universals of human language, vol. 4. Syntax, 249289. Stanford: Stanford University Press. Moravcsik, Edith. 2010. Conflict resolution in syntactic theory. Studies in Language 34:3. 636669. Nedjalkov, Igor V. 1992. Zalog, vid, vrema v tunguso-manurskix jazykax [Voice, aspect, tense in Tungusic languages]. St. Petersburg: University of St. Petersburg habilitation Nedjalkov, Vladimir P. (ed.). 1988. Typology of resultative constructions. Amsterdam & Philadelphia: John Benjamins. Nedjalkov, Vladimir P. 1991. Tipologija vzaimnyx konstrukcij [Typology of reciprocal constructions]. In Alexander V. Bondarko (ed.), Teorija funkcionalnoj grammatiki: personalnost, zalogovost [Theory of functional grammar: category of person and voice], 2837. St. Petersburg: Nauka. Nedjalkov, Vladimir P. 2007a. Overview of the research: definitions of terms, framework, and related issues. In Vladimir P. Nedjalkov (ed.), Reciprocal constructions, vol. 1. 3115. Amsterdam & Philadelphia: John Benjamins. Nedjalkov, Vladimir P. 2007b. Polysemy of reciprocal markers. In Vladimir P. Nedjalkov (ed.), Reciprocal constructions, vol. 1, 231335. Amsterdam & Philadelphia: John Benjamins. Neukom, Lukas & Manideepa Patnaik. 2003. A grammar of Oriya. Zrich: University of Zrich. Noonan, Michael. 1992. A grammar of Lango. Berlin & New York: Mouton de Gruyter. Siadhail, Michel. 1989. Modern Irish: Grammatical structure and dialectal variation. Cambridge: Cambridge University Press. Padueva, Elena V. 1996. Semantieskie issledovanija [Semantic studies]. Moscow: Jazyki russkoj kultury. Palmer, Frank R. 1986. Mood and modality. Cambridge: Cambridge University Press. Plank, Frans & Wolfgang Schellinger. 1997. The uneven distribution of genders over numbers: Greenberg Nos. 37 and 45. Linguistic Typology 1. 53101. Plank, Frans. 2003. Double articulation. In Frans Plank (ed.), Noun phrase structure in the languages of Europe, 337395. Berlin & New York: Mouton de Gruyter. Plank, Frans. 2006. The regular and the extended comitative reciprocal construction, illustrated from German. In Tasaku Tsunoda & Taro Kageyama (eds.), Voice and grammatical relations: In Honor of Masayoshi Shibatani, 247271. Amsterdam & Philadelphia: John Benjamins. Plungian, Vladimir. 2000. Obaja morfologija. [General morphology]. Moscow: Jazyki Russkoj Kuljtury. Poupynin, Youri A. 1999. Interaction between aspect and voice in Russian. Munich: Lincom. Pustejovsky, James. 1995. The generative lexicon. Cambridge, MA: MIT Press. Rennison, John R. 1997. Koromfe. London: Routledge. Rice, Keren. 2000. Morpheme order and semantic scope: word formation in the Athapaskan verb. Cambridge: Cambridge University Press. Riese, Timothy. 2001. Vogul. Munich: Lincom. Rombandeyeva, Evdokia I. 1973. Mansijskij (voguljskij) jazyk. [Mansi Language]. Moscow: Nauka. Saltarelli, Mario. 1988. Basque. London: Routledge. Sasse, Hans-Jrgen. 2002. Recent activity in the theory of aspect: accomplishments, achievements, or a non-progressive state. Linguistic Typology 6. 199271. erbak, Aleksandr M. 1981. Oerki po sravnitelnoj grammatike tjurkskix jazykov. Glagol. [Comparative grammar of Turkic languages: verbal categories]. Leningrad: Nauka. Schalley, Ewa. 2008. Imperatives: A typological approach. Antwerp: University of Antwerp dissertation.

282

A. Malchukov

Skinner, Margaret Gardner. 1979. Aspects of Paanci grammar. Madison, WI: University of Wisconsin dissertation. Smith, Carlota. 1991. The parameter of aspect. Dordrecht: Kluwer. Smith, Carlota. 1997. The parameter of aspect, 2nd edn. Dordrecht: Kluwer. Sohn, Ho-Min. 1994. Korean. London: Routledge. Sulkala, Helena & Merja Karjalainen. 1992. Finnish. London: Routledge. vedova, Natalja Ju. (ed.). 1970. Grammatika sovremennogo russkogo jazyka [Grammar of Modern Russian]. Moscow: Nauka. de Swart, Henritte. 1998. Aspect shift and coercion. Natural Language and Linguistic Theory 16. 347385. Tatevosov, Sergej. 2002. The parameter of actionality. Linguistic Typology 6. 317401. Taylor, Charles V. 1985. Nkore-Kiga. London: Croom Helm. Tereenko, Natalia M. 1947. Oerk grammatiki neneckogo jazyka [Grammar of Nenets]. Leningrad: Upedgiz. Tiersma, Peter M. 1982. Local and general markedness. Language 58. 832849. Trask, Robert L. 1979. On the origins of ergativity. In Frans Plank (ed.), Ergativity, 385407. London: Academic Press. Trubtskoj, Nikolay S. 1939/1969. Principles of Phonology. Berkeley: University of California Press. Tsunoda, Tasaku. 1981. Split case-marking in verb types and tense/aspect/mood. Linguistics 19. 389438. Vainikka, Anne. 1989. Deriving syntactic representations in Finnish. Amherst, MA: University of Massachusetts dissertation. Volodin, Alexander P. 1976. Itelmenskij jazyk [Itelmen language]. Leningrad: Nauka. van der Voort, Hein. 2004. Grammar of Kwaza. Berlin & New York: Mouton de Gruyter. Willet, Thomas. 1988. A cross-linguistic survey of grammaticalization of evidentiality. Studies in Language 12. 5197. Xrakovskij, Viktor S. 1990. Vzaimodejstvie grammatieskix kategorij glagola: opyt analiza [Interaction of verbal categories]. Voprosy Jazykoznanija 1990. 5. Xrakovskij, Viktor S. 1996. Grammatieskie kategorii glagola: opyt teorii vzaimodejstvija [Grammatical categories of the verb: towards a theory of category interaction]. In Alexander V. Bondarko (ed.), Mekategorialnye svjazi v grammatike [Category dependencies in grammar], 2243. St. Petersburg: Nauka. Xrakovskij, Viktor S. 2003. Grammatieskie kategorii glagola: svazi i vzaimodejstvie [Verbal categories: dependencies and interaction]. In Viktor S. Xrakovskij (ed.), Grammatieskie kategorii, ierarxii, svazi, vzaimodejstvie. [Grammatical categories: hierarchies, dependencies, and interactions. Proceedings of the St. Petersburg Conference], 156165. St. Petersburg: Nauka. Xrakovskij, Viktor S. (ed.). 1992. Tipologija imperativnyx konstrukcij [Typology of imperative constructions]. St. Petersburg: Nauka. Xrakovskij, Viktor S. (ed.). 2001. Typology of imperative constructions. Munich: Lincom. Xrakovskij, Viktor S. & Alexander P. Volodin. 1979. Ob osnovanijax vydelenija grammatieskix kategorij (vrema i naklonenie) [On identification of grammatical categories: tense and mood]. In Viktor S. Xrakovskij (ed.), Problemy lingvistieskoj tipologii i struktury jazyka [Problems of linguistic typology and language structure], 4254. Leningrad: Nauka. Xrakovskij, Viktor S. & Alexander P. Volodin. 1986. Tipologija imperativnyx konstrukcij: Russkij imperative [Typology of imperative constructions and imperative in Russian]. Leningrad: Nauka. Xrakovskij, Viktor S., Andrej L. Malchukov & Sergej Ju. Dmitrenko. (2010+). Grammatika akcionalnyx klassov [The grammar of actional classes]. In Alexander V. Bondarko (ed.), Teorija funkcionalnoj grammatiki: Kategorizacija semantiki [Theory of functional grammar: Semantic categorization]. St. Petersburg: Nauka.