Bilingual Sentence Processing - R.R. Heredia and J. Altarriba (Editors) 9 2002 Elsevier Science B.V. All rights reserved.
15
Code-Switching in Preschool Bilingual Children
Zehra F. Peynircio~lu American University Aydln Y. Durguno~lu University of Minnesota, Duluth
Abstract Code-switching refers to the alternating use of two or more languages, either within a sentence (intrasentential) or between sentences (intersentential). Because code-switching is governed by grammatical rules, both language systems are presumed to be active while producing mixed sentences. Codeswitching occurs in discourse of fluent and nonfluent bilinguals alike, although intrasentential switches are often thought to be illustrative of the level of bilingualism or comfort with another language (e.g., Poplack, 1980). This chapter focuses on code-switching in preschool children, with a special emphasis on the linguistic context, comparisons between intra- and intersentential switching, the effect of level of bilingualism, differences in production and comprehension of sentences, and experimental demands. To this end, we present new data from 36 Spanish-English bilingual children between the ages of 3 and 5, and tie the results to the theoretical framework of Kroll and her colleagues (e.g., Kroll & Stewart, 1994; Sholl, Sankaranarayanan, & Kroll, 1995) on conceptual and lexical links between words in sentence processing.
Overview Code-switching, or alternating between two or more languages in a given sentence or between sentences, is a common but intriguing phenomenon in bilingual or multilingual sentence processing. It is quite prevalent, especially in cultures where the use of multiple languages are commonplace and taken for granted, such as amongst the Hispanic populations in the U.S., the later generations of guest workers in industrialized European countries, and even students in schools where intensive foreign languages are taught. Interestingly, it is almost exclusively limited to spoken language, and in the rare written cases, it is often limited to conversation-like correspondence. Start of serious research on code-switching can be dated to the early 1950s as part of research on bilingualism (e.g., Weinreich, 1953), and the effort has proliferated since the 1960s. One of the problems with integrating so much research, however, has been the many different approaches to studying code-switching. Within the linguistic,
340
Zehra F. Peynircio~lu and Aydm E Durguno~lu
sociological, cognitive, and educational settings, similar questions have been explored from vastly different perspectives. For instance, whereas some social scientists have been interested in the group dynamics or identity that lead to code-switching or codeswitching as expression of position of power (e.g., Pandharipande, 1990; Scotton, 1988), others have focused on detailed analyses of lexical issues in code-switching (e.g., Sridhar & Sridhar, 1980; Yoon, 1993). In addition, not everyone agrees on exactly what code-switching should refer to. In the literature, sometimes detailed distinctions have been drawn between such terminology as code-mixing, codechanging, borrowing, deliberate vs. accidental code-mixing, etc. For instance, borrowing has sometimes been excluded from code-mixing or intrasentential codeswitching because theoretically monolinguals can also intersperse a single target word from another language without having any competence in that language (e.g., Pfaff, 1979). Similarly, situational switching (e.g., Blom & Gumperz, 1972; Lipski, 1978), or using different languages in different settings or when switching topics, has also sometimes been excluded because it does not capture the linguistic constraints highlighted when switching occurs in the same mindset. In this chapter, we define code-switching as any alternation between English and Spanish, including borrowings because we know that all of our participants in the experiment we report had at least a working knowledge of both languages. Thus, in this chapter, intrasentential code-switching refers to any switch from the starting language in a sentence (e.g., "swimming muy muy faster" or "she put her swimsuit para ira la playa"), and intersentential code-switching refers to switching languages between sentences, (e.g., from Spanish to English-," Ella esta llorando. He took her balloon" or from English to Spanish--"To go home. Para tomar jugo de naranja"). We concentrate on such code-switching from the perspective of sentence processing. First we briefly review results relevant to comprehension and production of code-switched speech in adults as well as results obtained in a few studies with children. Next we present data from a detailed study with bilingual preschool children, which shed some light on the prevalance of code-switching in such children under different circumstances as well as touch on developmental factors. We then compare our results with those from relevant studies and also speculate on the implications of these results on the cognitive representations of two or more languages and interactions between them.
Effects of Fluency, Context, and Age on Inter- and Intrasententiai Code-Switching
Code-switching occurs naturally in bilinguals. One immediate question that arises is whether a critical level of fluency needs to be reached in one's second language (L2) or in both languages for young children before code-switching can be engaged in. Or as an extension, whether code-switching behavior increases with increasing degree of balance between the two languages of a bilingual. From a purely cognitive, sentenceprocessing perspective (as opposed to say social or cultural demands, cf. Fishman, 1965), in nonbalanced bilinguals, when one language is definitely more dominant than the other, one may expect lots of code switches from L2 to L1 as people are speaking in L2 and cannot find words or phrases to express themselves or are uncomfortable
Code-Switching in Bilingual Children
341
with the syntactic structure of the sentence they have started. There should be fewer switches from L 1 to L2. In addition, code-switching has been shown to bridge not only lexical gaps but structural gaps, as well (e.g., Gawlitzek & Tracy, 1996), which would mean that more "need" might arise for switching to L 1 from L2 than vice versa. On the other hand, because nonbalanced bilinguals predominantly use L1 as their base language in natural conversational settings, most of their switches should be from L 1 to L2 (cf. Poplack, 1980). It has also been found to be easier for them to switch from L 1 to L2 (e.g., Meuter & Allport, 1999). In balanced bilinguals, the direction of code switches should be more unpredictable from a cognitive perspective, although the context or social factors might have an effect. Thus, ease of finding the words or phrases, or compensating for structural difficulty is only one factor in determining whether code switching occurs, and probably less of a factor with more fluent bilinguals (cf. Clyne, 1967). Comfort with and dominant use of one language are factors which create an artifactual preponderance of L1 to L2 switches for nonbalanced bilinguals. Whether the balanced or the nonbalanced bilinguals engage in more code switching then is not really a fair question and the answer depends on the reason and the direction of switching. It also depends on the content (e.g., whether it is of single words or interjections or whole phrases) and context (e.g., the topic being addressed and the linguistic setting) of switching. Whether the switch occurs intersententially or intrasententially, however, has been found to be highly indicative of bilingual fluency. Usually, balanced bilinguals have been found to engage in many more intrasentential switching than nonbalanced bilinguals, and nonbalanced bilinguals have been found to prefer intersentential switching (e.g., Muysken, Kook, & Vedder, 1996; Nortier, 1990; Poplack, 1980). The prediction then is that as the degree ofbilingualism increases, so should the differences between intra and intersentential switching preferences in favor of intrasentential switching. The reasoning is that balanced bilinguals feel equally at home with the syntactic constraints of both languages and can manipulate them both with equal ease, whereas nonbalanced biliguals feel more comfortable keeping to a single language within a sentence so as not to violate syntactic structures while determining compatibility. The underlying assumption is that at least in balanced bilinguals, both L1 and L2 are activated automatically regardless of which language is being spoken (cf., Macnamara, 1967). Code-switching has also been shown to be highly dependent on context (e.g., Heredia & Altarriba, 2001; Verhoeven & Boeschoten, 1986). For instance, Verhoeven and Boeschoten (1986) have found that for Turkish children living in the Netherlands, most code-switching occurred when describing life in general, followed by when describing a picture of a market place, and least code-switching occurred when describing a teacher's visit to their homes. Social and cultural demands aside, and only from the perspective of sentence processing, availability of and comfort with linguistic elements in L 1 and L2 in different topics or contexts is certainly a factor that affects code-switching frequencies. The context created by who is involved in the conversation is also important. For instance, if a person is addressed in a code-switched manner, he/she should be more likely to respond in like manner (cf. Martin-Jones, 1995). Finally, the context created by the type of discourse (e.g., formal interview versus
342
Zehra F. Peynirciofflu and Aydm Y. Durguno~lu
friendly joshing) also affects code-switching tendencies (cf. McClure, 1981). Children who are exposed to more than one language can differentiate between those languages at least by age two (Nicoladis & Genesee, 1997). Most code-switching studies, however, have been done with adult bilinguals, and, with a few exceptions, many of the studies with children have involved detailed case-studies, observational studies, or experiments with only a handful of participants. We know that just like adults, bilingual children also code-switch readily, both intersententially and intrasententially. Interestingly, although intrasentential switching in adults increases with degree ofbilingualism and is less frequent than intersentential switching for those not very fluent bilinguals, it is more common than intersentential switching in young children, and there is usually no relationship between code-switching in general and children's ages (cf. Redlinger & Park, 1980). If it is assumed that with increasing age comes increasing linguistic competence, and with increasing linguistic competence in two languages comes increasing intrasentential code-switching, then the finding that age and intrasentential code-switching are not related might seem to be at odds. But perhaps increasing linguistic competence in L2 to be able to cope with syntactic constraints in both languages is less important in intrasentential code-switching propensity than the degree of balance per se between L1 and L2, however impoverished both languages may be at the moment, especially for young children. For compound bilingual children, whose L 1 and L2 are developing together and at equal rates, the degree of balance is very high, and indeed code-switching is explained using the Unitary Language System Hypothesis (e.g., Genesee, 1989), which holds that the two languages for such children are really undifferentiated. Syntactic constraints start appearing later (cf. Meisel, 1989), but this does not pose a problem for compound bilingual children in terms of code-switching because they also develop together. Thus, if it is the degree of balance between the two languages that affects intrasentential code-switching, then the finding of no difference between rates of such code-switching and age is no longer surprising. The question remains, however, about whether young children with distinct L 1 and L2 systems and of varying degrees ofbilingualism show differences in code-switching behavior, especially in intrasentential code-switching behavior.
The Present Experiment We tested young children with varying degrees of bilingualism in a controlled experiment. We created three linguistic context situations such that the participants were more encouraged to speak in L1 in one, in L2 in another, and in a mixed fashion in the third. All participants participated in all situations. Given that more fluent or balanced bilinguals engage in more code-switching, at least intrasententially, but such switches have usually been observed mainly from L1 to L2 for all bilinguals (cf., Meuter & Allport, 1999; Poplack, 1980), we thought it would be interesting to create situations in which the context or presentation language was L2. In that case, from the perspective of ease 0fsentential processing, less balanced bilinguals might be expected to code-switch (L2 to L 1) more often than when the context/presentation language was
Code-Switching in Bilingual Children
343
L 1. That is, the gap between intrasentential code-switching tendencies might narrow between the more and less balanced bilinguals when the context/presentation language was L2. Likewise when the context/presentation language was itself mixed. In addition, all of the stimuli were rotated through each of the three presentation language conditions (L 1, L2, and Mixed) across participants so that the content of the story or the picture would not be a confound in determining language choice. This was important given the findings that code-switching in children was highly dependent on the topic or context (e.g., Verhoeven & Boeschoten, 1986). In the present experiment, because all children participated in all conditions and different topic contents were provided and held constant, it would be possible to observe code-switching behavior only as a function of changing language context and as a within-participant variable. For each presentation language condition, we presented two types of stimuli, a picture that participants were asked to describe, and a short story the participants were asked to retell and answer questions about. We assumed that whereas the story condition would look at both comprehension of L 1, L2, and Mixed presentations and production, the picture condition would encourage production without being dependent on correct comprehension. Drawing such a distinction between comprehension and production was important because in his studies, Kolers (1966, 1974) had shown that comprehension of mixed-language presentation was as good as comprehension of the average of L1 and L2 unilingual presentations (cf. Blair & Harris, 1981), whereas production of mixed-language took longer. Thus, he had concluded that the possible detrimental effects of code-switching in production was irrelevant in comprehension. In the present experiment, we added another variable, that of the level of bilingualism, while exploring such possible differences in code-switching between comprehension and production situations in young children.
Method Participants The participants were a total of 36 preschool children from two Spanish-English bilingual jump-start programs in the metropolitan Washington, D.C. area. They were all between the ages of 3 and 5 (with a mean age of 4.31); 31 were more fluent in Spanish than in English, 4 were more fluent in English than in Spanish, and 1 was equally fluent in the two languages. They were all rated by their teachers in terms of fluency in both languages separately on a scale of 1-10. The difference between the scores for the two languages constituted the primary measure of our bilingualism balance. We considered a child with a score of, for instance, 6 on the Spanish scale and 4 on the English scale (a difference of 2) to have the same degree of imbalance as a child with a score of, for instance, 9 on the Spanish scale and 7 on the English scale (again a difference of 2). The degree of imbalance ranged from 0 to 8 (mean of 2.67). These measures of participants' degrees of bilingualism were corroborated through several weeks of classroom observation of each of the participants before the experiment was conducted.
344
Zehra F. Peynircioglu and Aydm Y. Durgunoglu
Materials For the pictorial presentation, three busy scenes, one depicting the town, one the home, and one the beach, were chosen from children's books (see Appendix). For the verbal presentation, three short paragraphs, one about going to the zoo, one about going swimming, and one about going to the doctor were written by the experimenters (see Appendix). All paragraphs were written in English, in Spanish, and in two mixedlanguage versions; the two mixed versions differed from each other in that the words written in Spanish and English were completely interchanged. A tape recorder was used to record participants' speech. The tapes were transcribed later by two independent judges who were Spanish-English bilinguals. Design All participants were shown all the pictures and read all the stories, in two sessions. Half (18 participants) were shown the pictures in the first session and read the stories in the next session, and the other half did the reverse. The presentation order of both the pictures and the stories was the same for all participants. Within each half of the participants, however, the language of presentation was varied across three groups of participants such that all stories and pictures were presented in all languages (Spanish, English, Mixed) equally often. Thus, a third (6 participants) were shown the first picture and read the first story in English, the second picture and the second story in Spanish, and the third picture and the third story in mixed-language (with 3 participants getting one mixed version and the other 3 getting the complementary mixed version). For another third of the participants the presentation order was Spanish, Mixed, English, and for the final third, Mixed, English, Spanish. Procedure All participants' parents had signed parental consent forms presented in either English or in Spanish, and were told in detail about the experiment. The teachers had also rated each participant's English and Spanish competence on a scale from 1-10, and the experimenters had observed each participant in free play and classroom situations to make their own independent assessments. The experiment was conducted in a separate room during the participants' regular classroom sessions. Each participant was tested individually in two separate sessions, neither of which lasted more than 15 minutes. They were also given breaks in between stories or pictures. In the pictures condition, the participants were shown each picture and asked to tell what was going on in the picture. They could either describe the things in it or tell a story about it. The experimenters took care to speak only in English, only in Spanish, or in a code-switching fashion depending on the presentation language condition of the picture. Thus, for instance, if the first picture was in the English condition, the experimenter gave the instructions and interacted with the child only in English. The children were of course free to speak in any language they wished, although the experimenter would answer in only the language of the condition. Because the children
Code-Switching in Bilingual Children
345
were used to being addressed in both Spanish and English as well as in a mixed fashion during their days at the jumpstart program (by their teachers, friends, as well as the teachers' aides, two of whom also collected the data and did the classroom observations), they did not think it unusualto have the experimenter talk to them in the language of the condition and switch for the next condition. If the child was shy and did not speak, the experimenter would cue by pointing randomly and asking what he/she saw there. After each picture, a mini-break was given and the language switched for the next picture. In the stories condition, the procedure wa~s very similar except that the children were read the stories, then asked to relate what had happened in the story, again using only the language of the condition. Ifa child did not speak spontaneously, then a set of four questions were asked, such as "what animals did the children see at the zoo?" in the zoo story. Again, after each story, a mini-break was given and the language switched for the next story. Half of the participants had the pictures first and then the stories in the next session, and the other half had the reverse.
Results and Conclusions
In both pictures and stories, because we had both English-dominant and Spanishdominant participants, we report the results as a function of first language (L 1) and second language (L2), which in all but one case also coincides with the participants' dominant and weaker languages, respectively, rather than as a function of English and Spanish. For our only balanced bilingual participant, L1 was Spanish. Also, the "Mixed" condition refers to intrasentential code-switching (e.g., "she put her swimsuit para ir a la playa" or "swimming muy muy faster"). We should note that the number of intrasentential switches within a given sentence were not relevant; as long as there was any switch, that sentence was considered to belong to the Mixed condition. The direction of the switch was determined from the starting language. Thus, "este apples" would be a Spanish-English switch whereas "swimming muy muy faster" would be an English-Spanish switch. We report the intersentential code-switching results in comparison with intrasentential code-switching results separately, but when comparing the percentage of times participants spoke in L 1, L2, or Mixed conditions, the Mixed condition necessarily refers to intrasentential code-switching. There were no effects of order of presentation language or the version of Mixed presentation participants received in stories (i.e., which words were presented in Spanish and which words in English), and thus all data were collapsed across these variables. We first analyzed the results using a 2 x 3 x 3 ANOVA (pictures/stories x presentation language condition--L 1, L2, M x response language--L 1, L2, M). Overall, there was no main effect of the type of task (pictures vs stories) or of presentation language (L 1, L2, or M), both ps >. 10. There was a main effect of response language, however, F(2, 34) - 121.41, p < .01; L 1 was the preferred language. There was also an interaction between task and response language, F(2, 34) - 4.03, p < .05; more L1 r6sponses were given in the story condition than in the picture condition. Finally, and perhaps most importantly for present purposes, there was an interaction between
346
Zehra F. Peynircio~tu and Aydm E Durguno~lu
presentation language and response language, F(4, 32) = 19.66, p < .01. Participants changed their language depending on the situation and responded in the language that the materials were presented in. We then analyzed the pictures and stories separately. By and large, there were a great many more responses for the pictures overall (mean of 84.3) than for the stories (mean of 20.4). Thus, there were more data with the pictures, and the overall ANOVA had indicated that more L2 responses were given with these pictures. Looking at the participants' responses with the two types of materials in more detail, however, we saw that both the numbers of responses and code-switching proportions were positively correlated, rs = 0.44 and 0.46, respectively, both ps < .01. Thus, children who were more likely to speak or code-switch in the pictures condition were also more likely to speak or code-switch in the stories condition. It is just that they did both more frequently in the pictures condition when there was a lot more information that could be conveyed as well as when there was no comprehension involved but only production. We should also note that although the correlation between total number of responses and age was significant, r = 0.45, p < .01, that between mixed-language use percentages and age was not r = .05, p > . 10. Thus, older children did tend to speak more often and give more responses than younger children, but younger children were just as likely to switch codes intrasententially as the older children (cf. Redlinger & Park, 1980). In addition, the correlation between mixed-language use percentages and degree ofbilingualism was significant, r - 0.32, p < .05. Thus, although age per se did not influence intrasentential code-switching tendencies, consistent with previous research (e.g., Poplack, 1980), the more proficient bilinguals were more likely to engage in code-switching, at least intrasententially. Pictures
Table 1 summarizes the results of the pictures condition. As can be seen, collapsed across language of presentation condition, participants spoke most often in their L1, then in L2, and finally in a Mixed fashion (ts(35)= 2.39 and 5.19 between L1 and L2 and between L2 and Mixed, respectively, ps < .02). Thus, the results obtained with the overall ANOVA were mirrored with pictures alone, as well. In further analyses, however, we found that even though the number of times the participants spoke did not differ as a function of presentation language (M = 26.9 times when presentation language was L 1, 29.3 times when it was L2, and 28.0 times when it was Mixed), F(2, 105) = 0.58, p > .10, the percentage of responses given in each language did. Specifically, although participants gave more L 1 responses than L2 responses when the presentation language was L1, t(35)= 7.52, p < .01, when the presentation language was L2, their language of responding also changed and they gave more responses in L2 than in L 1, t(35) = 4.06, p < .01. Mixed-language responding was the lowest in both presentation-language conditions; it was significantly lower than the less preferred L 1 when the presentation language was L2, ts(35) = 3.36, p < .01, although the difference between the less preferred language and Mixed language conditions did not reach significance when the presentation language was L1, t(35)= 1.43, p > .10. Again, the
Code-Switching in Bilingual Children
347
results obtained with the overall ANOVA were mirrored with pictures alone. Interestingly, when the presentation language was itself Mixed, again L1 responses were the most frequent, followed by L2 responses, and then Mixed-language responses (ts(35) = 3.90 and 3.92 between L1 and L2 and between L2 and Mixed Language, respectively, ps < .01).
Table 1. Mean Percentages of Responses in Different Languages as a Function of Presentation Language for Picture Stimuli Presentation in
% of Responses in L1 L2 Mixed
L1
L2
Mixed
Overall
65.9 21.3 12.8
31.6 62.0 6.4
57.8 33.4 8.8
51.8 38.9 9.3
In addition, proportions of Mixed responses, or intrasentential code-switching, also changed as a function of language of presentation, F(2, 105) = 3.59, p < .03. Contrasts revealed that the main difference in intrasentential code-switching rates or proportion of Mixed responses was between L 1 and L2 presentation language conditions, t(3 5) = 2.66, p < .01; somewhat surprisingly, differences in intrasentential code-switching rates were not different between L 1 and Mixed and L2 and Mixed presentation conditions. That is, participants did not tend to code-switch more when the experimenter codeswitched and set the context. Indeed, Mixed presentation acted pretty much like L 1 presentation in determining language preference in responding and intrasentential codeswitching rates, implying perhaps that participants found it easier to function within L 1 as their base structure when they were confronted with code-switches. Given that our language-context change was successful in influencing the participants' language preferences in responding, we next looked at whether the degree of balance between the participants' languages affected the likelihood of their responding patterns in the different presentation language conditions. We divided the participants into two groups as a function of bilingual fluency, or balance. Those whose rating differences between the two languages were less than 3 were classified as "more" balanced and those whose rating differences were greater than 3 were classified as "less" balanced. There were 8 participants whose rating differences were exactly 3, and we assigned 4 of them to the more balanced and 4 of them to the less balanced group on the basis of complementary data from our classroom observations. Thus, there were 18 participants in the more balanced and 18 participants in the less balanced group. Of particular interest were any differences in mixed responding or
348
Zehra F. Peynircioglu and Aydm Y. Durgunoglu
intrasentential code-switching rates in the different conditions. The number of times the participants spoke was not influenced by degree of balance in bilingualism in any of the presentation language conditions (all 12s >. 10). However, even though the patterns of responding, both in general and as a function of presentation language, were the same for both the more and the less balanced group, overall, participants in the less balanced group tended to use L 1 more often than those in the more balanced group (54.7% tO 48.9%) and engage in intrasentential codeswitching less often than those in the more balanced group (7.7% to 11.6%), although these differences were not statistically significant. Interestingly, in both L2 and L1 presentation language conditions, it was the more balanced bilinguals who engaged in more intrasentential code-switching than the less balanced bilinguals (7.6% to 2.6% in L2 and 17.1% to 5.9% in L1). Thus, more balanced bilinguals engaged in more intrasentential code-switching, and it was easier for both groups to code-switch when the presentation context was in L1 than L2. In fact, the gap in the proportion of intrasentential code-switches did not close between the more and less balanced bilinguals as we had predicted when the general linguistic context was L2. That is, although less proficient in L2, the less balanced bilinguals preferred to speak in L2 at least as much as the more balanced bilinguals (62.8% vs 58.5%). Perhaps lexical difficulties that arose when speaking in L2 were not strong enough to overcome structural difficulties that arose when code-switching intrasententially. Age within each degree of balance group did not make a difference. Similar patterns were obtained for the younger more balanced bilingual children as the older more balanced bilingual children as well as for the younger and older less balanced bilingual children (all ps >. 10). Thus, as long as degree ofbilingualism was the same, syntactic competency that was assumed to have increased with age did not play a role in intrasentential code-switching rates. The results of the intersentential switches and in comparison with intrasentential switches are shown in Table 2 and as a function of degree of bilingualism. As can be seen, there were 6 possible types ofintersentential switches: L l-L2: "Ella est~i llorando. He t o o k h e r balloon"; L2-LI: "To go home. Para tomar jugo de naranja"; L2-M: "Swimming, swimming. Under de agua"; M-L2: "That's pap~i, nifio y mamfi, that's papfi y grama. He put the door on"; L l-M: "Un gato. Gato y clock"; M-L 1: "Perro con bunny rabbit y gato. Est~i dormiendo." Following a Mixed sentence by another Mixed sentence was not considered to be an intersentential switch even if the base languages were different (e.g., "El la de quiere con du con porque esta de swim? He gets in the water que se condo que tiene suefio") because by definition both sentences were uttered in the same language (i.e., Mixed). Overall, all participants code-switched intersententially much more often than intrasententially; in fact, this was true for 35 out of the 36 participants (p < .01 on a sign test). Similar results were obtained when the presentation language was L2 (27 showed this result, 1 showed the opposite results, and there were 8 ties), L 1 (30 showed this result, and there were 6 ties), or Mixed (30 showed this result, and there were 6 ties), all ps < .02 on a sign test. There were no differences as a function of whether the participants were more or less balanced bilinguals or as a function of their age (all ps > .10).
Code-Switching in Bilingual Children
349
Table 2. Inter- and Intrasentential Code-Switching Ratios in Percentages, the Directions of Intrasentential Switches, and the Rate and Directions of lntersentential Switches as a Function of Presentation Language Condition and Degree of Bilingualism for Picture Stimuli
Presentation Language L1
L2
Deg. of Biling. High Low M High Ratio of Intersentential 69 to Intrasentential 31 Switching
Mixed
Overall
Low
M
High Low M
High
Low
M
75
72
71
75
73
78
73
76
73
74
74
25
28
29
25
27
22
27
24
27
26
26
Intrasentential Switches 89 % L1 to L2 11 % L 2 t o L1
77 23
83 17
51 49
35 65
43 57
58 42
73 27
66 34
66 34
62 38
64 36
Intersentential Switching Rate Overall 36 L1 toL2 7.4 L2toL1 6.1 L2toM 4.0 MtoL2 3.7 L1 toM 7.3 MtoL1 7.0
21 4.6 4.3 1.3 1.5 4.9 4.8
34 6.0 5.2 2.7 2.6 6.1 5.9
18 3.1 4.5 2.7 2.2 2.3 3.0
15 2.5 3.1 3.4 3.5 0.9 1.8
17 2.8 3.8 3.1 2.9 1.6 2.4
29 7.3 6.7 2.6 3.0 5.0 4.4
28 7.2 8.3 1.8 1.3 4.4 5.2
29 7.3 7.5 2.2 2.2 4.7 4.8
28 5.9 5.8 3.1 3.0 4.9 4.8
21 4.8 5.2 2.2 2.1 3.4 3.9
25 5.4 5.6 2.7 2.6 4.2 4.4
In addition, when intrasentential code switching occurred, overall, it was easier to switch from L 1 to L2 than vice versa (24 participants switched more often from L 1 to L2, 7 switched more often from L2 to L1, and there were 5 ties), p <. 02) and there was no difference in this tendency as a function of whether they were more or less balanced bilinguals). However, when the presentation language was L2, the proportion of L1 to L2 and L2 to L1 switches were about equal for more balanced bilinguals and indeed, as expected, the proportion of L2 to L l switches dominated for less balanced bilinguals. Thus, even though the rate of intrasentential code-switching was always higher for more balanced bilinguals and regardless of presentation language (although a higher rate was obtained when the presentation language was L 1), as mentioned above, when code-switching did occur, its direction was indeed influenced by the presentation language. More code-switching occurred when the base language was L1, but the
350
Zehra F. Peynircio~lu and Aydm Y. Durguno~lu
direction of the switches changed when the base language was L2, at least for less balanced bilinguals. When we look at the more common intersentential code-switching results, we see that except when the presentation language was L1, both the more and the less balanced participants tended to have similar code-switching rates. In fact, overall about a quarter of their total utterances with the picture stimuli were intersententially codeswitched. The only influence of degree of bilingualism was apparent when the language of presentation was L 1, and the more balanced bilinguals code-switched more than did less balanced bilinguals. Age of participants did not play a role. Also of interest is the direction of the switches occurring between sentences. Because there were not enough data to yield meaningful statistical analyses, none were done. But simply looking at Table 2, it appears that intersentential code-switching occurred regardless of the base language of the preceding sentence. And although certainly not strikingly so, it also appears that the highest percentage of switches occurred between non-intrasententially code-switched sentences, and the lowest between L2 and intrasententially code-switched sentences. Stories
Table 3 summarizes the results of the stories condition. Overall, there was considerably less data because participants did not speak as much in response to the story stimuli. Many of the results were, however, similar to those obtained with pictures, and all of the results obtained with the overall ANOVA were mirrored with just the stories, as well.
Table 3. Mean Percentages of Responses in Different Languages as a Function of Presentation Language for Story Stimuli
Presentation in
% of Responses in L1 L2 Mixed
L1
L2
Mixed
Overall
79.8 7.2 13.0
29.5 60.9 9.6
65.0 19.6 15.4
58.1 29.2 12.7
Participants spoke most often in their L 1, then in L2, and finally in a Mixed fashion (ts(3 5)= 5.59 and 3.17 between L 1 and L2 and between L2 and Mixed, respectively, ps < .01, respectively). When we looked at the same results in the three presentation language conditions separately, again, we found that even though the number of times
Code-Switching in Bilingual Children
351
the participants spoke did not differ as a function of presentation language (a mean of 6.2 times when presentation language was L 1, 6.3 times when it was L2, and 7.6 times when it was Mixed), F(2, 105) = 1.59,.p > . 10, the percentages of responses given in each language did differ as a function of presentation language. Specifically, the same switch in preferred languages occurred as with the picture stimuli and participants gave more L 1 responses than L2 responses when the presentation language was L 1, t(35) = 14.31, p < .01, and they gave more L2 responses than L1 responses when the presentation language was L2, t(35) = 4.22, p < .01. When the presentation language was itself Mixed, again L 1 responses were the most frequent (t(3 5) = 6.83, p < .01, for the difference between L1 and L2), but this time there was no statistical difference between L2 and Mixed Language (t(3 5) = 0.32, p > . 10). Also, unlike with the picture stimuli, proportions of Mixed responses, or intrasentential code-switching, did not change as a function of language of presentation, F(2, 105) = 1.13, p > . 10.
Table 4. Inter- and Intrasentential Code-Switching Ratios in Percentages, the Directions of Intrasentential Switches, and the Rate and Directions of lntersentential Switches as a Function of Presentation Language Condition and Degree of Bilingualism for Story Stimuli
Presentation Language L1
L2
Deg. of Biling. High Low M High Ratio of Intersentential 67 to Intrasentential 33 Switching
Mixed
Overall
Low
M__
High Low M
High
Low
M
50
59
74
69
72
63
71
67
68
63
66
50
41
26
31
28
37
29
33
32
37
34
Intrasentential Switches 95 % L 1 to L2 5 % L2 to L1
66 34
81 19
33 67
17 83
25
74 26
91 9
83 17
67 23
58 42
63
75
Intersentential Switching Rate Overall 29 LltoL2 2.8 L2toL1 3.7 L2toM -Mto L2 2.3 LltoM 11.7 Mto L1 8.2
12 -1.1 1.1 -3.7 6.0
21 1.4 2.4 0.6 1.2 7.7 7.1
25 2.7 5.1 4.6 5.8 3.2 3.9
13 2.7 3.2 2.9 2.8 -1.8
19 2.7 4.2 3.8 4.3 1.6 2.9
29 3.4 3.4 3.0 2.7 10.0 7.0
20 3.7 4.2 -1.6 5.1 5.2
25 3.6 3.8 1.5 2,2 7.6 6.1
28 3.0 4.1 2.5 3.6 8.3 6.4
15 2.1 2.8 1.3 1.5 2.9 4.3
22
37
2.6 3.5
2.0 2.6 5.6 5.4
352
Zehra F. Peynircioglu and Aydm Y. Durgunoglu
Again, the number of times the participants spoke was not influenced by degree of balance in bilingualism in any of the presentation language conditions (all ps > . 10). Even though the pattems of responding were the same for both the more and the less balanced group, however, overall, participants in the less balanced group again tended to use L1 more often than those in the more balanced group (63.8% to 51.6%). Also, they engaged in intrasentential code-switching less often than the participants in the more balanced group (overall: 9.9% to 16.3%; when the presentation language was L 1: 6.6% to 16.9%; when the presentation language was L2: 5.8% to 10.8%). Thus, it was easier for both groups to code-switch when the presentation context was in L 1 than L2; and the gap in the proportion of intrasentential code-switches did not close between the more and less balanced bilinguals when the general linguistic context was L2. Once again, age did not influence any of these results. The results of the intersentential switches and in comparison with intrasentential switches and as a function of degree of bilingualism are shown in Table 4, and they pretty much mirrored those with pictorial stimuli. Most of the time, the differences that were observed were exaggerated even more. The only differences between the patterns of responding with the pictures and with the stories were that with stories, unexpectedly, less balanced bilinguals' intra- and inter-sentential code-switching rates were about the same in the L 1 presentation context, more balanced bilinguals, just like less balanced bilinguals, switched more often from L2 to L1 in the L2 presentation context, and a higher rate of switching seemed to occur between L1 and Mixed sentences rather than non-intrasententially code-switched sentences.
Grammatical Mistakes and Comprehension Mistakes We looked at grammatical mistakes while speaking in L 1, L2, and Mixed Language in response to both pictures and stories. Not surprisingly, in general, there were fewer such mistakes for both types of stimuli with increasing age, r = .37, p < .03 for pictures and r = .29,p < .09 for stories. The number of mistakes (means of 3.9 in L1,4.7 in L2, and .7 in Mixed with pictures and 1.1 in L1, 1.9 in L2, and .3 in Mixed) with pictures did change as a function of response language, F(2, 105)= 8.57 with pictures and F(2, 105) = 6.52 with stories, bothps < .01. This was true for both more balanced bilinguals (F(2, 51) = 3.40, p < .05 with pictures and F(2, 51) - 2.69, p < .08 with stories) and less balanced bilinguals (F(2, 51) = 5.60, p < .01 with pictures and F(2, 51) = 3.71, p < .05) with stories. With pictures, there was no difference between the number of mistakes made while speaking in L1 and in L2, t(35) - . 7 3 , p >. 10 whereas both L1 and L2 mistakes were greater than Mixed Language mistakes (ts(35) = 3.89 and 3.16, respectively, both ps < .01). Of course, for the most part this was probably due to the differences in the sheer numbers of times spoken in each language, but it might also have been reflective of code-switching in order to have an easier time in expressing oneself. With stories, more mistakes were made in both L2 and L1 compared separately with the Mixed Language condition, ts(35) = 1.81 and 3.61, and ps < .01 and .08, respectively. We also looked at comprehension of stories by looking at the number of conceptually wrong responses to questions. In general, although the number of wrong
Code-Switching in Bilingual Children
353
responses tended to decrease with increasing age, the correlation between age and number of wrong answers was not statistically significant, r = .22, p >. 10. There was also no effect of presentation language on the number of wrong responses, F(2, 105) = 1.14, p >. 10. The mean numbers of wrong responses after hearing the story in L 1 was 1.44, in L2 was 1.72, and in a Mixed Language was 1.97. Comprehension as measured by correct responses as a function of presentation language or the language in which the story was told was about the same for all languages. Thus, consistent with Kolers' (1974) observations, comprehension was not affected by code-switching in children, either. When we looked at the above results as a function of degree of balance in bilingualism, as well, we found that there were no differences whatsoever between the more and the less balanced bilinguals.
Conclusions
In this experiment we looked at children's intra- and intersentential code-switching tendencies as a function of age, degree of balance in bilingualism, and the context of language of presentation. By and large, the results were consistent with those found with adult bilinguals. Unlike the results reported in the Redlinger and Park (1980) study with young children, intrasentential code-switching was used less often than intersentential code-switching and the instances of switching (e.g., as a function of presentation language) were influenced by the degree of bilingualism; the more often used intersentential code-switching, however, was relatively independent of degree of bilingualism (cf. Meuter & Allport, 1999; Poplack, 1980). It appears that comfort with sentence processing, both lexically and syntactically, is related to the degree of intrasentential code-switching children will engage in, just as with adults. Switching between sentences on the other hand is easier because judging compatibility is no longer an issue, and thus less balanced bilinguals also do it quite readily. Interestingly, code-switching appeared to be independent of age as well as linguistic proficiency even in an age group (3 to 5) where language development is at a fast pace (cf. Redlinger & Park, 1980). The direction of intrasentential code-switching was another question of interest. Although somewhat counterintuitive in that it meant ostensibly going from the easier to the more difficult, children engaged in more L1 to L2 switches than vice versa. Consistent with those of other studies (e.g., Poplack, 1980), such a finding has usually been attributed to the greater difficulty faced in syntactic structure changes in L2. That is, it has been assumed that L2 to L1 switches are more difficult because the less balanced bilinguals are more uncomfortable with the word order or L2 and hence more uncomfortable with where and how to make the switches correctly. Another way of viewing the present results might be to connect these findings to those reported in translation studies in general. In exploring translation from one language to another with adults, usually, translation from L2 to L1 is faster and more accurate than translation from L1 to L2, and it has been proposed that translation from L2-L1 is lexically mediated whereas translation from L1 to L2 is conceptually mediated (e.g., Kroll & Stewart, 1994; but see Altarriba & Mathis, 1997). Within this framework,
354
Zehra F. Peynircio~lu and Aydm Y. Durgunoglu
Sholl, Sankaranarayanan, and Kroll (1995) showed that picture naming in either L 1 or L2, through priming the concept itself, helped only L 1 to L2 translation and not L2 to L1 translation because the latter was lexically mediated. As an extension, at the sentence level, participants in our study preferred L1 to L2 switches when codeswitching was intrasentential, possibly because speaking about both the pictures and the stories was conceptually rather than lexieally driven. Thus, even though we were not testing translation of equivalents but sentence processing, a larger unit than single words might have been activated at the conceptual level and made it easier to access compatible L2 phrases. Such a finding is also consistent with Meuter and Allport's (1999) findings of a lesser cost of switching in terms of response latencies from L 1 to L2. Thus, the preferred direction of code-switching in sentence processing appears to lend support to previous findings at the lexical level. The direction of intrasentential code-switching also depended on the linguistic context, however. When the presentation language was L2, participants code-switched more often from L2 to L1 than vice versa. Of course, such a result may not be surprising given that participants' base language became L2 and hence more switching would be expected from the base language to the other simply based on sheer usage frequency. But we had predicted that when the presentation language was L2, from the perspective of ease of sentential processing, the less balanced bilinguals might be expected to code-switch (L2 to L l) more often than when the presentation language was L l, and the gap between intrasentential code-switching tendencies might narrow between the more and less balanced bilinguals when the presentation language was L2. These predictions did not hold up. In addition, given that participants still preferred L 1 to L2 switches even when the presentation language was mixed such that the responses could be based on either language, it appears that the reversal of the preferred direction of switching when the presentation language was L2 was worthy of note and influenced by the context itself in addition to the sheer number of sentences started in one or the other language. Overall, the degree of balance in bilingualism had a moderating effect in that, especially with the stories, the more balanced bilinguals preferred the L l-L2 switch more than the less balanced bilinguals. Based just on comfort with and dominant use of L 1 in all situations, we should have expected the opposite--that the less balanced bilinguals would engage in a greater proportion of L1-L2 switches compared to more balanced bilinguals. But it appears that just like in other contextual domains, the linguistic context set by the experimenter also influenced code-switching tendencies and direction, especially for the less balanced bilinguals. Finally, consistent with results found with adult bilinguals (e.g., Blair & Harris, 1981; Kolers, 1966, 1974) comprehension as measured by being able to answer content questions about the stories correctly was not influenced by the language in which the story was presented. Production, however, as measured by both the preferred direction of intrasentential code-switching (L 1 to L2 vs. L2 to L l) and number of grammatical mistakes made while speaking in L1, L2, or Mixed Language, was influenced by presentation language or linguistic context as well as by the degree of balance in children's bilingualism. Future directions in studying code-switching behavior in young children might include conducting separate smaller scale experiments to explore the influences of linguistic proficiency and degree ofbilingualism more systematically, to
Code-Switching in Bilingual Children
355
concentrate on qualitative content analysis, especially within each presentation language context separately, and to investigate similar questions with bilinguals whose languages are more or less related to each other than English and Spanish.
Author Notes
We thank Megan Reilly and Daviana Greenberg for their help in data collection and Lisa Korenman for her help in statistical analyses.
References
Altarriba, J., & Mathis, K. M. (1997). Conceptual and lexical development in second language acquisition. Journal of Memory and Language, 36, 550-568. Blair, D., & Harris, R. J. (1981). A test ofinterlingual interaction in comprehension by bilinguals. Journal of Psycholinguistic Research, 10, 457-467. Blom, J. P., & Gumperz, J. J. (1972). Social meaning in linguistic structures: Codeswitching in Norway. In J. J. Gumperz & D. Hymes (Eds.), Directions in sociolinguistic: The ethnography of communication (pp. 407-434). New York: Holt, Rinehart & Winston. Clyne, M. G. (1967). Transference and trigerring. The Hague: Martinus Nijhoff. Fishman, J. A. (1965). Who speaks what language to whom and when? La Linguistique, 2, 67-88. Gawlitzek-Maiwald, I., & Tracy, R. (1996). Bilingual bootstrapping. Linguistics, 34, 901-926. Genesee, F. (1989). Early bilingual development: One language or two? Journal of Child Language, 16, 161-179. Heredia, R. R., & Altarriba, J. (2001). Bilingual language mixing: Why do bilinguals code-switch? Current Directions in Psychological Science, 10, 164-168. Kolers, P. A. (1966). Reading and talking bilingually. American Journal of Psychology, 79, 357-376. Kolers, P. A. (1974). Remembering trivia. Language and Speech, 17, 324-336. Kroll, J. F., & Stewart, E. (1994). Concept interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33, 149-174. Lipski, J. M. (1978). Code-switching and the problem of bilingual competence. In M. Paradis (Ed.), Aspects of bilingualism (pp. 250-264). Columbia, SC: Hornbeam Press. Macnamara, J. (1967). The bilingual's linguistic performance--a psychological overview. Journal of Social Issues, 23, 58-77. Martin-Jones, M. (1995). Code-switching in the classroom: Two decades of research. In L. Milroy & P. Muysken (Eds.), One speaker, two languages (pp. 90-112). Cambridge: Cambridge University Press. McClure, E. (1981). Formal and functional aspects of the codeswitched discourse
356
Zehra F. Peynircioglu and Aydm Y. Durgunoglu
of bilingual children. In R. P. Duran (Ed.), Latino language and communicative behavior (pp. 69-94). Norwood: Ablex. Meisel, J. M. (1989). Early differentiation of languages in bilingual children. In K. Hyltenstam & L. Obler (Eds.), Bilingualism across the lifespan: Aspects of acquisition, maturity, and loss (pp. 13-40). Cambridge: Cambridge University Press. Meuter, R. F. I., & Allport, A. (1999). Bilingual language switching in naming: Asymmetrical costs of language selection. Journal of Memory and Language, 40, 2540. Muysken, P., Kook, H., & Vedder, P. (1996). Papiamento/Dutch code-switching in bilingual parent-child reading. Applied Psycholinguistics, 17, 485-505. Nicoladis, E., & Genesee, F. (1997). Language development in preschool bilingual children. Journal of Speech-Language Pathology and Audiology, 21,258-270. Nortier, J. (1990). Moroccan Arabic-Dutch code-switching. Dordrecht: Foris. Pandharipande, R. (1990). Formal and functional constraints on code-mixing. In R. Jacobsen (Ed.), Code-switching as a worldwide phenomenon (pp. 15-31). New York: Peter Lang. Pfaff, C. W. (1979). Constraints on language mixing: Intrasentential code-switching and borrowing in Spanish/English. Language, 55, 291-318. Poplack, S. (1980). Sometimes I'll start a sentence in Spanish Y TERMINO EN ESPAlqOL: Toward a typology of code-switching. Linguistics, 18, 581-618. Redlinger, W., & Park, T .Z. (1980). Language mixing in young bilinguals. Journal of Child Language, 7, 337-352. Scotton, C. M. (1988). Self-enhancing codeswitching as international power. Language & Communication, 8, 199-211. Sholl, A., Sankaranarayanan, A., & Kroll, J. F. (1995). Transfer between picture naming and translation: A test of asymmetries in bilingual memory. Psychological Science, 6, 45-49. Sridhar, S. N., & Sridhar, K. K. (1980). The syntax and psycholinguistics of bilingual code mixing. Studies in the Linguistic Sciences, 1O, 203-215. Verhoeven, L. T., & Boeschoten, H. E. (1986). First language acquisition in a second language submersion environment. Applied Psycholinguistics, 7, 241-255. Weinreich, U. (1953). Languages in contact. The Hague: Mouton. Yoon, K. K. (1993). New perspectives on intrasentential code-switching: A study of Korean-English switching. Applied Psycholinguistics, 13, 433-449.