Transactions of the Philological Society Volume 107:2 (2009) 196–230 THE LOSS OF GRAMMATICAL GENDER IN CAPPADOCIAN GREEK1 By PETROS KARATSAREAS University of Cambridge Winner of the Fifth R. H. Robins Prize of the Philological Society ABSTRACT Cappadocian Greek is an extreme case of language change and dialectal variation among the Modern Greek dialects in having lost the tripartite grammatical gender distinction into masculine, feminine and neuter nominals, a distinction operative in Greek since its earliest recorded stages. In this paper, I argue that this linguistic innovation should not be viewed exclusively as the result of language contact with Turkish, as is most commonly assumed in the literature, but rather as the result of a series of language-internal analogical levellings of gender mismatches in polydefinite constructions, a process most probably accelerated by language contact but certainly not triggered by it. 1. A GENDERLESS GREEK VARIETY In Cappadocian Greek (henceforth Cappadocian), the tripartite grammatical gender distinction into masculine, feminine and neuter nominals has been lost: all nouns behave as neuters, in that they combine with the originally neuter forms of the various determiners and modifiers, which agree with them. In the variety of the dialect which was spoken in the village of Ulagha´tsh (UC), for example, the definite article (do for the singular; da for the plural) and the 1 I would like to thank Bert Vaux, David Willis and James Clackson as well as the anonymous reviewers for their valuable comments on earlier forms of this paper. Special thanks also go to my friend and fellow linguist Thanasis Giannaris for reading a first draft of this paper and for bringing most helpful references to my attention. Last but not least, I thank the editor of TPhS, Paul Rowlett. This research was supported by a scholarship from the Greek State Scholarships Foundation (Idqtla Jqasijèm !posqouièm – IJ! ) and by a bursary from the George and Marie Vergottis Fund of the Cambridge European Trust.  The author 2009. Journal Compilation  The Philological Society 2009. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ and 350 Main Street, Malden, MA 02148, USA.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 197 modifying adjective ‘good’ (kalo for the singular; kala for the plural) remain invariable in combination with head nouns whose Standard Modern Greek (SMG) cognates have three different gender values (1). In SMG, on the other hand, the definite article and the modifying adjective agree with their head nouns in gender and, therefore, appear in different forms when they combine with each of these nouns (2): (1) UC do kalon do andra ‘the good man’ do kalon do neka ‘the good woman’ do kalon do pei ‘the good child’ da kalan da andres ‘the good men da kalan da nekes ‘the good women’ da kalan da peija ‘the good children’ (Kesisoglou 1951: 29) (2) SMG oM kalosM anðrasM ‘the good man’ iF kaliF jinekaF ‘the good woman’ toN kaloN peðiN ‘the good child’ iM kaliM anðresM ‘the good men’ iF kalesF jinekesF ‘the good women’ taN kalaN peðjaN ‘the good children’2 The UC definite article forms do ⁄ da are cognates of the SMG neuter definite article forms to ⁄ ta. The appearance of the definite article in front of both the noun and the adjective in (1) (‘definiteness spreading’ or ‘polydefiniteness’) is obligatory in the Eastern Greek dialects.3 The final -n in UC kalon and kalan is 2 The following abbreviations are used throughout the text: 1: first person, 3: third person, AUG: augment, COP: copula, F: feminine, GEN: genitive, INT: interrogative, M: masculine, N: neuter, NOM: nominative, NONFIN: non-finite, PL: plural, PST: past, SG: singular. 3 Polydefinite constructions are also found in SMG: o kalos o anðras lit. ‘the good the man’, i kali i jineka lit. ‘the good the woman’, to kalo to peði lit. ‘the good the child’. Campos and Stavrou (2004) argue that these constructions have different syntactic, semantic and phonological properties from monadic constructions (i.e. constructions where the definite article appears only before the adjective, as in (2)) in SMG. For an account of the diachrony of polydefinite constructions in Greek, see Manolessou (2000). 198 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 conditioned by a morphophonological rule which inserts an [n] at the end of adjectives when followed by the definite article. The forms of these adjectives in other syntactic environments are kalo for the singular and kala for the plural, which are the SMG forms of the respective adjective. This paper brings forth an investigation of the factors that were operative in the language change process whereby grammatical gender distinctions were lost from Cappadocian varieties such as UC, exemplified in (1).4 In light of the lack of systematic accounts of the phenomenon and of various instances in the literature, where this loss is attributed to language contact with Turkish, I address the question whether the factors which were operative in this process were language-internal, that is, pertaining to the structure and the dynamics of the Cappadocian linguistic system itself, or language- external, that is, relating to language contact with Turkish. Following the examination of the agreement patterns in two Cappadocian varieties (UC and Axo´ Cappadocian), as well as in Pontic Greek, one of the closest cognate dialects of Cappadocian, I argue that the loss of grammatical gender distinctions in the most innovative Cappadocian varieties (always with respect to this feature) should not be considered exclusively as the outcome of language contact with Turkish. I show that this loss followed the emergence of an inflectionally active [±HUMAN] feature in the Cappadocian nominal inflection, and should rather be viewed as the outcome of a series of analogical levellings of gender mismatches in polydefinite constructions, a process most probably accelerated by language contact. Unlike previous considerations of the phenom- enon in the literature, I propose a chronology of the change in stages. The discussion of the various aspects of the change and of its interactions with other ongoing changes and structural features of the dialect further serves as a starting point for theoretical considerations of a wider scope, such as the diachrony of classification systems like gender and the [±HUMAN] feature or 4 Grammatical gender distinctions were lost from the majority of the Cappadocian varieties (Dawkins 1916: 87). Some traces of grammatical gender, however, are found in the varieties which were spoken in the villages of Sinaso´s, Zale´la, Pota´mia and Delmeso´. These are found in the use of the masculine and feminine forms of the accusative of the definite article ton and tin when the article immediately precedes the noun, i.e. without the use of an attributive adjective between the two, as in Pota´mia Cappadocian: (i) ipen ton deirmendZ i ‘she said to the.M.ACC. miller’ (Dawkins 1916: 456); (ii) na skotosun tin gata ‘that they kill the.F.ACC cat’ (Dawkins 1916: 464).  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 199 the interplay of language-internal and language-external factors in language change. The paper is structured as follows. Section 2 is a very brief introduction to the sociohistorical and linguistic background of Cappadocian, focusing on its linguistic isolation from other varieties of Modern Greek (MG) and the linguistic innovations that resulted. Section 3 serves as an illustration of the change undergone by Cappadocian by sketching the role of grammatical gender in SMG and Cappadocian. Section 4 presents other recorded cases of reduction and ⁄ or loss of grammatical gender distinctions from the history of other languages, followed by extensive discussion of the Cappadocian case. The concluding section 5 discusses which factors could be shown to have played a role in the process of gender loss in Cappadocian but were not addressed in the present paper, thus pointing towards some issues for future research. In accounting for the loss of grammatical gender distinctions in Cappadocian, I focus mainly on data from UC, but reference will often be made to the varieties of other Cappadocian villages, when appropriate. 2. CAPPADOCIAN GREEK ON THE LANGUAGE MAP 2.1. Historical and linguistic background Cappadocian is a MG dialect which belongs to the Eastern Greek dialectal branch along with Pontic Greek and the Greek of Mariupol (Anastasiadis 1995; Arapopoulou 2001; Dawkins 1910; 1916; 1937; 1940; Kontossopoulos 1994). Our knowledge of Cappadocian is almost exclusively based on R. M. Dawkins’ documentation and description of the dialect, published in 1916, and on a number of later descriptions of the varieties of specific Cappadocian villages by prominent Greek linguists (inter alia Andriotis 1948; Costakis 1964; 1968; Kesisoglou 1951; Mav- rochalyvidis & Kesisoglou 1960). The term ‘Cappadocian’ essentially describes a group of mutually related MG varieties which were spoken in a number of villages in Central Anatolia (contemporary Turkey) until 1924, when Greece and Turkey exchanged populations on the basis of religious identity as a criterion in accordance with the Treaty of Lausanne. Following 200 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 the relocation of Cappadocian-speaking populations to many different parts of continental Greece, and in light of the heavy standardisation pressures from SMG within the country, Cappa- docian was thought to have died out until very recently, when Mark Janse and Dimitris Papazachariou announced that they had discovered native speakers of Cappadocian in parts of Central and Northern Greece (Janse & Papazachariou 2007). Among these native speakers were middle-aged, third-generation Cappadocians who seem to have preserved Cappadocian thanks to positive attitudes towards the dialect. According to Janse and Papazachar- iou, the documentation of these people’s Cappadocian will result in a new grammar, dictionary and collection of texts, none of which (to the best of my knowledge), has been published to date. Long before the population exchange, the Greek-speaking people of the area had come into contact with Turkish-speaking people, dating back to the invasion of the Seljuk Turks in parts of Cappadocia even before the defeat of the Byzantine troops at Manzikert in 1071 (Janse 2002). The subsequent separation of the Cappadocian people from the rest of the Greek-speaking contin- gent, the consecutive dehellenisation of much of Asia Minor, and the further disintegration and fall of the late Byzantine Empire in 1453 resulted in Cappadocian developing in isolation for many centuries (Dawkins 1916: 1). 2.2. The outcomes of linguistic isolation The fact that Cappadocian developed in isolation compared to the vast majority of the MG dialects, in combination with the intense and long-standing contact with surrounding Turkish and concom- itant extensive Cappadocian–Turkish bilingualism in the Cappado- cian villages, brought about a number of innovative linguistic changes that significantly differentiate Cappadocian from SMG as well as from other MG dialects, even from those with which it is a close cognate, like Pontic. Such changes are found in all compo- nents of the grammar, from phonology and morphology to syntax, pragmatics and discourse. Language contact with Turkish is most commonly taken a priori as the principal cause of many of these changes (Thomason & Kaufman 1988: 215–22; Winford 2003: 83–4; 2005: 402–9). It is true that the synchronic presence of a certain number of linguistic  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 201 constructions and phenomena in the Cappadocian linguistic system can indeed be attributed to language contact. The Cappadocian pluperfect is a relevant example (see also Winford 2005: 405–6). In SMG, the pluperfect is formed periphrastically, and consists of the past tense of the auxiliary verb exo ‘to have’ with the so-called ‘non- finite’ form of the verb in question. In the pluperfect of SMG, it is the auxiliary verb which inflects for person, whereas the non-finite form remains unchanged (3a). In UC, the pluperfect is formed periphrastically and consists of the past tense of the verb in question and the 3SG form of the imperfect of the copular verb ime ‘to be’. In this construction, it is the past tense of the verb which inflects for person, whereas the copular verb form remains unchanged (3b). This innovative morphological pattern seems to have been modelled on the Turkish pluperfect in (3c), which originates in a periphrasis comprising the so-called ‘di-past’ and the 3SG form of the past of the copular marker. As in the Cappadocian case, in the Turkish pluperfect, it is the past tense of the verb which inflects for person, whereas the copular marker remains the same. (3) a. SMG Verb Pluperfect lino ixa lisei untie I.had untie-NONFIN ‘I had untied’ b. UC Verb Past Pluperfect lino e-lis-a e-lis-a iton untie AUG-untie-1SG.PST AUG-untie-1SG.PST COP.PST.3SG ‘I untied’ ‘I had untied’ (Kesisoglou 1951: 39) c. Turkish Verb Past Pluperfect c¸o¨z- c¸o¨z-du¨-m c¸o¨z-du¨-m idi (> c¸o¨zdu¨mdu¨) untie untie-PST-1SG untie-PST-1SG COP.PST.3SG ‘I untied’ ‘I had untied’ On the other hand, the trigger for an equally high number of language changes in Cappadocian was language-internal. The presence in the Cappadocian inventory of phonemes like the palato-alveolar fricatives ⁄ S ⁄ and ⁄ Z ⁄ (UC a'Sim ‘silver’ – cf. SMG 202 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 a'simi; Z i'mar ‘dough’, cf. SMG zi'mari), or the post-alveolar fricatives ⁄ tS ⁄ and ⁄ dZ ⁄ (UC tSi'lo ‘to roll’, cf. SMG ci'lo; dZ u'fali ‘head’, cf. SMG ce'fali), which also form part of the Turkish phonemic inventory, does not necessarily have to be attributed to language contact, as these phonemes are also found in a number of other MG dialects, where they most probably emerged language- internally after regular processes of phonological change (e.g. Cypriot Greek Serin ‘hand’, cf. SMG c¸eri; maxa'Z :a ‘shops’, cf. SMG maVa'ZJa; tSe'rin ‘candle’, cf. SMG ce'ri; ka'ndZ el:in ‘balus- ter’, cf. SMG ka¢elo).5 Finally, in numerous other cases the causal factors of change include both language-internal and language-external ones, as in the case of the Cappadocian head-final constructions (SOV, possessor–possessee word order), whereby previously marked constituent orders became the default (unmarked) orders under the influence of Turkish. So, into which of the above categories of language change does the loss of grammatical gender in Cappadocian fall? The search for an answer to this question begins with a brief description of grammatical gender as a morphosyntactic feature in SMG and Cappadocian. 3. ILLUSTRATING THE LINGUISTIC INNOVATION: GRAMMATICAL GENDER IN STANDARD MODERN GREEK AND CAPPADOCIAN 3.1. Standard Modern Greek In SMG as well as in the vast majority of the MG dialects, nouns, adjectives, determiners, a number of pronouns and a few numerals are marked for one of the three genders: masculine, feminine or neuter. As shown by Ralli (2002), gender is an intrinsic property of SMG noun stems and derivational affixes. Nouns are assigned a gender mainly by means of morphological information relating inter alia to the feature of inflectional class or to processes of word formation or, for noun stems which are underspecified for gender, 5 One anonymous reviewer notes that language contact with Romance should not be excluded as a possible explanation for the development of such phonemes in MG dialects like Cretan and Cypriot. As far as Cypriot Greek is concerned, the interested reader is referred to Davy and Panayotou (2001), who support the idea that the palatalisation of dental and velar stops and fricatives in the dialect pre-dates the Frankish period.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 203 by means of agreement in syntax. SMG, therefore, has a formal grammatical gender system. In SMG the three genders are primarily manifested in agreement patterns. The noun phrase forms the main agreement domain in the language, with agreement also extending to predicate argument constructions (4): (4) SMG a. Afti i teseris tixi this.M.NOM.PL the.M.NOM.PL four.M.NOM wall.M.NOM.PL ine kokini. are red.M.NOM.PL ‘These four walls are red.’ b. Aftes i teseris fustes this.F.NOM.PL the.F.NOM.PL four.F.NOM skirt.F.NOM.PL ine kokines. are red.F.NOM.PL ‘These four skirts are red.’ c. Afta ta tesera vivlia this.N.NOM.PL the.N.NOM.PL four.N.NOM book.N.NOM.PL ine kokina. are red.N.NOM.PL ‘These four books are red.’ In the noun phrase afti i teseris tixi ‘these four walls’ in (4a), the masculine head noun tixi acts as the controller6 which determines the grammatical genders, as well as the case and number, of the demonstrative afti, the definite article i and the numeral teseris. Beyond the domain of the noun phrase, tixi determines the gender of the predicate kokini. Similarly, the nouns fustes and vivlia in (4b) and (4c) determine the gender of their targets in their respective noun phrases and clauses. 6 For an extensive discussion of gender in the languages of the world and of the relevant terminology (‘controller’, ‘target’, etc.) see Corbett (1991). 204 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 3.2. Cappadocian As was seen in the noun phrases in (1), in Cappadocian, the forms of the definite article and the modifying adjective remain invariable in combining with head nouns whose SMG cognates have three different gender values (UC do kalon do andra ‘the good man’, do kalon do neka ‘the good woman’, do kalon do pei ‘the good child’; cf. SMG o kalos anðras, i kali jineka, to kalo peði in (2)). In UC, it is the neuter form which is used over the others in all the nominals which are marked for gender in SMG (Table 1). Gender agreement has been lost in UC in all the syntactic environments where it appears in SMG, namely both within the noun phrase and beyond it, in predicate–argument constructions (5): (5) UC Sano-ne mi ito do xerifos? crazy-COP.3SG INT this the man ‘Is this man crazy?’ (Kesisoglou 1951: 156) In the noun phrase ito do xerifos in (5), the demonstrative ito and the definite article do relate to the head noun xerifos. Notice that both targets appear in the invariable forms shown in Table 1. The head noun xerifos is a Turkish loan word (cf. colloquial Turkish herif ‘guy’) bearing the Greek inflectional ending -os, which is the inflectional ending most saliently related to the masculine gender in Table 1. Grammatical gender in SMG and UC SMG UC Adjectives: ‘small’ mikrosM mikriF mikroN mikro ‘black’ mavrosM mavriF mavroN mavro Participles: ‘hungry’ pinasmenosM pinasmeniF pinasmenoN pinasmeno ‘lost, dead’ xamenosM xameniF xamenoN xanimeno Articles: Definite oM iF toN do Indefinite enasM miaF enaN ena Pronouns: Proximal demonstrative aftosM aftiF aftoN ato, ito Indefinite ‘other’ alosM aliF aloN talo (< to alo) Numerals: ‘one’ enasM miaF enaN ena ‘three’ treisM,F triaN tria  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 205 SMG. It is reasonable, therefore, to assume that at some earlier point in the history of UC, xerifos was marked with masculine gender. Concerning the predicate–argument agreement, the pred- icate sano ‘crazy’ appears in the originally neuter form of the adjective (cf. Sı´ lli Greek tsannosM-tsanniF-tsannoN; Costakis 1968), illustrating the loss of agreement in this domain as well.7 4. THE INTERPLAY OF LANGUAGE-INTERNAL AND LANGUAGE-EXTERNAL FACTORS IN THE LOSS OF GRAMMATICAL GENDER 4.1. Grammatical-gender loss in the history of languages Cases of grammatical-gender reduction and ⁄ or loss such as the one observed in Cappadocian are not unheard of in the history of languages. Ibrahim (1973: 86) considers the loss of the various distinctive gender markers of the nouns (usually due to phonolo- gical changes), and of the inflections which mark agreement between nouns and other word classes which agree with the noun in gender, as the two major conditions necessary for the loss of gender distinctions in languages with formal grammatical-gender systems like Greek (see also Aikhenvald 2004). A familiar example is English. Old English had a formal grammatical-gender system, reminiscent of that of Greek. Nouns and other nominals were marked with one of the three gender values, masculine, feminine or neuter (6): (6) Old English se  cwen þæt s_cip cyning seo this.M king.M this.F queen.F this.N ship.N ‘this king’ ‘this queen’ ‘this ship’ 7 As can be seen from the noun xerifos in (5) and as will also be elaborated below (section 4.2.1), nouns in Cappadocian retain inflectional endings that are most saliently related to specific gender values in SMG and other MG dialects. It was already mentioned in section 3.1 that gender is an intrinsic property of SMG noun stems but not of inflectional endings. This explains why the same inflectional ending can be found in words that have three different gender values: tix-os ‘wall.M’ versus ipir-os ‘continent.F’ versus ðas-os ‘forest.N’. The gender of nouns is manifested in agreement patterns between the noun and its various determiners and modifiers (articles, pronouns, adjectives) within the noun phrase and in predicate–argument constructions. It is, therefore, already clear that the loss of inherent gender in Cappadocian nouns is closely related to the loss of grammatical gender agreement in the nouns’ modifiers. 206 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 During the Middle English period, the language lost the grammat- ical-gender system, which was replaced by a pronominal gender system (Corbett 1991: 5), a distinction preserved only in the third person personal and reflexive pronouns: nouns denoting male human entities are referred to by he, nouns denoting female human entities are referred to by she and nouns denoting non-human entities are referred to by it, irrespective of their biological sex. The formal-agreement system of Old English has been lost in Present- Day English, as shown by the glosses in (6). Kastovsky (2000) shows that the loss of grammatical-gender distinctions in Middle English nouns was a complex process involving a restructuring of the morphological system of the language triggered by the decay of nominal inflectional endings, itself caused by the phonetic attrition of word-final syllables. This was followed by the levelling of the inflectional endings of adjectives and other modifiers, which were no longer able to support a formal- agreement system (Curzan 2003: 44). In order to account for the fact that gender was lost only in English despite the fact that nearly all Germanic languages underwent similar phonological changes at some point in their linguistic history, historical linguists have pointed to contact with Old Norse (Curzan 2003: 48–54; Ibrahim 1973: 89–90).8 Bearing in mind, though, the language-internal processes already under way in the language, Curzan seems to be on the right track in assuming that ‘in the case of Middle English, the ‘‘creole-like’’ features of inflectional reduction and loss of grammatical gender seem to have been incipient in the language and accelerated by language contact’ (2003: 53).9 A similar change scenario will be shown for the case of gender loss in Cappadocian. 8 What has been suggested for the Middle English–Old Norse contact situation is a process of analogical levelling of different sets of inflectional endings which combined with cognate nouns in the two languages (Curzan 2003: 52; Ibrahim 1973: 90). 9 Apart from Old Norse, language contact with Anglo-Norman might also be responsible for some of these ‘creole-like’ features of Middle English. The author would like to thank the anonymous reviewer who brought this to his attention.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 207 4.2. Grammatical-gender loss in Cappadocian The loss of grammatical-gender distinctions in Cappadocian is found in various discussions of the contact-induced changes observed in the dialect which seemingly imply that language contact with Turkish was the decisive factor in this process of language change. Janse holds that ‘the loss of gender distinctions is due to Turkish influence, since Turkish has no grammatical gender’ (2002: 366), a view often encountered elsewhere in the literature: Dawkins considers the loss of grammatical gender which is almost complete in Cappadocia […] to be due to Turkish influence; Turkish has no gender. (Thomason & Kaufman 1988: 219–20) Again under Turkish influence, there was a progressive loss of gender distinctions, especially in South Cappadocian. (Winford 2005: 405) In most cases when gender was lost in Indo-European, its loss can be attributed to some substratum, or adstratum language […]. In other cases the influence of genderless languages are [sic] easier to prove: Turkish in the case of Asia Minor Greek [Cappadocian]. (Matasovic´ 2004: 77) The loss of gender as a nominal category has occurred […] dialectally, in Modern Greek [Cappadocian] due to contact with Turkish. (Igartua 2006: 56) The view echoed in the above quotations is probably based on the fact that Dawkins includes gender loss in a list of Cappadocian grammatical phenomena which he attributes to contact with Turkish (1916: 203). It is clear, however, from other parts of his description of the variety that he did not see contact with Turkish as the causal factor that set the language-change process of gender loss in motion, but rather as an accelerating factor that acted upon a change already initiated language- internally long before the Turkish invasion of Cappadocia (1916: 116). Given the attention that contact explanations for the changes observed in Cappadocian have received, they are treated first in section 4.2.1. 208 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 4.2.1. Language-external factors Contact-related explanations for the loss of grammatical gender in Cappadocian refer to the absence of grammatical-gender distinc- tions in Turkish (7): (7) Turkish a. Bu yas¸lı adam hasta. this old man ill ‘This old man is ill.’ b. Bu yas¸lı kadın hasta. this old woman ill ‘This old woman is ill.’ c. Bu yas¸lı ag˘ac¸ hasta. this old tree ill ‘This old tree is ill.’ (meaning ‘diseased’) In the noun phrases in (7), the proximal demonstrative bu and the modifying adjective yas¸ lı ‘old’ remain invariable in combination with head nouns denoting entities of different (or no) sex (adam ‘man’, kadın ‘woman’, ag˘ac¸ ‘tree’). The same holds for the adjectival predicate hasta ‘ill’.10 The Cappadocian data exhibit the same invariability, which is correctly used as evidence illustrating the loss of grammatical-gender distinctions in the variety. A language-contact scenario based on these data would assume extensive Cappadocian–Turkish bilingualism in the speech com- munity, a prerequisite for contact-induced changes to occur in any language-contact situation. Indeed, as already mentioned, bilin- gualism in the area is well established. As for UC, it is considered to be one of the varieties in which the Turkish influence was most pervasive owing to the large and increasing Turkish population in the village and the subsequent extensive bilingualism (Dawkins 1916: 18; Kooij & Revithiadou 2001). Bilingual speakers, and especially bilingual children, are seen as the agents of change by some current theories of contact-induced language change. As such, they are able to draw upon the resources, structures and elements of their two linguistic systems and to use them relatively freely. According to these theories, bilingual 10 Turkish does not have overt marking of the 3SG on nominal predicates of the ‘x is y’ type (Go¨ksel & Kerslake 2005).  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 209 speakers resort to this sort of language ‘mixing’ in an attempt to reduce the processing overload caused by the availability of two linguistic systems in their minds, which can differ in various grammatical aspects. To this end, they eliminate the linguistic elements or features which cause them cognitive inconvenience, in the sense of making it hard for them to differentiate between the two linguistic systems (Field 2002; Matras 1998; 2000; Matras & Sakel 2007).11 In the Cappadocian case, the cues that Cappadocian–Turkish bilingual children would have in order to establish the absence of grammatical-gender distinctions in Turkish would necessarily come from the invariability of elements modifying head nouns as in (7). The contact mechanism at hand, then, would be the non-acquisition of the [determiner ⁄ modifier + head noun] agreement rule of Greek in bilingual first-language acquisition. In other words, Cappado- cian–Turkish bilingual children would fail to acquire the Greek rule that conditions the agreement of determiners and modifiers of any sort with their head nouns in terms of gender, on account of the absence of such a rule from the Turkish grammatical system (see also Brendemoen 1999: 537). This contact-induced change process would be supported by the need of Cappadocian–Turkish bilingual children to reduce the processing overload caused by the differences in the underlying representations of nouns in the two languages. If the assumptions of the above-mentioned theories hold true, bilingual children presumably eliminated the gender feature from the underlying representations of the Cappadocian nouns to make them more 11 The setting in which such ‘mixing’ takes place is a bi- or multilingual linguistic community, where more than one language is spoken and is in everyday use. Most commonly, the different languages of a multilingual linguistic community will be spoken, at least at an incipient stage, by the members of different ethnic and demographic groups which come into cultural, economic, political and scientific contact within that linguistic community (Oksaar 1996). Thomason and Kaufman (1988) elaborated greatly on the notion of sociolinguistic dominance between the different linguistic groups, to which they assigned a central role in the process of contact-induced language change. In their framework, different types of change will occur depending on whether speakers of the sociolinguistically dominant or of the sociolinguistically dominated language will be the agents of change. The issue of social pressures and dominance within a multilingual community is of central importance to the study of contact-induced language change phenomena. The focus on the more structural and psychological approach to the Cappadocian case, however, as well as limitations of space, do not allow for further elaboration on this aspect of contact-induced language change in the present paper. 210 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 ‘uniform’ with the representations of Turkish nouns, which lack the gender feature. A view favouring a contact explanation for the loss of gender in Cappadocian could additionally be corroborated by the fact that the variety did not undergo any major phonological changes which could cause the sort of confusion and morphological restructuring observed in the cases of gender loss mentioned in section 4.1 (Middle English, other Germanic languages, Romance languages). The only phonological change worth mentioning that did have an impact on the variety’s structure was the loss of word-final unstressed high vowels according to the phonological rule in (8): (8) High Vowel Deletion Rule V  fi Ø ⁄ __# high stress This rule affected mainly the very large group of originally neuter nouns which ended in an unstressed ⁄ i ⁄ (9a), but also had an impact on a number of originally feminine nouns with the same ending (9b): (9) UC a. 'spiti ‘house’ > spit 'xteni ‘comb’ > xten b. 'nifi ‘bride’ > nif 'strosi ‘mattress’ > stroS (Kesisoglou 1951: 15) This phonological process caused confusion as to which gender words ending in a consonant belonged to, as shown by originally feminine nouns combining with neuter inflectional endings (10): (10) Malakopı´ Cappadocian Singular Plural Nominative Genitive Nominative ‘house’ spit spitju spitja ‘bride’ nif nifju nifja (cf. SMG nifis) (cf. SMG nifes) (Dawkins 1916: 115)  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 211 However, this ending was left unaffected in word-final stressed position, and so were the rest of the inflectional endings which are most saliently related to specific gender values in SMG and which are all preserved in Cappadocian. These are the endings -os for masculine nouns, -a and -i for feminine nouns, and -o for neuter nouns (Chila-Markopoulou 2003: 145). (11) SMG UC a. 'ipnosM ‘sleep’ 'jipnos 'ðjavolosM ‘devil’ 'javolos b. kar'ðjaF ‘heart’ kar'ja a'vli F ‘yard’ ne'vli c. 'ksiloN ‘wood’ 'ksilo xar'tiN ‘paper’ xar'ti Therefore, the conditions that Ibrahim considers to be decisive in language-internal cases of gender loss are not met by Cappado- cian. Given the absence of any major phonological changes in Cappadocian and the presence in its nominals of the most salient inflectional endings related to specific gender values in SMG and in other MG dialects, the selection of the neuter gender by Cappadocian–Turkish bilinguals (over the masculine or the feminine) as the new agreement controller could be accounted for on the basis of data coming from the bilingual SMG–Turkish speech of the Muslim community of the island of Rhodes. Georgalidou, Spyropoulos, Kaili & Revithiadou (2005) report on the confusion and avoidance of gender marking in SMG by SMG–Turkish bilingual speakers and on their use of the neuter, which they consider to be the default gender value in SMG (12):12 12 That neuter is the default gender value and is, therefore, unmarked in Greek can also be supported by the fact that (a) it is the value used in syntactic gender resolution when the conjoined nouns denote [–HUMAN] entities, irrespective of their grammatical gender; (b) it is the gender to which morphologically non-integrated loan words are generally assigned, unless they denote [+HUMAN] entities; (c) it is the gender used for nominalisations (Holton, Mackridge & Philippaki-Warburton 1997: 501, 250, 456–7). 212 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 (12) Rhodian Muslim Greek a. mecalo hia big.N aunt.F ‘the elder aunt’ b. irte skilos ... pinasmeno ita came.3SG dog.M hungry.N was ‘The dog came ... it was hungry.’ The data in (12) are reminiscent of the contact mechanisms proposed for Cappadocian, where gender agreement has been lost both within the noun phrase (12a) and beyond it, in predicate– argument constructions (12b). However, a contact explanation such as the one presented here, which resorts to postulating the failure of acquiring the [deter- miner ⁄ modifier + head noun] agreement rule and the subsequent loss of gender values from the feature bundles of Cappadocian nouns, does not account for the fact that the variety may have lost gender agreement between determiners ⁄ modifiers and head nouns but has retained number agreement, as shown in (1) (do kalon do andra ‘the good man’ vs. da kalan da andres ‘the good men’, etc.). Number agreement in noun phrases does not occur in Turkish. Modifying adjectives remain invariable when combining with their head nouns irrespective of number (13): (13) Turkish iyi adam ‘good man’ iyi adamlar ‘good men’ Agreement in number is also active in Cappadocian in copular constructions, as shown in the data from Arava´n Cappadocian in (14) (Dawkins 1916: 148): (14) Arava´n Cappadocian anastenar-me ‘I am ill’ anastenar-se ‘you (SG) are ill’ anastenar-ne ‘he ⁄ she ⁄ it is ill’ anastenarja-meste ‘we are ill’ anastenarja-ste ‘you (PL) are ill’ anastenarja-nde ‘they are ill’  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 213 In Turkish copular constructions of this type the predicate remains invariable across the paradigm (15): (15) Turkish hasta-yım ‘I am ill’ hasta-sın ‘you (SG) are ill’ hasta(-dır) ‘he ⁄ she ⁄ it is ill’ hasta-yız ‘we are ill’ hasta-sınız ‘you (PL) are ill’ hasta-(dır)lar ‘they are ill’ These data call for a readjustment of the contact hypothesis whereby Cappadocian–Turkish bilingual children would fail to acquire only part of the Greek morphosyntactic agreement rule, namely agreement in gender, but not in number. This, however, is not an economical explanation. As will be shown in the next subsection, a language-internal process of change leading to the decline of grammatical-gender distinctions in Cappadocian was already ongoing at the onset of the Cappadocian–Turkish contact, a process most probably accelerated by the subsequent language contact but certainly not triggered by it. 4.2.2. Language-internal factors Dawkins comments on the combination of the originally neuter forms of adjectives with head nouns of originally different gender by quoting Sarantidis (1899), who documented the following example from a proverb from Sinaso´s Cappadocian (16): (16) Sinaso´s Cappadocian to kalo o locos the.N good.N the.M speech.M ‘the fair speech’ (Sarantidis 1899: 150) In (16), the head noun locos is of masculine gender and so is its determiner o. The modifying adjective kalo, though, and its determiner to are both of neuter gender. According to Sarantidis, in Sinaso´s Cappadocian, nouns denoting [–HUMAN] entities combine with modifiers which take the neuter form whatever the grammat- ical gender of the head noun may be. In contrast, the modifiers of 214 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 nouns denoting [+HUMAN] entities agree normally with their head nouns in grammatical gender.13 Dawkins uses the Greek terms œlwtva [empsixa] and ¥wtva [apsixa] to denote [+HUMAN] and [–HUMAN] nouns respectively, and reports a similar situation in Pontic, which also forms part of the Eastern Greek dialectal branch. He correctly identifies a connection between the Pontic situation and Sarantidis’ Sinaso´s Cappadocian data: [I]t [the œlwtva versus ¥wtva distinction] is the stage which everywhere in Cappadocian preceded the present entirely genderless state of the adjectives. This entire loss of gender can hardly but be due to the influence of genderless Turkish. But the disuse of the m. and f. adjectival endings before œlwtva, but not before ¥wtva, in Pontos and, to judge from this evidence from Sinaso´s, in the least Turkised of the Cappadocian dialects, shews that the germ of this loss is involved in the distinction between œlwtva and ¥wtva, a distinction which is certainly not of Turkish origin. It would seem that the Turkish influence found already existing a loss of grammatical gender or at least a tendency to lose grammatical gender, and carried this further to its own conditions of total absence of any distinctions of gender. (Dawkins 1916: 116) It is clear, therefore, that Dawkins did not consider contact with Turkish as the initiating trigger for the loss of gender distinctions in the Cappadocian varieties. The connection between Pontic and Cappadocian regarding the decline of grammatical-gender distinctions is also hinted at by the Greek linguists who described the varieties of specific Cappadocian villages and other closely related Greek varieties in the 1950s and 1960s (Mavrochalyvidis & Kesisoglou 1951: 81 for Axo´ Cappado- cian; Andriotis 1948: 46 for Pha´rasa Greek). Horrocks considers the breakdown of the grammatical-gender distinction as an innovation of the Eastern Greek dialects (1997: 313–14), without taking a clear position as to whether this innovation was triggered language-internally or language-externally. Despite all these occur- rences in the literature, however, various historical linguists still talk 13 Henceforth nouns denoting [+HUMAN] entities will be referred to as ‘[+HUMAN] nouns’ and nouns denoting [–HUMAN] entities will be referred to as ‘[–HUMAN] nouns’.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 215 of gender loss in Cappadocian as being exclusively due to language contact with Turkish (see beginning of this section), whereas even those who recognise the connections between the various Eastern Greek varieties do not provide a systematic account of the change or some sort of chronology of it. The aim of the next subsection is to fill this gap in the history of Cappadocian. 4.2.1.1. œlwtva and ¥wtva in Pontic and Cappadocian Pontic is a MG dialect which was spoken widely in the Pontos area on what is today the Turkish coast of the Black Sea over about the same period as Cappadocian was spoken in Central Anatolia. Pontic is still spoken in areas of mainland Greece by the descendants of Pontic refugees, while some Pontic varieties are still spoken in areas of Turkey (Mackridge 1987; 1999). Pontic belongs to the Eastern Greek dialectal branch and is thus genetically related to Cappadocian. Like the other Eastern Greek varieties, Pontic developed in relative isolation from the rest of the Greek-speaking contingent and in contact with Turkish. Compared to Cappado- cian, however, language contact between Pontic and Turkish was not as intense or as long-standing, and was intensified only after the fall of the Empire of Trebizond in 1461 and the subsequent incorporation of Pontos into the Ottoman Empire. This, as well as other differences of sociohistorical nature between the Cappado- cian–Turkish and Pontic–Turkish cultural contacts, can explain the differences in the extent of linguistic ‘borrowing’ in the two MG dialects (in the sense of Thomason and Kaufman 1988). In Pontic, the tripartite grammatical gender distinction into masculine, feminine and neuter of SMG and of the vast majority of the MG dialects is preserved. Nouns, adjectives, determiners, a number of pronouns and some numerals are marked for gender. A selection of nominals is summarised in Table 2 (data from Papadopoulos 1955).14 Apart from gender, a further distinction based on the [±HUMAN] feature is operative in Pontic. The two features interact in agreement both within the noun phrase and in the clause domain. In this variety, [–HUMAN] nouns neutralise gender distinctions in the 14 The Pontic data show extensive intradialectal variation in some forms. Alternative forms of various of the nominals included in Table 2 were not included for simplicity reasons. 216 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 Table 2. Grammatical gender in SMG and Pontic SMG Pontic Nouns: ‘man’ anðrasM andrasM ‘woman’ jinekaF jinekaF ‘child’ peðiN peðinN Articles: Definite oM iF toN oM iF toN Adjectives: ‘good’ kalosM kaliF kaloN kalosM kalesaF kalonN ‘heavy’ varisM variaF variN varisM varesaF variN Pronouns: Proximal demonstrative aftosM aftiF aftoN autosM auteF autoN Indefinite ‘other’ alosM aliF aloN alosM aleF aloN Numerals: ‘one’ enasM miaF enaN enasM,F enanN ‘three’ trisM,F triaN trisM,F triaN plural, and behave as neuters in that they combine with the neuter form of the definite article in both the nominative and the accusative, thus exhibiting neuter-like syncretism (17): (17) Pontic a. Masculine b. Feminine c. Neuter [+HUMAN] [–HUMAN] [+HUMAN] [–HUMAN] ‘man’ ‘month’ ‘woman’ ‘chicken’ ‘village’ NOM.SG. o andras o minas i jineka i kosara to xorion ACC.SG. ton andran ton minan tin jinekan tin kosaran to xorion NOM.PL. i andres ta minas15 i jinekes ta kosaras15 ta xoria ACC.PL. tus andras ta minas ti jinekes ta kosaras ta xoria This has repercussions for the agreement between head nouns and their modifiers, as adjectives modifying [–HUMAN] nouns appear in their neuter form irrespective of the grammatical gender of their head nouns. This extends to both numbers, leading to a sort of 15 The differences in the inflectional endings between andr-es and min-as, on one hand, and jinek-es and kosar-as, on the other, are conditioned by a morphological rule whereby the nominative and accusative plural of [–HUMAN] nouns take the inflectional ending of the accusative plural of the respective inflectional paradigm (Spyropoulos & Kakarikos 2007). The syncretism of nominative and accusative in favour of the latter in the plural of [–HUMAN] nouns is considered by Horrocks as a sign of assimilation to neuter declensional patterns (1997: 315).  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 217 mismatch between the determiner of the modifying adjective, on one hand, and the determiner of the head noun, on the other, given that definiteness spreading is obligatory in Pontic (Drettas 1997; Tompaidis 1980) (18): (18) Pontic [–HUMAN] to kalon o minas ‘the good month’ to kalon i kosara ‘the good chicken’ ta kala ta minas ‘the good months’ ta kala ta kosaras ‘the good chickens’ In the noun phrases in (18), the adjectives appear in the neuter form because their head nouns denote [–HUMAN] entities. In the singular of these noun phrases, the definite article that appears before the adjective agrees with it in grammatical gender and is, therefore, in the neuter form to, and so does the definite article that appears before the head noun and is, therefore, in the masculine and feminine form o and i respectively. The agreement patterns in [+HUMAN] nouns present a slightly more complex picture. In the latest documented stage of Pontic, as is described by Papadopoulos (1955), Oikonomidis (1958) and Drettas (1997), adjectives modifying [+HUMAN] masculine and feminine nouns appear in their masculine form in the plural, as in (19): (19) Pontic i kali i andres ‘the good men’ i kali i jinekes ‘the good women’ This, however, seems to be a later development of Pontic, as there are no traces of such an agreement pattern or any similar phenomena to be found either in Cappadocian or any other MG dialect of the greater area of Asia Minor and Anatolia (Pha´rasa, Sı´ lli, Livisi, Demirdesi). It is, therefore, reasonable to assume that at an earlier stage in the history of Pontic, adjectives modifying [+HUMAN] nouns agreed ‘normally’ with their head nouns in grammatical gender in both numbers, a stage illustrated in (20): 218 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 (20) Earlier Pontic [+HUMAN] o kalon o andras ‘the good man’16 i kalesa i jineka ‘the good woman’ to kalon to peðin ‘the good child’ i kali i andres ‘the good men’ i kaleses i jinekes ‘the good women’ ta kala ta peðia ‘the good children’ Neuter agreement with [–HUMAN] nouns is also found in the predicate–argument domain (21): (21) Pontic a. I para en asimenon. the.F money.F is silver.N ‘Money is silver.’ b. I sevta-s en pola tranon. the.F love.F-your is very big.N ‘Your love is very big.’ (Anastasiadis 1995: 86) That the œlwtva versus ¥wtva distinction was active in Cappado- cian as it is in Pontic can be evidenced by the situation in Cappadocian varieties that have preserved it, even to a limited extent, along with an equally limited distinction based on gram- matical gender (for a discussion of animacy in Cappadocian, see Janse 2004). One such variety is Axo´ Cappadocian (AC) (Mavrochalyvidis & Kesisoglou 1960). In AC, [+HUMAN] masculine nouns do not take any form of the definite article either in the singular or in the plural. [–HUMAN] masculine nouns combine with the neuter form of the definite article in both numbers. Feminine nouns combine with a contracted form of the definite article t in the singular and with the neuter form of the definite article, irrespective of their [±HUMAN] specification. In AC nominal inflection, the following distinctions are therefore operative: 16 Masculine nouns and adjectives ending in -os change the nominative singular ending to -on when preceded by the definite article (Drettas 1997: 120; Oikonomidis 1958: 183).  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 219 (22) AC a. Singular [+HUMAN] masculine Ø arxopos ‘the man’ feminine t neka ‘the woman’ [-HUMAN] masculine ⁄ to jipnos ‘the sleep’ neuter to zevcli ‘the yoke’ b. Plural [+HUMAN] masculine Ø arxop(i) ‘the men’ all other ta jipnosja ‘the sleeps’ ta nekes ‘the women’ ta karjes ‘the hearts’ ta zevclja ‘the yokes’ (Mavrochalyvidis & Kesisoglou 1960: 40–41) Grammatical gender distinctions are neutralised in adjectival constructions and in predicate–agreement structures. Adjectives in AC have lost gender distinctions, and appear in the originally neuter form when modifying head nouns irrespective of their grammatical gender or [±HUMAN] feature, as in UC (23): (23) AC to kalo arxopos ‘the good man’ to kalo neka ‘the good woman’ to kalon to pei ‘the good child’ ta kala arxop ‘the good men’ ta kalan ta nekes ‘the good women’ ta kalan ta pedja ‘the good children’ (Mavrochalyvidis & Kesisoglou 1960: 42–3) The final -n in kalon and kalan is conditioned by the same morphophonological rule already encountered in UC. What calls for special attention in the AC case is the environments where definiteness spreading appears. As seen above, in UC and Pontic, definiteness spreading is obligatory in all environments. In AC, though, definiteness spreading is possible only when the forms of the two definite articles – the one agreeing with the head noun and the one agreeing with the modifying adjective – are identical, i.e. of neuter gender in form (Mavrochalyvidis & Kesisoglou 1960: 31) (24): 220 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 (24) AC a. to kalon to pei ‘the good child’ ta kalan ta nekes ‘the good women’ ta kalan ta pedja ‘the good children’ b. *to kalon t neka ‘the good woman’ *to kalon ton arxopo ‘the good man.ACC’ *ta kalan t arxopjus ‘the good men.ACC’ (Mavrochalyvidis & Kesisoglou 1960: 31–2) In light of the AC data, a language-internal hypothesis can be formulated according to which the mismatch between the forms of the definite articles of the modifying adjective and the head noun in constructions such as the ones found in Pontic, but also even in some Cappadocian varieties (cf. Sinaso´s Cappadocian (16)), was disallowed in the most innovative Cappadocian varieties such as AC and UC.17 This triggered a series of analogical levellings based on the forms of the definite articles and the modifying adjectives in adjectival constructions which in turn eventually led to the total loss of grammatical gender in the Cappadocian varieties like UC. Such a hypothesis would take the Pontic data as illustrating an earlier stage and the UC data as representing a later stage in the 17 As one anonymous reviewer correctly points out, the data from Sinaso´s Cappadocian in (16) seem to challenge the claim that the mismatch between the forms of the definite articles of the modifying adjective and the head noun in polydefinite constructions was disallowed in the most innovative Cappadocian varieties, as the specific variety supposedly preserves the mismatch when most other varieties appear to have lost it. Dawkins comments on Sarantidis’ description of Sinaso´s Cappadocian as being ‘professedly of a past state of things’ (1916: 27; see also quote from Dawkins above, section 4.2.2) and considers it similar to Pota´mia Cappadocian, which does not exhibit the sort of mismatch described above. However, even if one does not wish to discard this example on the basis of Dawkins’ remarks, thus considering it as truly depicting the synchronic state of the variety at the time of its documentation in 1899, it could well be the case that Sinaso´s Cappadocian was one of the least innovative Cappadocian varieties with respect to the loss of grammatical gender and never underwent this change characteristic of other Cappadocian varieties. The Sinaso´s data, then, could be thought of as illustrating an earlier stage in the whole process of language change.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 221 decline of grammatical-gender distinctions in the Eastern Greek dialects and could be illustrated in the following stages of analogical levelling:18 Stage 1 Nouns are marked for grammatical gender and the [±HUMAN] feature. Determiners of [+HUMAN] nouns, their modifiers and determiners of modifiers agree with their heads in grammatical gender in both numbers. Determiners of [–HUMAN] nouns agree with their heads in grammatical gender in the singular; in the plural, they take neuter agreement. Modifiers of [–HUMAN] nouns and their determiners take neuter agreement in both numbers. Definiteness spreading is obligatory with all nouns. a. [+HUMAN] (cf. Pontic (18)–(20)) o kalos o andras ‘the good man’ i kalesa i jineka ‘the good woman’ i kali i andres ‘the good men’ i kaleses i jinekes ‘the good women’ b. [–HUMAN] to kalon o minas ‘the good month’ to kalon i kosara ‘the good chicken’ ta kala ta minas ‘the good months’ ta kala ta kosaras ‘the good chickens’ Change 1 The mismatch between the forms of the definite article appearing before the adjective and that appearing before the head noun in the singular of [–HUMAN] nouns is levelled. 18 The series of changes postulated by the language-internal hypothesis formulated here are illustrated on the basis the Pontic data in (18) and (20). The nouns andras ‘man’ and jineka ‘woman’ are taken as examples of [+HUMAN] nouns and the nouns minas ‘month’ and kosara ‘chicken’ are taken as examples of [–HUMAN] nouns of masculine and feminine gender respectively. Dialectal variation (e.g. in UC the definite article appears as do ⁄ da) and phonological differences (e.g. the final -n insertion rule) are not taken into consideration. 222 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 Stage 2 Nouns are marked for grammatical gender and the [±HUMAN] feature. Determiners of [+HUMAN] nouns, their modifiers and determiners of modifiers agree with their heads in grammatical gender in both numbers. Determiners of [–HUMAN] nouns, their modifiers and determiners of modifiers take neuter agreement in both numbers. Definiteness spreading is obligatory with all nouns. Mismatch between the forms of the definite article in definiteness spreading is disallowed. a. [+HUMAN] o kalos o andras ‘the good man’ i kalesa i jineka ‘the good woman’ i kali i andres ‘the good men’ i kaleses i jinekes ‘the good women’ b. [–HUMAN] to kalon to minas ‘the good month’ (cf. AC (20)–(22)) to kalon to kosara ‘the good chicken’ ta kala ta minas ‘the good months’ ta kala ta kosaras ‘the good chickens’ Change 2 The contrast between grammatical gender and the [±HUMAN] feature is levelled in the modifiers. Neuter agreement in the modifiers is introduced for [+HUMAN] nouns. Stage 3 Determiners of [+HUMAN] nouns agree with their heads in grammatical gender. Determiners of [–HUMAN] nouns, and modifiers and determiners of modifiers of all nouns take neuter agreement. Defi- niteness spreading is obligatory with all nouns. Mismatch between the forms of the definite article in definiteness spreading is disallowed and the pre-nominal article then disappears in such cases. a. [+HUMAN] to kalon andras ‘the good man’ (cf. AC (20)–(22)) to kalon jineka the good woman’ ta kala andres ‘the good men’ ta kala jinekes ‘the good women’  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 223 b. [–HUMAN] to kalon to minas ‘the good month’ to kalon to kosara ‘the good chicken’ ta kala ta minas ‘the good months’ ta kala ta kosaras ‘the good chickens’ Change 3 Neuter agreement in the modifiers and their determiners with all nouns leads to the loss of grammatical gender. All nouns behave as neuters. Stage 4 Nouns have no grammatical gender. Determiners of all nouns, their modifiers and determiners of modifiers all take neuter agreement in both numbers. Definiteness spreading is obligatory with all nouns. No mismatch surfaces between the forms of the definite article in definiteness spreading. a. [+HUMAN] to kalon to andras ‘the good man’ (cf. UC (1)) to kalon to jineka ‘the good woman’ ta kala ta andres ‘the good men’ ta kala ta jinekes ‘the good women’ b. [–HUMAN] to kalon to minas ‘the good month’ to kalon to kosara ‘the good chicken’ ta kala ta minas ‘the good months’ ta kala ta kosaras ‘the good chickens’ 4.2.3. Summary The origins of grammatical-gender loss in Cappadocian can be traced back to the emergence of a [±HUMAN] feature which became active in the nominal inflection of Eastern Greek dialects. This feature was realised, among others, in agreement between head nouns and modifiers within the noun phrase domain and beyond it, in predicate–argument constructions, in that the modifiers and other agreeing nominals referring to [–HUMAN] nouns appeared in their neuter form. This, in combination with definiteness spreading, 224 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 which was obligatory in these dialects, resulted in a mismatch between the form of the definite article appearing before the adjective and those appearing before the head noun in polydefinite constructions, as the definite article appeared in the neuter form before the adjective but in the masculine or feminine form before the head noun. While this mismatch was allowed in Pontic, this was not the case in Cappadocian, where the mismatch was levelled at the expense of grammatical-gender distinctions. These processes of analogical levelling were probably aided and accelerated by Cappadocian–Turkish bilingualism and subsequent cross-linguistic influence from Turkish. The linguistic aspect that turned out to be the key in accounting for this change is agreement, especially within the noun-phrase domain. The series of analogical levellings progressed on the basis of [determiner ⁄ modifier + head noun] agreement, whereas the cues for the absence of gender distinctions in Turkish in the case of the cross-linguistic influence in bilinguals undoubtedly came from the invariability of determiners and modifiers, i.e. the absence of agreement within the noun phrase in the language. 5. IS THIS THE END OF THE STORY? The aim of this paper has been to identify the most salient linguistic factors in the process of grammatical-gender loss in Cappadocian varieties such as UC and to elaborate on one of them, namely the role of adjective–noun agreement in polydefinite constructions. In line with most current approaches to cases of language change, it has been argued that a combination of language-internal and language-external factors contributes to a more complete picture of the language change case in question. In an attempt to assess the validity of the language-internal and language-external factors that have been discussed here, it becomes clear that each of them allows us to account for different aspects of the Cappadocian data. Language-internal factors provide an enlightening overview of the processes and changes which were under way in the Eastern Greek dialects and which most probably pre-date the Turkish invasions in Asia Minor and subsequent language contacts between some of these dialects and Turkish. The most important of these features is neuter agreement with nouns of either masculine or feminine gender which is found, for example, in  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 225 Pontic. This is an important indicator of the situation that most probably preceded the complete loss of grammatical-gender distinctions in some Cappadocian varieties like UC, and also accounts for the retention of number agreement in [deter- miner ⁄ modifier + head noun] constructions – something which contact-related explanations were unable to deal with. Language-external factors, in contrast, help explain why the process of grammatical gender loss was completed in some Cappadocian varieties but not in other Eastern Greek dialects like Pontic, despite the fact that the structural conditions for the change are found in them as well. In the case of Pontic, recall that language contact between Cappadocian and Turkish was far more intense and long-standing than language contact between Pontic and Turkish. This resulted in extensive bilingualism in some Cappado- cian villages. This contact-related dimension of the change is further corroborated by the use of the neuter as the default gender value in SMG–Turkish bilingual speech. Overall, the examination of the language-internal and language- external factors that have been proposed for the explanation of grammatical-gender loss in Cappadocian varieties such as UC corroborates Dawkins’ claim that ‘the Turkish influence found already existing a loss of grammatical gender or at least a tendency to lose grammatical gender, and carried this further to its own conditions of total absence of any distinctions of gender’ (1916: 116). Of course, in order to elucidate the role of the [±HUMAN] feature and its realisations in Cappadocian, further research is required which will draw from data from more Cappadocian as well as other Eastern Greek varieties and will examine the issue from a wider perspective. One relevant aspect is the interaction of this feature with nominal inflection. Spyropoulos and Kakarikos (2007) show that the feature conditions the inflectional pattern of some nouns and the distribution of certain inflectional endings in Delmeso´, Axo´ and Ulagha´tsh Cappadocian. In an example from Delmeso´ Cappadocian, nouns of Greek origin ending in -as take different inflectional endings in the plural depending on their [±HUMAN] specification (25): 226 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 (25) Delmeso´ Cappadocian [+HUMAN] [–HUMAN] SINGULAR papa-s ‘priest’ kerata-s ‘snail’ PLURAL papað-es keratað-ja (Dawkins 1916: 109–10) According to their analysis, the -ja inflectional ending is specified as [–HUMAN], whereas -es is the default plural formative in this variety, as it appears with both [+HUMAN] and [–HUMAN] nouns in other inflectional classes. Such an interaction could also possibly contribute to grammatical gender loss in Cappadocian. In another example from Spyropoulos and Kakarikos, in UC, nouns ending in -os follow two different inflectional patterns based on their [±HUMAN] specification (26): (26) UC [+HUMAN] [–HUMAN] ‘man’ ‘wolf’ SINGULAR Nominative ⁄ Accusative xerif-os likos-Ø Genitive xerif-ju likos-ju PLURAL Nominative ⁄ Accusative xerif-ja likos-ja (Dawkins 1916: 102) According to Spyropoulos and Kakarikos’s analysis, [+HUMAN] nouns follow a synthetic (i.e. fusional) inflectional pattern, whereas [–HUMAN] nouns follow an agglutinative inflectional pattern. Irre- spective of the exact nature of the inflectional patterns in (26), in UC, the -ja inflectional ending also appears with [+HUMAN] nouns, unlike in Delmeso´ Cappadocian, where it only appears with [–HUMAN] nouns. It therefore needs to be examined whether the combination of an inflectional ending which is specified as [–HUMAN] in one Cappadocian variety with [+HUMAN] nouns in another Cappadocian variety could have been a factor operative in grammatical gender loss.19 19 That there can be a relation between inflectional patterning and gender is pointed out by Corbett (1991) in general, and by Ralli (2002) in particular for SMG.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 227 It would also be of great interest to investigate the conditions under which the [±HUMAN] feature emerged in Eastern Greek dialects. Horrocks suggests that this process was perhaps initiated by the local transfer in antiquity of unusually large numbers of masculine and feminine inanimates of the third declension to the neuter paradigm in -…m [-in], and subsequently accelerated by expanded use of neuter possessive adjectives, first with other inanimates regardless of gender, then more generally, a development perhaps prompted by the gender-invariant form of the corresponding genitive pronom- inal possessives: e.g. temo´n ⁄ teme´teron i nı´fe ‘the-my ⁄ our the daughter-in-law’. (1997: 314) The creation of subgenders based on features such as animacy or the [±HUMAN] feature is also reported for more or less all Slavonic languages. In Bosnian-Croatian-Serbian, for example, masculine nouns are further divided into animates and inanimates, a distinc- tion realised in terms of inflectional endings and agreement with various determiners (ovog studenta ‘this.M student.M’ versus ovaj zakon ‘this.M law.M’; Corbett 1991: 162). Finally, an issue not touched upon in the present paper concerns the invariability of modifiers for case in adjectival constructions of the polydefinite type. In Cappadocian, modifying adjectives do not agree with their head nouns in case, but remain invariable with respect to case across the inflectional paradigm. As also shown above, they only agree with their head nouns in terms of number (27): (27) AC ‘the good man’ SINGULAR Nominative to kalo arxopos Genitive t kalo arxop(u) Accusative to kalo arxopo PLURAL Nominative ta kala arxop(i) Genitive t kala arxoposju Accusative ta kala arxopjus (Mavrochalyvidis & Kesisoglou 1960: 43) 228 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 The invariability for case is reminiscent of Turkish, where modi- fying adjectives remain invariable for case across the inflectional paradigm as well. Whether the invariability in Cappadocian can be attributed to cross-linguistic influence from Turkish is another aspect that needs to be investigated. Department of Linguistics Faculty of Modern and Medieval Languages University of Cambridge Sidgwick Avenue Cambridge CB3 9DA United Kingdom Email: pk299@cam.ac.uk References AIKHENVALD, ALEXANDRA, 2004. ‘Gender and noun class’, in Geert E. Booij, Christian Lehmann, Joachim Mugdan & Stavros Skopeteas (eds.), Morfolo- gie ⁄ Morphology: Ein Handbuch zur Flexion und Wortbildung ⁄ A Handbook on Inflection and Word Formation, Berlin: Mouton de Gruyter, 1031–1045. ANASTASIADIS, VASILIS K., 1995. ‘Merikes omoiotites tis Pontiakis me Neoellinika Mikrasiatika glossika idiomata’ (Some similarities between the dialect of Pontos and the modern Greek Asian Minor lingual idioms), Archeion Pontou 46, 73–125. ANDRIOTIS, NIKOLAOS P., 1948. To Glossiko Idioma ton Farason, (The dialect of Pha´rasa), Athens: Ikaros. ARAPOPOULOU, MARIA, 2001. ‘Dialektikoi thylakoi tis ellinikis’ (Dialectal enclaves of Greek), in Anastasios Foivos Christidis (ed.), Egkyklopaidikos Odigos gia ti Glossa (Encyclopedic guide for language), Thessaloniki: Kentro Ellinikis Glossas, 175–179. BRENDEMOEN, BERNT, 1999. ‘Greek and Turkish language encounters in Anatolia’, in Bernt Brendemoen, Elizabeth Lanza & Else Ryen (eds.), Language Encounters Across Time and Space, Oslo: Novus Press, 353–378. CAMPOS, HE´CTOR & STAVROU, MELITA, 2004. ‘Polydefinite constructions in Modern Greek and in Aromanian’, in Olga Misˇ eska Tomic´ (ed.), Balkan Syntax and Semantics, Amsterdam: Benjamins, 137–179. CHILA-MARKOPOULOU, DESPOINA, 2003. ‘Genos kai symfonia sti nea elliniki’ (Gender and agreement in Modern Greek), in Anna Anastasiadi- Symeonidi, Aggeliki Ralli & Despoina Chila-Markopoulou (eds.), To Genos (Gender), Athens: Patakis, 132– 167. CORBETT, GREVILLE, 1991. Gender, Cambridge: Cambridge University Press. COSTAKIS, ATHANASIOS P., 1964. Le Parler Grec d’Anakou, Athens: Institut Franc¸ais d’Athe`nes. COSTAKIS, ATHANASIOS P., 1968. To Glossiko Idioma tis Sillis (The dialect of Sı´ lli), Athens: Institut Franc¸ais d’Athe`nes. CURZAN, ANNE, 2003. Gender Shifts in the History of English, Cambridge: Cambridge University Press. DAVY, JIM & PANAYOTOU, ANNA, 2001. ‘Strident palatals in Cypriot Greek’, Proceedings of the 4th International Conference on Greek Linguistics: Nicosia, September 17–19, 1999. Thessaloniki: University Studio Press, 338–345.  KARATSAREAS – GENDER LOSS IN CAPPADOCIAN GREEK 229 DAWKINS, RICHARD M., 1910. ‘Modern Greek in Asia Minor’, Journal of Hellenic Studies 30, 109–132. DAWKINS, RICHARD M., 1916. Modern Greek in Asia Minor: A Study of the Dialects of Sı´lli, Cappadocia and Pha´rasa with Grammar, Texts, Translations and Glossary, Cambridge: Cambridge University Press. DAWKINS, RICHARD M., 1937. ’The Pontic dialect of Modern Greek in Asia Minor and Russia’, Transactions of the Philological Society 36, 15–52. DAWKINS, RICHARD M., 1940. ’The dialects of Modern Greek’, Transactions of the Philological Society 39, 1–38. DRETTAS, GEORGE, 1997. Aspects Pontiques, Paris: Association des recherches pluridisciplinaires. FIELD, FREDRIC W., 2002. Linguistic Borrowing in Bilingual Contexts, Amsterdam: Benjamins. GEORGALIDOU, MARIANTHI, SPYROPOULOS, VASILEIOS, KAILI, HASAN & REVITHIADOU, ANTHI, 2005. ‘Linguistic and sociolinguistic aspects of a Rhodian Greek variety’, paper presented at the 6th International Linguistics Conference of ODEG, ‘The dialectal varieties of Greek from the ancient era till today’, Coriliano Otranto, 6–8 October. GO¨KSEL, ASLI & KERSLAKE, CELIA, 2005. Turkish: A Comprehensive Grammar, London: Routledge. HOLTON, DAVID, MACKRIDGE, PETER & PHILIPPAKI-WARBURTON, IRENE, 1997. Greek: A Comprehensive Grammar of the Modern Language, London: Routledge. HORROCKS, GEOFFREY, 1997. Greek: A History of the Language and its Speakers, London: Longman. IBRAHIM, MUHAMMAD HASAN, 1973. Grammatical Gender: Its Origin and Develop- ment, The Hague: Mouton. IGARTUA, IVA¨N, 2006. ’Genus alternans in Indo-European’, Indogermanische Forsch- ungen 111, 56–70. JANSE, MARK, 2002. ‘Aspects of bilingualism in the history of the Greek language’, in J. N. Adams, Mark Janse & Simon Swain (eds.), Bilingualism in Ancient Society: Language Contact and the Written Text, Oxford: Oxford University Press, 332– 390. JANSE, MARK, 2004. ‘Animacy, definiteness and case in Cappadocian and other Asia Minor Greek dialects’, Journal of Greek Linguistics 5, 3–26. JANSE, MARK & PAPAZACHARIOU, DIMITRIS, 2007. ‘Cappadocian: the resurrection of a ‘‘dead’’ language’, paper read at the 8th International Conference in Greek Linguistics, Ioannina, 31 August. KASTOVSKY, DIETER, 2000. ‘Inflectional classes, morphological restructuring, and the dissolution of Old English grammatical gender’, in Barbara Unterbeck & Matti Rissanen (eds.), Gender in Grammar and Cognition, I: Approaches to Gender; II: Manifestations of Gender (Trends in Linguistics: Studies and Monographs 124), Berlin: Mouton de Gruyter, 709–727. KESISOGLOU, I. I., 1951. To Glossiko Idioma tou Oulagats (The dialect of Ulagha´tsh), Athens: Institut Franc¸ais d’Athe`nes. KONTOSSOPOULOS, NIKOLAOS, 1994. Dialektoi kai Idiomata tis Neas Ellinikis (Dialects and lingual idioms of Modern Greek), Athens: Grigoris. KOOIJ, JAN G. & REVITHIADOU, ANTHI, 2001. ‘Greek dialects in Asia Minor: accentuation in Pontic and Cappadocian’, Journal of Greek Linguistics 2, 75–117. MACKRIDGE, PETER, 1987. ‘Greek-speaking Moslems of north-east Turkey: prole- gomena to a study of the Ophitic sub-dialect of Pontic’, Byzantine and Modern Greek Studies 11, 115–137. MACKRIDGE, PETER, 1999. ‘I ellinofonia stin periochi tou Ofi (Pontos)’ (The Greek spoken in the region of Ophis (Pontus)), in Anastasios Foivos Christidis (ed.), 230 TRANSACTIONS OF THE PHILOLOGICAL SOCIETY 107, 2009 Egkyklopaidikos Odigos gia ti Glossa (Encyclopedic guide for language), Thessa- loniki: Kentro Ellinikis Glossas, 101–105. MANOLESSOU, I., 2000. Greek noun phrase structure: a study in syntactic evolution, Ph.D. dissertation, University of Cambridge. MATASOVIC´, RANKO, 2004. Gender in Indo-European, Heidelberg: Winter. MATRAS, YARON, 1998. ‘Utterance modifiers and universals of grammatical borrow- ing’, Linguistics 36, 281–331. MATRAS, YARON, 2000. How predictable is contact-induced change in grammar?’, in Colin Renfrew, April McMahon & Robert Lawrence Trask (eds.), Time Depth in Historical Linguistics, vol. 1, Cambridge: McDonald Institute for Archaeological Research, 563–583. MATRAS, YARON & SAKEL, JEANETTE, 2007. ‘Investigating the mechanisms of pattern replication in language convergence’, Studies in Language 31, 829–865. MAVROCHALYVIDIS, G. & KESISOGLOU, I. I., 1960. To Glossiko Idioma tis Axou (The dialect of Axo´), Athens: Institut Franc¸ais d’Athe`nes. NEWTON, BRIAN, 1972. The Generative Interpretation of Dialect: A Study of Modern Greek Phonology, Cambridge: Cambridge University Press. OIKONOMIDIS, DIMOSTHENIS, 1958. Grammatiki tis Ellinikis Dialektou tou Pontou (Grammar of the Greek dialect of Pontos). Leksikografikon Deltion. Parartima 1. Athens: Academy of Athens. OKSAAR, ELS, 1996. ‘The history of contact linguistics as a discipline’, in Hans Goebl, Peter H. Nelde, Zdeneˇk Stary´ & Wolfgang Wo¨lck (eds.), Contact Linguistics: An International Handbook of Contemporary Research, Berlin: de Gruyter, 1–12. PAPADOPOULOS, ANTHIMOS A., 1955. Istoriki Grammatiki tis Pontikis Dialektou (Historical grammar of the Pontic dialect), Archeion Pontou, Supplement 1 (Athens). RALLI, ANGELA, 2002. ‘The role of morphology in gender determination: evidence from Modern Greek’, Linguistics 40, 519–551. SARANTIDIS, ARCHELAOS I., 1899. I Sinasos (Sinaso´s), Athens: Typografeion Ioannou Nikolaidou. SPYROPOULOS, VASIEIOS & KAKARIKOS, KONSTANTINOS, 2007. ‘Aspects of dialectal variation in the nominal inflection of Greek: a feature-based approach’, paper presented at the 6th Mediterranean Morphology Meeting, Ithaki, September. THOMASON, SARAH GREY & KAUFMAN, TERRENCE, 1988. Language Contact, Creolization and Genetic Linguistics, Berkeley: University of California Press. TOMPAIDIS, DIMITRIOS E., 1980. ‘Simvoli stin erevna tou onomatikou sinolou tis Pontiakis’ (Contribution to the study of the noun phrase of the dialect of Pontos), Archeion Pontou 36, 220–237. WINFORD, DONALD (2003) An Introduction to Contact Linguistics, Oxford: Blackwell. WINFORD, DONALD, 2005. ‘Contact-induced changes: classification and processes’, Diachronica 22, 373–427.