Codeswitching within prepositional phrases: Effects of switch site and directionality

Abstract

Aims and Research Questions:

Codeswitching (CS) was investigated among English-Hebrew bilingual preschool children in a sentence repetition task focusing on switching at different points in prepositional phrases (PPs). We asked the extent to which sentence repetition accuracy differed (1) as a function of the switch site in the PP and (2) as a function of directionality, English-to-Hebrew versus Hebrew-to-English CS.

Design/Methodology:

English/first language (L1)-Hebrew/second language (L2), sequential bilingual children (N = 65), ages 5;5–6;5, participated. Thirty-six English and 36 Hebrew stimulus sentences were matched for semantic content and syntax. English stimulus sentences contained switches to Hebrew; Hebrew stimuli contained switches to English. Six ‘switch’ conditions were examined: a single codeswitched noun (N), a determiner–noun switch (DET+N), a codeswitched preposition (P), a preposition–determiner switch (P+DET), a switch of the entire PP (P+DET+N), and a no-switch condition.

Data and Analysis:

Audio recordings were transcribed and coded. Full sentence repetition was coded as correct/incorrect. The number of errors and the proportion of CS errors were computed. A 6 × 2 repeated-measures analysis of variance examined the effects of switch site within the PP and directionality (L1-to-L2 versus L2-to-L1).

Findings/Conclusions:

Accuracy was highest for the non-switched, N, and P+DET+N conditions. Accuracy was lowest for DET+N switches in English sentences, and for P switches in Hebrew sentences, and these two conditions showed the highest proportion of CS errors. The findings show evidence for a hierarchy of processing costs and directionality differences, which are interpreted in terms of contrastive typological features, particularly definiteness marking in the two languages, English by a free morpheme, and Hebrew by a bound clitic.

Keywords

Codeswitching prepositional phrases child bilingualism preschool English-Hebrew bilinguals sentence repetition cognitive cost

Introduction: Codeswitching and codeswitching costs

The present study focuses on intra-sentential codeswitching (CS), the alternate use of two languages within a sentence or clause. While CS is widespread in some communities and seemingly effortless, under certain circumstances a switch in languages may incur a cognitive cost in comprehension and/or production. The cost may surface as longer processing time and lower processing quality due to cross-language competition and/or the need to inhibit the interfering language (Gollan & Goldrick, 2016; Van Hell et al., 2015). The cost may be higher when the codeswitch is ungrammatical or unnatural (Dussias, 2001; Pérez-Leroux et al., 2014). Conversely, there may be no cost when the switches correspond to switching patterns that are frequently encountered in a bilingual community (Adamou & Shen, 2019) or when the switches are more lexically accessible than their non-switched counterparts due to language proficiency or use (Chan et al., 1983; Dussias & Courtney, 1995). Only a handful of studies have examined the cost of CS in children (e.g. Gross & Kaushanskaya, 2015). These studies have not considered syntactic aspects of intra-sentential CS. The present study examines the cost of CS for different switch sites within prepositional phrases (PPs), linking cognitive cost to grammatical factors. It asks whether CS invokes processing difficulty and hence a CS cost under all circumstances, whether the cost mainly occurs for language switches under grammatically challenging CS conditions, and whether there is a hierarchy in CS costs. In addition, the study assesses cost effects in terms of directionality by comparing CS from L1/first language (English) to L2/second language (Hebrew) and vice versa. CS costs in previous research involve errors, dysfluencies such as lexical access difficulties, and slower response times. In the present study processing costs were operationalized in terms of accuracy rates and errors in a sentence repetition (SRep) task.

Rules and constraints in CS

Approaches concerning the linguistic conditions underlying CS vary widely. Among the earliest approaches, the Free Morpheme Constraint precluded CS between a free and bound morpheme, and the Equivalence Constraint stated that CS is not permitted at points where the two languages do not correspond in terms of word order (Poplack, 1980). Di Sciullo et al. (1986) argued that syntactic dependencies constrain CS. According to the Government Constraint, two elements in a government relationship – a lexical head and the ‘highest’ lexical element of its complement – must come from the same language. The Functional Head Constraint (Belazi et al., 1994) was also based on syntactic relations. In contrast to Di Sciullo et al. (1986), Belazi et al.’s (1994) account barred CS between a functional element (such as determiner (DET)) and its complement while permitting mixing between a lexical element (such as a preposition (P)) and its complement.

Other proposals argue that any constraint on CS should be grounded in principles that govern monolingual grammar, and that there is no need to involve a theory that pertains explicitly to CS (e.g. Chan, 2008; MacSwan, 2014). Mahootian (1996), for example, argues that the language of a head determines the syntactic properties of its complement in unilingual as well as codeswitched phrases. In contrast to the Government Constraint and the Functional Head Constraint, this claim does not imply that the language of the complement needs to match that of the head.

A different approach to CS is based on asymmetry, where different roles are assigned to the ‘matrix language’ (ML) and the ‘embedded language’ (EL), the EL supplying the switches and the ML providing the frame for the switches (Joshi, 1985; Myers-Scotton, 1993). In Joshi’s model, asymmetry is further evident in the constraint on switching closed class items. The Matrix Language Frame (MLF) model (Myers-Scotton, 1993) and its extended 4-M model (Myers-Scotton & Jake, 2000, 2017) expand on the notion of asymmetry. According to these models, there is opposition between the ML and the EL and between content and system morphemes. System morphemes are typically supplied by the ML, while content morphemes can come from either the ML or EL. Content morphemes are defined as morphemes that assign or receive thematic roles; system morphemes are those that do not.

CS of prepositions and prepositional phrases

Ps and PPs have several unique features that are of relevance here. Firstly, they are less commonly codeswitched than nouns (Ns) and noun phrases (NPs). Ps are considered a minor syntactic category and regularly co-occur with a subsequent NP, making them an interesting testing ground for CS models and constraints. Ps have both lexical and functional properties (Chan, 2008): they are considered lexical because they can assign theta roles, and they are considered functional in that they form a closed class. Finally, Ps are a well-attested source of difficulty for bilingual children and adults alike (e.g. Armon-Lotem, 2012; Song & Sardegna, 2014; Taliancich-Klinger et al., 2018).

Initially, Ps were claimed to be infrequently switched (Pfaff, 1979; Treffers-Daller, 1993, 2009). Treffers-Daller attributes this lower frequency to lack of cross-language congruency and integration of Ps in terms of morphology and syntax. Joshi’s (1985) model bars CS of Ps since they are closed class items, which according to his model cannot be switched.

Jake and Myers-Scotton (2009) and Myers-Scotton and Jake’s (2009) models allow P switching under most circumstances. They distinguish between two kinds of Ps: content morpheme Ps, such as locatives and temporals (e.g. inside, before, after), which are free to be codeswitched since they assign thematic roles, and three kinds of system morpheme Ps. The latter are more restricted in terms of CS: Early System Morphemes Ps, for example those occurring in phrasal verbs (Vs) (take off , come across ) are typically not switched; Bridge Morpheme Ps (e.g. destruction of the army) are rarely switched; and Outside Late System Ps (e.g. the animate object case assigner a in Spanish) are barred from switching in this model.

In addition to single P switching, the literature also discusses the switching of a complete PP. While some corpora have found limited examples of codeswitched PPs (Pfaff, 1979), others point to their (relatively) high incidence (Clyne, 1987; Hakimov, 2016; Zentella, 1997). Clyne argues that PPs constitute a semantically integral unit whose switching in his German/Dutch-English corpus was due to a trigger word, frequently the head of the phrase. Likewise, Hakimov asserts that codeswitched PPs are retrieved from memory as ‘holistic units’ (p. 120). In Myer-Scotton’s (1993) framework codeswitched PPs are called ‘islands,’ which are well-formed constituents that adhere to the grammar of the EL. Unlike the switching of PP internal elements (e.g. between the P and the DET+N), switching the entire PP does not demand the processing and integration of languages from two different grammars. In our estimate, switching a PP in its entirety should not pose particular difficulty.

Regarding the kind of switches predicted to occur within the PP, different models make different predictions. Switching between a P and the ‘highest’ lexical element of its complement, that is, a DET in a NP, is barred by the Government Constraint (Di Sciullo et al., 1986), but not by the Functional Head Constraint (Belazi et al., 1994), which treats Ps as lexical elements. As stated above, Joshi’s (1985) closed class constraint bars P switching since it is a closed class element, while Myers-Scotton’s models would permit it, particularly if the P is a content morpheme.

PPs are an excellent testing ground for the proposed asymmetry between CS functional and lexical categories. For instance, data in the CS literature show that the DET is generally not switched along with the N (Dussias, 2002; Jake et al., 2002). The MLF model provides an explanation for this fact, suggesting that Ns (content morphemes) may be switched freely, but that DETs (system morphemes) may not, since they set the morphosyntactic frame of the ML. However, the MLF model also allows ‘island’ switches, in which both DET and N may come from the EL.

The present study focuses on CS in PPs. The manner in which PPs are combined with head Ns is very much alike in English and Hebrew PPs, and both follow the same word order, namely P+DET+N (Wintner, 2000). Nevertheless, directionality effects in CS might surface due to the differential properties of the DET in the two languages. DETs in English are free-standing morphemes, but in Hebrew they are bound, that is, cliticized to the ensuing N, with no intervening elements. While in English a DET can be separated from the N by intervening modifiers such as adjectives and numerals, in Hebrew the DET must attach directly to the N. Thus, the prenominal DET in Hebrew combines with single words (Danon, 2001; Wintner, 2000). Several properties of the Hebrew DET support Wintner’s claim that it should be considered an affix. For one, the Hebrew DET never occurs in isolation, like bound morphemes. Phonologically, HA- (the) is not stressed except for contrastive purposes. Also, there are instances where the phonology of the N is changed as a result of the attached DET. In addition, the Hebrew DET HA- (the) is never subject to movement rules independent of the co-occurring N. Other arguments for treating the Hebrew DET HA- (the) as an affix are related to selectivity and semantic and N construct properties, as well as additional idiosyncrasies, which are beyond the scope of the present paper (see Wintner, 2000, for details). Importantly, the strong bond between the DET and the N in Hebrew makes the switch between a Hebrew DET and an English N less likely (along the lines of the Free Morpheme Constraint – Poplack, 1980). We further assume that while in English the head of the NP is D (Abney, 1987), in Hebrew the head of the NP is the D+N, and the D in Hebrew is a feature of the N (Borer, 1999; Danon, 2001, 2008). In terms of PPs (Di Sciullo et al., 1986; Mahootian, 1996; Mahootian & Santorini, 1996), it follows that in English the P selects the DET, while in Hebrew the P selects the DET+N. A codeswitch resulting in an English P with a subsequent Hebrew DET+N should be less likely than a Hebrew P with a subsequent Hebrew DET+N. For the P+DET+N construction, an English P selects a free-standing DET – the head of the NP. Due to the codeswitch, however, the free-standing head is now lacking and replaced by a cliticized Hebrew DET, which is a feature of the ensuing N rather than an independent head (Borer, 1999; Danon, 2001, 2008). We would argue that when the PP encodes definiteness, the speaker expects an English P to be followed by an independent, free-standing lexical item. Deviation from this expectation would result in an increase in processing cost. We expect the condition with a switch between a Hebrew P and an English DET+N to be easier and incur less CS cost. Firstly, unlike English, Hebrew has no indefinite article, and the occurrence of NPs without anything overt in the DET position is probably more likely than in English. It follows that processing a Hebrew P without a subsequent Hebrew article should be easier than processing an English P lacking an English article, since the Hebrew P only ‘expects’ a NP, with or without a DET. Moreover, since definiteness in Hebrew is a feature of the N (Borer, 1999; Danon, 2001, 2008), there is a lower expectation for a Hebrew P to select any particular type of definite article (cliticized or free-standing) than there is for an English P.

Codeswitching in bilingual children and grammatical constraints

Very little experimental research on processing codeswitched stimuli and CS costs has been conducted with children. However, processing mechanisms have been shown to operate differently in adults than children, and it is not clear whether the magnitude and nature of CS costs in children are similar to those of adults. Children are still developing their cognitive and linguistic systems and may be developing greater proficiency in the societal language (Davidson et al., 2006; Gross & Kaushanskaya, 2015; Kohnert, 2002). Moreover, children’s CS is occasionally viewed as a symptom of unbalanced and deficient language abilities rather than a sophisticated, rule-based system that bilinguals utilize (Bolonyai, 2009; Gross & Kaushanskaya, 2015).

Research on the cost of CS conducted with children has almost exclusively focused on single words rather than stimuli containing a richer linguistic context (but see Gross et al., 2019, for an exception). The present study, in contrast, examines switch costs using a sentence context. In line with previous research, we propose that CS that is grammatically anomalous, and that violates CS constraints, will incur a greater processing cost than CS that is part of the speakers’ linguistic experience (Beatty-Martínez et al., 2018; Dussias, 2001; Hamers & Blanc, 2000).

As for children’s linguistic experience, a few studies have demonstrated that children’s CS conforms to the same grammatical constraints and principles as that of adults (Cantone, 2007; Gutiérrez-Clellen et al., 2009). Studies have found support for Myer-Scotton’s MLF model (Paradis et al., 2000; Pert & Letts, 2006), Poplacks’s Free Morpheme and Equivalent Constraints (Gutiérrez-Clellen et al., 2009; Halpin & Melzi, 2018), and MacSwan’s Minimalist Approach (Cantone, 2005). As in studies on adult CS, violations of CS constraints, for example, the Government and Functional Head Constraints, have been reported (Cantone, 2005; Müller & Cantone, 2009).

Directionality in bilingual children’s codeswitching

Two main factors pertain to the issue of directionality in children’s bilingual CS, one internal and one external. The individual factor is related to the subject’s proficiency, while the external factor concerns the subject’s sociolinguistic background and exposure.

Sequential bilinguals, that is, those whose home language was acquired before the societal language, tend to become stronger in the societal language (Basnight-Brown & Altarriba, 2007; Ebert & Kohnert, 2016; Jia et al., 2006). They have been reported to codeswitch more from their less proficient to their more proficient language (Genesee et al., 1995; Lanvers, 2001; Ribot & Hoff, 2014; Smolak et al., 2020). For example, in a study of preschoolers’ performance on screening tests in Spanish and English, Greene et al. (2012) found a correlation between language dominance and CS direction, where Spanish dominant children switched more from English to Spanish, English dominant children switched more in the opposite direction, and balanced bilinguals switched in both directions.

Raichlin et al. (2019) studied Russian-Hebrew bilingual children with higher proficiency in Hebrew. Overall, children produced more CS from Russian to Hebrew and did so more for psycholinguistic motivations (to maintain fluency or to overcome difficulties in lexical access). The authors in these studies make reference to proficiency influences on directionality in CS, mostly in their interpretation of the findings. For example, Greene et al.’s (2012) statement that children’s code-mixing was motivated by a need to fill in gaps of lexical knowledge is an indirect reference to proficiency.

Other interpretations of asymmetries in CS directionality refer to sociolinguistic factors, such as the status of each of the two languages and whether it is a majority or minority language (Bosma & Blom, 2019; Smolak et al., 2020). Gutiérrez-Clellen et al. (2009) viewed their finding of directional asymmetry in sociolinguistic terms, arguing for children’s awareness of the ‘language prescribed by the majority culture’ (p. 105), English. Iluz-Cohen and Walters (2012) attributed CS asymmetry to the children’s ‘sociolinguistic sensitivity’ (p. 71). In addition, Montanari et al. (2019) found that CS was more frequent from Spanish to the majority language English, regardless of their participants’ proficiency.

Another factor that is related to directionality is language exposure. In Raichlin et al. (2019), the directionality findings were clearly weighted in favor of CS into L2/Hebrew, due to greater exposure to the societal language in the preschool classroom, in the playground, on the street in the children’s mixed neighborhoods, and in their homes, where older siblings and even parents all provided Hebrew input. Smolak et al. (2020) reported that for some groups of the Spanish-English bilingual children in their study, low exposure to Spanish caused them to codeswitch into English. Similarly, Ribot and Hoff (2014) noted that CS directionality differences in Spanish-English bilingual children were a function of L1 or L2 input: the children who had more English input switched more from Spanish to English, while the children with higher Spanish input switched more from English to Spanish.

The present study further examines CS directionality in bilingual children. Unlike previous research with children, we conducted a SRep experiment with codeswitches inserted in both the L1 and L2, allowing investigation of directionality effects for different CS sites within PPs. We argue that codeswitches in a direction less familiar to the speaker and not in line with their usage preferences may incur a larger cost than those that are part of their linguistic experience (Adamou & Shen, 2019; Beatty-Martínez et al., 2018; Dussias, 2001).

Sentence repetition

In order to assess production of CS, the present research uses a SRep task. In SRep, participants are instructed to listen to sentences and repeat them as accurately as possible. SRep involves auditory, memory, and production skills and taps into underlying linguistic representations (Riches, 2012). It has been shown to accurately reflect language abilities and is frequently a component of standardized language tests, for example, the TROG (Bishop, 1989), BESA (Peña et al., 2014), and CELF-2 Preschool for English (Wiig et al., 2004). SRep has been employed in a wide variety of populations (Vinther, 2002), including monolinguals (Komeili & Marshall, 2013), L2 learners (Rast & Dommergues, 2003), and bilingual children (Armon-Lotem, 2012; Chiat et al., 2013). The task provides insight into a child’s linguistic representations by targeting structures that are believed not to be intact, that is, markers of Developmental Language Disorder (Conti-Ramsden, 2003; Meir et al., 2015). SRep has also been used to assess the repetition of codeswitched sentences (Azuma & Meier, 1997; Lipski, 2017).

Two basic premises underlie the use of this task in the present study: (1) ungrammatical/less acceptable stimuli should result in more errors than more acceptable counterparts (Azuma & Meier, 1997; Lipski, 2017); and (2) structures not part of an individual’s linguistic representations should lead to more errors than those that are part of their repertoire (Vinther, 2002).

While SRep may be lacking in ecological validity, the task is experimentally flexible, allowing elicitation of structures that may be difficult to elicit through other experimental settings or through natural speech (Theodorou et al., 2017). Moreover, SRep data enable analyses of accuracy as well as errors and modifications made to the codeswitched target structures, and these are expected to shed light on the processing of codeswitched sentences (Lipski, 2017).

Aims and rationale

The present study operationalized the notion of CS costs in terms of accuracy and error rates in a SRep task. The study examined the following: (1) the extent to which the cost of CS is contingent on the locus of the codeswitch within a PP; and (2) the extent to which the cost of CS differs in English to Hebrew versus Hebrew to English CS and whether directionality differences are related to switch locus. Switches of and within PPs are controversial in the CS literature, and PP switches at different loci afford an opportunity to systematically examine structures with both lexical and functional properties. In parallel, a SRep task tests processing in preschool children, whose short-term memory limitations and general naïveté allow insight into processing that naturalistic data do not.

Research questions and predictions

1. Performance on repetition of sentences containing CS as insight to switch costs. To what extent does bilingual children’s SRep accuracy differ as a function of switching at different points in a PP (non-CS/N/DET+N/P/P+DET/P+DET+N)?

Repetition accuracy is expected to be highest for unilingual sentences, that is, those with no CS. Sentences with PPs containing a codeswitched N are predicted to show the next highest accuracy, followed by sentences containing a codeswitched PP (P+DET+N). The prediction that unilingual sentences will yield the highest level of accuracy implies a ‘cost of CS’ and follows from research on processing costs in codeswitched as compared to unilingual contexts (e.g. Van Hell et al., 2015). The prediction for accuracy with N switching follows from the high frequency of CS on single Ns reported in corpus studies (e.g. Marian, 2009). The third prediction regarding a relatively high accuracy rate for a P+DET+N codeswitch is based on the fact that PP switching does not involve manipulation of grammatical elements from two linguistic systems within the PP. Both the CS-N and CS-P+DET+N conditions do, however, involve switching twice and may lead to a processing cost due to cross-language competition and interference (e.g. Van Hell et al., 2015).

The three other conditions (a codeswitched DET+N/a codeswitched P/and a codeswitched P+DET) are expected to result in lower accuracies. The ‘closed class’ constraint, Government Constraint, and MLF/4-M models are all consistent (albeit in different ways) with the prediction for these conditions. The former two would predict particularly low accuracy for DET+N and P switches; the MLF/4-M would not. The Government Constraint would predict low accuracy for a switch between a P and its complement (head), while the MLF/4-M model would do so for a P+DET switch. The DET+N switch condition is of particular interest, since on the one hand this structure is a NP and in other contexts (e.g. as a NP subject) is readily switchable. On the other hand, corpus data attest to switching of Ns independently from their DETs (e.g. Dussias, 2002). Also, in the present study the DET+N is embedded in a PP, which may make switching DET+N more challenging. Beyond this, we hesitate to make more specific hypotheses regarding comparisons of switch costs in the DET+N, P, and P+DET conditions.

CS errors (i.e. modifications changing English to Hebrew, or vice versa) are a window to processing difficulty. When they are plentiful, the child is struggling with the grammar and CS to the other language may allow her a way to cope. Therefore, fewer CS errors are expected in the high-accuracy, non-CS, CS-N, and CS-P+DET+N conditions, and more CS errors in the other three conditions, CS-DET+N, CS-P+DET, and CS-P.

2. Directionality. To what extent does accuracy in repetition of codeswitched sentences differ as a function of directionality (English to Hebrew versus Hebrew to English) and to what extent does directionality interact with the CS condition?

One internal factor, proficiency, and two external factors, exposure and sociolinguistic influences, may play a role. For proficiency, we might expect more accurate repetition of codeswitched sentences from English to Hebrew than from Hebrew to English, since sequential bilinguals (even those of preschool age) generally become dominant in the new language (e.g. Basnight-Brown & Altarriba, 2007; Ebert & Kohnert, 2016). However, since the children in the present study were similarly proficient in their two languages (see the Method section), no prediction can be made for the direction of their CS. Exposure and sociolinguistic influences converge on a single prediction: the cost of CS is lower when the direction of the codeswitch corresponds to switching patterns that are familiar to the participants from day-to-day language use in their community. Hebrew is the societal language, and children are expected to have more exposure to the societal language due to input from peers, older siblings, and electronic media, such as television and internet. However, English, the home language, has vitality that other home languages do not have. The English-speaking population tends to be residentially enclaved, and English is often heard on the street and in the media that preschool children are exposed to. Even among native-born, monolingual Hebrew speakers, it is not uncommon to hear CS to English. Moreover, in sociolinguistic terms, while Hebrew is the majority language and English the minority language, English in Israel is the language of wider communication and enjoys high prestige (Altman et al., 2016). Therefore, in terms of exposure and sociolinguistic factors, we cannot predict that SRep accuracy will be higher in one direction or the other.

In terms of directionality differences for the CS conditions, we predict similar performance in both languages for the non-CS stimulus sentences as well as for the CS-N and CS-P+DET+N conditions, where we expect relatively high rates of accuracy. For the three more challenging CS conditions (CS-DET+N, CS-P, and CS-P+DET), however, we do expect directionality differences in SRep accuracy, due to the contrasting properties of the DET in English and Hebrew. More specifically, we expect the following:

Processing a switch between an English DET (a free morpheme) and a subsequent Hebrew N is expected to lead to higher accuracy rates and lower error rates than processing a switch between a Hebrew DET (a bound, cliticized morpheme) and a subsequent English N. Thus, better performance is expected on CS-P+DET in Hebrew sentences than in English sentences.

Processing a Hebrew P followed by an English DET is expected to be easier than processing an English P followed by a Hebrew DET. In an English P+DET+N construction, the P governs the DET, a free morpheme. When, however, the codeswitch is a Hebrew DET (a bound, cliticized morpheme), the expectation of a free-standing complement of the P is not met (Abney, 1987; Borer, 1999; Danon, 2001, 2008; Mahootian & Santorini, 1996). Thus, we expect better performance on CS-P in English sentences than in Hebrew sentences, and better performance on CS-DET+N in Hebrew sentences than in English sentences.

Method

Participants

Sixty-five bilingual children, 32 girls and 33 boys, participated in the study. The children were recruited from preschools in Israel, where the language of instruction was Hebrew. Background information on demographics and language history (see Table 1) were obtained from a parental questionnaire. Children were English/L1-Hebrew/L2 bilinguals who came from English-speaking homes. They had had at least 17 months of exposure to Hebrew.

Table 1.

Background information of the participants.

Characteristics		N
Age	5;10 (5;5–6;5)	65
Gender	Female	33
Gender	Male	32
Exposure to Hebrew	From birth	18
Exposure to Hebrew	After birth (0;3–4;9)	47
Mean length of exposure to Hebrew (years)	4;3 (1;4)
Mean age of onset of Hebrew (years)	1;7 (1;5)
Mother’s education (years)	16.98 (2.81)
English proficiency (z-score)	−0.50 (1.00)
Hebrew proficiency (z-score)	−0.88 (1.40)

All children were administered Raven’s Colored Progressive Matrices non-verbal IQ test (Raven, 1998). Only children who scored within the norm (a standard score above 90) were included in the study. Children with emotional, neurological, visual, oral, or auditory impairments, or recurrent otitis media, were excluded from the study.

The children were screened for language proficiency by means of standardized tests in both languages: the CELF-2 Preschool for English (Wiig et al., 2004) and the Goralnik Screening Test for Hebrew (Goralnik, 1995). The CELF includes sub-tests for sentence structure, word structure, expressive vocabulary, concepts and following directions, SRep, and word classes. The Goralnik test screens for vocabulary, SRep, listening comprehension, expression, storytelling, and pronunciation. Children who scored within the monolingual norms for these tests in at least one of their languages were included in this study. Cut-off scores were set as 1 SD below the monolingual mean in L1 English and 1.5 SD below the monolingual mean in L2 Hebrew. The latter cut-off point conforms to those used in other bilingual studies in Israel with similar age groups, which used 1.5 SD instead of 1 (e.g. Meir et al., 2015). Raw scores were converted to z-scores based on local norms for English-Hebrew bilingual children. Paired-sample t-tests conducted to examine the difference in language proficiency showed no significant difference between z-scores of the English (M = −0.50, SD = 1.00) and Hebrew tests (M = −0.88, SD = 1.40), t(64) = 1.67, p = 0.10, η² = 0.042. This suggests that overall the group did not show different proficiency in the two languages.

Design and materials

The study used a SRep task to assess participants’ ability to codeswitch within PPs. Each stimulus sentence included a NP subject, an intransitive V, a PP consisting of P+DET+N, and a temporal phrase (without a P). The following Ps were included in the stimulus sentences: around (SAVIV), inside (BETOX), over (ME’AL), next to (LEYAD), before (LIFNEY), and after (AXREY). All Ps were bisyllabic in English and Hebrew. Six sentences were created for each P, with a total of 36 stimulus sentences in English and 36 in Hebrew. Each sentence was then rearranged into six switch conditions: a codeswitched N (N), an N switched along with a preceding DET (DET+N), a codeswitched P (P), a P switched along with a subsequent DET (P+DET), a P switched along with a subsequent DET and N (P+DET+N), and a no-switch condition. An example of a set with each of the six conditions in English and Hebrew sentences appears in (1) and (2) respectively.

(1) English stimulus sentences with Hebrew switches

	Subject	Verb	Prepositional phrase	Temporal phrase
Non-CS	The builder	works	inside the house	every morning
CS-N	The builder	works	inside the BAYIT	every morning
CS-DET+N	The builder	works	inside HA- BAYIT	every morning
CS-P	The builder	works	BETOX the house	every morning
CS-P+DET	The builder	works	BETOX HA- house	every morning
CS-P+DET+N	The builder	works	BETOX HA- BAYIT	every morning

(2) Hebrew stimulus sentences with English switches

	Subject	Verb	Prepositional phrase	Temporal phrase
Non-CS	Ha- banay	‘oved	betox ha- bayit	kol boqer
CS-N	Ha- banay	‘oved	betox ha- HOUSE	kol boqer
CS-DET+N	Ha- banay	‘oved	betox THE HOUSE	kol boqer
CS-P	Ha- banay	‘oved	INSIDE ha- bayit	kol boqer
CS-P+DET	Ha- banay	‘oved	INSIDE THE bayit	kol boqer
CS-P+DET+N	Ha- banay	‘oved	INSIDE THE HOUSE	kol boqer

English and Hebrew sentences were matched for semantic content and syntax. Sentences contained early-acquired vocabulary, taken primarily from Hart and Risley (1995) and adapted to the Israeli context based on teachers’, parents’, and children’s feedback and pilot work. In order to minimize cross-linguistic activation effects (Van Hell et al., 2015), no English-Hebrew cognates were used.

Six experimental lists were created for each language with six items for each condition in English and six in Hebrew. Each list contained a total of 36 test items, that is, one sentence from each of the 36 sets. Thus, no participant was presented with the same test sentence in more than one condition. Each stimulus list also contained 24 filler sentences (which were not scored). The filler sentences were unilingual and differed syntactically from the test sentences. The six conditions and filler sentences appeared in the same order on each list. The order was pseudo-random, with sentences from the same CS condition as well as recurrent lexical items not appearing adjacent to one another. Each list began with one filler and one non-switched test sentence. Each participant was randomly assigned to one list in English and one in Hebrew.

Procedure

Approvals from the university Institutional Review Board and Ministry of Education were secured. Participants’ parents were informed about the purpose of the study and asked to sign consent forms. Each child was administered the non-verbal IQ test, the CELF English proficiency, and the Goralnik Hebrew screening test.

Stimuli were read by a fluent bilingual, female adult. They were recorded with an AKG C1000 condenser microphone in a sound-proof recording studio and digitized through an M Audio Delta interface. Recordings were integrated into a PowerPoint© presentation. Participants were tested individually in a quiet room in the kindergarten, where the stimuli were presented via a PowerPoint presentation administered by a laptop computer with audio-headphones. Children were instructed to repeat the sentences verbatim. To familiarize the children with the task, each session began with three practice sentences, with and without a codeswitched item. Children were rewarded with a sticker after each block of 10 sentences. Half of the children were tested first on the English sentence stimuli and then, in a separate session one week later, on the Hebrew stimuli; the other half were tested in Hebrew first. During English testing, all interactions between the tester and the child, including the instructions, took place in English, and vice versa for the Hebrew testing. Each session was audio-recorded using a Toshiba laptop supported with Audacity© software. During the session, responses were also recorded manually.

Transcription, coding, and data analyses

Audio recordings were transcribed and coded off-line. Responses were analyzed in terms of correct repetition of the full sentence (correct/incorrect). Correct responses were given a score of 1 and incorrect responses were given a score of 0. The proportion of correct responses was calculated for each condition (non-CS, CS-N, CS-DET+N, CS-P, CS-P+DET, CS-P+DET+N). For instance, for the CS-P condition, two correct repetitions out of a maximum of six items corresponded to a score of 0.33. A response was coded as incorrect when the participant did not repeat the stimulus sentence verbatim. Incorrect responses included omissions, lexical or grammatical substitutions, insertions, and word order errors. Substitutions included codeswitched elements that were not switched as well as non-elicited codeswitches (switches that were not present in the stimulus sentence). Dysfluencies such as hesitations, false starts, and repetitions were not counted as errors, nor were minor mispronunciations (e.g. kintchen for kitchen).

Responses were also scored for the total number of errors and the number of CS errors in the PP. The proportion of CS errors out of the total number of errors was calculated for each condition in order to determine if the errors were primarily CS errors (i.e. modifications changing English to Hebrew, or vice versa).

In order to assess the effects of CS condition (non-CS, CS-N, CS-DET+N, CS-P, CS-P+DET, CS-P+DET+N) and Language (English versus Hebrew stimulus sentences) on repetition accuracy, repeated-measures analyses of variance (ANOVAs) were conducted. Bonferroni-corrected post hoc tests were computed to test for significant differences. Errors were analyzed by non-parametric Friedman ANOVAs and post hoc Wilcoxon signed-rank tests.

Results

Findings are presented first for accuracy in each of the six conditions, then for directionality effects. The results section concludes with error analyses.

Sentence repetition accuracy

Table 2 presents the mean percentages for the full set of correctly repeated sentences for the six CS conditions for the English and Hebrew stimulus sentences. Figure 1 displays these accuracy scores graphically.

Table 2.

Mean percentages (SD) of correctly repeated sentences for the non-codeswitching (CS), CS-noun (N), CS-determiner (DET)+N, CS-preposition (P), CS-P+DET, and CS-P+DET+N conditions for English and Hebrew stimulus sentences.

CS condition	English	Hebrew
CS condition	N = 65	N = 65
Non-CS	0.62 (0.30)	0.64 (0.25)
CS-N	0.59 (0.30)	0.51 (0.27)
CS-DET+N	0.11 (0.14)	0.28 (0.26)
CS-P	0.37 (0.29)	0.15 (0.18)
CS-P+DET	0.31 (0.30)	0.43 (0.29)
CS-P+DET+N	0.48 (0.28)	0.52 (0.30)

Figure 1.

Mean percentages of correctly repeated sentences for six codeswitching (CS) conditions (non-CS, CS-noun (N), CS-determiner (DET)+N, CS-preposition (P), CS-P+DET, CS-P+DET+N) for English (a) and Hebrew (b).

To compare accuracy scores for the six CS conditions, a repeated-measures ANOVA was conducted with CS condition (non-CS, CS-N, CS-DET+N, CS-P, CS-P+DET, CS-P+DET+N) and Language (English, Hebrew) as within-subject factors. The ANOVA yielded a significant main effect for CS condition, F(5,320) = 106.72, p < .001, η² = .63 and a significant interaction between CS condition and Language, F(5,320) = 15.16, p < .001, η² = .19. The main effect for Language was not significant, F(1,64 = 0.08, p = .778, η² = .001.

In order to clarify the interaction between CS condition and Language, separate repeated-measures ANOVAs were conducted for English and Hebrew stimulus sentences on SRep accuracy with CS condition as a within-subject factor.

For the English sentences there was a significant main effect, F(5,320) = 61.33, p < .001, η² = .49. Bonferroni-corrected post hoc tests (p < .05/15 = .003) showed that all but two comparisons (non-CS/CS-N and CS-P/CS-P+DET) resulted in significant t-values (p < .003). At the upper end of accuracy, performance for the non-CS condition and for the CS-N condition were significantly better than for each of the other conditions (t values ranging from 3.25 to 15.52, df = 64, p < .003). There were no significant differences between the non-CS and CS-N conditions (t = 0.83, df = 64, p = .409). At the lower end of performance, CS-DET+N differed significantly from each of the other conditions (t values ranging from 6.68 to 15.52, df = 64, p < .001). Performance for the CS-P condition was significantly worse than performance for the CS-P+DET+N condition (t = 3.19, df = 64, p = .002), but not significantly different from the CS-P+DET condition (t = 1.45, df = 64, p = .153). Performance for the CS-P+DET+N condition, however, was significantly better than for the CS-P+DET (t = 4.86, df = 64, p < .001). In summary, accuracy scores were highest for the non-CS and CS-N conditions, followed by CS-P+DET+N. The CS-DET+N condition resulted in the lowest scores. These results are illustrated graphically in Figure 1(a).

For the Hebrew stimulus sentences, a repeated-measures ANOVA, too, showed a significant main effect for CS condition, F(5,320) = 52.62, p < .001, η² = .45. Bonferroni-corrected post hoc tests showed that performance for the non-CS condition was significantly better than all the other CS conditions (t values ranging from 3.32 to 15.47, df = 64, p < .003). At the weaker end of accuracy, performance for the CS-P condition was significantly worse than each of the other conditions (t values ranging from 3.64 to 15.47, df = 64, p < .003). With the exception of the CS-P condition, performance for the CS-DET+N condition was also significantly worse than for all the other conditions (t values ranging from 3.64 to 9.92, df = 64, p < .001). Performance for the CS-N and CS-P+DET+N was higher than for the CS-P+DET condition, the differences not reaching significance after Bonferroni corrections (t(64) = 2.40, p = .019 and t(64) = 2.81, p = .007, respectively). These results are illustrated graphically in Figure 1(b).

In summary, English performance on the non-CS and CS-N conditions was best, followed by CS-P+DET+N. The CS-DET+N condition resulted in the lowest accuracy scores. In Hebrew performance was significantly better on the non-CS condition than on all other conditions; CS-N and CS-P+DET+N were next highest in accuracy; and the CS-P elicited significantly lower scores than all the other conditions. Finally, the P+DET condition resulted in lower performance than the ‘easier’ CS-N and CS-P+DET+N condition in English, but not in Hebrew. The next section compares the findings for English to Hebrew (English stimuli) and Hebrew to English (Hebrew stimuli) CS.

Directionality

In order to assess directionality effects indicated by the significant interaction between Language and CS condition, Bonferroni-corrected paired-sample t-tests (p < .05/6 = .008) were conducted comparing accuracy on English and Hebrew stimulus sentences. No significant differences emerged for three conditions: non-CS, CS-N, and CS-P+DET+N. The non-significant difference found for the unilingual (non-CS) sentences (English, M = 0.62, SD = 0.30 and Hebrew, M = 0.64, SD = 0.25, t(64) = 0.57, p = .57, η² = .005) supports the data from the proficiency measures in English and Hebrew, arguing for similar proficiency in the two languages. The CS-N and CS-P+DET+N conditions did not show significant differences between the language pairs either, English (M = 0.59, SD = 0.30) and Hebrew (M = 0.51, SD = 0.27), t(64) = 1.78, p = .08, η² = .05, and English (M = 0.48, SD = 0.28) and Hebrew (M = 0.52, SD = 0.30), t(64) = 0.83, p = .409, η² = .01, respectively. Likewise, a paired-sample t-test comparing accuracy scores for all five CS conditions combined yielded no significant difference between English CS (M = 0.37, SD = 0.20) and Hebrew CS (M = 0.38, SD = 0.20), t(64) = 0.11, p = .92, η² = 0).

On the other hand, the t-tests showed significant differences between English CS-P (M = 0.37, SD = 0.29) and Hebrew CS-P (M = 0.15, SD = 0.18), t(64) = 6.02, p < .001, η² = .37) and between English CS-DET+N (M = 0.11, SD = 0.14) and Hebrew CS-DET+N (M = 0.28, SD = 0.26), t(64) = 5.07, p < .001, η² = .29). The CS-P+DET condition showed a near-significant difference between English, M = 0.31, SD = 0.30, and Hebrew, M = 0.43, SD = 0.29, t(64) = 2.64, p = .01, η² = .10.

Error analyses

For each CS condition, both non-CS errors and CS errors were examined in order to assess the extent to which the errors supported the accuracy data as well as what they could tell us about the findings for CS directionality. A CS error was defined as a codeswitched word or phrase within the PP from English to Hebrew or vice versa (e.g. inside the XEDER for the target inside HA- XEDER). A larger proportion of CS errors was expected to reflect greater processing costs, since it indicates that the children ‘modified’ the language of the target item.

Table 3 displays the mean frequencies and percentages of CS and non-CS errors for each of the six conditions for English and Hebrew stimuli. As can be seen in Table 3, the conditions with the highest accuracy scores (non-CS, CS-N, and P+DET+N) had low proportions of CS errors and high proportions of non-CS errors (60–95% of the errors for these conditions were non-CS errors). These high-accuracy conditions all showed the same error patterns, viz. more non-CS than CS errors.

Table 3.

Mean frequencies (SD) and proportions (SD) of non-codeswitching (CS) and CS errors for the non-CS, CS-noun (N), CS-determiner (DET)+N, CS-preposition (P), CS-P+DET, and CS-P+DET+N conditions for English and Hebrew prepositional phrases.

CS condition	English non-CS errors	English CS errors	Total English errors	Hebrew non-CS errors	Hebrew CS errors	Total Hebrew errors
Non-CS
Frequency	4.03 (4.81)	0.08 (0.27)	4.11 (4.86)	3.28 (2.89)	0.02 (0.12)	3.29 (2.93)
Proportion	0.98 (0.15)	0.02 (0.15)		0.9999 (0.01)	0.0001 (0.01)
CS-N
Frequency	3.77 (4.35)	0.46 (0.69)	4.23 (4.58)	3.20 (3.05)	1.11 (1.24)	4.31 (3.41)
Proportion	0.89 (0.20)	0.11 (0.20)		0.74 (0.33)	0.26 (0.33)
CS-DET+N
Frequency	3.58 (3.70)	4.98 (1.04)	8.56 (3.90)	3.09 (2.25)	3.59 (1.57)	6.68 (3.16)
Proportion	0.42 (0.22)	0.58 (0.22)		0.46 (0.23)	0.54 (0.23)
CS-P
Frequency	3.92 (4.44)	2.45 (1.58)	6.37 (4.84)	3.46 (3.06)	4.49 (1.38)	7.95 (3.49)
Proportion	0.62 (0.26)	0.38 (0.26)		0.44 (0.24)	0.56 (0.24)
CS-P+DET
Frequency	3.09 (3.57)	3.25 (1.85)	6.34 (4.67)	3.15 (2.66)	2.28 (1.71)	5.43 (3.86)
Proportion	0.49 (0.29)	0.51 (0.29)		0.58 (0.28)	0.42 (0.28)
CS-P+DET+N
Frequency	3.08 (1.32)	1.78 (1.32)	4.86 (4.05)	3.06 (2.60)	1.49 (1.38)	4.55 (3.36)
Proportion	0.63 (0.25)	0.37 (0.25)		0.67 (0.30)	0.33 (0.30)
Total
Frequency	21.48 (21.03)	13.00 (4.50)	34.48 (23.45)	19.25 (12.87)	12.97 (4.85)	32.22 (16.43)
Proportion	0.62 (0.17)	0.38 (0.17)		0.55 (0.14)	0.45 (0.14)

In contrast, the conditions with the lowest accuracy scores (CS-DET+N in both English and Hebrew stimulus sentences, and CS-P in Hebrew sentences) showed a high proportion of CS errors and a low proportion of non-CS errors. Moreover, for the CS-P condition in the English to Hebrew CS direction, non-CS errors were more plentiful than CS errors (M = 3.92 versus M = 2.45 for non-CS and CS errors, respectively). For the CS-P+DET condition, as for the low-accuracy conditions, there was a relatively high proportion of CS errors in English to Hebrew CS (M = 3.25 for CS errors versus M = 3.09 for non-CS errors). However, the P+DET condition in Hebrew to English CS showed an error pattern similar to those in the high-accuracy conditions (M = 2.28 for CS errors versus M = 3.15 for non-CS errors).

A Friedman ANOVA performed on the error data showed a significant difference in the proportion of CS errors among the six CS conditions for English sentences, (χ²(5) = 149.16, p < .001). Bonferroni-corrected post hoc Wilcoxon signed-rank tests (p < .05/15 = .003) indicated that the proportion of CS errors for the non-CS condition and CS-N condition were significantly lower than for the other four conditions. For the conditions with the larger numbers of errors, the CS-DET+N condition showed significantly more errors than four of the other five conditions (apart from CS-P+DET). These results show that for English stimuli, errors in the low-accuracy conditions involved more CS than in the high-accuracy conditions.

For the Hebrew stimuli, too, the Friedman ANOVA showed significant differences in the proportion of CS errors among the six CS conditions (χ²(5) = 124.11, p < .001). Bonferroni-corrected post hoc Wilcoxon signed-rank tests indicated that the proportion of CS errors for the non-CS condition was significantly lower than for all other conditions. The proportions of CS errors for CS-N and CS-P+DET+N were significantly lower than for CS-P and CS-DET+N. For the low-accuracy/high error rate conditions (CS-DET+N and CS-P), there were significantly more CS errors: CS-DET+N had a significantly higher proportion of CS errors than all other conditions (apart from CS-P), and the proportion of CS errors for the CS-P condition was significantly higher than all the other conditions (apart from CS-DET+N). There were no significant differences among CS-P+DET, CS-P+DET+N, and CS-N.

In summary, these findings indicate that for the stimulus sentences in both languages, errors in the low-accuracy conditions (DET+N and CS-P in Hebrew) involved more CS errors, while for the high-accuracy conditions (non-CS, CS-N, and CS-P+DET+N) there were fewer CS errors.

To complement the data based on percent accuracy and errors, we examined the CS modifications in the low-accuracy conditions, in English for the CS-DET+N and CS-P+DET condition and in Hebrew for the DET+N and P conditions. For the English DET+N sentences, children replaced a Hebrew DET (inside HA- XEDER) with an English DET (inside the XEDER). This pattern comprised more than 80% of their CS modifications, showing that the children avoided combining an English P with a Hebrew DET. The most frequent modifications for the CS-P+DET condition for the English stimulus sentences were (1) not switching the P+DET to Hebrew ( inside the bedroom instead of the elicited BETOX HA- bedroom), and (2) not switching the DET to Hebrew, instead repeating it in English, the language of the stimulus sentence (BETOX the bedroom). Together these two modifications constituted close to 50% of the repetition errors, showing that the children avoided the juxtaposition of a Hebrew DET with an English N by changing the Hebrew DET to English.

For the Hebrew CS-P sentences, children replaced a Hebrew DET (inside HA- XEDER) with an English DET (inside the XEDER), or they replaced an English P ( inside HA- XEDER) with a Hebrew P (BETOX HA- XEDER). These two modifications together comprised 80% of the children’s repetition errors, again suggesting that participants consistently avoided the juxtaposition of an English P with a Hebrew DET. For the CS-DET+N condition, children replaced the English DET (BETOX the bedroom) with a Hebrew DET (BETOX HA- bedroom) in over 80% of their modifications, demonstrating once more that switching a DET+N phrase is more difficult than switching just the N.

In summary, the error data corroborated the findings for accuracy, showing fewer errors for the non-CS, N, and P+DET+N conditions and more errors for the low-accuracy conditions. In addition, the proportion of CS errors was found to be higher for the low-accuracy conditions. These CS errors show avoidance of an English P with a Hebrew DET, avoidance of a Hebrew DET with an English N, and preference for switching N over switching DET+N. Error patterns also yielded important information about directionality: both the P condition and the P+DET condition showed contrasting error patterns for English and Hebrew. These asymmetries are examined in the Discussion section.

Discussion

The present study examined how preschool children process codeswitched sentences manipulated for the locus of the codeswitch within a PP and for CS directionality. The main findings, based on accuracy scores, error rates and the proportion of CS in PPs, showed the following.

Structural effects:

highest accuracy and lowest error rates for non-CS, N, and full PP (P+DET+N) switches;

lowest accuracy and highest error rates for DET+N switches.

Directionality effects:

lowest accuracy for a DET switched with N in English sentences and for a P switch in Hebrew sentences;

lower accuracy for CS N in Hebrew sentences, but not for the same structure in English sentences;

lower accuracy for CS a P+DET in English sentences than in Hebrew.

Quantitative and qualitative analyses of CS errors yielded important insights for both structure and directionality. The proportion of CS errors was higher for the low-accuracy conditions (CS-DET+N, CS-P in Hebrew, and CS-P+DET in English). Structurally, CS errors showed avoidance of an English P with a Hebrew DET, avoidance of a Hebrew DET with an English N, and preference of switching N over DET+N. In terms of directionality, the P+DET condition showed higher accuracy and fewer CS errors in Hebrew while the P condition showed lower accuracy and more CS errors in Hebrew.

The structural effects are discussed in terms of a hierarchy for processing structural features and the directionality differences based on contrasting features of English and Hebrew.

Grammatical constraints on CS in both languages

The aggregated data for both languages in the present study suggest a hierarchy in the cost of CS for grammatical switch points. Single Ns and full PPs are switched with lower cost in comparison to DETs and Ps. The high frequency of single N switching has been widely attested in corpus data and is extensively discussed in the CS literature (e.g. Halpin & Melzi, 2018; Myers-Scotton & Jake, 2014). The ease of single N switching may be attributed to several factors. First and foremost, since Ns in the two languages in question do not require morphosyntactic integration, they are grammatically less tied to a specific syntactic environment (Marian, 2009). Ns, unlike Ps, receive but do not assign semantic roles. Furthermore, since Ns are the most frequent lexical category, they provide more opportunities for CS than other constituents. Thus, the need to codeswitch for lack of word knowledge or the sense that something may be expressed better in a different language arises more frequently with Ns than other constituents (cf. Matras, 2009). Finally, in line with a usage-based approach, it has been argued that the more common a particular codeswitch is in natural discourse, the easier it is to activate and process (Dussias, 2001; Guzzardo Tamargo et al., 2016; Kheder & Kaan, 2016). Accordingly, Ns, which represent approximately 50% of most spoken corpora, should be relatively easily activated and processed.

The results also showed that switching the entire PP elicited relatively minimal cost. This finding aligns with our prediction that when the PP is codeswitched in its entirety, there would be no additional processing demand for integration into a sentence in the other language. Switching of PPs is attested in CS corpora data (Clyne, 1987; Hakimov, 2016; Zentella, 1997). Poplack (1980) noted that 60% of codeswitches in her data were intact syntactic constituents, such as NPs, PPs, or adverbial phrases. In Poplack’s view such constituents can be switched because they do not demand ‘a syntactic rule for generating the constituent’ that might not be ‘shared by both L1 and L2’ (p. 603).

A third finding that resulted for both languages was the relatively high cost of switching of a DET+N. This result contradicts the MLF/4-M models (e.g. Myers-Scotton & Jake, 2000), which suggests that DET+N phrases can be part of the EL as independent constituents (islands). This claim should make DET+N switches easier than P+DET switches. However, the high cost of DET+N (as well as P+DET) switching does find support in a variety of studies. For example, Myers-Scotton’s (1993) MLF model postulates that DETs set the morphosyntactic frame of the sentence and they are typically supplied by the ML. Different roles for functional (e.g. DETs) and lexical elements (e.g. Ns) in CS has also been argued elsewhere (Dussias, 2002; Jake et al., 2002; Smolak et al., 2020). Jake et al. (2002) account for this asymmetry by showing that if DETs were supplied by the EL, the ML would become less ‘transparent’ (p. 87). Alternatively, it may be the case that DETs are less involved in CS, since they do not reflect difficulties in lexical access. It follows, then, that the cost of processing such an infrequently encountered switch type would be relatively high. Regarding the lower cost for P+DET switches in comparison to DET+N, we offer two explanations. One possibility is that for these two switches, the children process the PP as a separate entity headed by the P. Thus, removing the embedded DET+N from the PP entails a higher processing cost than switching just the N (as in the CS-P+DET condition). A similar interpretation can be related to priming. When the child begins a sentence in one language, the default option would be to continue in the same language. The DET+N codeswitch appears relatively late in the stimulus sentence, after the child has already encountered six syllables and four grammatical categories in language X, making the switch to language Y particularly challenging.

In summary, switching single Ns involves little cost due to their semantic role and transferability, and their status as the most frequently codeswitched lexical category. Switching the entire PP is also relatively free of processing costs, since it does not involve manipulation of grammatical elements from two linguistic systems. Finally, we proposed two possible interpretations for the fact that DET+Ns are switched with difficulty and with error.

The hierarchical CS costs reported here are not supported by any one CS model. For instance, the finding that a P switch was more difficult than a single N switch supports Joshi’s (1985) constraint against switching closed class items and Di Sciullo et al.’s (1986) Government Constraint, but it does not follow from the MLF/4-M model prediction that content morpheme Ps should be easily codeswitched (e.g. Myers-Scotton & Jake, 2009). This model cannot account for the higher CS error rate in CS-P as compared to CS-P+DET either. Moreover, better performance for P+DET than DET+N switching is not explained by any of these models. This is further pursued below.

Directionality differences

Cross-language directionality differences for CS emerged for four conditions, one with a relatively low switch cost (CS-N) and three conditions with higher costs (CS-DET+N, CS-P, and CS-P+DET). While for single N switches, performance was similar for English to Hebrew and for Hebrew to English, the two languages showed a distinct pattern of differences between the CS-N condition and other conditions. For DET+N and P+DET switches, accuracy scores were better for Hebrew to English switching; and for the single P switch condition, accuracy was higher for English to Hebrew switching.

N switching resulted in relatively little cost and few CS errors. However, for Hebrew sentences, the CS-N condition resulted in lower accuracy than the non-CS condition and more CS errors, while for English sentences it did not. Two complementary explanations for this directionality difference are offered, one based on morphosyntactic differences between English and Hebrew, and one related to sociolinguistic factors. In English the DET the is an unbound morpheme, while in Hebrew the DET HA- is a bound morpheme, cliticized to the ensuing N. While in an English NP, intervening modifiers can separate the DET from the N, in Hebrew the DET must attach directly to the N (Danon, 2001; Wintner, 2000). The stronger ‘bond’ between the Hebrew clitic and its N, as compared to the English article, should make CS Ns into English more difficult, since this involves ‘detaching’ the N from the DET. This suggestion conforms with Poplack’s (1980) Free Morpheme Constraint, which precludes CS between free and bound morphemes. The asymmetry here between English to Hebrew and Hebrew to English switching supports the notion that typological differences of the two languages are crucial in processing codeswitched sentences.

Sociolinguistically, a switch into Hebrew is a switch into the societal language. The children in the present study grow up in a society where Hebrew is the language of day-to-day communication. Hebrew is also the language of their preschools, and often older siblings bring Hebrew into their English-speaking homes, where they may hear more or less CS, but usually from English to Hebrew. Thus, for these children, CS into Hebrew is more likely than CS into English. The finding is in line with other studies with sequential bilingual preschool children in Israel, which found more CS into the societal language Hebrew (Iluz-Cohen & Walters, 2012, for English-Hebrew bilinguals; Raichlin et al., 2019, for Russian-Hebrew bilinguals), as well as other research reporting on CS preference in the direction of the societal language (Adamou & Shen, 2019, for Romani-Turkish bilinguals; Dussias, 2001, and Gutiérrez-Clellen et al., 2009, for Spanish-English bilinguals). On the other hand, the sociolinguistic interpretation must be read with caution given the finding that CS in the other high-accuracy condition (P+DET+N), where the entire PP was switched, did not result in a cross-language difference, neither in accuracy scores, nor in error rate or the number of CS errors. It should be noted that proficiency differences cannot account for the directionality effect here either, since the participant group did not show evidence for higher proficiency in one of the two languages.

Finally, the cross-linguistic difference for N switching also has implications for the notion of a processing cost in CS. The fact that switching a single N from English to Hebrew did not result in lower performance than the unilingual, non-switched condition implies that not all CS entails a cognitive cost due to cross-language competition and interference.

At the lower end of accuracy, a cross-linguistic difference emerged for two of the high-cost CS conditions. For English stimulus sentences, lowest accuracies were found for the DET+N condition; for Hebrew sentences the CS-P condition showed the lowest level of accuracy. These two conditions resulted in scores that were significantly lower than all other conditions, and they resulted in the highest proportion of CS errors. We propose a single interpretation for both findings. Both switches involved difficulty in processing a Hebrew NP when it follows an English P. The Hebrew NP consists of the DET HA-, a bound morpheme, cliticized to the ensuing N. However, the English P is collocated with a free morpheme, the DET the, which is absent in both of these CS conditions:

English to Hebrew CS of DET+N

the girl played inside HA- XEDER all day

‘The girl played inside the bedroom all day’

Hebrew to English CS of P

HA- YALDA SIXQA inside HA- XEDER KOL HA- YOM

‘The girl played inside the bedroom all day’

The missing element (a free-standing DET) is what we argue makes processing these sentences particularly difficult.

This interpretation is also in line with the patterns of CS errors/modifications found in the present study. The vast majority of modifications were produced for the English DET+N condition and the Hebrew CS-P condition, showing that participants consistently avoided juxtaposition of an English P with a Hebrew DET, instead maintaining the P and the adjacent DET in a single language. This interpretation means that the connection between an English P and an English DET is stronger than the link between a Hebrew P and a Hebrew DET, a conclusion supported in the data, since the DET+N condition in English sentences (which contain an English P and a Hebrew DET+N) resulted in significantly higher switching costs than the Hebrew DET+N condition. This conclusion is also supported by the finding that for the CS-P condition in English sentences (which has a Hebrew P and an English DET), there were fewer errors than for the same condition in Hebrew, and the finding that the proportion of CS errors was lower than non-CS errors. This suggests less difficulty with switching between a Hebrew P and an English DET than between an English P and a Hebrew DET.

This interpretation also means that an account based on contrasting structural properties of the two languages prevails over accounts based on (a) Government (Di Sciullo et al., 1986), (b) contrasting roles for DET and N in CS (Jake et al., 2002), and (c) priming from the beginning of the sentence, since these accounts cannot explain the lower cost for the DET+N condition in Hebrew.

It could be argued that like the English P, the Hebrew P should also collocate with a Hebrew DET. We propose, however, that for Hebrew this result is less likely. Firstly, in Hebrew the P more frequently selects a DET-less NP than in English, since Hebrew does not have an indefinite article, and therefore indefinite NPs are not overtly marked. Moreover, the expectation of a Hebrew P selecting any particular type of definite article (cliticized or free-standing) is also lower than in the case of an English P due to the property of definiteness in Hebrew, which is a feature of the N (Borer, 1999; Danon, 2001, 2008). In line with a usage-based approach, it follows that the processing cost of a Hebrew P ‘lacking’ a Hebrew DET should be lower than the cost of an English P ‘lacking’ an English DET (Beatty-Martínez et al., 2018; Dussias, 2001).

The finding for the P+DET condition showed directionality differences, too, that is, greater difficulty for switching between a Hebrew DET and an English N in English sentences [the mouse ran SAVIV HA- table] than for an English DET and a Hebrew N in Hebrew sentences (HA- AXBAR RAC around the SHULXAN). As explained above, the tight connection between a Hebrew clitic and its subsequent N as compared to the looser connection between an English free DET and its N make the combination HA- table more costly than the SHULXAN, which does not involve ‘detaching’ the N from the DET.

CS models that bar switching of a particular constituent or between constituents, such as P, P+DET, and DET+N (e.g. Di Sciullo et al., 1986; Joshi, 1985; Myers-Scotton & Jake, 2009), cannot explain the cross-language differences found here for these constituents, viz. better performance for switching N and switching P within an English frame, and better performance for switching P+DET and DET+N within a Hebrew frame. These models do not have a mechanism to account for typological differences; nevertheless, the cross-language differences for P+DET and N switching are supported by Poplack’s (1980) Free Morpheme Constraint. Likewise, Mahootian’s (1996) proposal that the language of a constituent head determines the syntactic properties of its complement matches the attested difficulty of switching between an English P and a Hebrew DET, but it cannot explain the hierarchy in CS costs reported in the present study. We opt for an account based on the contrasting grammatical properties of the two languages, specifically the difference in definiteness marking in English and Hebrew. Due to the relatively strong bond between the Hebrew clitic HA- and its ensuing N, more difficulty is encountered in CS between these two elements than between an English DET and N, which are both free morphemes. Similarly, switching between an English P and a Hebrew cliticized DET+N complement ( inside HA- XEDER) is more challenging than switching between a Hebrew P and an English NP DET+N (BETOX the bedroom), since the syntactic requirement of the English P, which calls for a DET+N complement headed by a free morpheme (the), is not fulfilled.

Methodological issues, limitations, and future research

The present study operationalized the notion of CS costs in terms of accuracy and error rates in the investigation of children’s CS by means of a SRep task. The experimental approach adopted here pushed structural knowledge to its limits by presenting unlikely, sometimes ungrammatical, codeswitched structures to preschool children whose circumscribed memory capacity and naïve approach to the task allowed natural responses to the stimuli.

However, the study focused only on PPs and relied only on accuracy scores and errors. Additional studies should examine other grammatical structures and make use of response times and imaging data as complementary measures. Response time measures may be more sensitive to differences between the unilingual and CS conditions as well as among the various CS conditions. Similarly, comparison of SRep data with acceptability judgment and sentence correction tasks would be important tests for the CS hierarchy and cross-linguistic differences reported here. In addition, it would be edifying to conduct a study where the participants do not anticipate CS as much as in the present experiment, where 50% of the stimulus sentences contained a codeswitch and where the participants were instructed to repeat the codeswitches verbatim. This could be achieved by increasing the number of unilingual stimulus sentences and by not explicitly mentioning CS during the instructions. Furthermore, individual analyses of participants would be useful for assessing the relative contribution of language exposure and proficiency in CS. Finally, a further avenue for future research would be to replicate the study with L1/Hebrew-L2/English bilingual children residing in an English-speaking environment. This could shed more light on directionality effects, and particularly the question as to whether CS from English to Hebrew is facilitated by sociolinguistic as well as grammatical factors.

The relatively low repetition accuracy overall, even for the high-accuracy conditions, resulted in scores of no more than 64% correct. By way of explanation, the stimulus sentences all contained PPs, which are known to be difficult for bilingual children and adults alike (e.g. Armon-Lotem, 2012; Song & Sardegna, 2014). The relatively low overall accuracy may also be a function of the way correct repetition was scored. The analysis considered any deviation from the stimulus sentence as incorrect, including lexical-semantic, morphosyntactic, CS, word order, and insertion errors.

Summary and conclusions

By experimentally manipulating the switch site in PPs and directionality, the present study examined bilingual processing in preschool children. Experimental research, albeit less authentic, is crucial for clarifying claims about CS, since it can test hypotheses that may be impossible or take years to clarify based on examples selected from corpus studies, natural data, and negative evidence.

The study found strong evidence for a hierarchy in processing costs for different grammatical structures in PPs. The structural elements that involve less morphosyntactic integration resulted in lower CS costs. Those structures that involve more integration resulted in higher processing costs. Beyond the hierarchy for structures, the data also showed an interaction between structural features and CS directionality. This interaction implies that the grammatical properties of the two languages have to be taken into consideration when accounting for CS costs. The CS costs documented here imply that the notion of CS cost is not exclusively attributable to cognitive issues, such as language interference, competition, and inhibition. In other words, not all CS results in a cognitive cost. Therefore, structural features and directionality should be taken into account when examining the cost of CS.

The findings here do not support any single CS model. Pieces of the data conform to some models, but not others. We did not set out to provide an overall account of CS. Rather, we have focused on CS in PPs, known to be controversial in the CS literature and challenging for bilingual children as well as adults. We would argue that since an account of structural and directionality aspects of CS should relate to typological differences (e.g. Poplack, 1980) and incompatible features between participating languages, overarching linguistic approaches to CS are unlikely to bear fruit, as others have noted (Green & Wei, 2014; López, 2018; Vihman, 2018). The present paper focused largely on structural aspects of CS, but we would recommend taking some of the open issues raised here in the direction of more socio-pragmatic and usage-based approaches (Green & Wei, 2014; Wasserscheidt, 2014).

Footnotes

Acknowledgements

We would like to thank the preschool children for their participation, and Gabi Danon and two anonymous reviewers for their valuable comments and suggestions.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: a portion of this research was supported by the Israel Science Foundation (grant No.779/10).

ORCID iD

Aviva Soesman

Author biographies

Aviva Soesman is a lecturer at the Division of Languages, Tel Aviv University, Israel. Her current research focuses on codeswitching among typically developing and language impaired bilingual children, and language teaching in higher education.

Joel Walters research in bilingualism bridges sociopragmatics and psycholinguistics. His present work focuses on narratives and social identity in a variety of contexts, including bilingual children with typical language development and SLI, bilingual adults recovering from aphasia and narrative intervention with a variety of populations. His research has been funded by the Israel Science Foundation, the German-Israel Research Foundation (GIF), the German Ministry of Education (BMBF) and the Israel Ministry of Education. He is Professor Emeritus from Bar-Ilan University and currently head of the M.Ed. program in English as an International Language at Talpiot College.

References

Abney

(1987). The English noun phrase in its sentential aspect [Doctoral dissertation]. Massachusetts Institute of Technology.

Adamou

Shen

R. X.

(2019). There are no language switching costs when codeswitching is frequent. International Journal of Bilingualism, 23(1), 53–70.

Altman

Armon-Lotem

Fichman

Walters

(2016). Macrostructure, microstructure, and mental state terms in the narratives of English–Hebrew bilingual preschool children with and without specific language impairment. Applied Psycholinguistics, 37(1), 165–193.

Armon-Lotem

(2012). Between L2 and SLI: Inflections and prepositions in the Hebrew of bilingual children with TLD and monolingual children with SLI. Journal of Child Language, 41(1), 3–33.

Azuma

Meier

R. P.

(1997). Open class and closed class: Sentence-imitation experiments on intrasentential code-switching. Applied Psycholinguistics, 18, 257–276.

Basnight-Brown

D. M.

Altarriba

(2007). Code-switching and code-mixing in bilinguals: Cognitive, developmental, and empirical approaches. In Ardila

Ramos

(Eds.), Speech and language disorders in bilinguals (pp. 69–89). Nova Science Publishers, Inc.

Beatty-Martínez

A. L.

Valdés Kroff

J. R.

Dussias

P. E.

(2018). From the field to the lab: A converging methods approach to the study of codeswitching. Languages, 3(2), 19.

Belazi

H. M.

Rubin

E. J.

Toribio

A. J.

(1994). Code switching and X-bar theory: The functional head constraint. Linguistic Inquiry, 25(2), 221–237.

Bishop

D. V. M.

(1989). Test for reception of grammar (TROG). Medical Research Council.

10.

Bolonyai

(2009). Code-switching, imperfect acquisition, and attrition. In Bullock

B. E.

Toribio

A. J.

(Eds.), The Cambridge handbook of linguistic code-switching (pp. 253–269). Cambridge University Press.

11.

Borer

(1999). Deconstructing the construct. In Johnson

Roberts

(Eds.), Beyond principles and parameters (pp. 43–89). Kluwer Academic Publishers.

12.

Bosma

Blom

(2019). A code-switching asymmetry in bilingual children: Code-switching from Dutch to Frisian requires more cognitive control than code-switching from Frisian to Dutch. International Journal of Bilingualism, 23(6), 1431–1447.

13.

Cantone

K. F.

(2005). Evidence against a third grammar: Code-switching in Italian-German bilingual children. In Cohen

McAlister

Rolstad

MacSwan

(Eds.), ISB4: Proceedings of the 4th international symposium on bilingualism (pp. 477–496). Cascadilla Press.

14.

Cantone

K. F.

(2007). Code-switching in bilingual children. Springer.

15.

Chan

(2008). Code-switching, word order and the lexical/functional category distinction. Lingua, 118(6), 777–809.

16.

Chan

M. C.

Chau

H. L. C.

Hoosain

(1983). Input/output switch in bilingual code switching. Journal of Psycholinguistic Research, 12(4), 407–416.

17.

Chiat

Armon-Lotem

Marinis

Polišenská

Roy

Seeff-Gabriel

(2013). Assessment of language abilities in sequential bilingual children: The potential of sentence imitation tasks. In Mueller-Gathercole

(Ed.), Issues in the assessment of bilinguals (pp. 56–89). Multilingual Matters.

18.

Clyne

(1987). Constraints on code–switching – How universal are they? Linguistics, 25, 739–764.

19.

Conti-Ramsden

(2003). Processing and linguistic markers in young children with Specific Language Impairment (SLI). Journal of Speech, Language, and Hearing Research, 46(5), 1029–1037.

20.

Danon

(2001). Syntactic definiteness in the grammar of Modern Hebrew (2001). Linguistics, 39(6), 1071–1116.

21.

Danon

(2008). Definiteness spreading in the Hebrew construct state. Lingua, 118, 872–906.

22.

Davidson

M. C.

Amso

Anderson

L. C.

Diamond

(2006). Development of cognitive control and executive functions from 4 to 13 years: Evidence from manipulations of memory, inhibition and task switching. Neuropsychologia, 44(11), 2037–2078.

23.

Di Sciullo

Muysken

Singh

. (1986). Government and code-mixing. Journal of Linguistics, 22, 1–24.

24.

Dussias

P. E.

(2001). Psycholinguistic complexity in codeswitching. International Journal of Bilingualism, 5(1), 87–100.

25.

Dussias

P. E.

(2002). On the relationship between comprehension and production data in codeswitching. In Wiltshire

Camps

(Eds.), Romance phonology and variation (pp. 27–39). John Benjamins.

26.

Dussias

P. E.

Courtney

(1995). Qué es un good-codeswitch? Testing the Functional Head Constraint within noun phrases. Working papers of the interdisciplinary Ph.D. program in second language acquisition and teaching. University of Arizona.

27.

Ebert

Kohnert

(2016). Language learning impairment in sequential bilingual children. Language Teaching, 49(3), 301–338.

28.

Genesee

Nicoladis

Paradis

(1995). Language differentiation in early bilingual development. Journal of Child Language, 22(3), 611–631.

29.

Gollan

T. H.

Goldrick

(2016). Grammatical constraints on language switching: Language control is not just executive control. Journal of Memory and Language, 90, 177–199.

30.

Goralnik

(1995). Goralnik diagnostic test. Matan (in Hebrew).

31.

Green

D. W.

Wei

(2014). A control process model of code-switching. Language, Cognition and Neuroscience, 29(4), 499–511.

32.

Greene

K. J.

Peña

E. D.

Bedore

L. M.

(2012). Lexical choice and language selection in bilingual preschoolers. Child Language Teaching and Therapy, 29(1), 27–39.

33.

Gross

Kaushanskaya

(2015). Voluntary language switching in English-Spanish bilingual children. Journal of Cognitive Psychology, 27(8), 992–1013.

34.

Gross

M. C.

Lopez

Buac

Kaushanskaya

(2019). Processing of code-switched sentences by bilingual children: Cognitive and linguistic predictors. Cognitive Development, 52, 100821.

35.

Gutiérrez-Clellen

V. F.

Simon-Cereijido

Erickson Leone

(2009). Code-switching in bilingual children with specific language impairment. International Journal of Bilingualism, 12, 91–109.

36.

Guzzardo Tamargo

R. E.

Valdés Kroff

J. R.

Dussias

P. E

. (2016). Examining the relationship between comprehension and production processes in code-switched language. Journal of Memory and Language, 89, 138–161.

37.

Hakimov

(2016). Effects of frequency and word repetition on switch-placement: Evidence from Russian-German code-mixing. In Robinson

J. A.

Reif

(Eds.), Cognitive perspectives on bilingualism (pp. 91–125). De Gruyter.

38.

Halpin

Melzi

(2018). Code-switching in the narratives of dual language Latino preschoolers. International Journal of Bilingual Education and Bilingualism, 1–17.

39.

Hamers

J. F.

Blanc

(2000). Bilinguality and bilingualism (2nd ed.). Cambridge University Press.

40.

Hart

Risley

(1995). Meaningful differences in the everyday experience of young American children. Paul H. Brookes Publishing.

41.

Iluz-Cohen

Walters

(2012). Telling stories in two languages: Narratives of bilingual preschool children with typical and impaired language. Bilingualism: Language and Cognition, 15(1), 1–17.

42.

Jake

Myers-Scotton

(2009). Which language? Participation potentials across lexical categories in codeswitching. In Isurin

Winford

de Bot

(Eds.), Multidisciplinary approaches to code switching (pp. 307–342). Benjamins.

43.

Jake

Myers-Scotton

Gross

(2002). Making a minimalist approach to codeswitching work: Adding the Matrix Language. Bilingualism: Language and Cognition, 5(1), 69–91.

44.

Jia

Kohnert

Collado

Aquino-Garcia

(2006). Action naming in Spanish and English by sequential bilingual children and adolescents. Journal of Speech, Language, and Hearing Research, 49(3), 588–602.

45.

Joshi

(1985). Processing sentences with intra-sentential code-mixing. In Dowty

D. R.

Kartunnen

Zwicky

(Eds.), Natural language parsing (pp. 190–205). Cambridge University Press.

46.

Kheder

Kaan

(2016). Processing code-switching in Algerian bilinguals: Effects of language use and semantic expectancy. Frontiers in Psychology, 7, 248.

47.

Kohnert

(2002). Picture naming in early sequential bilinguals: A 1-year follow-up. Journal of Speech, Language, and Hearing Research, 45(4), 759–771.

48.

Komeili

Marshall

C. R.

(2013). Sentence repetition as a measure of morphosyntax in monolingual and bilingual children. Clinical Linguistics & Phonetics, 27(2), 152–161.

49.

Lanvers

(2001). Language alternation in infant bilinguals: A developmental approach to code switching. International Journal of Bilingualism, 5(4), 437–464.

50.

Lipski

J. M.

(2017). Language switching constraints: More than syntax? Data from Media Lengua. Bilingualism: Language and Cognition, 20(4), 722–746.

51.

López

(2018). Introduction: Theory and methodology in code-switching research. In López

(Ed.), Code-switching - experimental answers to theoretical question: In honor of Kay Gonzá́lez-Vilbazo (pp. 1–14). John Benjamins Publishing Company.

52.

MacSwan

(2014). Programs and proposals in codeswitching research: Unconstraining theories of bilingual language mixing. In MacSwan

(Ed.), Grammatical theory and bilingual codeswitching (pp. 1–33). MIT Press.

53.

Mahootian

(1996). Code-switching and universal constraints: Evidence from Farsi/English. World Englishes, 15(3), 377–384.

54.

Mahootian

Santorini

(1996). Code switching and the complement/adjunct distinction. Linguistic Inquiry, 27(3), 464–479.

55.

Marian

(2009). Language interaction as a window into bilingual cognitive architecture. In Isurin

Winford

De Bot

(Eds.), Multidisciplinary approaches to codeswitching (pp. 161–185). John Benjamins.

56.

Matras

(2009). Language contact. Cambridge University Press.

57.

Meir

Walters

Armon-Lotem

(2015). Disentangling SLI and bilingualism using sentence repetition tasks: The impact of L1 and L2 properties. International Journal of Bilingualism, 20(4), 421–452.

58.

Montanari

Ochoa

Subrahmanyam

(2019). A longitudinal investigation of language mixing in Spanish–English dual language learners: The role of language proficiency, variability, and sociolinguistic factors. Journal of Child Language, 46(5), 913–937.

59.

Müller

Cantone

K. F.

(2009). Language mixing in bilingual children: Code-switching? In Bullock

B. E.

Toribio

A. J.

(Eds.), The Cambridge handbook of linguistic code-switching (pp. 199–220). Cambridge University Press.

60.

Myers-Scotton

(1993). Duelling languages: Grammatical structure in codeswitching. Clarendon.

61.

Myers-Scotton

Jake

J. L.

(2000). Testing the 4-M model: An introduction. International Journal of Bilingualism, 4(1), 1–8.

62.

Myers-Scotton

Jake

J. L.

(2009). A universal model of code-switching and bilingual language processing and production. In Bullock

Toribio

(Eds.), The Cambridge handbook of linguistic code-switching (pp. 336–357). Cambridge University Press.

63.

Myers-Scotton

Jake

J. L.

(2014). Nonfinite verbs and negotiating bilingualism in codeswitching: Implications for a language production model. Bilingualism: Language and Cognition, 17(3), 511–525.

64.

Myers-Scotton

Jake

J. L.

(2017). Revisiting the 4-M model: Codeswitching and morpheme election at the abstract level. International Journal of Bilingualism, 21(3), 340–366.

65.

Paradis

Nicoladis

Genesee

(2000). Early emergence of structural constraints on code-mixing: Evidence from French–English bilingual children. Bilingualism: Language and Cognition, 3(3), 245–261.

66.

Peña

E. D.

Gutiérrez-Clellen

V. F.

Iglesias

Goldstein

B. A.

Bedore

L. M.

(2014). Bilingual English Spanish Assessment (BESA). AR Clinical Publications.

67.

Pérez-Leroux

A. T.

O’Rourke

Sunderman

(2014). Language dominance and codeswitching asymmetries. In MacSwan

(Ed.), Grammatical theory and bilingual codeswitching (pp. 283–312). MIT Press.

68.

Pert

Letts

(2006). Codeswitching in Mirpuri speaking Pakistani heritage preschool children: Bilingual language acquisition. International Journal of Bilingualism, 10(3), 349–374.

69.

Pfaff

C. W.

(1979). Constraints on language mixing: Intrasentential code-switching and borrowing in Spanish/English. Language, 55(2), 291–318.

70.

Poplack

(1980). Sometimes I’ll start a sentence in Spanish Y TERMINO EN ESPANOL. Linguistics, 18, 581–618.

71.

Raichlin

Walters

Altman

(2019). Some wheres and whys in bilingual codeswitching: Directionality, motivation and locus of codeswitching in Russian-Hebrew bilingual children. International Journal of Bilingualism, 23(2), 629–650.

72.

Rast

Dommergues

J.-W.

(2003). Towards a characterisation of saliency on first exposure to a second language. In Foster-Cohen

Pekarek-Doehler

(Eds.), EUROSLA yearbook: Papers from the annual conference of the European second language acquisition association (Vol. 3, pp. 131–156). John Benjamins.

73.

Raven

J. C.

(1998). Raven’s coloured progressive matrices. Psychological Corporation.

74.

Ribot

K. M.

Hoff

(2014). “¿Cómo estas?” “I’m good.” Conversational code-switching is related to profiles of expressive and receptive proficiency in Spanish-English bilingual toddlers. International Journal of Behavioral Development, 38(4), 333–341.

75.

Riches

N. G.

(2012). Sentence repetition in children with specific language impairment: An investigation of underlying mechanisms. International Journal of Language and Communication Disorders, 47(5), 499–510.

76.

Smolak

de Anda

Enriquez

Poulin-Dubois

Friend

(2020). Code-switching in young bilingual toddlers: A longitudinal, cross-language investigation. Bilingualism: Language and Cognition, 23(3), 500–518.

77.

Song

Sardegna

V. G.

(2014). EFL learners’ incidental acquisition of english prepositions through enhanced extensive reading instruction. RELC Journal, 45(1), 67–84.

78.

Taliancich-Klinger

C. L.

Bedore

L. M.

Peña

E. D.

(2018). Preposition accuracy on a sentence repetition task in school age Spanish-English bilinguals. Journal of Child Language, 45(1), 97–119.

79.

Theodorou

Kambanaros

Grohmann

K. K.

(2017). Sentence repetition as a tool for screening morphosyntactic abilities of bilectal children with SLI. Frontiers in Psychology, 8, 2104.

80.

Treffers-Daller

(1993). Mixing two languages: French-Dutch contact in a comparative perspective (Topics in Sociolinguistics, 9). Mouton de Gruyter.

81.

Treffers-Daller

(2009). Code-switching and transfer: An exploration of similarities and Differences. In Bullock

B. E.

Toribio

A. J.

(Eds.), The Cambridge handbook of linguistic code-switching (pp. 58–74). Cambridge University Press.

82.

Van Hell

J. G.

Litcofsky

K. A.

Ting

C. Y

. (2015). Intra-sentential code-switching: Cognitive and neural approaches. In Schwieter

J. W.

(Ed.), The Cambridge handbook of bilingual processing (pp. 459–482). Cambridge University Press.

83.

Vihman

V.-A.

(2018). Language interaction in emergent grammars: Morphology and word order in bilingual children’s code-switching. Languages, 3(4), 40.

84.

Vinther

(2002). Elicited imitation: A brief overview. International Journal of Applied Linguistics, 12(1), 54–73.

85.

Wasserscheidt

(2014). Constructions do not cross languages: On cross-linguistic generalizations of constructions. Constructions and Frames, 6, 305–337.

86.

Wiig

E. H.

Secord

W. A.

Semel

E. M.

(2004). Clinical evaluation of language fundamentals - preschool 2. Harcourt/Psych Corporation.

87.

Wintner

(2000). Definiteness in the Hebrew noun phrase. Journal of Linguistics, 36(2), 319–363.

88.

Zentella

A. C.

(1997). Growing up bilingual: Puerto Rican children in New York. Blackwell Publishers.