1 The "parsimony principle" Up until the late 20th century, the following assumption was predominant in linguistics:

Transcription

1 PRELIMINARY VERSION Identifying instances of contact-induced grammatical replication MPI Leipzig, May, 2007 Bernd Heine & Tania Kuteva, Abstract While it is fairly easy to establish that lexical material has been transferred from one language to another, this is hard when it comes to grammatical replication, where no form-meaning units are involved: For a number of reasons it is essentially impossible to "prove" that language A has replicated a grammatical structure on the model of language B. In this paper we argue that there nevertheless are ways of establishing that such processes have in fact taken place. Looking at a wider range of cases of contact-induced grammatical replication, we will propose a catalog of diagnostics that are of help in establishing whether or not a given linguistic grammatical change is or is not due to language contact. 1 The "parsimony principle" Up until the late 20th century, the following assumption was predominant in linguistics: (1) The "default hypothesis" (cf. Milroy 2003: 143) Change takes place in single languages unaffected by other languages, hence it is more satisfactorily explained internally than externally. Lass (1997: 199, 209) argues that whenever there is a possibility of dual or multiple origin for a given feature in a language, the most "parsimonious" explanation for a linguistic innovation is one in terms of internal development ("endogeny") rather than of language contact, that is, of external development: Endogenous changes, it is argued, occur in any case "whereas borrowing is never necessary"; Lass therefore sees justification for postulating a "law" or principle of parsimony (see Filppula 2003: 161-2). While the point made by Lass is well taken, considering that hypotheses on contact-induced grammatical replication are notoriously controversial, one may nevertheless wonder whether there is need to invoke such a principle or -- perhaps more appropriately -- what the theoretical significance of such a principle is. Given a situation such as the one alluded to by Lass, where some linguistic feature is equally amenable to an internal and an external explanation, obviously neither offers a sufficient basis for a viable hypothesis. So deciding on an internal hypothesis on the basis of the principle of parsimony in such a case would result in an analysis based on lack of evidence rather than on appropriate empirical evidence. Linguistic change is a historical event, and the task of the linguist is to reconstruct this process and to propose an explanatory account for it. In our view, reconstructing this process is not a matter of more vs. less parsimonious description; rather, what matters is the magnitude and the quality of the evidence that can be adduced for one's reconstruction, rather than whether some account is more parsimonious than another. We therefore side with Filppula (2003: 170) when he concludes: "Looking for the most parsimonious explanation in a given case may well be a virtue in itself, or produce the most satisfying solutions from an aesthetic perspective, but in my view it does not change the fact that the quest must always be for the best explanations whether more or less parsimonious." But there is another, more serious problem with Lass's stance. This stance echoes a widespread belief that there is a fundamental distinction in linguistic change, in that it is either internal or external. More recent work on language contact suggests that this stance is

2 2 in need of reconsideration, in that there is a third possibility, namely multiple causation, in that linguistic change can be simultaneously an internal and an external process (see, e.g., Thomason & Kaufman 1988; Thomason 2001b: 91; Heine & Kuteva 2003; 2005; 2006; see also Filppula 2003: 170). We will not pursue this issue further in this paper since it has been discussed extensively in the works just cited. The subject matter of this paper is limited in scope. As we observed above, whether a given transfer from one language to another has in fact taken place or not is frequently controversial since there are no appropriate criteria to decide on this issue. In the following we aim at searching for such criteria. By studying a wider range of cases of contact-induced grammatical replication we will be looking for diagnostics that are of help for establishing whether a given linguistic grammatical change is or is not due to language contact. 2 Diagnostics The perspective portrayed by Lass (1997) reflects what a number of other authors have written over the last decades: Linguistic transfer across languages was viewed as a phenomenon that is restricted largely or entirely to lexical and phonological borrowing while grammatical structure tended to be portrayed as being immune to transfer. In a monographic treatment of bilingualism, Oksaar (1972: 492) concluded that there are "no clear cases that would permit the generalization of statements that grammatical paradigms, bound morphemes, word order etc. can be subject to interference". As this assertion indicates, the problem addressed in such writings is an essential one, namely one of empirical evidence: Are there any "clear cases" of structural transfer? In the present paper we address this problem by looking for answers to the questions listed in (2). (2) Questions a What evidence is there for grammatical replication to have taken place? b If linguistic transfer has taken place, did it go from language A to B or from B to A? c Could that change have taken place without involving language contact? d Is it possible to prove that grammatical replication has taken place? Question (2a) is the subject of sections and 2.2.2, while (2b) is dealt with in 2.2.3, and in the final section 3 we will return to (2c) and (2d). 2.1 Grammatical replication Before dealing with these questions, a few words on the framework to be used are in order. Language contact manifests itself in the transfer of linguistic material from one language to another, typically involving the following kinds of transfer: (1) Kinds of linguistic transfer a Form, that is, sounds or combinations of sounds, b Meanings (including grammatical meanings) or combinations of meanings, c Form-meaning units or combinations of form-meaning units, d Syntactic relations, that is, the order of meaningful elements, e Any combination of (a) through (d). Our concern here is with (1b) and (1d), that is, with what following Weinreich ([1953] 1964: 30-1; see also Heine & Kuteva 2005; 2006) is called (contact-induced) grammatical replication, that is, a process whereby speakers create a new grammatical meaning or

3 3 structure in language R on the model of language M by using the linguistic resources available in R; traditionally, grammatical replication has been referred to with terms such as "structural borrowing" or "(grammatical) calquing". The terms model language (M) and replica language (R) are used for the languages being, respectively, the source (or donor) and the target (or recipient) of transfer. Replication contrasts with borrowing, which concerns the transfer of either physical substance (1a) or form-meaning units, that is, words and morphemes (1c). 1 The data to be discussed in this paper are taken from attested cases of language contact, involving two kinds of languages, namely a model language M and a replica language R, as well as a linguistic property P R belonging to R that is suspected to have been replicated from a corresponding property P M in M. Note that M and R are not fixed entities; rather, one and the same language can at the same time be model and replica language. Studies on language contact in the past have focused in particular on sprachbunds. While it is true that these are paradigm products of language contact, for an analysis of grammatical replication they are as a rule of limited value, for the following reason: Sprachbunds, irrespective of whether they concern the Balkans, Meso-America, Ethiopia, or South Asia, are the result of a long and complex history, involving a range of different languages and of historical processes that took place at different periods in the development of the sprachbund (see e.g. Tosco 2000 on the Ethiopian sprachbund), and it remains in many cases unclear which of the factors, historical processes and/or languages exactly contributed what to some particular change. Another important source of information has been seen in creoles, which, like sprachbunds, owe their existence to language contact. But like that of sprachbunds, the history of creoles is the result of an interaction of a variety of different factors, namely (a) the "substrate" language(s) of its speakers, (b) one or more "superstrate" language(s) as well as possible "adstrate" languages, (c) an interaction of (a) and (b), and (d) a sequence of historical events with each characterized by a different sociolinguistic situation. Determining what each of these factors contributed to produce a given grammatical change is more often than not near to impossible. Accordingly, we will have little to say on sprachbunds and creoles and rather concentrate on cases of language contact that took place more recently, and where it is fairly uncontroversial which the model and which the replica language was. 2.2 Evidence for contact Still, even in the case of the last mentioned situations it turns out to being hard to establish whether a given grammatical change is due in some way to language contact. A starting point for hypotheses on GR (grammatical replication) is frequently provided when there is a property P r in a given language R associated with speakers who are or have been in contact with another language M, while that property is absent in speakers of R who have not participated in that contact, especially if a corresponding property P m is found in language M. Such a situation, however, provides neither a necessary nor a sufficient basis for establishing GR. Heine and Kuteva (2005) propose the following heuristic to identify instances of contactinduced linguistic transfer in the absence of any other evidence: (3) Identifying instances of contact-induced transfer 1 There are many alternative terminologies; for example, Thomason and Kaufman (1988) or Thomason (2001b: 93) uses borrowing, source language and receiving language for both kinds of transfer.

4 4 If there is a linguistic property x shared by two languages M and R, and these languages are immediate neighbors and/or are known to have been in contact with each other for an extended period of time, and x is also found in languages genetically related to M but not in languages genetically related to R, then we hypothesize that this is an instance of contactinduced transfer, more specifically, that x has been transferred from M to R. (Heine & Kuteva 2005) This heuristic takes care of a number of cases of presumed replication but -- unfortunately -- it is far from being an exhaustive generalization A catalogue In the following we will provide a catalog of diagnostics that we argue are helpful for identifying instances of GR. None of these criteria in itself is sufficient to prove language contact, but the more of them apply, the stronger a case can be made that there is contactinduced grammatical change. D1: Intertranslatability. P R is immediately intertranslatable with P M in M and, provided other factors, in particular genetic relationship, can be ruled out, the only reasonable explanation for this fact is one in terms of language contact. D1 refers to a phenomenon that is frequently described with reference to higher degree of either structural isomorphism or intertranslatability between the languages in contact. Intertranslatability may concern (i) the morphosyntax, (ii) the meaning, (iii) the contextual structure, or (iv) any combination of these, and it may involve single grammatical categories, phrasal structures, sentence structure, or discourse structures beyond the clause. The more of these factors are concerned the more plausible a hypothesis on language contact can be made. It is not possible to deal here with each of these phenomena separately, so we will be restricted to a few examples of what may be typical cases of presumed intertranslatability. Morphosyntactic examples of D1 are provided, e.g., by phrase and sentence structure alignment in the Takia-Waskia contact situation on Karkar Island off Papua New Guinea (Ross 1996; 2001; 2003), the word-for-word structure in the Guernésiais - English contact on the Channel island of Guernsey (Ramisch 1989; Jones 2002), or the morph-for-morph structure in Kupwar village of India (Gumperz & Wilson 1971: 165). The following example of D1 concerns syntax relative clause formation: Yiddish speakers at The Beach in Venice along the coast next to Los Angeles, replicate both properties of English relatives in Yiddish and properties of Yiddish relatives in their variety of English (Rayfield 1970: 69), leading to a certain amount of intertranslatability, especially but not only when the head noun is taken up anaphorically by a possessive or prepositional participant in the relative clause, as in the following examples: There is an invariable relative clause marker followed by a full main clause. (4) Yiddish speakers in Venice, Los Angeles (Rayfield 1970: 69) a Yiddish: di froy, vos bey ir bin ikh geven the woman REL at her am I stayed 'the woman with whom I was staying' b Yiddish English: one of the members that we're very proud of her

5 5 That the existence of a high degree of intertranslatability is due to language contact is suggested in particular by cases where both morphosyntax and meaning are involved. An example from varieties of the Indo-Aryan language Romani spoken in Bulgaria may illustrate this (Boretzky & Igla 1999: 719). Like Bulgarian, these varieties use the following construction to form a negative future tense: ['have' + NEG - complementizer - Verb + person/tense inflection]. Since genetic relationship can be ruled out as a contributing factor, such a similarity in morphosyntactic structure and grammatical meaning is likely to have been induced by contact (As we will see below under paired grammaticalization, there is additional evidence to strengthen a contact hypothesis). (5) Bulgarian (Boretzky & Igla 1999: 719) njama da otida. not.have to go.1.sg.pres 'I will not go.' (6) Romani varieties influenced by Bulgarian (Boretzky & Igla 1999: 719) naj/nane te ker- av. have.not that do- 1.SG.PRES 'I will not do.' In the following example from northwest Amazonia, intertranslatability also concerns both morphosyntax and meaning. The East Tucanoan language Tucano and the North Arawak language Tariana are presumably genetically unrelated but are known to have been in close contact. Aikhenvald (2002: 130) found examples such as the ones in (7) and (8), where intertranslatability relates to the following factors: Both languages exhibit the order verbimperative form, both have a category that is crosslinguistically unusual (see below), namely what Aikhenvald calls the secondhand imperative (IMP.SEC), used to convey the meaning 'Do this because someone else told you to!', and both use this category for second and third person referents. Thus, the two languages exhibit a fairly high degree of syntactic and semantic intertranslatability, even if there is no structural identity (as can be seen in the presence of a second person prefix in Tariana, which does not have an equivalent in Tucano). (7) Tucano (East Tucanoan; Aikhenvald 2002: 130) a»ti- ato. come- IMP.SEC 'Come (on his order)!' (8) Tariana (North Arawak; Aikhenvald 2002: 130) pi- nu- pida. 2.SG- come- IMP.SEC 'Come (on his order)!' The following example also involves a single grammatical category. Like most other Meso- American languages, Pipil, an Aztecan language of El Salvador now being replaced by Spanish, has traditionally neither prepositions nor postpositions, but it has relational nouns instead. Under the influence of the model language Spanish, Pipil speakers have grammaticalized some of their relational nouns to Spanish-type prepositions, thereby creating a structurally equivalent category (Campbell 1987; Harris & Campbell 1995: 126-7; see also Heine & Kuteva 2003: 535). Thus, the presence of prepositions in both Spanish and Pipil in a

6 6 linguistic area where prepositions are not really a commonality can be taken as evidence for contact influence. A higher degree of intertranslatability can also arise due to structural assimilation in the direction of the model language. The Balto-Finnic language Estonian is characterized by a highly productive pattern of nominal compounding, compounds having three or more components being quite common. In Russian on the other hand, compounding is far less common, compounds of more than two components being rare. Russian spoken in Estonia exhibits a clearly higher amount of compounds than Russian spoken in Russia and, as the examples by Verschik (2004) suggest, Estonian Russian appears to be more readily intertranslatable with Estonian. Genetic patterning An important prerequisite for identifying instances of contact-induced change is to determine the genetic relationship patterns among the languages concerned, e.g. whether genetic relationship might not be causally responsible for suspected cases of transfer from one language to another. Genetic relationship among the languages concerned can make it difficult to discover instances of replication, but it can as well offer clues for contact-induced change. D2: Genetic patterning. P R is not found in other dialects or languages closely related to R, while the corresponding category P M of M does not show such restrictions. This diagnostic is discussed by Heine and Kuteva (2005, 1.4.1) under the label "genetic patterning". Eastern Yiddish, a language historically derived from medieval German, has been spoken for centuries in a Slavic-speaking environment. Russian has a focus construction illustrated in (9). As Prince (1998) argues convincingly, this construction has been replicated in Yiddish, as in (10). The similarities between the two constructions, referred to respectively as the eto- and the dos-constructions, are described by Prince (1998: 340) thus: "both are simple sentences (i.e. without subordinate clauses), each has an expletive NP in initial position, in both cases the expletive NP is the neuter demonstrative pronoun, and both are roughly translatable by an English it-cleft". (9) Russian (Prince 1998: 340) Eto L'eon'id uv'id'el Er'iku. this.n Leonard.M.NOM saw Erica.F.ACC 'It's Leonard who saw Erica.' (10) Yiddish (Prince 1998: 340) Dos hot Leyb gezen Erike- n. this.n has Leonard.M.NOM seen Erica- F.ACC 'It's Leonard who saw Erica.' This construction has no obvious analog in Germanic language other than Yiddish, but there are analogs in other Slavic languages -- hence, replication from Russian to Yiddish is the most plausible hypothesis. Our second example also concerns Germanic-Slavic language contacts. The German reflexive pronoun sich is confined to third person subject referents. Influence of Slovenian on German

7 7 in Trieste had inter alia the effect that the Slovenian pattern of reflexive marking was replicated to some extent by German speakers (Morfill [1885] 1971: 269): The German reflexive pronoun sich came to be used in new contexts on the model of the Slavic reflexive marker se, being extended to second and first persons, e.g., wir waschen sich (we wash REFL) 'we wash ourselves', that is, its use spread across all three categories of personal deixis. Similarly, the Yiddish reflex six of the German third person reflexive sich is said to have been extended to all persons and numbers under Slavic influence (Thomason & Kaufman 1988: 82). This usage is uncommon in German but common in Slavic languages, hence it is suggestive of transfer from Slavic languages to Yiddish and German. The Finnic language Estonian has grammaticalized the verb tulema 'to come' to a modal auxiliary for the deontic modality of necessity ('must', 'have to'), presenting the agent in the stative-locative adessive case (ADE), an oblique case form. Much the same is found in the Baltic language Latvian, which also has grammaticalized the verb for 'come' in its reflexive form to a modal auxiliary (na#kties), also expressing the deontic modality of necessity, with the agent being encoded by means of an oblique case marker, the dative (DAT) (Stolz 1991: 79-80). The striking similarity between these two constructions, as well as the fact that a grammaticalization of verbs for 'come' to deontic modals appears to be crosslinguistically rare, can be interpreted meaningfully only by assuming that the constructions are historically related. Estonian and Latvian are genetically unrelated, so genetic relationship can be ruled out, and the only reasonable hypothesis is one in terms of contact. That it was Finnic language that provided the model is suggested by the following: Finnish, a language closely related to Estonian, also uses the verb for 'come' as a modal auxiliary for deontic modality and an oblique case, the genitive, for presenting the agent, while in Baltic, such a grammaticalization is absent except for Latvian -- that is, it is not found in Lithuanian. In view of this genetic patterning we follow Stolz in arguing that this is a case of contact-induced transfer. 2 Presence of a copula appears to be a genetically determined characteristic of Tucanoan languages in northwest Amazonia; all Tucanoan languages have one. On the other hand, the majority of North Arawakan languages do not have any copula verb, but Tariana is a noteworthy exception: The copula alia of this North Arawakan language shows the same structure of marking locative-existential clauses as the corresponding East Tucanoan categories do, and its use is expanding to also mark identity and equation clauses, like those of East Tucanoan languages. This genetic distribution suggests that Tariana may have replicated its copula from Tucanoan languages, and exactly this is what appears to have happened: Tariana has been massively been influenced by East Tucanoan languages, and the creation of a copula on the model of the latter languages is one of the many manifestations of this process (Aikhenvald 2002: 153-4). 2 The most plausible hypothesis is that this transfer proceeded from a Finnic to a Baltic language, that is, from Estonian to Latvian, rather than the other way round. But there is at least one more possibility for a contact explanation of this particular case, not mentioned by Stolz. According to this possibility, the model language is Russian, while Finnish, Estonian, and Latvian are the replica languages. The arguments in support of such a hypothesis would be: First, Russian has a pattern equivalent to the one that Latvian has for expressing the deontic modality of necessity, and second, Russian is geographically contiguous with Finnish, Estonian and Latvian, but not with Lithuanian (Östen Dahl, p.c.).

8 8 Another example of genetic patterning is provided by Savala (2002) on language contact between Mayan languages and languages of the Mixe-Zoquean family in the Mexican state of Chiapas, where the following observations suggest that in this case there was directionality of semantic and structural transfer from the latter to the former languages: Given that the verbal compounds with predicate serialization is a pattern commonly found in the Mixe-Zoquean family and the only Mayan languages that have developed that pattern are those that surround the Mixe-Zoquean family, it is clear that incorporating secondary predication is an areal phenomenon and that [the Mayan languages] Chol and Huastec borrowed it from the Mixe-Zoquean languages with which they are or were in contact at some point in time. (Zavala 2002: 184; translated by Gast and van der Auwera 2006: 13). The diagnostic of genetic pattern can also be framed in terms of genetic inheritance, in that P M can be reconstructed back to earlier stages of M but P R cannot similarly be reconstructed in R. For example, grammatical categories of evidentiality are found in many languages extending from southeastern Europe to central Asia, both in Turkic and non-turkic languages; note that evidentials in these languages are of a special type, referred to as "indirectives", where the speaker has not witnessed the narrated event directly and the source of information is mostly unspecified. In Modern Bulgarian the evidential category indicates that the relation between the speaker and the event s/he is reporting is not direct but mediated by a third party, i.e. the speaker is reporting hearsay information. In other words, along with verbal forms for events which are reported as first hand (which is also the case with most Indo-European languages), Bulgarian has a special set of grammatical forms which encode the fact that the event denoted by the verb is presented as hearsay and not as something which has been witnessed directly by the speaker. Examples (11) and (12) capture the relevant difference: (11) Bulgarian a Declarative "first-hand" information Ne iskas da í pises. NEG want.2.sg.pres to her write.2.sg.pres 'You don't want to write to her.' b Evidential hearsay information Ne si iskal da í pises. NEG be.2.sg.pres want.aor.prtc.m.sg to her write.2sg.pres 'They say/i hear, you don't want to write to her.' As the above examples indicate, the evidential form in (b) is a complex auxiliary structure consisting of the auxiliary si 'be' and the aorist participle of the main verb inflected for the masculine gender and the singular number, iskal. We argue that the Bulgarian evidential is the result of language contact (Mirc&ev 1963: 208; Kuteva 2001). Old Bulgarian had no gram for the evidential. Moreover, apart from closely related Macedonian, there is no other Slavic language in which an evidential has been attested either diachronically or synchronically. Therefore it is unlikely for the evidential in Bulgarian to be a genetically inherited Slavic feature. In fact, Old and Middle Bulgarian texts reveal that the first indications for the existence of this category in Bulgarian appear after the end of the 14 th century. On the other

9 9 hand, Turkish did have an evidential category before the Ottoman Empire was established, i.e. before the end of the 14 th century. Therefore the most plausible explanation for the genesis of the Bulgarian evidential is that it appeared as replication of the evidential in Turkish, especially given the socio-linguistic aspect of the contact situation of the Balkan sprachbund, where Bulgarian like its neighbouring Balkan countries was strongly colonized by Turkish speakers and a great number of Bulgarian speakers were living in conditions of bilingualism. In fact, a number of such evidential categories can be shown to belong to the genetic heritage of the Turkic language family, while in non-turkic languages they do not seem to have been genetically inherited. On the basis of this observation one is justified to argue in favor of a Turkic origin and of contact-induced transfer, as Johanson (1996) in fact does: "Features of Turkic evidential systems have proven highly attractive in contact situations and have exerted considerable influence on non-turkic contact languages of Asia and Europe, e.g. the Balkans, Anatolia, the Caucasus region, the Volga region and Central Asia". D3: Rare grammatical category. If two neighboring, genetically unrelated (or only remotely related) languages share a grammatical category that is crosslinguistically highly unusual, then there is some probability that this commonality is due to language contact. We saw one example in the preceding section: As Aikhenvald (2002: 130) observes, the East Tucanoan language Tucano and the North Arawak language Tariana of northwestern Amazonia both have a modal category that is crosslinguistically unusual, namely what she calls the secondhand imperative. Since these two neighboring languages are presumably genetically unrelated but are known to have been in close contact, an account other than in terms of contact appears to be not very plausible. As Aikhenvald demonstrates in fact, there has been massive grammatical replication in Tariana on the model of East Tucanoan languages. "Rare category" may also be a relative notion, it can be one that, although being found in a wider range of languages, is uncommon in some specific language family. If a language has such a category although other members of that family do not, and that language is known to have been in contact with languages that also have the category, then a case for contactinduced transfer can be made. For example, Romance languages have no fully grammaticalized category of an evidential, but some varieties of Portuguese spoken in northwest Amazonia do, for the following reason: The North Arawak language Tariana of northwest Amazonia has an obligatory paradigm of four clitics for tense and evidentiality, distinguishing between visual, non-visual, inferred, and reported evidence. Tariana speakers use Portuguese, the official language of Brazil, as an important lingua franca, and in their use of Portuguese, they tend to replicate their evidentiality system by drawing on lexical expressions of Portuguese, using them more frequently and developing them into what appear to be incipient categories for which there is no equivalent in Standard Portuguese (Aikhenvald 2002: , 315-6; Heine & Kuteva 2005, 2.4.1). Thus, the fact that Tariana Portuguese is presumably the only variety of Portuguese to have such a paradigm of four evidential structures can be accounted for meaningfully only with reference to language contact. D4: Paired structural similarity. There is a set of two or more properties shared by M and R whose presence cannot be coincidental nor can it be due to shared genetic relationship.

10 10 While it is frequently hard to make a convincing case of change when a single property is involved, a much stronger case can be made when R and M share a property that incorporates a set of two or more interrelated properties. Molise Croatian, in short Molisean, is the language of a community of Croatian speakers from the Hercegovinian Neretva Valley who emigrated around 1500 because of the Turkish invasion on the Balkans, settling in areas of southeastern Italy that were sparsely inhabited due to earthquakes and epidemics; today, Molisean is spoken only in two villages, Acquaviva and Montemitro, of Molise Region in the Campobasso Province. After contact both with the local varieties and with Standard Italian over a period of half a millennium, their language has been massively influenced by this Romance language (for a survey, see Breu 1998; see also Breu 1999; 2003a; 2003b; 2003c; 2004). One of the many commonalities found in Molisean and the local colloquial dialects of Italian spoken in the Molise region concerns comparative and superlative constructions (see Breu 1999 for more details): In both of them there is a marked development from synthetic and suppletive to analytic, periphrastic comparative constructions. This fact in itself is not really telling since similar processes have occurred and are still occurring widely in European languages, including various Slavic languages (see Heine & Kuteva 2006, section 2.6). Even the fact that in this process, Molisean has acquired a type of comparatives that is unique among the Slavic languages, differing drastically from those found in Standard Croatian, the closest relative of Molisean, is not sufficient to argue that contact-induced grammatical transfer was involved. But as Breu (1999) shows, the structure of the Molisean comparatives contains some idiosyncratic features that have their parallels in the local Italian dialects and can only be accounted for with reference to an Italian model; these are in particular: In both Molisean and these Italian dialects it is exactly the same kind and number of suppletive forms that were spared out from loss. Second, in both languages in contact, adjectival comparatives are replaced by adverbial ones (e.g. Molisean bolji (adj.) > bolje (adv.) 'better' vs. local Italian migliore (adj.) > meglio (adv.) 'better') and, third, both languages in contact are also experiencing a replacement of the superlative by the comparative construction. We therefore follow Breu in maintaining that such specific shared developments are sufficient to support a contact hypothesis. An especially common kind of paired structural similarity can be seen in cases of shared polysemy, where R and M exhibit the same polysemy pattern. "Polysemy" can be found in particular in cases of auxiliation where R and M exhibit the same kind of main verb-auxiliary pairing; we will return to this case below (shared grammaticalization). The Persian verb goza:s&tan 'put' is also used as a modal verb meaning 'let, allow', and the same situation exists in the Turkic language Azerbaijanian, where the verb goy- is also used as a verb for 'put' and as a modal expressing 'let, allow' (KIral 2005: 290). Since this particular kind of "polysemy" is hard to find in other languages but occurs in these two neighboring languages known to share a long history of contact, there is reason to assume that it is due to language contact. A special kind of shared polysemy is offered by the contact situation between the French/Norman dialect Guernésiais of Guernsey Island and English 3 (Ramisch 1989; Jones 3 Guernésiais (Guernsey) is a Norman French dialect that has been spoken on the Island of Guernsey of the Channel Islands archipelago for more than a thousand years but is now moribund: After the World War II, when many island inhabitants who had been evacuated to

11 ). One of the salient features of Guernsey English is its frequent use of the interrogative interjection (or discourse particle) eh (typically pronounced as a diphthong [ei]; Ramisch 1989: ). Since it is also frequently found in Guernésiais, this is a good candidate for a contact-induced transfer in this variety of English. However, there is a problem with the contact hypothesis: This interjection is also common in other varieties of English, and it has been described as a particular feature of Canadian English: Not only is it mentioned by US Americans when asked about characteristics of Canadian English but Canadians themselves regard it as typically Canadian. That the frequent use of eh in Guernsey English nevertheless is the result of contact with this French/Norman dialect Guernésiais is suggested by the fact that eh shows the same polysemy in both varieties, having the following three meanings (Ramisch 1989: ): It is (i) a request to repeat an utterance, (ii) a tag added to a statement to induce the hearer to express his opinion on what is said by the speaker, and (iii) a phatic element used by the speaker to secure the hearer's attention. A convincing case for transfer from one language to another can be made when the polysemy pattern shared by R and M is crosslinguistically rare. The Maltese locative preposition fuq 'on top of, upon' occurs in some contexts where in English the locative preposition on would be used, as in the following examples: fuq it-tv 'on television', fuq btala 'on holiday', fuq parir ta» 'on the advice of', fuq talba ta» 'on the request of' (Drewes 1994: 101). It would probably be hard to find another language using a superessive locative adposition ('on') in exactly these contexts. Since genetic relationship can be ruled out, contact provides the most reasonable hypothesis; hence, Drewes (1994) argues convincingly that this a case of transfer from English to Maltese (Heine & Kuteva 2006, chapter 2). Examples of structural doublets can also be found in contact between Uzbekistan Arabic (Bukhara and Qashqa Darya dialects) on the one hand and Turkic and Iranian languages on the other. In Arabic, an indefinite noun remains unmarked whereas a definite noun is marked by the article al-, l- or il-, cf. (12a). In the latter languages, the opposite pattern is found: The indefinite noun receives an indefinite marker while the definite noun is unmarked, as in the Turkish example (12c). Uzbekistan Arabic shows exactly the same pattern as the Turkic and Iranian languages, as (12b) shows. In view of this double correspondence, it is likely that Uzbekistan Arabic has replicated this pattern by creating a new indefinite article fat (historically derived from Classical Arabic fard 'single, individual' (Jastrow 2005: 135), while it lost the definite article. (12) Arabic-Turkic-Iranian language contact (Jastrow 2005: 135) a Damascus Arabic ka#n fi# mara lmara»a#let there.was woman the.woman asked b Uzbekistan Arabic fat mara ko#net mara qo#let England during the war returned back home, English gradually began to replace this Norman dialect, a process that appears to be ongoing -- Guernésiais is described by Jones (2002) as being obsolescent: Between 1981 and 1996, the number of Guernésiais speakers plummeted from 11 % to 3 % of the island population of Guernsey (Ramisch 1989; Jones 2002: 164). The linguistic history of this island has been characterized by grammatical replication in both directions, from Guernsey to English and vice versa (Jones 2002).

12 12 c Turkish bir kadin vardi kadin dedi. a woman there.was woman asked 'There was a woman.' 'The woman asked.' Another case of doublets can be seen in the following example from language contact between Mayan and Mixe-Zoquean languages in the Mexican state of Chiapas (Zavala 2002; Gast & van der Auwera 2006: 20ff.). Both the Mayan language Tzotzil and the Mixe- Zoquean language Zoque have grammaticalized a verb meaning 'to want, to love' (Tzotzil k»an, Zoque sun) in its impersonal use (with a third person ergative prefix) to a construction of the deontic modality of necessity ('have to do X'). The fact that this grammaticalization is found in other Mayan languages as well but not really in Mixe-Zoquean languages would suggest that there was transfer from Tzotzil to Zoque, rather than the other way round (see genetic patterning). But of interest for the present purpose is the fact that in both languages this grammaticalization has given rise to two distinct but semantically largely equivalent constructions: on the one hand to a finite-verb complement construction (a), and on the other hand to an auxiliary construction (b). (13) The encoding of 'NECESSARY (HAPPEN (X)' in Tzotzil and Zoque (Gast & van der Auwera 2006: 20-1) a 3.ERG-WANTS [that X happens] b 3.ERG-WANTS [X to happen] The fact that in these two neighboring languages belonging to different language families one and the same grammaticalization process has led to a doublet of constructions is hard to account for other than in terms of contact-induced transfer. D5: Frequency of use. P R occurs more frequently with speakers of R being in contact with speakers of M than in speakers of R who are not or less exposed to contact with language M, and M has a property P m that is equivalent to P r. According to this diagnostic, differences in the relative frequency of constructions may also provide a diagnostic for grammatical replication. In Finnish (and Finland Swedish) the genitive modifier precedes its head in attribute possession while in Russian the genitive modifier follows the head, though in modern colloquial Russian the reverse order is said to be more common when the possessor is a specific person and/or is in focus. As Leisiö (2000) shows, Russian speakers in Finland who have had a long history of contact with Finnish use the Finnish-type modifier-head order in genitive constructions distinctly more frequently (89 %) than Russian speakers in Finland who do not have such a contact history (41 %). The Austronesian language Tigak does not have obligatory plural marking on nouns, but under the influence of English, which has obligatory plural marking and is a secondary lingua franca for Tigak speakers, plural marking appears to have risen. Jenkins (2002: 260) found in her transcribed Tigak material that young educated Tigak increase the use of their overt plural morpheme, use their plural morpheme with 46 % of all plural nouns whereas traditional, conservative Tigak speakers use the plural marker only with 19 % of plural nouns. Thus, frequency of use offers a cue to hypothesize that nominal plural marking has been affected by (the relative degree of) language contact.

13 13 Turks in Germany have been found to use their conjunction ve 'and' and their plural marker much more frequently than Turks in Turkey, and Johanson (1992: 182-3) takes this to reflect German influence, drawing attention to the fact that Turkic languages that have been influenced by Indo-European languages generally show an increased frequency of use of conjunctions. D6: Demographic variables. Language contact in general and grammatical replication in particular affect certain social or demographic groups more than others; accordingly, demographic categories that are more strongly affected are likely to show more pronounced effects of contact-induced change. Certain kinds of language contact can be especially intense in specific social categories of a speech community, such as speakers having had some formal education, people of some particular profession, of the upper class rather than of lower classes, or males rather than females, etc. Or contact may be a recent phenomenon that has affected only younger generations. Accordingly, grammatical replication can be restricted to or more pronounced in such categories, and appropriate sociolinguistic information may assist in formulating hypotheses on contact. That sociolinguistic factors, in particular by demographic variables such as age, sex, profession, etc. provide cues for identifying instances of grammatical replication has been demonstrated in particular by Aikhenvald (2002; see also Heine and Kuteva 2005, 1.4.4). One example volunteered by Aikhenvald (2001; 2002: 183) is the following: The North Arawak language Tariana of northwestern Brazil is in close contact with Portuguese, the official language of Brazil, and has been influenced by the latter in various ways. In Portuguese, interrogative pronouns are also used as relative clause markers and Tariana speakers also use their own interrogative pronouns as markers of clause subordination on the pattern of the model language Portuguese, retaining their own relative construction and simply adding their interrogative pronoun (e.g. kwana 'who?'). As Aikhenvald points out, it is young and innovative speakers of Tariana that have created this new structure, and the structure seems to be confined to this social group. The example of the situation on the island New Ireland in Papua New Guinea that was sketched above also illustrates this factor. The Austronesian language Tigak does not have obligatory plural marking on nouns, but under the influence of English, which has obligatory plural marking and is a secondary lingua franca for Tigak speakers, educated young Tigak increase the use of their overt plural morpheme (Jenkins 2002: 260). Thus, once again, it is one particular segment of the population that provides clues of an ongoing contact-induced change Grammaticalizing behavior A number of examples that we discussed in the preceding section are the product of grammaticalization processes (Heine & Kuteva 2003; 2005; 2006). In the present section now we will be more centrally concerned with clues that grammaticalizing behavior may offer for identifying instances of contact-induced grammaticalization. D7: Relative degree of grammaticalization: P R differs from P R used by R speakers or speakers of languages closely related to R that are not, or less, exposed to language contact by being more grammaticalized.

14 14 That replica categories are as a rule -- though not invariably -- less grammaticalized than the corresponding model categories can be demonstrated with the Portuguese-Tariana example of language contact in the Vaupés region of northwest Brazil that we saw already above: The four evidential categories that Tariana speakers developed in their variety of Portuguese are only weakly grammaticalized, lexical constructions, whereas the four corresponding Tariana model categories are more strongly grammaticalized (Aikhenvald 2002: , 315-6; Heine & Kuteva 2005, 2.4.1). Another example is the following. According to Rayfield (1970: 69), Yiddish speakers in Venice along the coast next to Los Angeles, most of whom are strongly bilingual in English, created a future tense on the model of the English be-going-to future, illustrated in (14). (14) Yiddish of Venice, California (Rayfield 1970: 69) All right, ge ikh kumen bald. all right go 1.SG come.inf soon 'All right, I'm going to come in a minute.' As is to be expected, however, the replicated future category is less strongly grammaticalized than its English model. In particular, it is not normally acceptable when the andative deixis of ge- 'go' is violated, e.g., when serving as an auxiliary for kumen 'to come'. Thus, while in English it is possible to say He's going to come, a Yiddish speaker at the meeting of a Yiddish cultural group in Venice was reproved for saying er geet kumen (he goes to.come) because of the conflict of spatial deixis expressed by 'go' and 'come'. At a more advanced stage of grammaticalization, after of centuries of language contact, replica categories can come to be virtually indistinguishable from the model categories. For example, Slavic languages do not normally distinguish articles, but Upper Sorbian 4 in eastern Germany, having had a long history of contact with German, an article language, has grammaticalized its numeral 'one' to an indefinite article jen- which is, as Breu (2003a) shows convincingly, equally grammaticalized as its German model category ein-: Both are used as presentative (stage 2), specific indefinite (stage 3), and as non-specific indefinite articles (stage 4); 5 consider the following example of the generic stage 4 article in Upper Sorbian and German. (15) Upper Sorbian (Slavic; Breu 2003a: 46) Upper Sorbian Jen Serb nebc&i. a Sorbian not.lies German Ein Sorbe lügt nicht. a Sorbian lies not 'A Sorbian never lies.' Breu (2003a: 61) therefore concludes that the two are equally fully developed and, in fact, without any knowledge about the history of this situation it would therefore seem hard to 4 What we have to say in the following applies exclusively to non-standard varieties of Upper Sorbian, not the standard language (see Breu 2003a; 2004). 5 See Heine and Kuteva (2006, chapter 3) for the five-stage scenario of the grammaticalization of indefinite articles.

15 15 decide which is the model and which is the replica. Still, even at such an advanced stage of grammaticalization there remain some clues to solve this problem: There are usually specific contexts and use patterns which tend to be spared out in the process and which therefore can bear witness to the direction of transfer, as the case of the Upper Sorbian indefinite article shows: First, the replica category has not been extended to a number of idiomatic expressions where the model language would require the indefinite article. And second, there are a number of contexts, involving in particular generic concepts, where there must be an indefinite article in German, while in Upper Sorbian the indefinite article is either optional, as in (a), or is disallowed, as in (b). (16) Upper Sorbian (Breu 2003a: 44; no glosses are provided by the author)$ a Upper Sorbian Tón jo tak sylny kaj O / jen elefant. German Er ist so stark wie *O / ein Elefant. 'He is as strong as an elephant.' b Upper Sorbian German Ja sym ódny kaj O / *jen law. Ich bin hungrig wie *O / ein Löwe. 'I am hungry like a lion.' To conclude, in spite of the fact that the Upper Sorbian indefinite article has become a nearly complete replica of the German model category, having reached the same general stage of grammaticalization, there remain a number of contexts where the replica category is less grammaticalized than the model. D8: Rare grammaticalization: R and M share a grammaticalization process that is crosslinguistically uncommon. There are certain grammaticalization processes that are crosslinguistically widely attested, hence their diagnostic value for identifying grammatical replication is low. For example, if both R and M have developed a de-andative future, where a verb 'go to' is grammaticalized to a future tense marker, then this is not strong evidence for a hypothesis to the effect that R replicated this process from M: Since this is a highly common process, it is well possible that it took place in R independent of M. But even in the case of future tenses it is sometime possible to use grammaticalization as a diagnostic tool for proposing a contact hypothesis. In Romani dialects spoken in Russia there is a future tense using the verb l-av 'take' as an auxiliary (e.g., l-av te xav 'I am going to eat'). A grammaticalization process from a verb 'take' to future tense is crosslinguistically quite unusual; but Ukrainian has two future tenses, and one of them also uses the verb 'take' as a future auxiliary. Boretzky (1989: 369) suggests that Vlach Romani speakers acquired their 'take'-future when they crossed Ukrainian territory (Heine & Kuteva 2006, chapter 2). That two neighboring languages, these Romani dialects on the one hand and Ukrainian on the other, would have undergone the same unusual process independent of one another is not very likely; hence, a hypothesis in terms of contact-induced transfer is plausible. The grammaticalization from possession (e.g., He has no car) to possessive perfect (He has gone), while widespread in Europe, is fairly rare crosslinguistically. Thus, if two neighboring languages have both undergone this process, then the chances that contact played a role are higher (assuming that shared genetic relationship can be ruled out). Most Slavic languages