The Typology of Morphological Ergativity in Neo-Aramaic EDIT DORON & GEOFFREY KHAN The Hebrew University of Jerusalem & University of Cambridge Abstract Morphological ergativity is attested in all Neo-Aramaic dialects of the North-Eastern Neo- Aramaic (NENA) group, which comprises over a hundred different dialects spoken by Jewish and Christian communities originating in the border areas of Turkey/ Iraq/ Iran. Historically, Aramaic is nominative-accusative, and ergativity developed through contact with ergative Iranian languages, especially Kurdish, which is spoken by the Muslim population of the region. Ergativity developed in the perfective aspect only, and is marked by verb-agreement rather than Case. We divide NENA dialects into three types according to their degree of ergativity, reflected by differences in the distribution of the ergative marking of intransitive verbs. In dialects exhibiting the highest degree of ergativity, which we call Split-S, the ergative marker is restricted to transitive and unergative verbs, and is not found with unaccusative verbs. In a second type of dialect, which we call Dynamic-Stative, the ergative marker is also optionally found with unaccusative verbs. Dialects exhibiting the lowest degree of ergativity, Extended- Ergative, mark all intransitive subjects as ergative. This is surprising from the perspective of theories of ergativity, since it contradicts Marantz's Generalization, and suggests that ergative Case is not inherent but structural, and, specifically, that it is assigned by v and not by T. We show that the parametric variation between the different dialects reduces to the distribution of v. v is obligatory in Extended-Erg dialects, and assigns ergative Case to its argument if it has one, or to the internal argument otherwise. In Dynamic-Stative dialects, the presence of v is optional. In Split-S dialects – v is obligatorily missing; this is nevertheless compatible with verbs having an external argument, since ergative languages allow the merge of the external argument as an adjunct. 1. Neo-Aramaic Neo-Aramaic is unique among the Semitic languages in that many of its dialects have ergative- absolutive morphology (Khan 2007a). What makes it particularly interesting is that these dialects developed historically from documented forms of earlier Aramaic, which are nominative accusative. Neo-Aramaic is divided into four different groups of dialects: (i) Western group spoken in Maʿlula and various other villages in the region of Damascus (ii) Ṭuroyo group, spoken in Ṭur ʿAbdin and in the village of Mlaḥso in south-eastern Turkey (iii) Mandaic, spoken in the cities of Ahwaz and Khorramshahr, Iran (iv) North-Eastern Neo-Aramaic (NENA). The Neo-Aramaic dialects that now exhibit ergativity belong to the Ṭuroyo and North-Eastern Neo-Aramaic groups. These dialects are currently spoken, or at least have been spoken until recently, in south-eastern Turkey, northern Iraq and north-western Iran. The majority belong to North-Eastern Neo-Aramaic (NENA), the most diverse of all Neo-Aramaic dialect groups, which contains over a hundred dialects spoken by Jewish and Christian communities in villages and towns lying East of the Tigris river (Khan 2007b). 1 Fig. 1. The North-Eastern Neo-Aramaic dialect area The NENA dialects vary in lexicon, morphology and syntax (e.g. some dialects are predominantly SOV and others are predominantly SVO), but are all ergative. Ergativity in NENA developed historically due to contact with ergative Iranian languages, especially Kurdish, which is spoken by the Muslim population of the region (Khan 2007c). The historical evolution of ergativity in NENA involved the disappearance of original Aramaic tensed finite verbal forms. New tense/ aspect forms developed on the basis of the original participles. The active participle became the stem of the NENA imperfective, while the passive participle evolved into the NENA perfective. Ergativity developed only in the perfective, where the subject, the original agent phrase of the passive participle, was expressed as an oblique with dative Case. As in many other languages (also in English), in the earlier form of Aramaic from which the modern NENA dialects developed, there was an overlap in the morphological marking of dative and accusative case. When the dative agent phrase was reanalyzed as the subject of the Neo-Aramaic perfective, it had exactly the same form as the accusative object of the imperfective. Eventually, this identity in Case turned into identity in agreement, through the cliticization of both subject and object pronouns to each of the perfective and imperfective stems. The characteristic features of NENA regarding ergativity, therefore, are that it is split-ergative (in the perfective aspect only), and that its ergativity is marked by verb-agreement rather than Case. 2. Split ergativity and Agreement Inversion 2.1. S-suffixes and L-suffixes We begin with a discussion of transitive verbs, postponing intransitive verbs to section 3. Transitive verbs in NENA, whether imperfective or perfective, co-occur with two types of 2 affixes, in a fixed order, conventionally labelled S-suffixes and L-suffixes: (S- stands for Simple, as it is the unmarked inflection, while the L-suffix is the inflected accusative/ dative preposition l-) (1) a. V – S-suffix – L-suffix IMPF b. V – S-suffix – L-suffix PERF For imperfective verbs, the S-suffix cross-references the subject (we shall call it NOM), and the L-suffix cross-references the object (we call it ACC), whereas for perfective verbs, this is reversed, the S-suffix cross-references the object (we call it ABS), and the L-suffix cross- references the subject (we call it ERG): (2) a. V – NOM-suffix – ACC-suffix IMPF b. V – ABS-suffix – ERG-suffix PERF In other words, imperfective verbs are NOM-ACC, whereas perfective verbs are ERG-ABS. NOM and ABS suffixes have the same exponents, S-suffixes , and so do ACC and ERG suffixes – with L-suffix exponents. As an example, we list the verbal suffixes of one dialect: (3) Jewish Sanandaj (Khan 2009) SSSS----ssssuuuuffffffffiiiixxxxeeeessss LLLL----ssssuuuuffffffffiiiixxxxeeeessss NOM/ABS ACC/ERG 3MS -∅ -le 3FS -a -la 3PL -i -lu 2MS -et -lox 2FS -at -lax 2PL -etun -lăxun 1MS -na -li 1FS -an -li 1PL -ex -lan The inversion of the cross-referencing of subject and object between the perfective and the imperfective is illustrated below. In (4) and (5), in the imperfective (a) examples, the S-suffix (-a and -i respectively) cross-references the subject and the L-suffix (-lu and -la respectively) cross-reference the object. This is reversed in the perfective (b) examples, where the S-suffix cross-references the object, and the L-suffix cross-references the subject.1 (4) Jewish Sanandaj a. Imperfective baxt-ăke barux-ăwal-i garš-á-lu woman-DEF friend-PL-my pull -NOM.3FS-ACC.3PL IMPF 'The woman pulls my friends.' b. Perfective barux-ăwal-i baxt-ăke gərš-á-lu friend-PL-my woman-DEF pull -ABS.3FS-ERG.3PL PERF 'My friends pulled the woman.' 1 Examples throughout the article are either from published grammatical descriptions of the various NENA dialects, or from unpublished data gathered by the authors from native speakers. 3 (5) Jewish Sanandaj a. Imperfective barux-ăwal-i baxt-ăke garš-í-la friend-PL-my woman-DEF pull -NOM.3PL-ACC.3FS IMPF 'My friends pull the woman.' b. Perfective baxt-ăke barux-ăwal-i gərš-í-la woman-DEF friend-PL-my pull -ABS.3PL-ERG.3FS PERF 'The woman pulled my friends.' 2.2. The nature of L- vs. S-suffixes It should be noted that L-suffixes (and inflected prepositions) follow the anteriority auxiliary -wa 'PAST', whereas S-suffixes precede the auxiliary: (6) Christian Urmi (Khan, in prep) a. jrəš-wa-lə qatox pull -PAST-ERG.3MS P .2MS PERF ACC b. jriš-ət-wa-lə pull -ABS.2MS-PAST-ERG.3MS PERF both: ‘He had pulled you’ This demonstrates the nature of L-suffixes as clitics, in contrast to the inflectional nature of S- suffixes, which inflect the verbal stem. The inflectional nature of the S-suffix vs. the clitic nature of the L-suffix is further demonstrated by the fact that at most a single S-suffix is found in the verb, whereas in some dialects, two L-suffixes can be combined with the stem, one of them ergative and the other accusative:2 (7) Jewish Urmi (Khan 2008b) xze-le-le see -ERG.3MS-ACC.3MS PERF ‘He saw him.’ 2.3. ERG-suffixes cross-reference the subject We start by arguing that the ergativity found in NENA is not syntactic ergativity. In syntactically ergative languages, the absolutive argument in transitive clauses is the syntactic subject (Dixon 1994: 12). We claim that NENA is morphologically ergative, i.e. the absolutive argument in transitive clauses is the syntactic object. This is contrary to Polotsky 1996, who treats perfective sentences as passive, not ergative, and accordingly views the argument cross- referenced by the ABS-suffix as subject. This view however is actually equivalent to syntactic ergativity. Here we demonstrate that the ergativity found in NENA is morphological, and that it is actually the argument cross-referenced by the ERG-suffix which has all the subject properties (see also Goldenberg 1992 who disagrees with Polotsky's view). 2 In this paper, we do not further discuss cases where ACC replaces ABS in perfective verbs. 4 2.3.1. Insensitivity to definiteness Subject agreement in the imperfective is found in the verb irrespective of definiteness/ referentiality, whereas object agreement is only present if the object is definite. In the perfective, we find that the argument cross-referenced by ABS-affixes must be definite, which is an indication that it is the object. This is the case in all NENA dialects. We present examples from the Jewish Sanandaj and Christian Barwar dialects. In (8), the verb agrees with the definite object, but not in (9), where the object is indefinite. The argument cross-referenced by the ERG-suffixes in (8)—(9) can be indefinite and non-referential, even downward entailing in (8). This indicates that the argument cross referenced by the ERG-suffix is the subject. (8) Jewish Sanandaj a. hič-kas baxtăke gərš-a-le NEG.person woman-DEF pull -ABS.3FS-ERG.3MS PERF 'Nobody pulled the woman.' b. baṣor naše baxtăke gərš-a-lu few people woman-DEF pull -ABS.3FS-ERG.3P PERF 'Few people pulled the woman.' (9) Christian Barwar (Khan 2008a) xa naša qṭil-le raba kalwe a man kill -ERG.3MS many dogs PERF ‘A man killed many dogs.’ 2.3.2. Raising The argument cross-referenced by the ERG-suffix is the one that undergoes raising (or copy-raising in the sense of Potsdam and Runner 2001), indicating that it is the subject: (10) Christian Barwar a. priq-la xil-la finish -ERG.3FS eat -ERG.3FS PERF PERF 'She finished eating.' (see Khan 2008a: 941) b. la mṣe-la xil-la NEG can -ERG.3FS eat -ERG.3FS PERF PERF 'She could not eat.' (see Khan 2008a: 940) priq- 'finish' and mṣe- 'can' are indeed raising verbs rather than control verbs. In dialects such as Jewish Sulemaniyya and Jewish Urmi, where the subject of an unaccusative verb is inflected with absolutive rather than ergative agreement, these verbs have absolutive agreement, as can be seen in (11a-b): (11) a. Jewish Sulemaniyya priq-a m-xala finish -ABS.3FS from-eat PERF INF 'She finished eating.' (see Khan 2004: 301) 5 b. Jewish Urmi la mǝss-a axl-a NEG can -ABS.3FS eat -NOM.3FS PERF IRREALIS 'She has not been able to eat.' (see Khan 2008b: 137) 2.3.3. Clausal coordination In a conjunction of two clauses where the second clause has subject agreement but no overt subject, the argument cross-referenced by the ERG-suffix of the first clause is treated as subject by the predicate of the second clause (12a). In syntactically ergative languages, on the other hand, it is the argument cross-referenced by the ABS-suffix which is treated as subject of the second clause (Dixon 1994). In the NENA dialects, an overt pronoun must be introduced in the second clause to allow the argument cross-referenced by the ABS-suffix in the first clause to be interpreted as the subject of the second clause (12b): (12) Christian Barwar a. ʾɛ-brata muxl-a-la ʾu zil-la the-girl feed -ABS.3FS-ERG.3FS and leave -ERG.3FS PERF PERF 'She fed the girl and left' b. ʾɛ-brata muxl-a-la ʾu ʾay zil-la the-girl feed -ABS.3FS-ERG.3FS and she leave -ERG.3FS PERF PERF 'She fed the girl and she (the girl) left' 2.3.4. Ā-extraction The argument cross-referenced by an ERG-suffix can be Ā-extracted. In (13a) such an argument is questioned, and in (13b) it is relativized: (13) Christian Barwar a. ʾεni qṭil-a-le ʾay-baxta who.MS kill -ABS.3FS-ERG.3MS that-woman PERF 'Who killed that woman?' b. gawrə-t qṭil-a-le ʾay-baxta man-CRel killPERF-ABS.3FS-ERG.3MS that-woman 'the man who killed that woman' In some syntactically ergative languages such as Dyirbal (Dixon 1979), the ergative-marked argument cannot be Ā-extracted, whereas the absolutive-marked argument can. This would indicate that the ergative-marked argument is below the absolutive-marked argument on the Keenan-Comrie noun-phrase accessibility hierarchy (Keenan and Comrie 1977). Since the subject is the top of the hierarchy, the ergative-marked argument is not the subject in Dyrbal. The Ā-extraction of the ergative-marked argument in NENA, therefore, is consistent with our claim that this argument is the subject. 6 2.3.5. Anaphor binding There is a cross-linguistic tendency for reflexive anaphors to be bound only by the subject. In NENA, the argument cross-referenced by an ERG-suffix can bind an absolutive reflexive anaphor:3 (Notice that reflexives in NENA govern feminine agreement.) (14) Christian Barwar qṭil-a-le gyane kill -ABS.3FS-ERG.3MS himself PERF 'He killed himself.' 3. Variation in ergativity The NENA dialects exhibit varying degrees of ergativity. This is reflected by differences in the distribution of the ergative marker on intransitive verbs. The dialects exhibiting a higher degree of ergativity (Split-S and Dynamic-Stative) mark some intransitive subjects as absolutive, whereas dialects exhibiting a lower degree of ergativity (Extended-Erg) mark all intransitive subjects as ergative. The variation in the marking of intransitive subject agreement is linked to variations in the extent to which the dialects exhibit other properties that are known to correlate with ergativity. Of particular significance is the extent to which the various NENA dialects are consistent with Mahajan's generalization (Trask 1979, Mahajan 1994, 1997) that the verb in ergative languages is clause-peripheral. Dialects that we identify as exhibiting a higher degree of ergativity are mostly SOV, while the dialects that we identify as exhibiting a lower degree of ergativity are mostly SVO. Nevertheless, we will show that all NENA dialects exhibit some degree of ergativity. 3.1. Split-S dialects In dialects with the highest degree of ergativity, the ergative marker is restricted to transitive and unergative verbs (15a-b), and is not found with unaccusative verbs (15c). We call such dialects "split-S dialects" (following Dixon's 1994 terminology): (15) Jewish Sanandaj a. barux-ăwal-i brat-i gərš-a-lu friend-PL-my daughter-my pull -ABS.3FS-ERG.3PL PERF ‘My friends pulled my daughter.’ b. kalba nwəx-le dog bark -ERG.3MS PERF 'The dog barked.' c. brat-i qim-a daughter-my rise -ABS.3FS PERF ‘My daughter rose.’ 3.2. Extended-Erg dialects In dialects with a low degree of ergativity, the ergative suffix has been extended to unaccusative verbs as well. We call such dialects "Extended-Ergative dialects". This is the 3 It is true that in some syntactically ergative languages as well, it is the ergative element which binds anaphors, cf. Falk 2006, Aldridge 2008. We include the argument from binding for the sake of completeness. 7 most surprising type of dialect from the perspective of theories of ergativity, since they contradict Marantz's generalization (Marantz 1991) from which it follows that unaccusative subjects are never ergative. In Extended-Erg dialects, there is ergative agreement marking for the subject of transitive verbs (16a), for the subject of unergative verbs (16b), and also for the subject of unaccusative verbs (16c): (16) Christian Barwar a. xawr-ăwaθ-i brat-i griš-a-la friend-PL-my daughter-my pull -ABS.3FS-ERG.3PL PERF ‘My friends pulled my daughter.’ b. kalba nwix-le dog bark - ERG.3MS PERF 'The dog barked.' c. brat-i qim-la daughter-my rise -ERG.3FS PERF ‘My daughter rose.’ Despite the fact that intransitive subjects are marked like transitive subjects, we argue that these dialects are ergative-absolutive, and should not be analysed as nominative-accusative, for the following reasons. 3.2.1. Morphological Markedness Subject agreement in the perfective is marked and is different from subject agreement of the imperfective, which is not. The term "Extended- Ergative" for such languages is due to Dixon 1979:77. Dixon proposes a criterion for determining whether one type of agreement is less marked than another, namely if it has more ∅ affixes (null affixes). As was shown in the list of S- and L-affixes in (3) above, ∅ is only found in NENA among S-affixes, not L-affixes, which makes subject agreement in the perfective marked. 3.2.2. Syntactic Markedness Subject agreement in the perfective can be dropped when the subject is not overt, whereas imperfective subject agreement can never be dropped. In the perfective examples in (17), only absolutive agreement is expressed, while ergative agreement is missing. This construction should not be analysed as passive, despite the English translation, since the verb-stem is the same as that of an active verb. Yet it lacks the L-suffix. We take this construction to be active, with an unspecified subject associated with the context but not fully recoverable from it, which we call 'impersonal'. For example, (17a) should literally be translated as They killed the man, with an 'impersonal' reading of they, which is phonologically null in Barwar: (for examples from Jewish Zakho see Gutman 2008) (17) Christian Barwar a. gawṛa qṭil-∅ man kill -ABS.3MS PERF ‘The man was killed.’ b. baxta qṭil-a woman kill -ABS.3FS PERF ‘The woman was killed.’ 8 c. naše qṭil-i people kill -ABS.3PL PERF ‘The people were killed.’ The basic perfective stem qṭil (missing also the S-suffix) can be used with indefinite 3FS or 3PL arguments, which is a property of object agreement rather than subject agreement. This further demonstrates that the clauses in (17) have an object and not a subject: (18) Christian Barwar prim ʾərwe slaughter sheep.PL PERF ‘Sheep were slaughtered.’ (Khan 2008a: 750) Null impersonal ergative subjects are also documented in other ergative languages (Comrie 1988). The drop of the agreement marking for such subjects is due to the clitic nature of L- suffixes (cf. Preminger 2009). Interestingly, null impersonal subjects occur in Barwar only with 3rd person objects. The forms in (17) – (18) above are only available with 3rd person objects, 1st-2nd person objects do not surface despite the fact that the subject is null. A contrastive pair is shown in (19). This is a significant fact related to the PCC, to which we presently turn. (19) Christian Barwar a. brat-i griš-a daughter-my pull -ABS.3FS PERF ‘My daughter was pulled.’ b. * ʾana griš-ən I pull -ABS.1MS PERF ‘I was pulled.’ 3.2.3. The Person-Case Constraint (PCC) The Person/Case Constraint (PCC) was formulated by Bonet 1991 as a universal constraint: "In a combination of a weak direct object and an indirect object [clitic, agreement marker or weak pronoun], the direct object has to be third person." (Bonet 1991: 182). The PCC was later reformulated by Anagnostopoulou 2003, 2005 and Béjar & Rezac 2003, 2009 as a consequence of relativized minimality: A Person/Case relation between a head and a dependent is barred by an intervener, and accordingly, the dependent cannot realize 1st/2nd person features, but must realize 3rd person, which is not a person feature but rather default morphology. In all SOV NENA dialects, and in most SVO dialects as well, the ergative subject counts as a PCC intervener. We call these dialects PCC-abiding, and the dialects where the subject is not a PCC intervener – PCC-obviating. In the PCC-abiding dialects, ABS agreement across an ergative subject is restricted to 3rd person, whether or not that subject is overt: (For further details, including discussion of repair strategies in PCC-abiding dialects, see Doron and Khan, in preparation) (20) Christian Barwar a. transitive 3ms. gríš-∅-le ‘He pulled him’ 3fs. griš-á-le ‘He pulled her’ 9 3pl. griš-í-le ‘He pulled them’ etc. 2ms. * griš-ət-le 1ms. * griš-ən-le 2fs. * griš-at-le 1fs. * griš-an-le 2pl. * griš-itu-le 1pl. * griš-əx-le b. transitive with null impersonal subject 3ms. gríš-∅ ‘He was pulled’ 3fs. gríš-a ‘She was pulled’ 3pl. gríš-i ‘They were pulled’ etc. 2ms. * griš-ət 1ms. * griš-ən 2fs. * griš-at 1fs. * griš-an 2pl. * griš-itu 1pl. * griš-əx Under accepted theories of Case, the intervening element could not be a nominative subject, i.e. a dependent of the T node, the highest position in the clause, since it must intervene between the object and the head that the object depends on, i.e. its position must be structurally lower than that head. This argues strongly for the non-nominative nature of the subject, even in Extended-Erg dialects. The construction with null impersonal subjects found in Extended-Erg dialects should be distinguished from the anticausative construction of Split-S dialects. The latter dialects allow anticausative verbs productively, practically with every transitive verb. Anticausative verbs enter the derivation with no external argument, unlike transitive verbs with null impersonal subjects, which enter the derivation with an external argument, albeit null. The stems of anticausative verbs can be distinguished by their template, which differs from that of the template of their transitive and unergative counterparts (transitive/unergative CCəC vs. unaccusative CCiC). Though it is the case that all Split-S dialects are PCC abiding, e.g. (21a), 1st and 2nd person objects surface in the anticausative construction (21b), since anticausative verbs have no external argument, and hence no PCC intervener. This distinguishes the anticausative construction from the 'null impersonal subject' construction in (20b) above, where 1st and 2nd person objects do not surface: (21) Jewish Sanandaj a. transitive (stem grəš) 3ms. grəš-∅-le ‘He pulled him’ 3fs. gərš-á-le ‘He pulled her’ 3pl. gərš-í-le ‘He pulled them’ etc. 2ms. * gərš-et-le 1ms. * grəš-na-le 2fs. * gərš-at-le 1fs. * gərš-an-le 2pl. * gərš-etun-le 1pl. * gərš-ex-le b. anticausative (stem griš) 3ms. gríš-∅ ‘He got pulled’ 3fs. gríš-a ‘She got pulled’ 3pl. gríš-i ‘They got pulled,’ etc. 2ms. gríš-et 1ms. gríš-na 2fs. gríš-at 1fs. gríš-an 2pl. gríš-etun 1pl. gríš-ex 10