AAA - Arbeiten aus Anglistik und Amerikanistik Band 35 (2010) Heft 2 Gunter Narr Verlag Tübingen “Ich spreche Englisch, ich spreche Deutsch, aber was ist das für eine Sprache? ” Emigranto - a grammatical approach to code-switching Eva Duran Eppler This paper presents the first grammar of Emigranto, the German/ English mixed code spoken by Jewish refugees residing in London, UK. It goes back to Gardner-Chloros & Edwards’ (2004) paper by showing that the German/ English bilinguals possess two identifiable linguistic systems or languages, each with its grammatical rules and lexicon, and that the mixed variety results from the interaction between lexical elements and grammatical rules from these languages. It supports Gardner-Chloros and Edwards’ argument that grammar is distinct from the processes driving language production, but shows that knowledge about both can be derived from an analysis that combines syntactic relations with processing features. The results of this analysis are captured in the Distance Hypothesis, which states that greater dependency distance increases the chances of code-mixing (Duran Eppler 2010). The paper concludes that there is more variation in bilingual than in monolingual data, but it argues that (a.) this does not preclude grammatical approaches to code-switching - they just have to be probabilistic - and (b.) variation has to be tackled if we want to identify grammatical regularities in natural speech. 1. Introduction The title quote is taken from a recording of a seventy-four-year-old Austrian Jewish refugee who has been living in London, UK, since 1939. In this extract DOR describes how she and her friends speak. *DOR: and # natuerlich wir sprechen alle Deutsch . *DOR: wir sprechen ein bisschen Englisch ein bisschen Deutsch . *DOR: wir waren [/ / ] in Spanien waren wir vor zwei jahren . *DOR: und wir haben gesprochen so a [: ein] bissl Englisch # so gemischt" nicht ? Eva Duran Eppler 166 *DOR: and ist neben uns ein Wiener gesessen . *DOR: sagt er “ich spreche Englisch # ich spreche Deutsch # aber was ist das fuer eine sprache? ” IBron line 134-160 The quote raises two main issues that will be dealt with in this paper. One is whether Emigranto is a ‘Sprache’, i.e. a language with a grammar. The second one is whether the bilingual or code-switched speech results from the interaction between lexical elements and grammatical rules from the contributing languages; in other words, whether the syntax of Emigranto is based on the two monolingual grammars. These are the same questions Gardner-Chloros and Edwards (2004) deal with when they attend to grammatical approaches to code-switching. Code-switching is the use of lexical items and grammatical features from two languages in one utterance or sentence. Gardner-Chloros and Edwards (2004: 103) identify the following assumptions as underlying grammars of code-switching: 1. informal speech can be adequately and appropriately described in terms of grammar 2. deep, rather than surface, structures are involved in code-switching 3. one language is the base or “matrix” 4. constraints derived from existing data are universal and predictive They question these suppositions on several grounds: i. grammar is distinct from the processes driving speech production ii. the role of grammar is mediated by the variable, poly-idiolectal repertories of bilingual speakers iii. the notion of base system is either irrelevant or fails to explain the facts iv. no principles proposed to date account for all the facts, i.e. are absolute Not all researchers working on structural aspects of bilingual speech and code-switching share assumptions 1.-4. In the introduction I will provide background information on the issues raised in these suppositions and outline the position I take on them. Regarding the first assumption, that informal speech cannot be adequately and appropriately described in terms of grammar, Gardner-Chloros and Edwards (2004) quote Alvarez-Caccamo’s (1998: 36) ‘challenge’ that “[i]n order to argue convincingly for or against the existence of code-switching constraints and code-switching grammars based on the two monolingual ones […] research should first convincingly prove that a) speakers who code-switch possess two identifiable linguistic systems or languages, each with its identifiable grammatical rules and lexicon; and b) code-switched speech results from the predictable interaction between lexical elements and “Ich spreche Englisch, ich spreche Deutsch …” 167 grammatical rules from these languages.” The present study, like several others before it (Backus 1996, Boumans 1998, Sankoff & Poplack 1982, Treffers-Daller 1994) will show that (a.) and (b.) are valid assumptions. As regards (a.) the present study adopts the position that approaches to codeswitching which posit specific mechanisms and principles which cannot be independently motivated in the analysis of monolingual speech are conceptually flawed (cf. McSwan 2009). As regards (b.) the present study will show that code-switched speech does indeed result from the interaction between lexical elements and grammatical rules from the contributing languages. Syntactic approaches to code-switching conducted in a Government & Binding framework had to assume that deep structures are involved in codeswitching (cf. assumption (2.); see McSwann 2009 for a survey). Romaine, on the other hand, already argued in 1986 that CS is a surface phenomenon. The present study uses a grammatical framework that allows a complete surface analysis, viz. Word Grammar (Hudson 2007, 2010). Studies based on data where language contact has led to insertional patterns of code-switching (Muysken 2000, Deuchar, Muysken & Wang 2007) tend to subscribe to assumption three, i.e. that one language is the base or “matrix” language of the mixed code (e.g. Joshi 1985, Myers- Scotton 2009). In other communities, like the Puerto Rican Spanish/ English bilinguals in New York (e.g. Poplack 1982) or the German/ English bilinguals in London (Eppler 1999), language contact results in more alternating patterns of code-switching. Studies based on such data tend to assume a less asymmetric relation between the languages involved in mixing. The period during which some syntactic constraints on code-switching were assumed to be universal and absolute (cf. assumption (4.)) lasted for about ten years, from the mid 1980s to the mid 1990s, but was clearly over when Gardner-Chloros and Edwards wrote their paper. Too many counterexamples from too many language pairs had been published (e.g. Mahootian 1993) for this claim to be defendable. This, however, does not mean that code-switching is parochial and code-switching patterns are random. Different grammatical structures are switched with different frequencies, and these can be used to make predictions about the bilingual speech produced by the speakers the switch probabilities were obtained from. This is, in a very narrow context, restrictions derived from existing data are predictive (cf. assumption (4.)). Gardner-Chloros and Edwards (2004) question these assumptions on several grounds. Their first argument is that grammar is distinct from the processes driving speech production. This is undoubtedly the case, but grammatical speech production and successful syntactic parsing require speakers to have and apply knowledge of grammar in language use. Their second contention, that the role of grammar is mediated by the variable, poly-idiolectal repertories of bilingual speakers, is also undoubtedly Eva Duran Eppler 168 1 Lüschner (2008) and Davies (2010) could assign a matrix language (Myers-Scotton 1993) to 80-94% of utterances from some files from the Emigranto corpus. true, but all speech (mono-, biand multilingual) is individual and variable, and as bilingual speech tends not to be codified, it may vary more than monolingual speech. Furthermore, every monolingual speaker’s repertoire is - to a certain extent - idiolectal; so every bilingual speaker’s repertoire will necessarily be bior poly-idiolectal. Many researchers agree with Gardner-Chloros and Edwards’ (2004) third and fourth arguments: Sankoff & Vanniarajan (1990) and McSwann (2009), for example, concur with the view that in many instances of code-switching the notion of base system is either irrelevant, or fails to explain the facts. The present study will show that this is also true for many of the mixed utterances in the German-English corpus. The informants’ L1, German, however, seems to function as a kind of base language, but in a very different sense to the notion of “matrix” language suggested by, for example, Joshi (1985) and Myers-Scotton (1993, 1995, 2002). 1 Gardner-Chloros & Edwards’ last contention with grammatical approaches to code-switching, that no principles proposed to date account for all the facts of syntactic code-switching, has been accepted by pretty much everybody working in grammatical code-switching research (e.g. Clyne 1987, Poplack & Sankoff 1988, Mahootian 1993, McSwan 2009). For principles to account for all the facts of a certain phenomenon is rare in linguistics and other disciplines; this tends to encourage rather than discourage further research. Gardner-Chloros and Edwards (2004) conclude from this that rather than seeking universal, predictive grammatical rules, research on CS should focus on the variability of bilingual grammars. The following paper will do exactly this. Within a variationist framework (Sankoff & Labov 1979) it addresses the following questions: Do these bilinguals possess two identifiable linguistic systems, each with its identifiable grammatical rules and lexicon? Which language, if any, functions as a base for the mixed variety? Can grammatical approaches to code-switching inform our understanding of bilingual syntactic parsing? Which grammatical relations are frequently switched and which ones are rarely switched? Does bilingual speech elude grammatical description because it is variable (i.e. is the blueprint a red herring)? Not all of Gardner-Chloros and Edwards reservations against grammatical approaches to code-switching can receive equal coverage in this paper; the main focus will be on (ii.) and (iv.). With regard to (ii.) it will be demonstrated that bilingual speech can be described grammatically; with regard to (iv.) it will be argued that grammatical restrictions on code-switching are “Ich spreche Englisch, ich spreche Deutsch …” 169 2 The long account is Duran Eppler (2010). 3 A notational summary is presented in Appendix 1. 4 The word on which a word depends syntactically is its head. probabilistic (and not absolute). The Distance Hypothesis, furthermore, sheds light on syntactic processes driving bilingual speech production. Section two of this paper provides information about data collection and the corpus. The main section presents a short 2 variationist account of the grammar of a German-English mixed code. The study is synchronic, i.e. it presents the lexical and syntactic composition of the mixed code at a specific point in time (1993). The analysis is carried out in a lexically based dependency grammar (Hudson 2007). 3 I will focus on monolingual dependency relations (where both word A and word B are from the same language) and mixed dependency relations (where words A and B are from different languages) and will compare and analyse the variability in switch frequencies and dependency distances (the number of words between a word and its head 4 ) of these syntactic relations. Standard grammatical terminology is used throughout the paper. In the final section I will compare the problems Gardner-Chloros and Edwards (2004) have with grammatical approaches to code-switching with the results of this (and other) grammatical studies of code-switching. This will lead to the conclusion that grammatical approaches to code-switching are not futile but can reveal new insights into monoand bilingual speech, i.e. the blueprint is not a red herring. 2. The Emigranto data The empirical study I am going to present is based on a corpus of German- English monolingual and code-mixed discourse. The data is drawn from a community of Austrian Jewish refugees from the National Socialist regime who settled in London in the late 1930s. The L1 of the informants is Austrian German. Yiddish was not part of the linguistic profile of the so-called assimilated Jewish community in Vienna of the time and none of the speakers included in this study speaks Yiddish. The age of onset of the L2, British English, was during adolescence (15-21 years) for all speakers included in this study. At the time the audiorecordings were made (1993) all informants were in their late sixties or early seventies. A bilingual mode of interaction, called Emigranto developed among a close-knit network of community members. Linguistically the mixed code is characterised by frequent switching at speaker turn boundaries and heavy intra-sentential code-switching. Eva Duran Eppler 170 5 The LIDES Coding Manual (2000) and http: / / www.ling.lancs.ac.uk/ staff/ mark/ lipps/ lipps.html. 6 I will use a conservative count of code-switches (1,274) as the basis for all further calculations. 7 The total is higher than the sum of word tokens from both languages, and some of the percentages do not add up to 100% because the total includes words which could not be assigned to either German or English on an unequivocal basis. The basis for this study is a corpus (48,467 words) of ten and a half hours of monolingual and code-mixed speech transcribed in the LIDES format. 5 It consists predominantly of group recordings of casual speech involving the central informant (DOR), three of her friends from the refugee generation (TRU, MEL and LIL) and the researcher. 3. A grammatical approach to Emigranto 3.1 Quantitative background: distribution of languages and frequency of code-switching This section presents background information to the grammar of the mixed code, i.e. frequency data on the distribution of German and English in the corpus and by individual speakers, the frequency of mixed utterances and the frequency of code-switches. German (28,596 words) is more heavily represented than English (19,446 words) in the corpus this study is based on, and code-switching is asymmetrical, i.e. there are 50% more switches from German to English (813) than from English to German (543). 6 The distribution of word tokens from each language and the fact that there are more code-switches from German to English may be interpreted as a first vague indicator that the German language plays a more basic or prominent role in the data than English. The distribution of languages, however, varies considerably among the individual speakers. German English Word tokens Percent Word tokens Percent 7 Tokens *MEL Overall 3,451 50% 3,407 48% 6,947 c-s 794 50% 739 47% 1,561 *TRU Overall 6,564 56% 4,984 43% 11,639 c-s 881 60% 568 39% 1,469 “Ich spreche Englisch, ich spreche Deutsch …” 171 German English Word tokens Percent Word tokens Percent 7 Tokens *LIL Overall 3,069 61% 1,845 37% 4,999 c-s 476 65% 244 33% 733 *DOR Overall 11,331 80% 2,615 19% 14,049 c-s 1,928 72% 719 27% 2,691 Table 1: Distribution of languages in the overall corpus and in code-switched utterances by speaker *MEL is quite balanced in the number of word tokens she selects from both her languages. *TRU, *LIL and *DOR, on the other hand, are increasingly more German dominant. Interestingly, this correlates exactly with the age at which they arrived in the United Kingdom. Only *DOR uses significantly less German in her mixed utterances than in her overall speech. Relating these findings to Gardner-Chloros & Edwards’ (2004) paper, these frequency data show that the repertoires of the bilingual speakers involved in this study are indeed poly-ideolectal and that socioand psycholinguistics factors, like age of onset of the L2, do play a role, but neither of these findings necessarily mean that the role of grammar is mediated or overridden by them. Following this overview of the overall proportions of German and English in the corpus and among individual speakers, I will now focus on codeswitching. The corpus contains 52% monolingual German, 37% monolingual English and 11% mixed utterances. Proportional to their total number of utterances, the individual informants produce roughly the same number of mixed utterances (10%). In mixed utterances the switch frequency is 5.0, this is, in intra-sententially code-switched utterances every fifth words is in a different language to the preceding one. The switch frequency in mixed utterances is not significantly different between the individual informants (× 2 = 1.74; df = 4; p > 0.5). This is important for the analysis as the speakers represented in the data thus produce • roughly the same number of mixed utterances, and • the intra-sentential mixing of my informants is not significantly different. The data chosen as a basis for analysis are therefore similar enough to warrant a comparison; and, within these texts, it is the intra-sententially mixed utterances, i.e. the main focus of this study, that are most similar. These results relate to Gardner-Chloros & Edwards’ paper in two ways: first, they show that when my informants switch within an utterance, they switch several times and thus produce patterns that are unlike insertion and Eva Duran Eppler 172 8 In insertion, lexical items or entire constituents from one language are embedded into a structure from the other language. In the alternational mixing pattern, bilinguals oscillate between structures from at least two languages. When material from different lexical inventories is used in grammatical structures that are shared between the languages involved in intra-sentential switching, Muysken (2000) talks of congruent lexicalization. 9 A comparison with monolingual Viennese German from the early 20 th century and contemporary London English is unfortunately not feasible because of a lack of comparable corpora. The monolingual sample is large enough for quantitative syntactic analysis because grammatical feature counts tend to be stable across samples half the size of what is used here (Biber 1998). more akin to alternation and congruent lexicalization. 8 In these types of code-switching the notion of base system makes little sense as neither of the two languages establishes a matrix language frame (Myers-Scotton 1993) into which elements of the other language are inserted. Second, the switch frequencies can be combined with the relative switch frequencies of individual syntactic relations to form a probabilistic grammar of code-switching (akin to Sankoff & Poplack 1982). The next section presents the analysis of grammatical functions in the German/ English corpus. The mixed syntactic functions are systematically compared to a random sample of monolingual syntactic functions from the corpus (2,025 words) to show how often a particular phenomenon occurs relative to other possible realisations (cf. the principle of accountability, Labov 1972, Poplack 1990). 9 3.2 Quantitative analysis of grammatical functions A grammatical function is a specific type of syntactic (dependency) relation, e.g. an object or an adverbial. Each dependency relation is controlled by at least one syntactic rule that allows that particular pair of words to occur as dependent and head (Hudson 1990). In monolingual dependencies both word A and word B are from the same language. If words A and B are from different languages, we are looking at a mixed dependency. The null hypothesis for this study requires each word in a dependency to satisfy the constraints imposed on it by its own language. It therefore constrains codemixed and monolingual dependencies alike. Before I proceed with the grammatical analysis I need to state how the relation between (conversational) utterances and (grammatical) sentences is handled in this study. This is important in relation to Gardner-Chloros & Edwards’ paper, because they question whether informal speech can be adequately and appropriately described in terms of grammar. A sentence can be defined as any string of words held together by syntactic relations. I did not edit the transcribed speech before the grammatical analysis (unlike many of the quoted examples). I left any material that could not be taken as “Ich spreche Englisch, ich spreche Deutsch …” 173 10 The null hypothesis this study is based on does not a priori ascribe a more prominent role to heads. a token of a word-form of either German or English in the texts, but if it could not be linked to other elements in the utterance via a dependency relation, it was not included in the syntactic analysis. Thus all the words in a transcribed utterance that are related to other words by syntactic relationships constitute the sentences on which the following grammatical analysis is based. I will first outline the more general picture, frequency distributions and mean distances of monolingual and mixed dependencies, before focusing on individual grammatical functions. The distribution of monolingual German and English dependency relations is 5 : 4. In the mixed dependencies the number of heads per language is even more skewed towards German (76% German: 24% English) than the number of words per language (63% German: 36% English). German thus is not only numerically dominant in the speech of my informants, but also contributes a significantly higher proportion of heads to code-mixed utterances than English. If we assume that heads play a more prominent role in syntactic relations than dependents, then the lexical and syntactic properties of the numerically dominant German heads should have more influence over code-mixing patterns in Emigranto than English heads. 10 This finding may again indicate that German functions as a default or base language in mixed utterances. The dependency distance of a word is the number of words between it and its head. Table 2 presents the mean dependency distances and standard deviation in monolingual and mixed dependencies (listed by the language of the head). German English Total Monolingual 0.87 (ó = 0.78) 0.49 (ó = 0.41) 0.68 Mixed with head… 0.85 (ó = 0.81) 1.26 (ó = 1.08) 1.06 Table 2: Mean distances (and standard deviation) in monolingual and mixed dependencies These results generate hypotheses which will be tested in the next section. Table 2 shows that: 1. Monolingual German dependencies are longer than English ones. This is a direct result of the word order properties of the two languages; e.g. V2 vs. SVO, Verbalklammer. 2. Mixed dependencies with a German head are shorter than mixed ones with an English head. This is surprising if we assume (a.) that code- Eva Duran Eppler 174 mixing is based on the two monolingual grammars (cf. (1.)) and (b.) that heads play a more prominent role in syntactic relations than dependents (see above). 3. Monolingual German and mixed dependencies with a German head are of a similar length. This similarity may suggest that mixed dependencies with a German head are not very different from monolingual German ones, at least as far as distance is concerned. 4. Monolingual English dependencies are considerably shorter than mixed dependencies with an English head. In contrast to German heads, English heads thus seem to enter into ‘looser’, literally more remote syntactic relations with German dependents. We would then expect English words to enter more dependency relations that are characterised by long distances, e.g. adjunct, extractee and extraposee relations, and German dependents of English heads to be more frequently located at the clause periphery (cf. Treffers-Daller 1994). 5. The difference in distances between monolingual and mixed dependencies is highly significant (× 2 = 18.6, df = 1, p < 0.001); and the standard deviation from the mean is higher for mixed dependencies; i.e. there is more variation in the distances of mixed dependencies and there are more mixed outliers. 6. The mean distance of mixed dependencies (1.06) is longer than that of both English and German monolingual dependencies. If we assume that the influence of a word’s language on that of its dependent will decrease with increased distance (speakers ‘forget’ the language of the head), mixed dependencies may be the result of distance. By their very nature, long distance dependents are more likely to be located at the clause periphery (cf. Treffers-Daller 1994, Muysken 2000). The long dependency distances of mixed syntactic relations may point towards a processing motivation behind code-switching: the influence of a word’s language on that of its dependent may decrease with increased distance. This would then mean that the longer the dependency distance, the more likely we are to encounter an other-language dependent, i.e. a code-switch. Gardner-Chloros and Edwards (2004) stress that grammar is distinct from the processes driving speech production, but speech production and comprehension does not and cannot happen without grammar (syntactic parsing). The Distance Hypothesis combines aspects of grammar (syntactic relations) with a psycholinguistic processing feature (dependency distance). “Ich spreche Englisch, ich spreche Deutsch …” 175 11 Sharer, the WG term for xcomp or incomplement, is the type of verb complement which shares its subject with the head verb. Sharers are found with auxiliary, copula raising and control verbs. 12 Prepositionals and particles are treated as separate syntactic functions, but their numbers are so low that I conflated the figures in Tables 3-8. 13 < and > are used in WG notation for the direction of the relation, >x = ’following’, x< = ’preceding’. Distance Hypothesis Greater dependency distance of syntactic relations increases the chances of code-mixing. (Duran Eppler 2010) Because the Distance Hypothesis combines syntax with parsing, evidence in its support potentially sheds light on both grammatical and psycholinguistics aspects of code-switching. The next section presents the analysis of individual monolingual and mixed grammatical functions. An analysis of monolingual and mixed dependency relations will (a.) establish whether the “speakers who code-switch possess two (or more) identifiable linguistic systems or languages, each with its identifiable grammatical rules and lexicon” (Alvarez-Caccamo 1998: 36, quoted in Gardner-Chloros 2004: 104); (b.) highlight the similarities and differences of the two languages involved in bilingual production in this particular incident; and (c.) indicate where, i.e. in which dependency relations, the syntax of German and the syntax of English allow mixing. Because dependency distance is a property of all dependency relations, this analysis will also establish whether the German/ English data support the Distance Hypothesis. 3.3 Monolingual syntactic relations A comparison of the sample of monolingual German and English grammatical functions, disregarding the position of the dependent, shows that there is no significant difference for most functions (c = complement, s = subject, a = adjunct, r = sharer 11 , o = object, >x = extraposee, x< = extractee, n = negative, p = prepositional/ particle 12 ). This is a good indicator that German and English are typologically similar languages. Differentiated by position 13 , a comparison between the monolingual German and English dependency relations highlights the main word order differences between the two languages. Eva Duran Eppler 176 14 This comparison does not include extractees because they are neither adjuncts nor valents. >c s< >s >a a< >r r< >o o< >x x< >n n< >p p< Total German 21% 19% 6% 13% 11% 9% 2% 7% 5% - 3% 2% 1% - 1% 754 English 22% 22% 1% 12% 7% 16% 0% 14% 0% - 1% 4% 0% 1% 0% 596 p 0.575 0.588 0.518 0.013 0.000 0.000 0.000 0.000 0.405 0.004 0.009 0.000 0.497 Table 3: Comparison of monolingual dependencies differentiated by position: percentages, p-values (significant differences are in bold), Four columns have a 0 entry for English, which means there are no predependent sharer, object, negative, particle and prepositional relations in the monolingual English of my informants. These syntactic functions are postdependency relations of verbs in English, but pre-dependents in German when their heads are clause final verbs. This result therefore shows that, when in monolingual English mode, my informants adhere to English syntactic rules. In other words, they “possess two identifiable linguistics systems [...] , each with its identifiable grammatical rules” (Gardner-Chloros 2004: 104). From this comparison and the null hypothesis we would expect no, or hardly any, code-switches across the dependency relations that are not shared by the two grammars. In the next section I will contrast the monolingual dependencies from each language with the correlating mixed ones. This will show whether the mixed variety results from the interaction between lexical elements and grammatical rules from the two languages involved in code-switching in Emigranto. The analysis of the mixed syntactic relations will also allow us to evaluate the universality of various constraints on where code-switching can occur in a sentence. 3.4 Comparison of monolingual and mixed syntactic relations The code-switching literature (Mahootian & Santorini 1996, Treffers-Daller 1994) proposes that switching of head-complement relations is more constrained than the switching of adjunction structures. In the corpus the present study is based on the difference between monolingual and switched complement and adjunct relations 14 is highly significant (× 2 = 6.82, df = 1, p = 0.009), but there are more switched complements than adjuncts. Overall, Mahootian & Santorini’s (1996) and Treffers-Daller’s (1994) hypotheses that adjuncts are more easily switched than complements are thus not substantiated by the present analysis, but a more fine-grained analysis of grammatical relations is required for a clearer picture to emerge. “Ich spreche Englisch, ich spreche Deutsch …” 177 15 In WG determiners are the head in determiner-noun relations. The comparison between monolingual German and mixed syntactic relations with a German head is presented in Table 4. It shows that codeswitched speech results from the interaction between lexical elements (German heads and English dependents) and grammatical rules (functions) from the bilinguals’ two linguistic systems. The code-switching is predictable in that it only occurs between lexical elements from the two languages that enter recognisable syntactic relations which are based on grammatical rules. >c s< >s >a a< >r r< >o o< >x x< >n n< >p p< Total Mono G 21% 19% 6% 13% 11% 9% 2% 7% 5% 0% 3% 2% 1% 0% 1% 754 Mix h G 59% 2% 0% 7% 5% 12% 1% 6% 4% 0% 3% 0% 1% 0% 0% 525 P 0.000 0.000 0.000 0.002 0.000 0.195 0.383 0.272 0.427 0.447 0.874 - - - - Table 4: Comparison of monolingual German and mixed dependencies with a German head: percentages, p-values (significant differences are in bold) Of the most common syntactic relations (complements, subjects, adjuncts, sharers and objects) in this category, three show a significant difference between their monolingual and their mixed frequencies in the data. There are significantly more mixed complements with a German head than monolingual German ones. The most frequent complement relation is the one between nouns and their heads, 15 and nouns are the most frequently switched (or borrowed) word class in all borrowability hierarchies listed in the code-switching literature (van Hout & Muysken 1994 for a summary), an observation confirmed by the data. If I distinguish between code-switches and borrowings and classify all English nouns that enter syntactic relations with German heads as borrowings, the complement versus adjunct distinction simply becomes non-significant (× 2 = 0.416, df = 1, p = 0.838). So even if we take all potential borrowings out of the equation, Mahootian & Santorini’s (1996) and Treffers-Daller’s (1994) hypotheses that adjuncts are more easily switched than complements are not substantiated by the present analysis. It therefore supports Garnder-Chloros and Edwards’ (2004: 110) argument that hypothesis formulated on the basis of a particular data set may not apply to all code-switching situations. Subjects also show a significant difference in tokens between their monolingual German occurrences and the mixed ones with a German head - not because there are so many, but because there are so few (12 tokens). This finding is in accordance with other studies of code-switching which indicate that subjects, and particularly subject pronouns, are infrequently Eva Duran Eppler 178 switched (Gumperz & Hernandez-Chavez 1971, Timm 1975, Treffers-Daller 1994). My mixed corpus contains four English subject pronouns depending on German heads. One example is (1) *LIL: you kannst # jauchzen . (English/ German) When arguing against the tradition of universal and predictive constraints on code-switching, Garnder-Chloros and Edwards (2004) invoke Muysken’s (2000) argument that bilingual speech may contain too much variation for a single set of rules to account for code-mixing. Muysken, however, maintains that it is legitimate to describe code-mixing in terms of the grammatical regularities which characterise it. One of these regularities is that subjects and subject pronouns are significantly less likely to be switched than other syntactic categories. Absolute constraints on switching subjects and subject pronouns are therefore not supported by the data, but probabilistic ones are. Table 3 furthermore shows that adjuncts form the second most frequently mixed syntactic relation after complements (the borrowed English nouns). Speaker *MEL, for example, switches adjuncts frequently. (2) *MEL: nein # ich bin draussen ## as per usual. (German/ English) Sharers, objects, extraposees and extractees show no significant difference between their monolingual German occurrences and those with German heads and English dependents. For sharers this result is, on the one hand, surprising, because Timm (1975) proposes that auxiliaries and main verbs are only found in unilingual constructions (although she did find a few counterexamples to this constraint in her own data). My corpus contains only eight switches between German auxiliaries/ modals and English verbs (many morphologically integrated) and thus lends probabilistic support to this constraint. If we focus on a different word-class-pair that can be involved in sharer relations, i.e. verbs and predicative adjectives, this finding is less surprising. Pfaff (1979), Poplack (1980), Treffers-Daller (1994) note that predicative adjectives are frequently switched in their data. This is indeed the sharer relation that contributes most mixed tokens to this category (i.e. sharers with a German head and an English dependent) in my corpus, namely 32. (3) *MEL: die [gay men] sind meistens nice-looking . (German/ English) The switching of sharer relations is therefore not restricted by an absolute constraint. The present and the referenced studies, however, show that there are grammatical regularities that characterise code-switching across different language pairs, and we can therefore predict that the sharer relation between auxiliaries predicative adjectives is more frequently switched than the sharer relation between auxiliaries and main verbs (whether this prediction holds is an empirical issue). “Ich spreche Englisch, ich spreche Deutsch …” 179 That objects are more frequently switched than subjects is a fairly wellestablished fact in the grammatical literature on code-switching. That there is no significant difference between monolingual German and mixed ones with a German head (p = 0.272 for >o, and p = 0.427 for <o) is interesting because objects of German clause final verbs tend to have fairly long dependency distances. This should increase chances of code-switching according to the Distance Hypothesis. The result of objects therefore corroborates the Distance Hypothesis. Monolingual and mixed extrapositions and extractions (see Example 1) show no significant difference in this category, i.e. monolingual German vs. mixed with a German head, but given the small sample size, there are quite a few (17). These two syntactic relations involve dislocated constituents, i.e. (groups of) words that have been shifted out of their ‘normal’ position towards the right (extraposition) or the left periphery (extraction) of a clause. Treffers-Daller (1994: 207) reports 21 switched dislocated constituents in her Dutch-French data; my German-English corpus contains 14 switched extractees and 3 switched extraposees with a German head. An example illustrating an extracted English objects is (4) *DOR: +” one club +” hab(e) ich gemeint sagt man wenn man nicht +/ . The relatively high proportion of mixed dislocated constituents in my data thus supports Treffers-Daller’s (1994) proposal that switching is favoured for dislocated constituents at the clause periphery. Again, this is not an absolute rule, but seems to constitute one of the four ‘primitives’ of code-switching (Muysken 2000). Extractees and extraposees also have particularly long dependency distances and the relatively large number of mixed ones in the data therefore support the Distance Hypothesis. I will return to this point in the discussion of Table 5. The results from this section show that the code-switched speech results from the interaction between lexical elements and grammatical functions from German and English. The analysis furthermore revealed no significant difference between the number of German and English dependents for most syntactic relations German heads enter. This finding supports the hypothesis that the dependency relations German heads enter with English words are not very different to the ones they enter with same-language words. If we furthermore assume that heads determine the syntactic properties of the constructions they head, this result may also point towards German as a base language of the mixed code. Table 2 demonstrates that the mean distance of mixed dependency relations with a German head is similar to the one of monolingual German dependencies (0.85 versus 0.87). Table 5, however, reveals that most mixed grammatical relations (subjects, adjuncts, pre-dependent sharers and postdependent objects) are actually longer than their monolingual German Eva Duran Eppler 180 equivalents. The slightly shorter mean distance of mixed dependencies is mainly attributable to the large number of English complements with a German head, i.e. the English noun borrowings. >c s< >s >a a< >r r< >o o< >x x< >n n< mean German 0.65 0.54 0.07 1.1 0.37 1.64 0.07 0.78 0.83 2.16 0.33 0.87 Mix h G 0.1 0.7 0.5 2.9 0.52 0.95 0.29 1.38 0.5 0.33 2.07 0.33 0.85 Table 5: Comparison of the mean distances of monolingual German and mixed dependencies with a German head The fact that English (post-dependent) adjuncts are almost three times as far away from their German head as monolingual ones supports Treffers- Daller’s (1994), Mahootian & Santorini’s (1996) and Muyskens’ (2000) assumption that code-mixing is favoured in adjoined peripheral positions and the Distance Hypothesis. The next paragraphs compare monolingual English with mixed syntactic relations with an English head. Table 6 and the subsequent analysis reveal that the bilingual speech produced by the Austrian Jewish refugees living in London is a product of the interaction between words and grammatical rules from both their languages, i.e. English and German. Like the comparison between monolingual German and mixed syntactic relations with a German head, this section demonstrates that code-switching is not random, but that constraints derived from existing data are probabilistic rather than universal. c s >a a< r o >x x< n p Total E 22% 23% 12% 7% 16% 14% 0% 0% 4% 1% 596 h E 27% 7% 11% 22% 4% 11% 4% 8% 2% 0% 165 p 0.249 0.000 713 0.000 0.001 0.397 0.000 0.000 0.000 - Table 6: Comparison of monolingual English and mixed dependencies with an English head: percentages, p-values (significant differences are in bold) As for German, we get a highly significant result for subjects, with disproportionately fewer mixed ones (eight, including three subject pronouns). (5) *DOR: die don’t mind ## aber I do . (German/ English) Subjects are thus rarely switched in both directions. Absolute constraints on switching subjects or subject pronouns (Gumperz & Hernandez-Chavez 1971, Timm 1975) are therefore not supported by my data, but subjects are significantly less likely to be switched in both directions than many other syntactic relations. “Ich spreche Englisch, ich spreche Deutsch …” 181 16 The examples are clause objects which only count as extraposees because of an intervening post-modifying adverb. Table 6 furthermore demonstrates that my informants like switching for German adjuncts that precede their English heads. The difference is highly significant and, proportionally, there are more German pre-adjuncts of English words than English ones. (6) *LIL: also die dings ... hat es # in high heaven gelobt . (German/ English) The results for sharers are also highly significant, but for the opposite reason. Hardly any English verb shares its subjects with a German word. Hawkins (1986) assumes that English is less resistant to sharer structures than German because of generally increased semantic diversity of basic grammatical relations in English. If he is right, the low sharer numbers with English heads and German dependents in my data may indicate that the dependent rather than the head inhibits switching (see footnote 10). In this category, i.e. monolingual English vs. mixed with an English head, there are significantly more mixed extraposees and extractees than monolingual ones. The extraposee examples are not clear-cut, 16 but the extractee ones are. Example (7) illustrates a German long-distance (distance = 7) extraction. (7) *MEL: was die Dorit wieder geschmissen hat , I [/ ] I would have liked. In its default position the German object would have a distance of zero; because it has been dislocated to the clause periphery at the front, the dependency distance between like and was increases to seven. The significantly larger number of mixed extractions supports the Distance Hypothesis and Treffers-Daller’s (1995) findings and establishes code-switching in peripheral positions as a potential grammatical ‘primitive’ of code-switching (Muysken 2000). There is no significant difference between the monolingual English complements and the German complements of English heads in the data. This indicates that there are not many ‘borrowed’ German nouns in my corpus. As in the German case, objects are relatively easily switched and the difference is not significant (p = 0.397). The comparison between monolingual English dependencies and mixed ones with an English head shows that the words that enter mixed dependencies are either English or German, and that the type of syntactic relations they enter are also established by the grammatical rules of English and German. This is, the mixed variety results from the interaction between lexical elements and grammatical rules from the speakers’ two linguistic systems. The results furthermore confirm the hypotheses formulated on the basis of the data presented at the beginning of this section, i.e. that English Eva Duran Eppler 182 heads preferably enter into rather ‘loose’ syntactic relationships with their German dependents. This result lends even stronger support to the findings of Treffers-Daller (1994), Mahootian and Santorini (1996) and Muysken (2000) that code-switching is favoured in peripheral and/ or adjoined positions than the data presented in the previous section on German. English heads encourage switching to pre-adjuncts and extractees (and extraposed objects), but discourage switching to subjects (as with German heads) or sharers (unlike German heads). The comparison of the mean distances (Table 7) moreover strongly supports the assumption that the influence of a word’s language on that of its dependent seems to decrease with increased distance. >c s< >a a< >r >o >x x< >n mean English 0.22 0.07 1.26 0.38 0.53 0.5 - 0 0 0.49 Mix h E 0.84 0.9 1.33 0.78 2.12 0.18 0.45 3.5 - 1.26 Table 7: Comparison of the mean distances of monolingual English and mixed dependencies with an English head Table 7 shows that all mixed dependencies with an English head (apart from objects) are longer than their monolingual English counterparts. This result clearly supports the Distance Hypothesis. 3.5 Comparison of mixed syntactic relations I have already discussed all mixed dependency relations in comparison with their monolingual equivalents and will therefore only briefly comment on the comparison of mixed dependency relations to complete the grammar of the mixed code. c s >a a< r O >x x< N Total h G 59% 2% 7% 5% 13% 9% 0% 3% 1% 525 h E 27% 7% 11% 22% 4% 11% 4% 8% 2% 165 p 0.000 0.006 132 0.000 0.002 0.551 0.001 0.001 0.000 Table 8: Comparison of mixed dependency relations with a German and an English head: percentages and p-values (significant differences are in bold) Table 8 confirms that even in intra-sententially mixed utterances my informants hardly ever violate the rules of the two monolingual grammars they mix. There are no pre-dependent sharers, and only one German object and two German negatives in, for English, ungrammatical word order positions. For the aspects of the bilingual speakers’ grammar investigated in this study, “Ich spreche Englisch, ich spreche Deutsch …” 183 17 Notable exceptions to this generalisation are ‘borrowed’ English noun complements and predicative adjectives. i.e. grammatical functions, the research conducted convincingly proves that (a.) the speakers who code-switch possess two identifiable linguistic systems or languages, each with its identifiable grammatical rules and lexicon (Table 3); and (b.) code-switched speech results from the interaction between lexical elements and grammatical rules form these languages (Tables 4-8) (cf. Gardner-Cholrs & Edwards 2004). The scope of the present study does not allow for an in-depth analysis of all aspects of the bilingual speakers’ linguistics systems. Grammatical functions, however, are a fundamental aspect of every linguistic system, and the results obtained from the analysis of the German English bilinguals speech shows that it does not elude grammatical description (Gardner-Chloros 2004: 125). Table 8 furthermore illustrates that mixed dependency relations deviate much more from the expected distribution than their monolingual counterparts (more differences are significant). This supports Gardner-Chloros & Edwards’ (2004) assumption that there is more variation in mixed than in monolingual utterances. To summarise the results from the more detailed analysis of individual syntactic relations presented in Sections 3.3-3.5, they confirm what was hypothesised on the basis of the less detailed analysis at the beginning of that section. The finding that German words more frequently act as heads of mixed dependency relations than English ones points towards German as a base language, especially if we assume that heads determine the grammatical character of the words they establish syntactic relations with. German heads in mixed dependencies furthermore act much more as they do in monolingual ones, but ‘loose’ long-distance relations are also favoured for switching, that is, pre-adjuncts and extractees. 17 In this respect the complement versus adjunct distinction proposed in one form or another by Mahootian and Santorini (1996), Treffers-Daller (1994), and Muysken (2000) appears to have some validity. When English words do function as heads of other language dependents, they predominantly enter syntactic relations that are not essential for building sentence structures, that is, adjunction and extraction (and extraposition). These grammatical relations are also associated with long distances. The fact that mixed syntactic functions with an English head deviate more from the monolingual English ones than their German counterparts furthermore seems to indicate that German grammar contributes more to the mixed code than the L2 grammar. The analysis of the dependency distances of the individual syntactic relations reveals that the trend emerging from the rough analysis, i.e. that mixed dependency distances are longer than monolingual ones, is actually Eva Duran Eppler 184 a far more robust finding. With two exceptions all dependency relations are longer than their monolingual counterparts. This finding therefore strongly supports the hypothesis that greater dependency distance increases the chances of code-switching. 4. Conclusion This paper follows Gardner-Chloros & Edwards’ (2004) suggestion that, rather than seeking universal, predictive grammatical rules, research on code-switching should focus on the variability of bilingual grammars. The quantitative background to the grammatical analysis presented in Section 3.1 has shown that the repertoires of the bilingual speakers involved in this study are indeed poly-ideolectal (Gardner-Chloros & Edwards 2004: 103): the amounts of German and English in their bilingual speech differs significantly from speaker to speaker, as do the overall switch frequencies. Some socioand psycholinguistics factors, e.g. age of onset of the L2, seem to play are role, while others, e.g. number of years of exposure to English and membership in the same close-knit network of speakers, do not. Neither of these findings, however, means that the role of grammar is mediated or overridden by these factors. This becomes apparent from the grammatical analysis presented in the main sections of this paper (Sections 3.3-3.5). They present a quantitative account of the grammar of the German- English mixed code spoken by Jewish refugees from the Holocaust living in London. The syntactic analysis in terms of monoand bilingual grammatical functions demonstrates that, contrary to Gardner-Chloros & Edwards’ (2004) claims, informal bilingual speech can be adequately and appropriately described in terms of grammar, and that the linguistic behaviour of codeswitching speakers does not elude grammatical description, although their mixed utterances are found to be more variable than their monolingual ones (Table 8). Deuchar, Muysken and Wang (2007: 298), working on codeswitching of entirely different language pairs, come to a similar conclusion. They state that “bilingual speech, just as monolingual speech, shows variation, but this variation is patterned or structured, not random.” The comparison of monolingual and mixed syntactic relations has shown that the bilingual informants possess two identifiable linguistic systems, each with its grammatical rules and lexicon (Sections 3.3 and 3.5), and that the mixed variety results from the interaction between lexical elements and grammatical rules from these two languages (Section 3.4). Only three codeswitches in the overall corpus violate syntactic rules of one of the two languages involved in this contact situation. Absolute constraints on where in a sentence code-switching can occur are not supported by my data (cf. Gardner-Chloros & Edwards 2004: 103, 110), but there is strong probabilis- “Ich spreche Englisch, ich spreche Deutsch …” 185 tic support for syntactic restrictions on code-mixing. Subjects, for example, are significantly less likely to be switched in both directions than adjuncts, extractees or extraposees. Gardner-Chloros and Edwards (2004) furthermore question the validity of the notion of a base system as either irrelevant, or failing to explain the facts. The notion of a “matrix” language as suggested by, for example, Joshi (1985) and Myers-Scotton (1993) does not seem to be relevant to the mixed utterances in the German/ English data. The quantitative analysis has, however, revealed that German contributes more word tokens and syntactic heads to the mixed variety than English, that German words behave more uniformly in monolingual and mixed dependencies than English words, and that code-switching is asymmetrical in the corpus, with more switches from German to English than vice versa. These results were interpreted as the informants’ L1, German, possiby functioning as a kind of base language, but in a very different sense to the notion of “matrix” language suggested by, for example, Joshi (1985) and Myers-Scotton (1993). Gardner-Chloros and Edwards (2004: 103) challenge grammatical approaches to code-switching on the ground that grammar is distinct from the processes driving speech production. A feature of the syntactic theory used for the grammatical analysis, i.e. dependency distance (the number of words between a head and a dependent) revealed a syntactic processing factor as a facilitator of code-switching. Variation in the overall dependency distances of monolingual and mixed syntactic relations led to the hypothesis that the influence of a word’s language on that of its dependent seems to decrease with increased distance. The more fine-grained examination of individual syntactic functions in my data supports this proposed explanation, i.e. the longer the distance, the more likely we are to encounter an other-language dependent, i.e. a code-switch. This finding is in line with previous research, which revealed a propensity for switching for dislocated constituents at the clause periphery (Treffers-Daller 1994, Muysken 2000). The Distance Hypothesis (Duran Eppler 2010), which states that greater dependency distance increases the chances of code-mixing, captures this fundamental principle of code-mixing on a more general processing level. So is the blueprint a red herring? In response to this question I would like to quote Newport and Aslin (2000: 3), who state that a corpus provides a rich source of information for identifying grammatical regularities. At the same time, however, access to the distributional properties of a corpus and the calculations one might perform on it only worsens the theoretical problem of limiting hypotheses: corpora provide yet larger infinities of things to calculate. The blueprint, which I take to be bilingual speech corpora, is not a red herring. Bilingual corpora, like all corpora, are messy in that they suggest Eva Duran Eppler 186 many analyses, but they also contain the information for identifying grammatical regularities. Grammatical approaches to code-switching can help us find them (despite variation). References Biber, D. / S. Conrad / R. Reppen (1998). Corpus linguistics. Investigating language structure. Cambridge: CUP. Deuchar, M. / P. Muysken / S. Wang (2007). “Structured Variation in Codeswitching: Towards an Empirically Based Typology of Bilingual Speech Patterns.” International Journal of Bilingual Education and Bilingualism 10/ 3. 298-338. Duran Eppler, E. (2010). Emigranto. The syntax of German / English code-swtiching. Vienna: Braumüller. Eppler, E. (2004). “… because dem Computer brauchst’ es ja nicht zeigen.”: because + German main clause word order.” International Journal of Bilingualism 8. 127-143. Eppler, E. (2010). “German/ English LIDES database.” Talkbank. http: / / talkbank.org (July 2010). Gardner-Chloros, P. / M. Edwards (2004). “Assumptions Behind Grammatical Approaches to Code-Switching: When the Blueprint Is a Red Herring.” Transactions of the Philological Society 102. 10-29. Gumperz, J.J. / E. Hernandez-Chavez (1971). “Cognitive aspects of bilingual communication.” In: W.H. Whiteley (ed.). Language use and social change. Oxford: OUP. 111-125. Hawkins, J.A. (1986). A comparative typology of English and German. London: Croom Helm. Hudson, R.A. (2007). Language networks: the new Word Grammar. Oxford: OUP. Hudson, R.A. (2010). An Introduction to Word Grammar. Cambridge: CUP. Mahootian, S. / B. Santorini (1996). “Code-switching and the complement/ adjunct distinction.” Linguistic Inquiry 27. 464-479. McSwann, J. (2009). “Generative approaches to code-switching.” In: B.E. Bullock / A.J. Toribio (eds.). The Cambridge Handbook of Linguistic Code-Switching. New York: CUP. 309-338. Muysken, P. (2000). Bilingual speech. Cambridge: CUP. Myers-Scotton, C. (1993). Dueling Languages. Grammatical Structure in Codeswitching. Oxford: Clarendon. Newport, E.L. / R.N. Aslin. (2000). “Innately Constrained Learning: Blending Old and New Approaches to Language Acquisition.” In: S.C. Howell / S.A. Fish / T. Keith-Lucas (eds.). Proceedings of the 24 th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press. 1-21. Pfaff, C. (1979). “Constraints on language mixing: intrasential code-switching and borrowing in Spanish/ English.” Language 55. 291-318. Poplack, S. (1980). “Sometime I’ll start a sentence in Spanish y terminó en español: toward a typology of code-switching.” Linguistics 18. 581-618. Romaine, S. (1986). “The syntax and semantics of the code-mixed compound verb in Punjabi-English bilingual discourse.” In: D. Tannen / J. Alatis (eds.). Language and Linguistics. The Interdependence of Theory, Data and Application. Washington, D.C.: Georgetown University Press. “Ich spreche Englisch, ich spreche Deutsch …” 187 Sankoff, D. / W. Labov (1979). “On the use of variable rules”. Language in Society 8. 189-222. Timm, L.A. (1975). “Spanish-English code-switching: el porque y how - not - to”. Romance Philology 28. 473-482. Treffers-Daller, J. (1994). Mixing two languages: French-Dutch contact in a comparative perspective. Berlin: de Gruyter. Van Hout, R. / P. Muysken (1994). “Modeling lexical borrowability.” Language Variation and Change 6. 39-62. Eva Duran Eppler English Language and Linguistics Roehampton University, London