On the Gradual Development of Polysynthesis in Nahuatl

16 On the Gradual Development of Polysynthesis in Nahuatl Jason D. Haugen 16.1 Introduction One of the widest-known and best-elaborated proposals for a macroparameter within the Principles and Parameters approach to syntactic variation is Baker (1996, 2001)’s ‘Polysynthesis Parameter’, which is very closely related to Jelinek (2002)’s ‘Pronominal Argument Parameter’. In both of these proposals certain languages require some kind of marking for both subject and object arguments on the main verb of a sentence. The macroparametric approach has been criticized along several lines, not the least of which is that it seems to presuppose an abrupt acquisition of the parameter in question, i.e. a ‘catastrophic’ change in grammar brought on by a new generation of speakers (acquirers) of a given language; see, for example, Trask (2002). This approach to parametric change is perceived to be inconsistent with other, more clearly- established aspects of historical syntax, such as the gradual grammaticalization of case-markers from adpositions. Furthermore, the macroparametric view posits certain factors that correlate with the parameter in question, but in most cases it has not been clearly established whether those correlations are caused by or lead to the development of that parameter. In this paper I consider one well-known polysynthetic Uto-Aztecan language, Nahuatl, in comparative and historical perspective. Although Nahuatl is in fact polysynthetic by the criteria set forth by Baker, it is unusual in this regard with respect to other Uto-Aztecan languages. Based on considerations from comparative Uto-Aztecan morphosyntax I will show that the development of polysynthesis in this language was most likely gradual (i.e. stepwise) rather than catastrophic. However, the comparative Uto- Aztecan evidence suggests that there were a series of stages in the development of full-blown polysynthesis, and each of these could have been catastrophic if a single generation of speakers made obligatory that which had previously been optional. The key factors in this gradual development were the ordered grammaticalization of previously optional subject and object clitics to obligatory pronominal affixes bound to the verb, and syntactic noun incorporation (NI). This paper is structured as follows. In section 2 I discuss the notion of the ‘polysynthesis parameter’, as presented by Baker (1996), and I make a distinction between subject and object pronominal ‘polysynthesis’, as suggested and rejected by Baker (2001), using evidence from Uto-Aztecan and elsewhere. In section 3 I give a brief overview of the structure of Nahuatl, and in sections 4 and 5 I discuss the comparative Uto-Aztecan evidence for the gradual development of subject and object pronominal affixes and syntactic noun incorporation, respectively. Section 6 offers my proposal for the various stages of grammaticalization involved in the transition from Proto-Uto-Aztecan (PUA) to polysynthesis in Nahuatl, and section 7 concludes. 16.2 On the Polysynthesis Parameter Baker (1996) establishes the ‘polysynthesis parameter’ as the setting of one single essential quality that he terms the ‘Morphological Visibility Condition’, defined as follows: (1) Morphological Visibility Condition (MVC) (Baker 1996: 486) A phrase X is visible for θ-role assignment from a head Y only if it is coindexed with a morpheme in the word containing Y via: (i) an agreement relationship; or (ii) a movement relationship. The MVC is intended to be a single property (i.e. a macroparameter) that explains a range of attributes that follow from the setting of this parameter. These attributes are the following: (2) Major characteristics of polysynthetic languages (Baker 1996: 498-9) a. Syntactic noun incorporation (NI) b. Object agreement is obligatory c. Free pro-drop d. Free word order e. No NP reflexive f. No true quantifiers g. Obligatory wh-movement h. N agrees with R argument i. No true determiners j. N agrees with possessor k. Restricted morphocausative l. NI or agreement in PP m. CP arguments only if nominal n. No infinitives Using evidence from noun phrases (NPs) in Southeast Puebla Nahuatl, MacSwan (1998) argues that two of the implicational relationships posited by Baker, (2)d and (2)f, do not hold up in that language. In Baker’s (1996) theory NPs must be adjuncts, and this theoretical point is supposed to explain both their optionality and their free word order. However, MacSwan claims that NPs are in argument positions in SE Puebla Nahuatl, and that evidence from weak cross-over effects indicate that this variety also has true quantifiers. MacSwan concludes that: the basic clause structure of Nahuatl, and perhaps that of other polysynthetic languages, is not radically different from English, Spanish or other well-studied languages. For this reason, while it may be useful for some purposes to refer to polysynthetic languages informally as those languages which are characterized by rich agreement morphology on the verb (sometimes called head-marking languages) plus productive NI, it does not seem likely that polysynthetic languages exist as a formal typological class, contrary to Baker’s proposals. (pp. 112-3) Accepting MacSwan’s conclusion, I apply the term ‘polysynthetic’ to languages that exhibit obligatory subject and object agreement plus syntactic NI, i.e. those languages which manifest (2)a and (2)b, as well as an additional stipulation that there be subject agreement on the verb. Baker himself has recently presented a similar definition of ‘polysynthesis’, as given in (3): (3) Polysynthesis Parameter (Baker 2001: 111) Verbs must include some expression of each of the main participants in the event described by the verb (subject, object, and indirect object). Baker also raises the possibility that this parameter might best be broken down into two separate parameters, one requiring subject agreement and one requiring object agreement, as in (4) and (5): (4) The Subject Polysynthesis Parameter (Baker 2001: 148) The subject of a verb must be expressed in that verb (e.g. Mohawk, Chichewa, Slave). (5) The Object Polysynthesis Parameter (Baker 2001: 149) The object of a verb must be expressed inside that verb (e.g. Mohawk) Baker (2001) presents Chichewa (Bantu) as a language that only has obligatory subject agreement on its verbs, and in this language subject NPs are optional. Object NPs can be deleted if there is object agreement, but if an object NP does appear without this agreement then it has to appear immediately after the verb. With object agreement, the word order of both the subject and object NPs is completely flexible. Other languages like this are attested, including Slave (Northern Athabaskan), which has an SOV order when object agreement is not present. Such languages as Chichewa and Slave would be ‘subject polysynthesis’ languages, given the bifurcated parameter suggested in (4) and (5). Baker rejects the partition in (4) and (5), however, because of the perceived lack of a language with object agreement without subject agreement, a language that Baker calls ‘Reverse Chichewa’. ‘Reverse Chichewa’ would be a language ‘in which the verb always agreed with the object, so that the object could be omitted or put on either edge of the sentence. But the verb would not have to agree with the subject, which would have to be present and in its usual position before the main verb or tense auxiliary’ (Baker 2001: 147). However, it is supposedly the case that no such language exists. To the contrary, though, at least one Uto-Aztecan language, Tohono O’odham (aka Papago) shows this property, although it does have an obligatory subject agreement marker cliticized to the second-position AUX element, as shown in the following data:1 (6) ’a:cim ’ ac g wisilo ø-ceposid (Zepeda 1983: 33) 1.PL.NOM. AUX DET calf 3.SG.OBJ.-branding ‘We are/were branding the calf’ (7) ’a:ñi ’añ g wipsilo ha-cecposid (Zepeda 1983: 33) 1.SG.NOM AUX DET calf.PL 3.PL.OBJ.-branding.PL ‘I am/was branding the calves’ The full pronominal paradigms of O’odham are given in Table 1: <<Insert Table 1 About Here>> According to Steele (1977)’s diachronic account of the development of these pronominals and auxiliaries, the Proto-Uto-Aztecan subject pronominal has fused with the tense/aspect AUX element in exactly the position that Baker suggests that subjects should appear in a ‘Reverse Chichewa’, ‘before the main verb or tense auxiliary’ (Baker 2001: 147). In addition to this object agreement without (verbal) subject agreement in O’odham, other Uto-Aztecan languages also demonstrate object-marking without subject agreement, for example in Hopi (a Northern Uto-Aztecan isolate), with obligatory object agreement on what K. Hill (2003) calls ‘pronominal verbs’, as in (8), as well as in Yaqui (Taracahitic), where third person object agreement is optional, as in (9):2 (8) Hopi a. Tiyòoya paavay àa-pungyala (Hill and Black 1998: 881) little.boy older.brother 3rd.SG.-keep.wanting.to.hang.around.with ‘The little boy wants to hang around with his older brother.’ b. Nu’ nuunukpàntuy amùu-piyna 1sg evil.ones-ACC 3rd.PL-make.leave ‘I made the bad guys go away.’ (9) Yaqui a. Inepo Hose-ta (aa)-vicha-k (Maria Florez Leyva, p.c.) 1.SG Hose-ACC (3.SG.OBJ)-see-PERF ‘I saw Hose’ b. Huan Hose-ta into Maria-ta (am)-vicha-k Huan Hose-ACC and Maria-ACC (3.PL.OBJ)-see-PERF ‘Huan saw Hose and Maria’ If we accept that O’odham is a language with (5) applied without (4), then we have reason to suppose that Baker’s proposal in (4) and (5) is in fact necessary to differentiate Chichewa/Slave- style languages from O’odham-style languages. Following Mithun (1991), I take it to be the case that the linear order of the agreement morphemes on the verb reflects the chronological order in which these elements were grammaticalized, in addition to reflecting the synchronic syntax of the language in question. Evidence in support of this view comes from languages that have morphologized the agreement patterns in different orders. Nahuatl patterns with Mohawk and most other polysynthetic languages in having its subject pronominals precede object pronominals; see §3 below. In the Athabaskan languages, however, object pronominals precede subject pronominals (Mithun 1991, Rice 2000). We can even see variation in the order of subject and object pronominals within Uto-Aztecan itself, since two Takic languages also have object-marking outside of subject-marking. In Cahuilla, transitive verbs are obligatorily marked for subject and object. As (10) illustrates, the object prefix precedes the subject prefix within the verb complex: (10) Cahuilla pe-n-téew-qal (Seiler 1977: 134 [156i]) 3.SG.OBJ-1.SG.SUBJ-STEM.trans-SUFF.DURAT. ‘I see him’ The situation in Cupeño is slightly more complex. In this language the past tense is marked by means of an obligatory subject prefix on the verb. The use of the third person plural prefix for past tense is shown in (11)a, contrasted with (11)b, which has a future tense verb form and therefore does not have a subject prefix:3 (11) Cupeño a. túku=’ep pem-chíx (J. Hill 2003: 210 [6a]) yesterday=R 3.PL-die.PL  ‘Yesterday they died’ b. tukumáy m=el=pe chíx (J. Hill 2003: 210 [6b]) tomorrow PL=3.PL.ABS=IRR die.PL ‘Tomorrow they will die’ Prefixal object clitics are optional for verbs in any tense. The use of the object clitic with a past tense form is given in (12); note that the object clitic appears before the subject prefix in this example: (12) túkup=’ep ’i-ché’-max (J. Hill 2003: 211 [7]) yesterday-R 2.SG.OBJ-1.PL-give ‘yesterday we gave it to you’ (2 43 480) This ordering of subject and object pronominals presents an interesting challenge to current theories of morphosyntax; see J. Hill (2003) for a plausible analysis. These examples from Cahuilla and Cupeño are quite interesting in that they clearly illustrate different degrees of grammaticalization of subject and object pronominals in two relatively closely related languages—both are within the Cupan group of the Takic sub-branch, which itself is a sub-branch of a larger grouping, Northern Uto-Aztecan. For our purposes here, I merely want to claim that the evidence suggests that subject agreement has been grammaticalized for the past tense in Cupeño independently from object agreement. Thus, to use Baker (2001)’s terminology, Cupeño has developed the ‘subject polysynthesis’ parameter for verbs in the past tense. Object prefixes are also possible in this language, although since they are not required no ‘object polysynthesis’ parameter has been set. Cahuilla, on the other hand, seems to have developed obligatory ‘subject polysynthesis’ for verbs in all tenses, and ‘object polysynthesis’ as well, although Cahuilla does not have syntactic noun incorporation and is thus not a typical polysynthetic language as discussed by Baker (1996). Crucially, this variation shows that there are two independent parameters for marking agreement on a verb. In theory, a future generation of Cupeño speakers could possibly reanalyze the object clitics as obligatory, and if this were to occur then the addition of the object polysynthesis parameter to the subject polysynthesis parameter would yield something like Baker’s polysynthesis parameter, but with a morpheme order opposite to that which we see in Nahuatl. Having justified a distinction between subject and object polysynthesis, we now turn to a brief discussion of the structure of Nahuatl (§3) before establishing the Uto-Aztecan bases for Nahuatl’s polysynthetic characteristics in §4-5. 6.3 On the Structure of Nahuatl Most varieties of Nahuatl, including Classical Nahuatl, are unambiguously polysynthetic in Baker’s syntactic sense: verbs in this language must be inflected for both subject and object agreement, and this inflection takes the form of prefixes.4 The relevant prefixes for Classical Nahuatl are given in (13) and (14), respectively:5 (13) Classical Nahuatl Subject Prefixes (Newman 1967: 193) Sg. Pl. 1st person: ni- ti- 2nd person: ti- am- 3rd person: ø- ø- (14) Classical Nahuatl Object Prefixes (Newman 1967: 193) Sg. Pl. 1st person: nēch- tēch- 2nd person: mitz- amēch- 3rd person: k(i)- k(i)- The reduced forms of the subject pronominals are obviously related to the independent pronouns, at least in the first and second person: (15) Classical Nahuatl Independent Pronouns (Steele 1977: 571 [15]) sg. pl. 1st person: newātl tewān(tin) 2nd person: tewātl amewān(tin) 3rd person: yewātl yewān(tin) In addition to marking a verb with one of the direct object prefixes in (14), object-marking can also occur via the incorporation of a nominal root which is ostensibly the direct object of the verb. Merlan (1976) discusses the pragmatics behind such incorporation in Huauhtla Nahuatl (HN). In general, noun incorporation is used to maintain topicality, i.e. definiteness in discourse. As Merlan explains, ‘In contrast with complete pronominalization in HN, the nature of incorporation is to permit preservation of the lexical properties of NP adjuncts previously introduced, and thus definiteness of discourse reference is maintained’ (p. 185). In Merlan’s terminology this applies only to ‘contextual’ or ‘discourse-determined’ NI; there are also cases of what Merlan refers to as ‘lexical’ NI, which is the obligatory use of NI in certain constructions. This ‘lexical’ NI involves idiomatic meanings induced by noun incorporation. For example, in (16), a literal interpretation occurs with a non-incorporating structure (16)a, but an idiomatic reading is generated with NI, as in (16)b: (16)a. ø-neč-maka-ø-k patli (Merlan 1976: 184) 3SG-1SG-GIVE-PAST-SG medicine ‘He gave me medicine’ b. ø-neč-pa-maka-ø-k (Merlan 1976: 184 [4]) 3SG-1SG-medicine-GIVE-PAST-SG ‘He doctored me’ (lit. ‘He medicine-gave me’) In addition to the incorporation of direct objects, unaccusative subjects and instruments may also incorporate in Nahuatl, as in (17) and (18), respectively: (17) tla-ā-weci-ø-ø (Merlan 1976: 183 [1]) INDEF-water-fall-PRES-SG ‘It is raining’ (18) Ya ø-ki-kočillo-teteki panci (Merlan 1976: 185 [10B]) he 3.SG.S-3.SG.OBJ-knife-cut bread ‘He cut the bread with a knife’ In this paper we will only concern ourselves with the incorporation of direct objects, which is common across Uto-Aztecan (Haugen 2008, and §5 below). We now examine the Uto-Aztecan context for the crucial aspects of polysynthesis in Nahuatl: subject and object clitics and agreement (§4) and noun incorporation (§5). 6.4 Subject and Object Agreement in Uto-Aztecan Most Uto-Aztecan languages have subject and object clitics or affixes of some kind. As we saw above, the order of subject and object affixes can vary, and I take this to be evidence of the different chronological ordering of the morphological fusion of these affixes to the verb in the languages in question. In the usual case, the object marker has grammaticalized first, and some languages have set the ‘object polysynthesis parameter’ independently from ‘subject polysynthesis’. Evidence for this comes from Tohono O’odham (Tepiman), where all transitive verbs must be marked for the direct object (as in 6 and 7 above). Additionally, certain verbs in Hopi require direct object pronominal marking (as in 8), and transitive verbs in Yaqui can optionally be marked for the direct object (as in 9). By hypothesis, PUA had similar patterns of optional pronominal object marking, and over time Pre-Nahuatl (PN) gradually grammaticalized obligatory object agreement for all transitive verbs that did not have overt noun incorporation, as is currently the case in O’odham. Thus, at some point PN set the ‘object polysynthesis parameter’, and at this stage PN was like Tohono O’odham in having obligatory object agreement. At this stage this parameter could be satisfied by either the incorporation of a nominal root or a direct object pronominal into the verb. I assume that at this stage subject agreement was not obligatory. With respect to subject pronominals, most of the Uto-Aztecan languages have subject clitics or affixes, and an examination of these morphemes shows that they are clearly cognate. Examples from O’odham were given above in (Table 1), and from Nahuatl in (13). Some other representative examples are given below: (19) Tübatulabal6 (Steele 1977: 567 [4]) Clitic Pronouns Independent Pronouns sg. dl. pl. sg. dl. pl. 1 g(i) gil(a) (gi)luuc nik inggila inngiluuc (incl.) gilaang inggilaang (excl.) 2 b(i) bu(u)m(u) imbi imbuumu 3 ø da in inda (20) Serrano (Steele 1977: 567 [5]) Clitic Pronouns Independent Pronouns sg. pl. sg. pl. 1 n čim n ačam 2 č cm mi m 3 v m ivi iim (near) pta pm (close) ama aam (far) (21) Yaqui (Steele 1977: 569-70 [11]) Clitic Pronouns Independent Pronouns sg. pl. sg. pl. 1 ne te, itom inepo itepo 2 e, en em, eme empo eme?e 3 ø ø aapo bempo (22) Huichol (Steele 1977: 570-1 [14]) Clitic Pronouns Independent Pronouns sg. pl. sg. pl. 1 ne te nee taame 2 pe ze eek zeeme 3 ø we/me iik iime (near) mk mm (distant) iya (general) With respect to the distribution of these clitics within the sentence the Uto-Aztecan languages differ: some languages have second position clitics, some have pre-verbal clitics or affixes, and some have both. Steele (1977) presents the following table showing the distribution of subject clitics across a representative sample of the Uto-Aztecan languages: <<<Insert Table 2 About Here>>> Langacker (1977) reconstructs subject and object clitics for Proto-Uto-Aztecan (PUA), stating that ‘it seems fairly clear that subject clitics, but not subject verb prefixes, can be reconstructed for [PUA]’ (p. 126). He offers the following as speculative reconstructions, ‘offered mainly as a basis for research and discussion’ (p. 126): (23) PUA Subject Clitics (Langacker 1977: 126) SG PL 1P *=n *=ta 2P *=’ *=’-m 3P *(=p) *=(p)-m (24) PUA Independent Definite Pronouns (Langacker 1977: 124) SG PL 1P *(i-)n *(i-)ta(-m) 2P *(-m) *’-m 3P-H *p *p-m 3P-NH *a *a-m Based on the extant distribution of subject clitics in the family, it appears that subject clitics in PUA could either be in second position or prefixed to the verb, and this property has been retained in several languages, including Tarahumara:7 (25) Tarahumara čú mu šika ké mu nakí muhé ko ba (Steele 1977: 553 [30]) WH CP WH NEG CP want you emphatic emphatic ‘Why don’t you want it?’ Following Langacker (1977), Steele (1977), and much other work, I assume that the pronominal prefixes of Nahuatl developed from these PUA clitics. If the suggestion that separating subject polysynthesis and object polysynthesis is correct, then it is plausible that it was only after PN already had obligatory object-marking (i.e. ‘object polysynthesis’, as in Tohono O’odham) that these pre-verbal subject clitics were reanalyzed as obligatory subject agreement. It would have been then that they ‘fused’ to (i.e. became morphologically a part of) the verbal complex. It was with this reanalysis that PN then had obligatory subject agreement and thus the ‘subject polysynthesis parameter’ had been set. In this way we can think of the setting of the polysynthesis parameter, as defined in (3), as an ordered sequence of the independent setting of separate sub-parameters, namely, (4) and (5). In this section we have accounted for the gradual development of the weaker of Baker’s two definitions of the polysynthesis parameter (cf. 3). In the next section we will consider the stronger version, in which polysynthesis is also related to syntactic noun incorporation (cf. 2a). 5. Syntactic Noun Incorporation in Uto-Aztecan In addition to obligatory object agreement (2b), the second critical element to Baker (1996)’s definition of the polysynthesis parameter is syntactic noun incorporation (2a). Syntactic noun incorporation is usually taken to be the type of noun incorporation that occurs when a noun-verb complex retains its transitivity, a key diagnostic for which is the presence of external elements (determiners, adjectives, etc., often referred to as ‘stranded modifiers’) modifying any incorporated nouns; this has been termed Classifier NI (Rosen 1989) or Type 4 / classificatory NI (Mithun 1984). In some cases syntactic noun incorporation takes a full DP complement, where the head noun of the complement NP is hyponymous with respect to the incorporated noun. Examples of such syntactic NI are evident in Hopi, which allows for both ‘modifier stranding’ (26) and hyponymous object arguments (27): (26) Hopi Naat itam pu-t qatsi-yese. (K. Hill 2003: 234 [98]) still we that.-acc life-sit(PL) ‘We are still living that life.’ (27) Pam tsiili-t nakwa-’y-ta-ngwu. (K. Hill 2003: 236 [112]) he chile-ACC feather.worn.on.head-POSS-DUR-HAB ‘He (the Hehey’a kachina) wears chili pepper (‘chile as a feather’) on his head.’ Syntactic NI can be contrasted with another kind of NI, that which creates an intransitive verb. This type of NI is associated with compounding and has been referred to as compound NI (Rosen 1989), or ‘Type 1 NI’ in the typology of Mithun (1984).8 The Southern Uto-Aztecan language Yaqui has some examples of intransitivizing NI, although this process is not productive in this language: (28) Yaqui a. aapo maso-ta peu-ta-k (Jelinek 1998: 213 [48]) 3sg deer-ACC butcher-TRAN-PERF ‘He butchered a deer’ b. aapo maso-peu-te-n 3sg deer-butcher-INTR-PAST ‘He was deer butchering’ c. *aapo bwe’uu-k maso-peu-te-n 3sg big-ACC deer-butcher-INT-PAST [*‘He was [big deer]-butchering’] or [*’He was deer-butchering a big one’] Example (28)a shows the use of the verb peu- as a standard transitive verb, which is marked with the transitivity suffix –ta, and in this sentence the direct object maaso ‘deer’ appears in object position and takes accusative case (marked with the accusative case suffix –ta). In (28)b maaso is incorporated into the verb, and thus is not marked with case. The intransitive verb that is formed is marked with the intransitive verb marker –te. (28)c shows that the incorporated nominal cannot be modified by adjectives, determiners, or the like (i.e. there is no ‘modifier- stranding’). It is unclear from the extant grammatical descriptions which of these two types of NI is currently more general in Uto-Aztecan with respect to lexical verbs. Following Hale and Keyser (1993 and subsequent work), though, Haugen (2008) takes the position that denominal verbs involve obligatory incorporation of the head nominal in the verbal complement, by claiming that denominal verb morphemes are bound affixes that are base-generated in v.9 Denominal verbs do generally show the characteristics of syntactic NI across the Uto-Aztecan languages. For example, ‘modifier-stranding’ seems to be typical of denominal verbs in all of the Uto-Aztecan languages. Some representative examples are given below. (29) Gosiute Shoshone (Miller 1996: 706) N wahatti kahnikantn I two-OBJ house-HAVE ‘I have two houses’ (30) Comanche (Charney 1993:205: [78]) [suRIse nknu tammati sonipIφai] suti=-se n-knu tamma=-ti soni-=ph-pai that.one-CNTR my-father’s.father a.lot-OBJ grass, hay-ABS-pai ‘My grandfather had hay (for horses)’ (31) Southern Paiute (Sapir 1930: 134) wa·΄q·utcan qava·΄x·Α two-OBJ-preterite-I horse-GET ‘I received two horses’ (32) Tohono O’odham (Saxton 1982: 141) n-t wo ha’i kii-ki-t  I-TNS FUT some RED-house-MAKE ‘I’m going to build some houses’ (33) Southeastern Tepehuan (Willett 1991: 64 [141]) day ma’n tu-sa’ua-h’iñ only one OWN-blanket-1s ‘I have only one blanket’ (34) Western Tarahumara (Burgess 1984: 28) bilé mahtá-ga-me one metate-STAT-PRTC ‘I have one metate’ (35) Yaqui 'aapo hiva tu’ii-k te-tekil-e (Jelinek 2003: 207 [24]) 3sg always good-ACC RED-job-ITER-IMP ‘He always has a good job’ Uto-Aztecan denominal verbs can also often take hyponymous arguments, thus instantiating ‘classificatory noun incorporation’: (36) Tümpisa Shoshone (Dayley 1989: 91 [123]) Nümmü so’oppüh putish pungkupaimmippühantü we(exc) many burro pet-HAVE-HAB-PAST ‘We used to have many burro pets’ (37) Gosiute Shoshone (Miller 1996: 706) Isapaipph sukka ponaiha taipai Coyote that-OBJ Mouse- OBJ brother-HAVE ‘Coyote has Mouse for a younger brother’ (38) Cupeño (Jane H. Hill, p.c.) túku=’ep ne-’ásh-lyu awá-l-i yesterday=r 1s-pet-VB dog-NPN-O ‘Yesterday I had a dog’ (39) Yaqui ’uka ’ ili chu’u-ta ne=vuk-ek (Haugen 2004: 207 [13]) det.ACC little dog-ACC 1.sg=pet-PERF ‘That little dog is my pet’ (lit. ‘I pet-have that little dog.’) Based on such evidence as that given above, Haugen (2008) reconstructs syntactic NI, including modifier-stranding and hyponymous object arguments, for PUA. The crucial point that I would like to make here is that, just as there are non-polysynthetic languages with subject or object marking on their verbs, there are also non-polysynthetic languages that have syntactic NI, e.g. the Northern Uto-Aztecan language Hopi. I conclude from this discussion that the two key aspects of Baker (1996)’s definition of the polysynthesis parameter (i.e. obligatory object-marking and syntactic NI) are independent variables, and thus this ‘macroparameter’ does not necessarily entail a catastrophic development. In the next section I will propose a likely scenario for the step-wise development of the various stages that Pre- Nahuatl must have gone through en route to full-blown polysynthesis, each of which themselves could have involved the setting of a single parameter by a single generation of speakers. The instructive aspect of the comparative Uto-Aztecan evidence is that each of these proposed stages has independent attestation elsewhere in the family. 16.6 A Grammaticalization Cline for Polysynthesis in Nahuatl I propose that the ‘polysynthesis parameter’ developed in Nahuatl in the following gradual stages. First, Pre-Nahuatl (PN) descended from a language, Proto-Uto-Aztecan (PUA), which had both subject and object clitics, as argued in §4, as well as syntactic NI, as argued in §5. By hypothesis, the subject clitics of PN could appear either in second position or pre-verbally, as I have argued to have been the case for PUA, and as is still the case in many extant Uto-Aztecan languages, e.g. Tarahumara. Presumably, object clitics were pre-verbal, as in Hopi, Yaqui, and other Uto-Aztecan languages. We can refer to this initial period, inherited directly from PUA, as Stage One. Stage Two would have involved the setting of the object polysynthesis parameter to ‘on’, and transitive verbs would have required either object agreement or noun incorporation. The ‘object polysynthesis property’ is also attested in Tohono O’odham and other Tepiman languages. The difference between the hypothetical PN Stage Two and Tepiman, however, is that, with the exception of denominal verbs, the Tepiman languages do not have noun incorporation at all, but all transitive verbs must agree with their objects. That is, the Tepiman languages are Pronominal Argument languages, in the terms of Jelinek (2001). It is my hypothesis that the loss of NI in favor of obligatory object agreement in these languages was the result of the reanalysis of an incorporated pronominal element as a general classifier, meaning something like ‘thing’ or ‘it’, which eventuality lost its referentiality and became just a marker of agreement. At Stage 2 of PN there would have been no requirement that verbs show agreement with their subjects, although subject clitics could have optionally attached to the object-verb complex. Subsequent to the fixing of the object polysynthesis parameter, the reanalysis of optional pre- verbal subject clitics as obligatory subject agreement markers would have constituted Stage Three of PN, which I assume led to Proto-Aztec, and this property has been retained in all of the varieties of Nahuatl that descended from that common ancestor. It was at this stage that the subject clitics ‘fused’ to the object-verb complex, and it was here that ‘full-blown polysynthesis’ occurred. Each of the changes that occurred to bring about the subsequent stages could conceivably have occurred in a single generation, thus each could have been a ‘catastrophic’ change brought about by the setting of a single parameter. Each of these parameter-settings would have involved the reanalysis of a grammatical option as an obligatory requirement in a particular individual’s grammar. This view is supported by the synchronic optionality of subject and/or object pronominal markers in various languages, a feature which I assume is conservative with respect to those Uto-Aztecan languages that have obligatory agreement. In this way it can be argued that the polysynthesis parameter itself did not arise, and could not plausibly have arisen, as a single change, but was, rather, the eventual result of a gradual series of changes that occurred in distinct stages. The grammaticalization trajectory that I have proposed for the polysynthesis parameter in Nahuatl is summarized below in Table 3. <<<Insert Table 3 About Here>>> A crucial aspect of my analysis is that subject agreement is completely independent of the object incorporation referred to in Table 3. Although Nahuatl has developed obligatory subject agreement, languages like Tohono O’odham (and other Tepiman languages) show us that this is not a necessary development. 7. Conclusion In this paper I have shown that multiple non-polysynthetic Uto-Aztecan languages display properties that have been attributed to polysynthesis, and I have suggested that these properties led to polysynthesis in Nahuatl, rather than the reverse. I have argued that polysynthesis developed in a discrete series of stages, each of which has attestation elsewhere in the family. The proposal made here is consistent with divergent work in both grammaticalization and more formal approaches to language change. The gradualist approach links a gradual syntactic change, the development of obligatory NI with the fixing of subject clitic placement, to the abrupt change required by approaches to diachronic syntax focused on child language acquisition (e.g. Lightfoot 1991). If this gradualist approach is correct for the polysynthesis parameter, then it could probably also be applied to other proposed macroparameters as well. Thus, this line of research suggests the need to search for potential ‘precursors’, in the form of retentions, in other languages and language families with no written record, in order to find (and reconstruct) likely sources for the ‘grammaticalization’ of particular ‘macroparameters’. Table 1. Tohono O’odham Object Prefixes and Free Object Pronouns (Zepeda 1983: 35) Object Prefixes (Optional) Free Object Pronouns Singular Plural Singular Plural 1st person: ñ- t- ’a:ñi ’a:cim 2nd person: m- em- ’a:pi ’a:pim 3rd person: ø- ha- hegai hegam Table 2. Subject clitics across Uto-Aztecan (Steele 1977: 541, 542-3 [II, III]) Second Position: Comanche, S. Paiute, Chemehuevi, Tübatulabal, Serrano, Luiseño, Tohono O’odham Proclitic/Prefix: Kitanemuk, Cahuilla, Huichol, Classical Nahuatl, Pochutla Both: Cupeño, Tepecano, Tarahumara, Yaqui, Cora Neither: Northern Paiute, Shoshone, Hopi Table 3. Grammaticalization trajectory of (object) polysynthesis Syntactic and Optional Object Obligatory Object Pronominal Classificatory NI Polysynthesis Polysynthesis Objects Only Incorporation of No nominal direct object Optional Incorporation is incorporation; nominals and pronominal required: either Pronominals pronouns; agreement markers pronominal or (or classifiers) hyponymous nominal only objects Hopi; Pre-Nahuatl; Nahuatl Tohono O’odham Proto-Uto-Aztecan Yaqui 1 Nivkh (aka Gilyak), a language isolate of Russia, may be another example of a ‘Reverse Chichewa’. According to Mattissen (2003), ‘one undergoer is obligatorily marked on the verb, either via dependent-head synthesis of the undergoer noun and the verb or via undergoer prefixation’ (p. 203). Subjects (‘actors’) are only marked under certain conditions. 2 This optional agreement is only available for third person. Generally speaking, our consultants disprefer the co-occurrence of object agreement with an overt NP in object position. Such pronominals are obligatory, however, when the object NP is right-dislocated. In this respect Yaqui seems to be like Chichewa, although it is unlike Chichewa in not having subject agreement. 3 Subject pronominals are prefixed to the verb root only in the ‘zero class’. There are two other large classes, the –in and –yax classes, wherein the subject pronominal appears between the verb root and the suffixal thematic elements that are indicated in the names of the classes: (i) yút-ne-n (ii)hét-pe-yax (Barragan 2003: 143 [5], [6]) raise-1.SG-IN crouch-3.SG-YAX ‘I raised’ ‘He crouched’ See J. Hill (2005) for details, and Barragan (2003) for a Distributed Morphology analysis of these constructions where he proposes that the thematic affixes are located in v, and the shifting of the subject is claimed to be the result of Local Dislocation at PF. These cases also clearly have object-agreement external to subject-marking on the verb; see below. 4 Some modern varieties of Nahuatl are losing some of Classical Nahuatl’s polysynthetic characteristics and taking on a more analytic nature due to a long contact period with Spanish; see Hill and Hill (1986) and Flores Farfán (2004) for further discussion of these varieties. 5 All Nahuatl orthography in this section, which has been collected from various sources from different periods, has been standardized with respect to vowels: a macron indicates a long vowel. In this usage I follow the practice of Andrews (2003). 6 In Tübatulabal, ‘independent subject pronouns are formed from the clitic pronouns with the prefixation of –in; the first singular form is, however, irregular’ (Steele 1977: 567). 7 This claim contrasts with that of Steele (1977), who argues that pre-verbal clitics are secondary developments from second position clitics in PUA; see Haugen (2007) for a critique of this view. 8 An independent issue is the syntactic status of NI. Rosen (1989) gives an analysis where both classifier and compound NI are lexical processes (i.e. occur ‘in the Lexicon’, prior to syntax), whereas Baker (1995) argues that some languages have lexical NI and some have non- lexical (i.e. purely syntactic) NI. For the purposes of this paper I remain agnostic on this issue. 9 Hale and Keyser reject their 1993 analysis of denominal verbs in Hale and Keyser (2002), largely due to the ‘hyponymous object problem’: non-cognate nominals appearing in the complement position in such constructions as I danced a jig or He wears a chili pepper (‘chile as a feather’) on his head (see the Hopi example in (26)) seem to suggest that the incorporated nominal could not have originated from that position. However, see Haugen (2008, 2009) for a solution to this problem using the notion of Late Insertion of lexical material within the framework of Distributed Morphology.