COGNATE EFFECTS ON LANGUAGE CONTROL ...

RUNNING HEAD: COGNATE EFFECTS ON LANGUAGE CONTROL

Cognates Facilitate Switches and then Confusion: Contrasting Effects of Cascade versus Feedback on Language Selection Chuchu Li, Tamar H. Gollan University of California, San Diego

Acknowledgement: This research was supported by grants from the National Institute on Deafness and Other Communication Disorders (011492) and the National Science Foundation (BCS1457519). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the NIH or NSF.

Please send correspondence to: Dr. Chuchu Li Department of Psychiatry University of California, San Diego 9500 Gilman Drive La Jolla, CA 92093-0948, USA Email: [email protected]

PHONOLOGY INFLUENCES LANGUAGE CONTROL

2

Abstract The current study investigated the hypothesis that cognates (i.e., translation equivalents that overlap in form, e.g., lemon is limón in Spanish) facilitate language switches. SpanishEnglish bilinguals were cued to switch languages while repeatedly naming pictures with cognate vs. non-cognate names in separate (Experiment 1) or mixed blocks (Experiments 2 and 3). In all three experiments, on the first presentation of each picture, cognates elicited significantly smaller switch costs, and were produced faster than non-cognates only on switch trials. However, cognate switch-facilitation effects were eliminated (Experiment 2) or reversed (i.e., larger switch costs for cognates than noncognates, in Experiment 3) in mixed blocks with the repeated presentation of a stimulus, largely due to the increasingly slower responses for cognates on switch trials. Cognates may facilitate switches due to increased dual-language activation, which is inhibited on non-switch trials. With repeated presentation of the same pictures, dual-language activation may feed back up to the lexical level, increasing competition for selection. In contrast, when naming pictures in a cognate block, bilinguals may avoid discrimination problems at the lexical level by adaptively focusing less on activation at the phonological level. Cross-language overlap in phonology appears to influence language selection at both phonological and lexical levels involving multiple cognitive mechanisms and reflecting both automatic processes and rapid adaptation to contextual variations in the extent of dual-language activation.

Keywords: bilingualism, switching, cognate, separate and mixed block, repetition


3

Introduction Bilinguals have powerful language control mechanisms that prevent interference from the nontarget language and either restrict their utterances to one language, or allow them to switch between languages when they need or wish to do so. Such control mechanisms are essential given numerous studies showing that both languages are active even when bilinguals speak in just one language (e.g., Colomé, 2001; Costa, Caramazza, & Sebastián-Gallés, 2000; Gollan, Sandoval, & Salmon, 2011; Hermans, Bongaerts, De Bot, & Schreuder, 1998; Hoshino & Kroll, 2008; Moon & Jiang, 2012; Poulisse & Bongaerts, 1994; Schwartz & Kroll, 2006). Language control has been widely studied, most commonly with the language switching paradigm in which bilinguals respond more slowly when switching languages than when using the same language on consecutive responses (Meuter & Allport, 1999; for reviews see Bobb & Wodniecka, 2013; Declerck & Philipp, 2015a). Surprisingly, few studies have asked how cross-language overlap in phonology affects switch costs, even though this question has obvious implications for understanding mechanisms of language control in bilingual language production. Spoken word production begins with formation of a nonlinguistic concept, proceeds to assembly of syntactic information and selection of lexical representations, or lemmas, followed by access of sound representations at the phonological level, and articulation (Levelt, Roelofs, & Meyer, 1999). Many models of bilingual language production specify a language node level, that enables language selection by modulating the extent of activation in the target language (Hartsuiker, Pickering, & Veltkamp, 2004; Hermans, 2000; La Heij, 2005; Poulisse & Bongaerts, 1994), without entirely preventing activation of representations in both languages (Hermans, 2000; Poulisse & Bongaerts, 1994; for similar notions in bilingual language comprehension see Dijkstra & Van Heuven, 1998, 2002; Grainger & Dijkstra, 1992). The extent of dual-language


4

activation may vary with form-overlap between translation equivalents. For example, cognates are translation equivalents that share etymological origin. As a result, they overlap in form (e.g., baby-bebé in English and Spanish), whereas noncognates differ substantially or entirely in form across languages (e.g., apple-manzana). Loaned words may have similar effects especially once they are fully integrated into the recipient language (e.g., many languages adopted a word similar to television and its Spanish equivalent televisión, with slight modifications to accommodate language specific phonology; Poplack, 1988). Bilinguals produced picture names that are cognates more quickly than those that are noncognates in both highly similar languages (e.g., Spanish and Catalan, Costa et al., 2000), and in formally quite distinct languages (e.g., Japanese and English, Hoshino & Kroll, 2008). Cognate facilitation effects have also been found in bilinguals with aphasia, who could name more pictures with cognate than non-cognate names (Kohnert, 2004; Roberts & Deslauriers, 1999), and in tip-of-the-tongue states (TOTs) in cognitively intact bilinguals who report fewer TOTs for pictures with cognate than noncognate names (Gollan & Acenas, 2004). These cognate facilitation effects illustrate how cross-language overlap in phonology speeds production of speech in bilinguals, likely by increasing duallanguage activation. During bilingual language production, activation cascades from the conceptual level to the phonological level in both languages (much as activation is assumed to cascade forward in monolingual production; Cutting & Ferreira, 1999; Jescheniak & Schriefers, 1998; Peterson & Savoy, 1998). Thus, cognates receive extra activation from phonology in the non-target language, either at the lemma or the phonological level, or both. Even without assuming cascading activation, if language selection occurs at a relatively late processing stage, presentation of a picture could lead activation to flow in both languages all the way to the phonological level


5

(Colomé, 2001), thus facilitating lexical selection (but see Hermans et al., 1998). Cognate facilitation effects can also be explained in other ways (for review see Costa, Santesteban, & Caño, 2005). For example, cognates may simply be higher frequency words for bilinguals than noncognates, being updated for frequency in both languages each time they are used in just one language (Titone, Libben, Mercier, Whitford, & Pivneva, 2011). Consistent with this view, cognates are associated with an ERP signature that matches that found for frequency effects (Strijkers, Costa, & Thierry, 2010). Cognate status could easily affect both lexical and phonological processing stages (Christoffels, Firk, & Schiller, 2007; Strijkers et al., 2010) – i.e., different accounts of cognate facilitation effects are not mutually exclusive. While cognate facilitation effects are well established, much less is known about whether and how cognates might affect lexical selection itself, and whether or not cognates can trigger spontaneous language switching (Broersma & De Bot, 2006; Broersma, 2009; Clyne, 1967, 1980). If cognates could reduce (or increase) the magnitude of language switching costs, this could be termed cognate switch-facilitation (or inhibition) effects. Such effects would need to arise at the level of language selection, and since cognates are distinguished from noncognates only by virtue of overlap at the phonological level, that could most easily be explained by a feedback mechanism in which activation flows from phonology back up to affect the preceding processing stage. Feedback has been proposed in models of language production to explain effects observed within production of a single language (Dell, 1988; Santésteban, Pickering, & McLean, 2010) and in bilinguals managing activation of two languages (Costa, Roelstrate, & Hartsuiker, 2006; Declerck & Philipp, 2015b; see also Bernolet, Hartsuiker, & Pickering, 2012). In one such study, bilinguals were instructed to describe a picture starting with Dutch but ending in English (i.e., they needed to switch the language at some location); if a cognate was included


6

in a previous language switching sentence that the bilingual heard and repeated, it primed a switch in the same location of the following sentence even when they didn’t repeat the same target word (Kootstra, van Hell, & Dijkstra, 2012). Cognates might also influence language switching costs (without positing feedback) if language selection is, or can in some contexts be, delayed until the phonological level. Consistent with this hypothesis, bilinguals produce words with a more pronounced accent when switching languages, particularly with cognates (Goldrick, Runnqvist, & Costa, 2014; Olson, 2013). Additional evidence comes from the observation of reversed language dominance effects on production of accent errors in which bilinguals produce an entire word correctly but with the accent of the nontarget language (when normally they speak without an accent; Gollan, Schotter, Gomez, Murillo, & Rayner, 2014; Kolers, 1966). Reversed language dominance refers to a phenomenon sometimes found in mixed-language contexts, in which bilinguals appear to respond less fluently (more slowly and with increased errors) in their otherwise usually more proficient language (Christoffels et al., 2007; Costa & Santesteban, 2004; Costa, Santesteban, & Ivanova, 2006; Gollan & Ferreira, 2009; Gollan et al., 2014; Gollan & Goldrick, 2016; Santesteban & Costa, 2016; Verhoef, Roelofs, & Chwilla, 2009, 2010). Reversed dominance effects might reflect inhibition of the dominant language (Green, 1998) – which might be separately applied to the lexical and/or phonological levels to explain how reversal can affect both accent-only errors, and full language intrusion errors (e.g., producing pero instead of but; Gollan et al., 2014). Note that, while language switching costs likely reflect transient language control processes (i.e., trial-by-trial modulated control), reversed dominance effects more likely reflect global and sustained control (i.e., inhibition of the dominant language as a whole across the entire testing block). These two processes may overlap but are at least partially dissociable,


7

given that modulation of switch costs by language dominance, and reversal of dominance effects do not always occur simultaneously (see Declerck & Philipp, 2015a for a review). Few studies have directly investigated how cognates might affect language switching costs. One study investigated if language switch costs varied as a function of whether the target stimuli were pictures or digits (Declerck, Koch, & Philipp, 2012). Multiple careful controls were included to pinpoint possible ways in which the numbers might differ from pictures including a manipulation of cognate status (because number names are cognates in many languages). German-English bilinguals named nine targets with 12 repetitions in each of the following four conditions: digits (1-9; all of which are German-English cognates), nine pictures with GermanEnglish cognate names (e.g., house is haus in German), 9 semantically related pictures (body parts), and a control condition with 9 pictures that were not semantically related and had noncognate names (e.g., horse is pferd in German). During the test, a trial began with a fixation point (+) for 400 ms, followed by a color square (green or blue, serving as the language cue) that was presented for 1000 ms. After the language cue, the picture or digit was presented until a response was given. Of great interest, in this study, switch costs in both languages were significantly smaller in the digit set compared to the semantically related set and the control set, while they did not differ from the cognate pictures. Since the focus of this study was on the digit set, no direct comparison was conducted between the cognate pictures and semantically related pictures or the control set. However, the indirect evidence still suggested cognates decrease the switch costs. However, the evidence for a modulation of switch costs by cognate status is decidedly mixed. Verhoef et al. (2009) reported cognate facilitation effects, but no cognate switchfacilitation effects, in a study in which Dutch-English bilinguals switched languages while


8

naming 48 pictures (24 cognates, 24 non-cognates) with a total of 32 repetitions (1,536 trials) and 50% switch rate. In this study language cues were presented for 250 ms, followed by a 500 ms or 1500 ms blank, and then target pictures were presented for 250 ms. Testing sessions lasted approximately 4 hours (including questionnaire, instructions, cap application for the ERPs measure, formal testing, and breaks). Santesteban and Costa (2016) also showed no cognate switch-facilitation effects among Spanish-Catalan bilinguals and Catalan L2 learners who were native Spanish speakers. In this study, bilinguals named 20 pictures (10 cognates, 10 noncognates) in lists that varied in length from 5-14 trials. Overall, the switch rate was 30% and each picture was presented 47 or 48 times. Each participant was presented with 950 trials. The language cue (a blue or red circle) was presented for 2000 ms, followed by the target picture that stayed on the screen for 2000 ms or after a response was given, and the trial interval was 1150 ms. Yet another result was reported by Christoffels et al. (2007) who found bigger switch costs for cognates than noncognates, i.e., a cognate switch-inhibition effect. In this study, GermanDutch bilinguals named 48 pictures (24 cognates, 24 non-cognates) with a total of 12 repetitions and also 50% switch rate. A fixation was presented for a variable duration between 300 ms and 600 ms, followed by the target picture in red or green (the color served as the language cue) that stayed on the screen for 2000 ms or after a response was given. Replicating cognate facilitation effects, bilinguals produced pictures with cognate names more quickly than pictures with noncognates, however, switch costs were significantly larger for cognates than for non-cognates. Numerous between-study differences in small but possibly critical methodological details in the studies just reviewed make it difficult to reach any firm conclusions about how and why cognate status may affect switching, whether reducing, increasing, or having no effect on switch costs. Across studies, bilinguals of different language combinations were tested (including


9

German-English, German-Dutch, Dutch-English, English-French, Spanish-Catalan, and ItalianEnglish bilinguals). Different languages share different proportions of cognates overall which could lead to strategic differences in the extent to which bilinguals allow both languages to remain active during speech production. Other details in design and procedure also varied across studies, such as the number of items, the preparation time (i.e., the interval between the language cue and the target picture), and the number of repetitions. Longer preparation time is known to reduce switch costs (e.g., Costa & Santesteban, 2004; Fink & Goldrick, 2015; also see Mosca & Clahsen, 2016 for the absence of switch costs with long preparation time), and could leave less room for modulation of switch costs by various factors including cognate status. One possibly critical difference across studies was that cognates and non-cognates were presented in separate blocks in Declerck et al. (2012) while they were mixed together in a single testing block in Christoffels et al. (2007), Santesteban and Costa (2016), and Verhoef et al. (2009). Consistent with this hypothesis, larger switch costs were found for cognates relative to noncognates when cognates and non-cognates were mixed in a single testing block in word naming (i.e., reading aloud; Filippi, Karaminis, & Thomas, 2014) and visual word lexical decision (Thomas & Allport, 2000). However, cognate interference effects on switch costs in these tasks could reflect competition between languages in processing the written words. Declerck and Philipp (2015a) suggested that reduction of switch costs for cognates when these are presented in a separate block from noncognates could reflect a persisting block-wide relaxation of the extent to which the nontarget language is inhibited at the phonological level, while the larger switch costs on cognates in a mixed block might be a result of control processes that are modulated trial to trial.


10

The present study provided a more systematic test of if and how cognate status influences language switching – with the goal of clarifying the possible theoretical implications of cognate effects on switching costs for models of bilingual language production. Spanish-English bilinguals were cued to name pictures in English or Spanish. Target pictures had cognate or noncognate names, and were presented in separate blocks in Experiment 1, and in mixed blocks in Experiments 2 and 3. Participants were recruited from the same population across experiments which all presented the same materials, and had virtually identical design and procedures (e.g., same switch: non-switch trial ratio; same preparation time). The only difference across experiments was whether cognates and noncognates were presented separately or intermixed (comparing Experiment 1 to Experiment 2), and the number of items presented (comparing Experiment 2 to Experiment 3). Based on prior studies we expected to find reduced switch costs for cognates relative to noncognates in Experiment 1 (with cognates and noncognates presented in separate blocks), but no difference or the opposite result (i.e., bigger switch costs for cognates than noncognates) when cognates were intermixed in Experiments 2 and 3. If blocked versus mixed presentation were the primary factor influencing the presence and direction of cognate effects, this would require an account of bilingual language control that adaptively adjusts the nature of control processes at the level of phonology based on context. Repetition could also be a significant factor modulating the nature of cognate facilitation effects obtained across studies, which would more directly implicate the degree of activation in lexical and phonological representations (which is increased by recent or frequent repetition). Finally, given that previous reports together included every possible pattern in switch costs, including cognate facilitation, cognate inhibition, and no cognate effects, these might all have been Type 1 errors in which case


11

no consistent pattern should emerge when cognate effects on switch costs are investigated systematically. Experiment 1 Our primary goal in Experiment 1 was to replicate Declerck et al.’s (2012) finding of cognate switch-facilitation effects with presentation of either cognates or noncognates in separate testing blocks. One unavoidable difference between our study and theirs was that we tested Spanish-English bilinguals (who are readily accessible in San Diego; Declerck et al. tested German-English bilinguals). Additionally, to maximize the possibility of observing the effects of cognate status on switch costs, we eliminated the 1000 ms preparation time used by Declerck et al. and instead presented the pictures and the language cues simultaneously. In addition to maximizing switch costs, the elimination of preparation time could also allow other factors that modulate the size of switch costs to appear. For example, Declerck et al. (2012) did not observe reversed language dominance effects, or greater switch costs for the dominant than the nondominant language (e.g., Meuter & Allport, 1999), but these might be found in Experiment 1 (given the absence of long preparation times). Participants Thirty-two unbalanced Spanish-English bilingual undergraduates at University of California, San Diego participated for course credit. Table 1 shows self-reported participant characteristics and Multilingual Naming Test scores in both languages (MINT; Gollan et al., 2012). All participants were English-dominant (higher scores in English than in Spanish; two Spanish-dominant bilinguals were replaced), although all of them reported acquiring English later than Spanish or acquiring the two languages simultaneously. All participants (for all the


12

experiments in this study) signed the consent form that was approved by the institutional review board. Design and Procedure Pictures were two sets of nine color pictures (size 400 × 400 pixels) that were selected to be easy to name in both languages (see Appendix). One set had nine pictures with cognate names and the other set had nine pictures with non-cognate names. Cognate and noncognate picture names were matched for frequency and length, and pictures were matched for visual complexity (see Table 2; cognates and noncognates did not differ significantly on any of the characteristics listed, all ts ≤ 1.08, all ps ≥ .30). In addition, in each language, onsets in each set of nine picture names included two voiced stop sounds, two voiced liquid sounds, three voiceless stop sounds, one voiceless fricative sound, and one voiced nasal sound. Bilinguals completed two blocks of cued picture naming—in one block pictures with cognate names were presented, and in the other block pictures with noncognate names were presented. Block order was counterbalanced between subjects. In each block, bilinguals were instructed to name pictures in either English or Spanish, as cued by a red or blue color square (size 160 × 106 pixels). The color-cue to language assignment was constant throughout the experiment for each bilingual but was counterbalanced between participants. In each block, each picture was repeated six times with each language cue (i.e., a total of 12 repetitions of each picture), three times on switch trials (i.e., the target language was different from the previous trial) and three times on non-switch trials (i.e., the target language was same as the previous trial). Thus, as in Declerck et al. (2012), language switches occurred 50% of the time, and there was equal power for examining switch versus non-switch trials in each language. The items were presented in a pseudo-randomized order so that the same picture was never


13

presented on consecutive trials. The maximum number of consecutive switch trials or non-switch trials was 6. For each block, we prepared four different pseudorandomly ordered experimental lists that followed above restrictions. Participants were randomly assigned to one of the lists. In this way, participants assigned o different lists would see the first presentation of each picture in different conditions. For example, for the first presentation of the picture tooth, bilinguals would name it in English as a switch trial in List A, in Spanish as a non-switch trial in List B, in English as a non-switch trial in List C, and in Spanish as a switch trial in List D. Each trial began with a fixation point (+) for 400 ms, followed by a cue (i.e., the color square) and the target picture that were presented for a maximum of 3000 ms, or disappeared when a response was registered. The next trial began 150 ms after the picture disappeared. Pictures were presented at the center of the screen, and the cue was presented above the picture. Before the experimental trials, 16 practice trials were presented (using eight non-critical pictures). After a break in which participants were informed that the practice had ended, a familiarization block followed in which all 9 critical pictures that would be included in the following critical blocks were presented twice, once with a cue to name the picture in English, and once with a cue to name it in Spanish. Different from the practice and critical blocks, after the picture in the familiarization block disappeared, the screen stayed blank until the experimenter pressed a button to initiate the next trial. During the blank, the participants were corrected if the name they provided was different from the target name or failed to provide an answer, which rarely occurred.1 After the familiarization phase, one dummy trial (i.e., a nonFor example saying bunny instead of rabbit. One bilingual did so for 6 out of the 36 names (18 items* 2 languages), 6 bilinguals did for 4/36 names, 9 did for 3/36 names, 9 did for 2/36 names, and 4 did for just 1/36 names. We repeated the analyses reported below after excluding three items (i.e., rabbit/bunny, piano/no responses, and map/world) for which 8-13 of 32 bilinguals provided unexpected answers, or did not provide an answer in the familiarization block in either language, and this did not change the pattern of results. Also, note that in some cases bilinguals 1


14

critical trial) was presented, followed by the 108 critical trials of the first block. Another familiarization block that only included the critical items in the second critical block was presented (whichever item type, namely cognate vs. noncognate, was presented in the first block), then the second block of critical trials was presented following another dummy trial. In each critical block, the first critical trial was always a non-switch trial. The dummy trial before the cognate block was always a non-critical target cognate (i.e., pear-pera) and the dummy trial before the non-cognate block was always a non-critical target non-cognate (i.e., gloves-guantes). Thus, not including the familiarization blocks (one before the cognates block, and one before the noncognates block), each bilingual completed a total of 216 critical trials, with 108 in each block. Results All analyses were carried out in R, an open source programming environment for statistical computing (R Core Team, 2013) with the lme4 package (Bates, Maechler, Bolker, & Walker, 2013) for linear mixed effects modeling (LMM) and general linear mixed effects modeling (GLMM), both with the “bobyqa” optimizer. Response times (RT) data for incorrect responses were excluded. Correct RTs were trimmed if any one or more of the following conditions were met: hesitation, disfluency, the correct answer failed to trigger the voice key, or an error was produced on the previous trial. Responses less than 200 ms were removed (as were any responses above 3,000 ms; these were not recorded). We intended to remove responses that were 4SDs larger or smaller than the means for each subject (collapsing all conditions), but found no outliers (i.e., all trials were within 4SDs). Contrast-coded fixed effects included cognate status (cognates vs. non-cognates), language (Spanish vs. English), trial type (switch vs. may have failed to provide an answer in the familiarization block because of noise that triggered that voice key so that the picture disappeared before they could see it clearly.


15

stay), and all the two-way and three-way interaction of these factors. Subject and item/picture were entered as two random intercepts with related random slopes (i.e., cognate status, language, trial type and all their two-way and three-way interactions as the slopes for subject; language, trial type, and their two-way interaction as the slopes for picture). The same fixed effects and random intercepts were included in the logistic regression for error rates analyses, but random slopes were removed due to the failure to converge. The significance of each fixed effect was assessed via likelihood ratio tests (Barr, Levy, Scheepers, & Tily, 2013). Figure 1 shows the mean response time with the 95% confidence intervals of each condition. Bilinguals responded significantly more slowly overall on switch than non-switch trials (mean 938 ms vs. 866 ms, β = 74.31; SE β = 9.22; χ2 (1) = 36.94, p < .001), produced pictures with cognate names more quickly than noncognates (mean 871 ms vs. 933 ms, β = 60.46; SE β = 18.76; χ2 (1) = 9.39, p = .002), and tended to respond more slowly overall in English (the dominant language) than in Spanish, though this main effect of language was just marginally significant (mean 917 ms vs. 886 ms, β = 33.28; SE β = 18.00; χ2 (1) = 3.47, p = .062). Of great interest, cognates reduced the magnitude of switch costs, but only in the dominant language, English, a significant three-way interaction between cognate status, trial type, and language (β = -58.56; SE β = 28.31; χ2 (1) = 4.20, p = .040). Planned comparisons revealed significantly smaller switch costs when bilinguals produced cognate relative to noncognate names in English (mean 44 ms vs. 91 ms, β = -48.25; SE β = 22.09; χ2 (1) = 4.67, p = .030), but numbers actually trended slightly in the opposite direction in Spanish (mean 81 ms vs. 72 ms), though not significantly so, (β = 10.61; SE β = 16.09; χ2 < 1). None of the other fixed effects were significant (ps ≥ .21).


16

Table 4 shows the mean error rates (which were low overall, 3.67%) and standard deviations by condition (i.e., trial type, cognate status and language). Overall, bilinguals produced more errors on switch than on non-switch trials (mean 4.7 % vs. 2.7%; β = .69; SE β = .14; χ2 (1) = 24.00, p < .001) and in English than in Spanish (mean 4.8% vs. 2.5%; β = .71; SE β = .14; χ2 (1) = 25.46, p < .001). In the dominant language (English), bilinguals produced fewer errors on cognates than noncognates (mean 3.8% vs. 5.9%; β = - .48; SE β = .18; χ2 (1) = 2.64, p = .008), but in the nondominant language (Spanish) error rates were the same for cognates and non-cognates (mean 2.5% vs. 2.5%; β = - .06; SE β = .29; χ2 < 1), though this interaction between language and cognate status was just marginally significant (β = - .51; SE β = .29; χ2 (1) = 1.78, p = .077). None of the other main effects or interactions were significant (ps ≥ .12). First-presentation trials only. To consider if the results were dependent on the substantial repetition of items (a total of 12) in Experiment 1, we asked whether the general pattern of results held when considering only the first presentation of each picture in the block (i.e., only 18 trials per participant).2 First-presentation responses revealed a very similar pattern of results (note that interactions between random effects were not included due to the failure to converge); Figure 2 shows the mean RTs with the 95% confidence intervals of condition 2

The plan to consider first presentation only was was not built into the experimental design a priori but was motivated by our discovery that language switch costs can be eliminated by instructing bilinguals to name each picture in just one language (Kleinman & Gollan, 2016). Thus, the items varied with respect to the number first-presentations that were switch vs. nonswitch trials. In both Experiments 1 and 2, there were 18 items, of these 8 items (4 cognates and 4 noncognates) had equal numbers of switch vs. non-switch trials. For the first presentations of the remaining 10 items (5 cognates and 5 noncognates), switch trials were 3 times as many as stay trials (for 2 cognates and 3 noncognates), or vice versa (for 3 cognates and 2 noncognates). In Experiment 3, there were only 8 items, of these 3 (1 cognate and 2 noncognates) had equal numbers of switch vs. non-switch trials. For the other 5 items (3 cognates and 2 noncognates), switch trials were 3 times as many as stay trials (for 1 cognate and 2 noncognates), or vice versa (for 2 cognates and 0 noncognates). However, linear mixed effects models are well suited for analysis of data without perfectly equal numbers of data points in different conditions (Baayen, Davidson, & Bates, 2008; Gelman & Hill, 2007, p.254).


17

(accuracy was high overall, 98.19%, and the analysis of errors revealed no significant effects, ps ≥ .11).

There were significant switch costs with slower responses on switch than on non-switch

trials (mean 933 ms vs. 854 ms; β = 71.63; SE β = 21.61; χ2 (1) = 11.08, p < .001). Of greatest interest, switch costs were numerically smaller for cognates than noncognates in English (mean 22 ms vs. 125 ms), and numerically larger for cognates than noncognates in Spanish (mean 138 ms vs. 85 ms), in first-presentation trials. These two individual comparisons were not statistically significant on their own in this subset of the data (ps ≥ .23). However, cognates still tended to reduce the magnitude of switch costs in English but not in Spanish, reflected by the numerically different switch cost values and a marginally significant three-way interaction between trial type, cognate status, and language (β = -134.86; SE β = 74.79; χ2 (1) = 3.17, p = .074). Other fixed effects revealed no significant results (ps ≥ .15). These analyses demonstrate that the cognate switch-facilitation effects in Experiment 1 should not be attributed to massive repetition of the pictures, as the same pattern of results was found in first-presentation of each picture alone. Discussion The results of Experiment 1 replicated and extended several effects previously reported in the literature including switch costs (bilinguals responded more slowly and less accurately on switch than on non-switch trials), cognate facilitation effects (bilinguals produced pictures names that are cognates more quickly than those that are noncognates), and reversed language dominance effects (bilinguals named pictures in their dominant language more slowly than in their nondominant language); though this effect was just marginally significant overall and was not significant in first-presentation trials.


18

Of greatest interest in the present context, switch costs were significantly smaller for cognates than for noncognates, namely, cognate switch-facilitation effects, replicating Declerck et al. (2012), though in the present study this modulation of switch costs was observed only in the dominant language. As outlined above, the latter difference might be attributed by our choice to eliminate preparation time (i.e., with no preparation time switch costs are larger, perhaps leaving more room for modulation of switch costs by various factors). In Declerck et al. (2010) German-English bilinguals responded more quickly overall in their dominant language, German, than in English,3 but language dominance did not affect the magnitude of switch costs, i.e., there was no switch-cost asymmetry. We also did not observe a switch-cost asymmetry, however, we reported reversed dominance effects, a different signature of inhibitory control of the dominant language (Bobb & Wodniecka, 2013; Guo, Liu, Misra, & Kroll, 2011; Philipp, Gade, & Koch, 2007; Philipp & Koch, 2009). Of course many other methodological differences could have influenced the results (we studied bilinguals of different language combinations, and participants’ immersion context differed across studies) and we cannot arrive at definitive conclusions as to why we obtained the cognate switch-facilitation effects only for the dominant language in Experiment 1. To further explore the nature of this result, we examined whether asymmetrical cognate switch-facilitation effects might have been affected by block order. In this analysis we compared bilinguals who completed the cognate block first to those who completed the noncognate block first. Importantly, these two groups of bilinguals did not differ significantly in their objectively Declerck et al. (2012) had four separate sets of items and they did not separately compare cognates to each other item set. Instead, they focused on the digits set which they compared with other item sets. They observed significant language dominance effects (i.e., faster RT in German) when comparing digits with cognates, and digits with semantic controls. However, when comparing digits with non-cognates, language dominance effects were not significant, though numbers trended in this direction (faster responses in German than in English). 3


19

measured proficiency in each language (i.e., MINT scores, ps ≥ .48). Repeating our analysis including all data points and that with just first-presentation trials, but including block order as an additional fixed effect, revealed no significant differences in the pattern of results reported above. Critically, the three-way interaction between cognate status, language, and trial type still held (ps ≤ .041). In addition, block order revealed no significant main effects or interactions (ps ≥ .20),

except that the reversed language dominance effect (English was always produced more

slowly) was always stronger in the block that was completed later (mean English disadvantage difference ≥ 27 ms between the first vs. second block, for both cognate-first and noncognate-first blocks), a significant interaction between block order, language, and cognate status (β = -70.69; SE β = 24.42; χ2 (1) = 7.89, p = .005). Having found some evidence for a reduction of switch costs for cognates when cognates and noncognates were presented in separate blocks, in Experiment 2 we asked if cognate switchfacilitation effects might disappear when cognates and noncognates were presented intermixed in a single testing block as done in one previous study that reported no cognate switch-facilitation effects (Santesteban & Costa, 2016; Verhoef et al., 2009) and another that reported significant cognate switch-inhibition effects (Christoffels et al., 2007). Experiment 2 Experiment 2 was a replication of Experiment 1 with one critical difference, which was that cognates and noncognates were intermixed in the same testing block, instead of in separate blocks. In all other aspects (e.g., participants’ characteristics, items, procedure) Experiment 2 was the same as Experiment 1. Based on previous studies, we anticipated that the cognate switch-facilitation effects observed in Experiment 1 might be absent (Santesteban & Costa,


20

2016;Verhoef et al., 2009), or could even reverse in direction (Christoffels et al., 2007), in Experiment 2. Participants Thirty-two Spanish-English bilingual undergraduates who did not participate in Experiment 1 but were recruited from the same subject pool participated for course credit. Table 1 shows self-reported participant characteristics and MINT scores in English and Spanish. Again, all the participants were English-dominant (five Spanish-dominant bilinguals were replaced), and participants from Experiments 1 and 2 did not differ in either self-reported or objectively measured proficiency levels of the two languages, or any other variables (ps ≥ .23; see Table 1). Design and Procedure The stimuli, design, and procedure were same as those in Experiment 1,4 except that: a) after the practice, a familiarization block followed in which all 18 critical pictures were presented twice, once with a cue to name the picture in English, and once with a cue to name it in Spanish, and thus there was only one familiarization block in this experiment; b) after the familiarization phase, one dummy trial was presented, followed by the 108 critical trials of the first block in which all the 18 critical pictures were included, then the second block of another 108 critical trials followed another dummy trial was presented. The dummy trial before each of two mixed blocks was pear and gloves (as in Experiment 1), and the order of the dummy trials was counterbalanced between participants. Results The procedure of data exclusion and analyses was same as in Experiment 1, but correlations between random effects were not included in the analysis of RTs due to the failure to Similar to Experiment 1, we repeated the analyses reported below after excluding three items for which 8-14 of 32 bilinguals provided unexpected answers, or did not provide an answer in the familiarization block in either language, and this did not change the pattern of results. 4


21

converge.5 Figure 1 shows the mean response times with the 95% confidence intervals by condition. Bilinguals responded more slowly overall on switch than on non-switch trials (mean 1041 ms vs. 943 ms, β = 94.77; SE β = 8.44; χ2 (1) = 43.80, p < .001), produced pictures with cognate names more quickly than noncognates (mean 965 ms vs. 1017 ms, β = -49.96; SE β = 14.68; χ2 (1) = 9.29, p = .002), and responded more slowly in English than in Spanish (mean 1008 ms vs. 975 ms, β = 33.52; SE β = 15.83; χ2 (1) = 4.46, p = .035). Of great interest, switch costs were not modulated by cognate status or language, and none of the interactions were significant (ps ≥ .22). Table 4 shows the mean error rates (5.1% overall) and standard deviations for each condition by language (note that no random slope was included in the logistic regression analysis due to failure to converge). Bilinguals produced significantly more errors on switch than on nonswitch trials (mean 6.4% vs. 3.8%; β = .01; SE β = .12; χ2 (1) = 27.24, p < .001), and in English than in Spanish (mean 6.3% vs. 4.0%; β = .53; SE β = .12; χ2 (1) = 21.04, p < .001). None of the other fixed effects were significant (ps ≥ .18). First-presentation trials only. To again consider possible effects of substantial repetition, we conducted further analyses when considering only the first presentation of each picture in the block (as in the analysis of all data points, correlations between random effects were removed due to the failure to converge). Figure 2 shows the mean RTs with the 95% confidence intervals of condition (accuracy was high overall, 94.3%, and the analysis of errors revealed no significant effects, ps ≥ .10). There were significant switch costs with slower responses on switch trials than on non-switch trials (mean 1008 ms vs. 906 ms, β = 91.80; SE β = Given this difference, we re-ran the analyses in Experiment 1 without the correlation between random effects, and the pattern of the results remained the same. The main effects were significant (i.e., switch cost, cognate facilitation, and reversed dominance effect; ps < .05), as was the three-way interaction between cognate status, trial type, and language (p = .032). Again, cognates significantly reduced switch costs in English (p = .026) but not in Spanish (p = .500). 5


22

25.37; χ2 (1) = 11.03, p < .001), and faster responses with cognates than noncognates (mean 916 ms vs. 1002 ms, β = -80.48; SE β = 29.90; χ2 (1) = 6.41, p = .011). Importantly, switch costs were smaller with cognates than non-cognates (mean 25 ms vs. 170 ms, β = -87.85; SE β = 44.75; χ2 (1) = 3.87, p = .049). Planned comparisons revealed significant switch costs for non-cognates (β = 136.65; SE β = 36.43; χ2 (1) = 10.91, p < .001) but not for cognates (β = 43.63; SE β = 38.51; χ2 (1) = 1.32, p = .249). Additionally, bilinguals produced pictures with cognate names 145 ms more quickly than non-cognates on switch trials (β = -127.17; SE β = 39.70; χ2 (1) = 7.75, p = .005), while no such effect was found on nonswitch trials (only a 3 ms difference; β = -29.30; SE β = 39.61; χ2 (1) = .55, p = .459). All other fixed effects failed show significant results (ps ≥ .57). These results both resemble and differ those of first-presentation trials in Experiment 1. In both experiments we observed cognate switch-facilitation effects on first-presentation trials. On one hand, switch-facilitation effects seemed to be more robust in Experiment 2 given that both Spanish and English exhibited the effect (whereas in Experiment 1 cognate switchfacilitation effects were found only in the dominant language). On the other hand, cognate switch-facilitation effects seemed more robust in Experiment 1 given that they survived repetition and were present in the overall analysis (whereas in Experiment 2 they were present only in first-presentation trials). Taken together, repetition appears to be an important factor modulating the presence or absence of cognate switch-facilitation effects, but blocked versus mixed presentation might also be critical. In particular, cognate switch-facilitation effects might be wiped out by repetition particularly when cognates and noncognates are intermixed in the same testing block (as in Experiment 2). In contrast, cognate switch-facilitation effects might be less influenced by


23

repetition when cognates are presented in separate blocks from noncognates (as in Experiment 1). To examine this possibility we conducted a further analysis comparing repetition effects across experiments. Stimulus repetition effects. In both Experiments 1 and 2, each picture was repeated 12 times. To compare repetition effects across experiments we compared bilinguals’ performance on switch trials in the first and last thirds of trials, that is, the first four presentations of each picture (i.e., presentation 1-4) to the last four presentations of each picture (i.e., presentation 912). The results of this analysis are shown in Figure 3. These data were analyzed with the factors experiment (Experiment 1 vs. 2), cognate status (cognates vs. non-cognates), language (English vs. Spanish), and stimulus repetition (the first four presentations vs. the last four presentations) entered as the fixed effects, while subjects and items were entered as random intercepts with related random slopes. To simplify the models, we conducted this analysis for switch vs. nonswitch trials separately. For brevity, we present here only the critical interactions involving repetition. For switch trials, stimulus repetition produced just one significant effect. Specifically, bilinguals produced cognates more slowly in the last four presentations than in the first four presentations in Experiment 2 but not in Experiment 1, and noncognates did not exhibit this effect. This three-way interaction between experiment, cognate status, and stimulus repetition was significant (β = 88.09; SE β = 37.99; χ2 (1) = 5.12, p = .024). Follow-up comparisons revealed different effects of stimulus repetition for cognates vs. noncognates in Experiment 2, there was a significant interaction between cognate status and stimulus repetition (β = 68.42; SE β = 30.93; χ2 (1) = 4.72, p = .030), but in Experiment 1 there was no such effect (β = -18.19; SE β = 30.86; χ2 < 1). Specifically, in Experiment 2 with cognates and non-cognates intermixed in the


24

same testing block, bilinguals produced cognates increasingly more slowly with repeated presentation of a stimulus (mean 992 ms vs. 1037 ms in the first four vs. last four presentations respectively; β = 57.15; SE β = 25.27; χ2 (1) = 4.54, p = .033), i.e., a repetition interference effect. In contrast, bilinguals produced non-cognates slightly but not significantly more quickly with repeated presentation of a stimulus (mean 1090 ms vs. 1079 ms in the first four vs. last four presentations; β = -9.89; SE β = 20.80; χ2 < 1). Additionally, in Experiment 1 with cognates and non-cognates presented in separate blocks, bilinguals named pictures with both cognate and noncognate names slightly more quickly with repeated presentation stimuli (cognates: mean 919 ms vs. 895 ms in the first four vs. last four presentations; non-cognates: mean 976 ms vs. 950 ms in the first four vs. last four presentations; ps ≥ .16). Similar analyses of error rates (all ps ≥ .11), and RTs and error rates on non-switch trials (all ps ≥ .36) showed no significant effects involving stimulus repetition. Discussion In both Experiments 1 and 2 we observed switch costs, cognate facilitation, and reversed dominance effects. Of greatest interest, in both experiments, cognates significantly reduced switch costs on first-presentation trials (only in the dominant language in Experiment 1, in both languages in Experiment 2). In addition, comparing switch trials across Experiments 1 and 2, cognates were produced more slowly with repetition in Experiment 2 but not in Experiment 1, and repetition had no effect on noncognates in either experiment. A final question we asked concerned why repetition caused cognate interference effects on switch trials in mixed (Experiment 2) but not in blocked (Experiment 1) presentation of cognates and noncognates. Specifically, we considered if the critical difference between Experiments 1 and 2 might have been the number of items within the mixed block, since with a


25

larger number of items there would be greater lag between repetitions of the same stimulus. Long-lags between repetitions reduce repetition facilitation effects in monolingual speech production (Wheeldon & Monsell, 1992), and similar effects might be found in the context of language switching. To investigate the possible role of lag length between repetitions, we conducted Experiment 3, which differed from Experiment 2 only in the number of items included in the mixed block. That is, Experiment 1 had only 9 items repeating in each block (9 cognates and 9 noncognates presented in separate blocks), whereas in Experiment 2 all 18 items were intermixed. In Experiment 3 we aimed to rule out this possibly critical difference between Experiments 1 vs. 2 (as the possible factor eliciting cognate interference effects on switch trials in Experiment 2 but not 1) by intermixing a smaller number of cognates and noncognates in a single testing block. Experiment 3 Experiment 3 was a replication of Experiment 2 with one critical difference, which was that only four cognates and four non-cognates were included. All eight items were selected from the 18 items that were used in Experiments 1 and 2. In all other aspects, Experiment 3 was the same as Experiment 2. If cognate interference effects would be found with repetition on switch trials in Experiment 3 (as found in Experiment 2) this would imply that the critical difference between Experiments 1 versus 2 and 3 was indeed the mixed presentation of cognates and noncognates within the same block (not the total number of items in the block). Participants Thirty-two Spanish-English bilingual undergraduates who had not participated in Experiments 1or 2 but were recruited from the same subject pool participated for course credit. Table 1 shows self-reported participant characteristics and MINT scores in English and Spanish.


26

Again, all the participants were English-dominant (one Spanish-dominant bilingual was replaced), and participants from Experiments 2 vs. 3 did not differ significantly in either selfrated or objectively measured proficiency levels of the two languages (ps ≥ .17; see Table 1). Design and Procedure The design and procedure were same as in Experiment 2, except that only four cognates and four non-cognates were included (i.e., fewer items than Experiment 2; see Appendix). As a result, only 96 critical trials (8 pictures crossed in a 2 languages, by 2 trial types, by 3 repetitions) were included in the two blocks in Experiment 3 (whereas Experiment 2 had 216 critical trials). The frequency and number of syllables of the four cognate and four non-cognate names, and the visual complexity of the pictures used to elicit their production, were not significantly different from each other (ps ≥ .40; see Table 3). Results The procedure of data exclusion and analysis was the same as in Experiment 2. Figure 1 shows the mean response times by condition with 95% confidence intervals when all trials were included. Similar to Experiment 2, bilinguals named pictures more slowly overall on switch than on non-switch trials (mean 924 ms vs. 864 ms, β = 59.85; SE β = 8.26; χ2 (1) = 22.37, p < .001), responded more slowly in English than in Spanish (mean 920 ms vs. 867 ms, β = 52.68; SE β = 12.96; χ2 (1) = 11.09, p < .001), and produced pictures with cognate names slightly though unlike in Experiments 1-2 not significantly more quickly than noncognates (mean 882 ms vs. 904 ms, β = -24.30; SE β = 21.60; χ2 (1) = 1.38, p = .240). Of great interest, and different from Experiment 2, switch costs were significantly larger for cognates than non-cognates (mean switch cost 83 ms vs. 36 ms; i.e., a cognate switch-inhibition effect, as reported by Christoffels et al., 2007), a significant interaction between cognate status and trial type (β = 42.91; SE β = 17.03; χ2 (1) =


27

5.99, p = .014). Lastly, bilinguals seemed to benefit more from cognates in English than Spanish, a marginally significant interaction between language and cognate status (mean cognate facilitation effect 46 ms vs. 0 ms; β = 47.36; SE β = 26.73; χ2 (1) = 3.25, p = .071). All other fixed effects failed to reach significance (ps ≥ .11). Table 4 shows the mean error rates (low overall, 3.7%) and standard deviations by condition and language. Bilinguals produced significantly more errors on switch than non-switch trials (mean 5.0% vs. 2.4%; β = .87; SE β = .22; χ2 (1) = 12.67, p < .001), and marginally more errors in English than in Spanish (mean 4.2% vs. 3.2%; β = .39; SE β = .22; χ2 (1) = 3.36, p = .067). No other fixed effects were significant (ps ≥ .10). The mean error rate of first time trials only was low overall (5.1%) and revealed neither significant main effects nor interactions (ps ≥ .44).

First-presentation trials only. Figure 2 shows the mean response times by condition with 95% confidence intervals when only the first-presentation trials were included. Overall, bilinguals responded more slowly in English than in Spanish (mean 930 ms vs. 858 ms, β = 87.75; SE β = 40.75; χ2 (1) = 4.85, p = .028), and marginally more slowly with noncognates than cognates (mean 910 ms vs. 876 ms, β = -51.47; SE β = 28.36; χ2 (1) = 3.46, p = .063). However, switch costs were not significant overall, bilinguals named pictures on non-switch vs. switch trials equally quickly (mean 898 ms vs. 887 ms, β = -19.13; SE β = 29.28; χ2 < 1). Of great interest, bilinguals exhibited switch costs for noncognates but switch benefits with cognates, a significant interaction between cognate status and trial type (β = -243.44; SE β = 57.38; χ2 (1) = 11.39, p < .001). Bilinguals responded more slowly on switch trials than non-switch trials with noncognates (mean 961 ms vs. 863 ms; β = 100.54; SE β = 44.10; χ2 (1) = 4.13, p = .042), but significantly more quickly on switch trials than non-switch trials with cognates (mean 801 ms vs.


28

929 ms, β = -137.46; SE β = 47.58; χ2 (1) = 5.60, p = .018; error analyses tended towards switch costs for both cognates and noncognates (mean ≥ 2.5%), but these effects were not significant (ps ≥ .30), thus the switch benefit for cognates in RTs likely could not be attributed merely to a speed-accuracy trade-off). No other fixed effects were significant (ps ≥ .25). Stimulus repetition effects. Similar to Experiments 1 and 2, we further investigated stimulus repetition effects by comparing bilinguals’ naming latency on switch trials in the first four vs. last four presentations of each item (see Figure 3). Note that the number of data points available for analysis in Experiment 3 was only 44% of that in Experiment 2 (because of the smaller number of items). As in Experiment 2, for brevity, we report only the critical interactions involving repetition. The analyses revealed just one significant result, which was that naming latencies decreased with the repeated presentation of stimuli (the first four vs. last four presentations means respectively were 959 ms and 896 ms, β = -64.93; SE β = 34.33, χ2 (1) = 4.13, p = .042). This main effect of stimulus repetition was absent from both Experiments 1 and 2 (ps ≥ .22). All other fixed effects involving repetition failed to reach significance (ps ≥ .28). Recall that in Experiment 2, cognates, but not noncognates, were produced increasingly more slowly with the repeated presentation of stimuli. Though these repetition interference effects in cognates were not significant in Experiment 3, in Figure 3 cognates do appear to benefit less from repetition than noncognates in both languages (though not significantly so, p = .208). Similar analyses of error rates (ps ≥ .10), and RTs and error rates on non-switch trials (ps ≥ .41) showed no significant effects involving stimulus repetition. Discussion Experiment 3 replicated switch costs and reversed dominance effects, and overall cognate facilitation effects were not as robust as in Experiments 1 and 2. More critically, cognates


29

significantly increased switch costs in both languages overall in Experiment 3, but on the first presentation trials cognates significantly reduced switch costs, and even showed significant switch benefits (i.e., faster responses on switch than on non-switch trials – the opposite of switch-costs).6 Additionally, on switch trials in Experiments 2 and 3, noncognates tended to be produced more quickly with repetition, whereas cognates were produced significantly more slowly (Experiment 2) or appeared to benefit less from repetition than noncognates (though not significantly so in Experiment 3). Note that the repetition facilitation effects for noncognates on switch trials were stronger in Experiment 3 with shorter lags between repetitions than in Experiment 2 with longer lags (repetition facilitation was significant only in Experiment 3). Differences in lag between repetitions across experiments might also explain why cognate switch-inhibition effects were significant in Experiment 3 but not in Experiment 2 with repetition-- it seemed to be easier to switch to noncognates with short than long lags between repetitions, so that switch costs of noncognates with short lags (in Experiment 3) were reduced more than with long lags (in Experiment 2). In summary, though the results of Experiments 2 and 3 differed in some ways, both experiments revealed significantly smaller switch costs for cognates than noncognates in first-presentation trials, but repeated presentation of stimuli eliminated or even reversed (in Experiment 3) cognate switch-facilitation effects. General Discussion 6

We examined whether these unexpected switch benefits were unique to the 8 items in Experiment 3, by re-analyzing the data from Experiment 2 but including only the 8 items that were presented in Experiment 3. In the first presentation trials, cognates showed switchfacilitation effects in Spanish, and no switch effects (neither switch costs nor benefits) in English. With repetition, switch costs emerged for both cognates and noncognates, in both languages (i.e., when including all presentation trials). These results provide some evidence that switch benefit for cognates found in Experiment 3 might reflect some property specific to the 4 cognates chosen for study in Experiment 3 (i.e., bottle, camera, lemon, and map). Also, note that when only including these 8 items, the critical results in Experiment 2 still held (i.e., cognate switch-facilitation in first-presentation trials but not when combining all presentations).


30

The present study investigated how cross-language overlap in translation equivalents at the phonological level influences bilingual language control. Spanish-English bilinguals named pictures with cognate vs. non-cognate names in separate blocks (Experiment 1) or mixed blocks (Experiments 2 and 3, with Experiment 3 having fewer total items than Experiment 2). Results replicated several results often reported in the literature on bilingual language switching including switch costs (slower response times when switching than when not switching languages), cognate facilitation effects (faster responses for cognates than noncognates), and reversed dominance effects (slower responses in the dominant than the non-dominant language). In addition, in all 3 experiments we found that cognates reduced switch costs. In Experiment 1 this was true in the dominant language overall and on first-presentation trials, in Experiment 2-3 cognates reduced switch costs on first-presentation trials in both languages. In other analyses cognates had no effect on switch costs (in Experiment 2 overall), or increased switch costs (in Experiment 3 overall). Thus, we replicated in our one study all previously reported effects including cognate switch-facilitation, switch-inhibition, and no effect of cognate status on language switching. Of greatest interest was our finding in the analysis of first-presentation trials which consistently revealed a significant (or marginally significant, in Experiment 1) reduction in switch costs for cognates relative to noncognates, i.e., a cognate switch-facilitation effect in all three experiments. Below, we first consider how cognates might facilitate a language switch prior to repeated presentation of the targets. Subsequently, we consider why the switch advantage for cognates disappeared (Experiment 2) or even reversed significantly (Experiment 3), whereas repeated production of noncognates did not increase switching difficulty. Together these results imply that cognates generally facilitate language switches regardless of context, i.e.,


31

in both separate blocks and when cognates and noncognates are intermixed, but that repetition modulates cognate switch-facilitation effects differently across separate versus mixed blocks. A minor qualification to these observations was that in Experiment 1 cognate switch-facilitation effects were significant only for the dominant language (both overall and in first-presentation trials only), whereas in Experiments 2 and 3 both languages exhibited the switch cost reduction for cognates. How Do Cognates Facilitate Language Switching? Two different accounts might explain how translation equivalent overlap in phonology facilitates switching, though differentiating between them is difficult. First, as suggested by Decklerck et al. (2012), cognates might reduce switch costs strictly because of dual-language activation of phonology, and differences in task demands on non-switch versus switch trials. On this view, when bilinguals need to plan a switch, co-activation of phonology would speed responses because activation of the switched-to language could begin even before the switch is planned. In contrast, on a non-switch trial, increased dual-language activation might slow responses (because the goal is not to switch). Together, speeding of switch trials, and slowing of non-switch trials, would reduce the difference between switch and non-switch trials, i.e., the size of switch costs. By contrast, with non-cognates, dual-language activation would function in the same way on switch vs. non-switch trials. Consistent with this view, cognate facilitation effects were larger on switch trials in Declerck et al. (2012), implying that cognates were easier to produce especially when a switch was being planned. A second possibility is that the extent of dual-language activation might itself be different on non-switch versus switch trials, due to inhibitory control of the non-target language on nonswitch trials. On switch trials the non-target language has been recently activated (since


32

bilinguals just produced words in that language and it might take some time before inhibition can be implemented). However, on non-switch trials, the non-target language might already be successfully inhibited, and if so the extent to which cognates should benefit from dual-language activation would be reduced (i.e., reduced cognate facilitation effects on non-switch trials). Thus, both accounts predict that switching will magnify cognate facilitation effects on switch trials, but only the second account predicts reduced cognate facilitation effects on non-switch trials. Figure 2 reveals a pattern of cognate switch-facilitation effects that seems to implicate non-switch rather than switch trials (i.e., no cognate facilitation effects on non-switch trials). In this line of reasoning we assume that under normal circumstances, without burdening the language control system in a switching task, cognates will be produced more quickly than noncognates (for one or more reasons – dual-language activation at the level of phonology, higher baseline frequency for cognates at the lexical level, or because of feedback from phonology to the lexical level). Figure 2 shows that cognates were indeed produced more quickly than noncognates but only on switch trials. On non-switch trials there was no cognate facilitation effect (noncognates were produced as quickly as cognates). This pattern of results seems more consistent with the second account, which assumes that dual-language activation is suppressed on the non-switch trials. Additional research will be needed to confirm this pattern, as in Declerck et al. (2012) cognates were produced faster than non-cognates on both switch and nonswitch trials, as found here in Experiment 1 when combining all presentations (see Figure 1). It is not clear how these should be compared because, as noted above, Declerck et al. had long preparation times, and did not consider repetition effects in their study. Other possibly important differences between the accounts of cognate switch-facilitation effects can be considered. The task-demands account assumes that activation of shared


33

phonology can sometimes lead to interference (even for cognates, i.e., when the goal is not to switch languages), whereas the second explanation makes no such assumption – though it does invoke the notion of inhibitory control. Additionally, the task-demands account focuses on how phonology affects responses with different goals specified; i.e., activation of shared phonology facilitates switches, while non-switch responses are more difficult when dual-language activation is maximized by overlapping phonology. The inhibition account is more activation based, assuming a modulation of the extent of dual-language activation on switch versus non-switch trials. Previous research suggested that dual-language activation can influence language switching at multiple processing levels. As described above, cognates increased the likelihood of switching at the same position as in a prime sentence (Kootstra et al., 2012). Such structural priming effects are also boosted by shared word order between two languages (Kootstra, van Hell, & Dijkstra, 2010) and lexical repetition (Kootstra et al., 2012) between prime and target sentences. These effects can be explained by assuming co-activation of the two languages at both syntactic and lexical levels. Related to the present study, co-activation of the two languages at the syntactic level (like syntactic cognates) facilitates naturally occurring switches in sentence production. However, where cognates are concerned, co-activation occurs at the phonological level, which is at a later planning stage than syntactic planning. Facilitation effects at this later stage might be weaker than those that happen in earlier processing stages, prior to language selection. Dual-language activation at the phonological level may also facilitate language switching after processing of cognates (the triggering hypothesis; Broersma & De Bot, 2006; Broersma, 2009; Clyne, 1967, 1980). In natural speech, bilinguals are more likely to switch languages after


34

producing a cognate (Broersma, 2009), though this triggering effect was not found on cognate verbs in sentence comprehension task (Bultena, Dijkstra, & van Hell, 2015a, 2015b). In cued language switching with picture naming, Costa and Santesteban (2016) did not find triggering effects. However, preparation times in Costa and Santesteban (2016) were relatively long (as long as 2000 ms), which may leave very limited room for cognate status to modulate switch costs. In addition, the trials were not fully counterbalanced in each condition (e.g., for half of the participants, the cognate status of target word was always same as the preceding trial, while for the other half of the participants it was never the same). In Experiment 2 of the present study, a post-hoc analysis to examine possible triggering effects also failed to reveal faster switch responses after cognates, either in first-presentation trials only or when all trials were combined (ps ≥ .20). However, our study was not designed to address triggering effects and as such cannot provide definitive conclusions (e.g., the number of data points in each cell was small and unbalanced; for example, on average, we had only one and 1.4 data points on average per bilingual when the previous trial was a cognate and the target word was an English noncognate versus a cognate in first-presentation switch trials). Additional work is still needed to examine the triggering effects. At the heart of any account for cognate switch-facilitation effects will need to be a mechanism that allows activation to flow automatically from concepts to phonology in both languages (Colomé, 2001; Costa, Miozzo, and Caramazza, 1999; Costa et al., 2000; Kroll, Bobb, & Wodniecka, 2006). A true switch-facilitation pattern of data would require an additional assumption to explain why between-language similarity at the phonological level does not make it more difficult for bilinguals to distinguish which activated phonemes belong to the target language, or at the very least that any discriminability problems caused by similarity at the


35

phonological level are offset completely by facilitation at a different level of processing. Repetition effects in the present study and their differential effects with blocked versus mixed presentation of cognates and noncognates in the present study shed some light on this question. More detailed consideration of different types of cognates might also be revealing. For example, cognate pairs may differ in terms of the degree of the nature of form overlap and semantic similarity (Allen & Conklin, 2013; Dijkstra, Miwa, Brummelhuis, Sappelli, & Baayen, 2010; Schwartz & Kroll, & Diaz, 2007; Taylor, 1976; Titone et al., 2011; Van Hell & De Groot, 1998), which could also modulate cognate effects on switching. Why Does Stimulus Repetition Abolish Cognate Reduction in Switch Costs? Instead of asking if cognates facilitate a language switch, a priori we might even have considered the opposite possibility. That is, why doesn’t cross-language similarity at phonological level actually make it harder to switch languages because of increased difficulty with discriminating which representations belong to the target language? This question was also raised by previous reports of cognate switch-inhibition effects (Christoffels et al., 2007) and our finding that cognate switch-facilitation effects disappeared with repetition in Experiment 2. The latter result suggests that with repeated production of names in both languages, the discrimination problem indeed became more and more difficult but only for names that are phonologically similar across languages (i.e., for cognates but not for noncognates), and only on switch trials. Neither cognates nor noncognates exhibited slowed responses with repeated presentations of stimuli on non-switch trials. In other words, on non-switch trials, the discrimination problem would be less severe if dual-language activation is weaker due to successful inhibition of the nontarget language – again implicating non-switch trials as being critical for the patterns observed herein.


36

To explain the increasing discrimination difficulty for cognates on switch trials with repetition it is necessary to assign cognate facilitation and cognate inhibition effects to different processing levels. While facilitation effects on switch trials might arise at the phonological level (via cascading activation from the lexical level), discrimination difficulties could arise at the lexical level due to feedback from the phonological level back up to the lexical level. A further assumption we make here is that ultimately language selection must also occur at the lexical level (i.e., even if activation cascades forward automatically from lexical representations in both languages to the phonological level, ultimately a single lexical representation must be selected for production at the lexical level). With increasing repetition, activation would flow more rapidly and automatically in all directions – including both cascade forward and feedback. However, repetition would magnify feedback effects more – because these start out weaker than cascade effects, increasing discriminability problems at the lexical level. Without repetition, feedback effects would be relatively weak, taking more time to emerge, and ultimately influencing selection much less than cascade effects (Cutting & Ferriera, 1999; Damian & Martin, 1999). Indeed the shift towards slowed responses for cognates on switch trials with repetition is particularly striking given that repetition usually decreases naming latencies (i.e., repetition priming effects, see Francis, & Sáenz, 2007; Gollan, Montoya, Fennema-Notestine, & Morris, 2005; Ivanova & Costa, 2008; but see Waszak, Hommel, & Allport, 2003 for an interference effect). Repetition priming effects might have been relatively weak in the present study because of the familiarization phase of the procedure, but importantly noncognates and cognates on nonswitch trials did at least exhibit means in this direction (i.e., tended to be produced more quickly with repetition, possibly reflecting strengthened association between lemmas and their


37

phonological forms; Barry, Hirsh, Johnston, & Williams, 2001). Additionally, in Experiment 3, with a smaller set of items and therefore reduced lag between successive repetitions of each item, noncognates in fact revealed a precipitous and significant quickening of response times with repetition (but cognates did not). On this view, the discrimination problem would still have been in play in Experiment 3, but was weaker (no repetition inhibition on cognates as shown in Experiment 2) because of more recent repetition and shorter lags between items. A further question that arises is why cognates became increasingly difficult to produce on switch trials but only when mixed in the same testing block with noncognates? Similar sounds can, under some circumstances elicit stronger competition for selection than dissimilar sounds, for example in the tongue twister paradigm in which a small set of targets are produced repeatedly (Wilshire, 1998, 1999). Previous studies also included repetition when cognates and non-cognates were inter-mixed (Christoffels et al., 2007; Verhoef et al., 2009), and the rate of switches in these two studies was same as in the present study and Declerck et al. (2012). Christoffels et al. had 12 repetitions of each item, and target pictures were presented at the same time as the language cues, leaving no preparation time, just as in the present study. Interestingly, Christoffels et al. found cognate switch-inhibition effects as in our Experiment 3. In Verhoef et al. there were 32 repetitions, but long preparation times (750 ms or 1500 ms), which might explain the absence of cognate switch-inhibition effects in that study. Similarly, in Santesteban and Costa (2016) there were more than 40 repetitions and the preparation time was even longer (2000 ms). Another way of framing the blocked versus mixed contrast is to ask why phonological competition effects become more robust with repeated presentation of stimuli and when bilinguals did not know on a given trial if they will or will not face a difficult discrimination


38

problem – i.e., when cognates and noncognates were intermixed in the same block. Bilinguals in the present study knew which items would be presented in the critical trials from exposure during the familiarization phase. This knowledge could have elicited a mode of processing that focuses selection, control, or monitoring processes (or some combination thereof) more carefully at the phonological level on every trial, so that feedback would lead to stronger discrimination problems at the lexical level. This in turn might have obscured or eliminated discriminability problems in the cognates only block. However, such a strategy (or special focus) may be generally difficult and as such would be implemented only when truly necessary (i.e., when cognates and noncognates are intermixed so that some of the trials, namely cognates, pose a discrimination problem, while other trials do not). On this view, the blocked presentation of cognates in Declerck et al. (2012) revealed cognate switch-facilitation effects for different reasons than they might arise normally (as in first-presentation trials in Experiments 2-3). Other aspects of the data also hint at the possibility that cognate switch-facilitation effects in Experiment 1 might reflect a different underlying cognitive mechanism than found in Experiments 2-3. As noted above, only English exhibited cognate-switch facilitation effects in Experiment 1, whereas facilitation was found in both languages in Experiments 2-3 (additional work will be need to confirm the reliability of these possibly different patterns). An alternative possible explanation involves repetition of the same stimulus in what might be considered two different tasks, that is, naming the same picture in English versus in Spanish. Each time the picture is presented, both task contexts may be retrieved thus competing and leading to an interference effect. The possibility of competition between different stimulustask bindings has been used to explain the modulation of task switching costs in monolinguals


39

(Waszak et al., 2003).7 In the present study, the repeated presentation of a picture may activate the previous episodes that this picture was encountered and also the language in which it was previously named, leading to interference. However, this task-binding theory could not explain why we did not find repetition interference effects in Experiment 1, when cognates and noncognates were presented in separate blocks. In addition, other data imply that language switching reflects language selection more than task or response selection (i.e., binding). Slevc, Davey, and Linck (2016) had Chinese-English bilinguals switch between languages in response to inherently univalent stimuli (written English words and Chinese Characters) as well as lexically univalent, but orthographically bivalent, stimuli (English words and Chinese Pinyin). That is, a written Chinese character or its orthography provides an unambiguous language cue due to its unique script (e.g.

, meaning sparrow, cannot cue a response in English).

However, Pinyin is a writing system that represents Chinese words with Roman alphabets, thus is orthographically bivalent (e.g. the orthographic strings máquè or sparrow could be mapped Chinese or English phonemes. Significant switch costs were found in naming both orthographically bivalent stimuli, and importantly, even with inherently univalent stimuli. Waszak et al. (2003) presented pictures with semantically related words superimposed (pictureword interference) and asked participants to switch between naming the pictures and naming the words. Additionally, some of the words had never appeared previously on picture naming trials, while others had been previously presented. In this paradigm, task switching costs for word naming were significantly larger for stimuli that had previously been presented on picture naming trials. In addition, this interference effect remained robust even when subjects had performed more than 100–200 intervening trials between the competing task contexts for the same stimulus, suggesting that the effect was not a result of persisting inhibition of the distractor representation (see Houghton & Tipper, 1994). The interference effect also could not be explained by the competing stimulus-response association, as it was robust even in a congruent response condition (i.e., when the same item was always associated with the same response; e.g., the item cap was presented as a picture in the picture naming trial, then was presented as a word in the word naming trial, so that the response cap was produced in both cases, even though it appeared in both picture naming and word naming tasks). Therefore, as the stimulus-task binding theory predicted, when the same stimulus was bound with different tasks (i.e., picture vs. word naming), the task switching cost might result from interference from the previous task context. 7


40

Therefore, switch costs were not a result of the difficulty of response selection. As such, the discrimination difficulties account may better explain why repetition shifted cognate effects from facilitation to null or inhibition but only in mixed blocks, but further research is needed to verify the patterns we observed and determine which explanation works best. Conclusions Language switching is facilitated when the translation equivalent of the intended target is phonologically similar. This facilitation may reflect dual-language activation, which is strong on switch trials but inhibited on non-switch trials. However, when targets are produced in both languages repeatedly, switching becomes increasingly difficult when translations overlap in form because of magnified effects of feedback to the lexical level, and increased competition for selection at this processing level. This increased difficulty with switching on cognates with repetition can be circumvented if all the targets in a block of trials are cognates, suggesting that dual-language activation can be modulated across a block of trials to weight less heavily the effects of feedback on selection (for similar modulation of control of dual-language activation in different contexts see Green & Abutalebi, 2013). Together these results suggest that activation flows automatically from lexical representations in both languages to the phonological level (i.e., cascaded activation), and that the extent to which activation at the phonological level affects lexical selection can be modulated by context –perhaps strategically (see similar effects in comprehension of language switches, cognate effects were modulated by sentence context in some cases, see Kroll, Bobb, Misra, & Guo, 2008; Libben & Titone, 2009; Schwartz & Kroll, 2006; Titone, et al., 2011; but see Van Assche, Drieghe, Duyck, Welvaert, & Hartsuiker, 2011). When bilinguals must choose a single lexical representation in speech production, repetition makes it increasingly more difficult to


41

select between phonologically similar translation equivalents (i.e., cognates) at the lexical level, but only when these are intermixed with phonologically dissimilar translations (i.e., noncognates). Importantly, the cognates-only block is not typical, and neither is massive repetition of items, nor production of single words out of context. A future question to address will be whether cognates facilitate language switching in connected and more spontaneous speech. In such productions, it might be easier for bilinguals to switch on cognates, regardless of target language, degree of repetition, and the proportion of cognates in speech.


42

References Allen, D. B., & Conklin, K. (2013) Cross-linguistic similarity and task demands in JapaneseEnglish bilingual processing. PLoS ONE, 8(8), 1-14, DOI: 10.1371/journal.pone.0072631 Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59(4), 390-412. Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255-278. Barry, C., Hirsh, K. W., Johnston, R. A., & Williams, C. L. (2001). Age of acquisition, word frequency, and the locus of repetition priming of picture naming. Journal of Memory and Language, 44(3), 350-375. Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). lme4: Linear mixed-effects models using Eigen and S4. R package version 1.1–7. 2014. Institute for Statistics and Mathematics of WU website. http://CRAN. R-project. org/package= lme4. Accessed March, 18. Bobb, S. C., & Wodniecka, Z. (2013). Language switching in picture naming: What asymmetric switch costs (do not) tell us about inhibition in bilingual speech planning. Journal of Cognitive Psychology, 25(5), 568–585. Bultena, S., Dijkstra, T., & van Hell, J. G. (2015a). Switch cost modulations in bilingual sentence processing: Evidence from shadowing. Language, Cognition and Neuroscience, 30, 586–605.


43

Bultena, S., Dijkstra, T., & van Hell, J. G. (2015a). Language switch costs in sentence comprehension depend on language dominance: Evidence from self-paced reading. Bilingualism: Language and Cognition, 18(03), 453-469. Broersma, M. (2009). Triggered codeswitching between cognate languages. Bilingualism: Language and Cognition, 12, 447–462. Broersma, M., & de Bot, K. (2006). Triggered codeswitching: A corpus-based evaluation of the original triggering hypothesis and a new alternative. Bilingualism: Language and Cognition, 9, 1–13. Christoffels, I. K., Firk, C., & Schiller, N. O. (2007). Bilingual language control: An eventrelated brain potential study. Brain Research, 1147, 192–208. Clyne, M. (1967). Transference and triggering: Observations on the language assimilation of postwar German-speaking migrants in Australia. Den Haag: Martinus Nijhoff. Clyne, M. (1980). Triggering and language processing. Canadian Journal of Psychology, 34, 400–406. Colomé, À. (2001). Lexical activation in bilinguals' speech production: Language-specific or language-independent? Journal of Memory and Language, 45(4), 721-736. Costa, A., Caramazza, A., & Sebastián-Gallés, N. (2000). The cognate facilitation effect: Implications for models of lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(5), 1283–1296. Costa, A., Miozzo, M., & Caramazza, A. (1999). Lexical selection in bilinguals: Do words in the bilingual's two lexicons compete for selection?. Journal of Memory and Language, 41(3), 365-397.


44

Costa, A., & Santesteban, M. (2004). Lexical access in bilingual speech production: Evidence from language switching in highly proficient bilinguals and L2 learners. Journal of Memory and Language, 50(4), 491-511. Costa, A., Santesteban, M., & Caño, A. (2005). On the facilitatory effects of cognate words in bilingual speech production. Brain and language, 94(1), 94-103. Costa, A., Santesteban, M., & Ivanova, I. (2006). How do highly proficient bilinguals control their lexicalization process? Inhibitory and language-specific selection mechanisms are both functional. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32(5), 1057-1074. Cutting, J. C., & Ferreira, V. S. (1999). Semantic and phonological information flow in the production lexicon. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(2), 318. Damian, M. F., & Martin, R. C. (1999). Semantic and phonological codes interact in single word production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(2), 345-361. Declerck, M., Koch, I., & Philipp, A. M. (2012). Digits vs. pictures: The influence of stimulus type on language switching. Bilingualism: Language and Cognition, 15(04), 896-904. Declerck, M., & Philipp, A. M. (2015a). A review of control processes and their locus in language switching. Psychonomic Bulletin & Review, 22(6), 1630-1645. Declerck, M., & Philipp, A. M. (2015b). The unusual suspect: Influence of phonological overlap on language control. Bilingualism: Language and Cognition, 18(04), 726-736. De Groot, A. M., & Christoffels, I. K. (2006). Language control in bilinguals: Monolingual tasks and simultaneous interpreting. Bilingualism: Language and Cognition, 9(02), 189-201.


45

Dijkstra, T., Miwa, K., Brummelhuis, B., Sappelli, M., & Baayen, H. (2010). How crosslanguage similarity and task demands affect cognate recognition. Journal of Memory and language, 62(3), 284-301. Dijkstra, A., & Van Heuven, W. J. B., (1998). The BIA model and bilingual word recognition. In J. Grainger & A. Jacobs (Eds.), Localist Connectionist Approaches to Human Cognition (pp. 189–225). Hillsdale, NJ: Lawrence Erlbaum Associates Dijkstra, T., & Van Heuven, W. J. (2002). The architecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition, 5, 175– 197. Filippi, R., Karaminis, T., & Thomas, M. S. (2014). Language switching in bilingual production: Empirical data and computational modeling. Bilingualism: Language and Cognition, 17(02), 294-315. Fink, A., & Goldrick, M. (2015). Pervasive benefits of preparation in language switching. Psychonomic Bulletin & Review, 22(3), 808-814. Francis, W. S., & Sáenz, S. P. (2007). Repetition priming endurance in picture naming and translation: Contributions of component processes. Memory & Cognition, 35(3), 481-493. Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevel/hierarchical models. Cambridge, UK: Cambridge University Press. Goldrick, M., Runnqvist, E., & Costa, A. (2014). Language switching makes pronunciation less native-like. Psychological Science, 25, 1031–1036. Gollan, T. H., & Acenas, L. A. R. (2004). What is a TOT? Cognate and translation effects on tipof-the-tongue states in Spanish-English and tagalog-English bilinguals. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30(1), 246-269.


46

Gollan, T. H., & Ferreira, V. S. (2009). Should I stay or should I switch? A cost–benefit analysis of voluntary language switching in young and aging bilinguals. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35(3), 640. Gollan, T. H., & Goldrick, M. (2016). Grammatical constraints on language switching: Language control is not just executive control. Journal of Memory and Language, 90, 177-199. Gollan, T. H., Montoya, R. I., Fennema-Notestine, C., & Morris, S. K. (2005). Bilingualism affects picture naming but not picture classification. Memory & Cognition, 33(7), 12201234. Gollan, T. H., Sandoval, T., & Salmon, D. P. (2011). Cross-language intrusion errors in aging bilinguals reveal the link between executive control and language selection. Psychological Science, 22, 1155– 1164. Gollan, T. H., Schotter, E. R., Gomez, J., Murillo, M., & Rayner, K. (2014). Multiple levels of bilingual language control: Evidence from language intrusions in reading aloud. Psychological Science, 25, 585-595. Gollan, T. H., Weissberger, G. H., Runnqvist, E., Montoya, R. I., & Cera, C. M. (2012). Selfratings of spoken language dominance: A multi-lingual naming test (MINT) and preliminary norms for young and aging Spanish-English bilinguals. Bilingualism (Cambridge, England), 15(3), 594-615. Grainger, J., & Dijkstra, T. (1992). On the representation and use of language information in bilinguals. In R. J. Harris (Ed.), Cognitive processing in bilinguals. Amsterdam, The Netherlands: North Holland. Green, D. W. (1998). Mental control of the bilingual lexico- semantic system. Bilingualism: Language and Cognition, 1, 213–229.


47

Green, D. W., & Abutalebi, J. (2013). Language control in bilinguals: The adaptive control hypothesis. Journal of Cognitive Psychology, 25(5), 515-530. Guo, T., Liu, H., Misra, M., & Kroll, J. F. (2011). Local and global inhibition in bilingual word production: fMRI evidence from Chinese–English bilinguals. NeuroImage, 56(4), 23002309. Hermans, D., Bongaerts, T., De Bot, K., & Schreuder, R. (1998). Producing words in a foreign language: Can speakers prevent interference from their first language? Bilingualism: Language and Cognition, 1, 213–229. Hoshino, N., & Kroll, J. F. (2008). Cognate effects in picture naming: Does cross-language activation survive a change of script? Cognition, 106, 501–511. Houghton, G., & Tipper, S. P. (1994). A model of inhibitory mechanisms in selective attention. In D. Dagenbach & T. H. Carr (Eds.), Inhibitory processes in attention, memory, and language (pp. 53–112). San Diego, CA: Academic Press. Ivanova, I., & Costa, A. (2008). Does bilingualism hamper lexical access in speech production? Acta Psychologica, 127(2), 277-288. Jescheniak, J. D., & Schriefers, H. (1998). Discrete serial versus cascaded processing in lexical access in speech production: Further evidence from the coactivation of nearsynonyms. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24(5), 1256-1274. Kleinman, D., & Gollan, T. H. (2016). Speaking two languages for the price of one: Bypassing language control mechanisms via accessibility-driven switches. Psychological Science, 27, 700-714.


48

Kohnert, K. (2004). Cognitive and cognate-based treatments for bilingual aphasia: A case study. Brain and Language, 91(3), 294-302. Kolers, P. A. (1966). Reading and talking bilingually. The American Journal of Psychology, 79(3), 357-376. Kootstra, G. J., van Hell, J. G., & Dijkstra, T. (2010). Syntactic alignment and shared word order in code-switched sentence production: Evidence from bilingual monologue and dialogue. Journal of Memory and Language, 63, 210–231. Kootstra, G. J., van Hell, J. G., & Dijkstra, T. (2012). Priming of code-switches in sentences: The role of lexical repetition, cognates, and language proficiency. Bilingualism: Language and Cognition, 15(04), 797-819. Kroll, J. F., Bobb, S. C., Misra, M., & Guo, T. (2008). Language selection in bilingual speech: Evidence for inhibitory processes. Acta psychologica, 128(3), 416-430. Kroll, J. F., Bobb, S. C., & Wodniecka, Z. (2006). Language selectivity is the exception, not the rule: Arguments against a fixed locus of language selection in bilingual speech. Bilingualism: Language and Cognition, 9(02), 119-135. Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33(2), 149-174. La Heij, W. (2005). Selection processes in monolingual and bilingual lexical access. In J. F. Kroll & A. M. B. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 289– 307). Oxford, England: Oxford University Press. Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Brain and Behavioral Sciences, 22, 313- 335.


49

Libben, M. R., & Titone, D. A. (2009). Bilingual lexical access in context: evidence from eye movements during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35(2), 381-390. Meuter, R. F. I., & Allport, A. (1999). Bilingual language switching in naming: Asymmetrical costs of language selection. Journal of Memory and Language, 40, 25–40. Misra, M., Guo, T., Bobb, S. C., & Kroll, J. F. (2012). When bilinguals choose a single word to speak: Electrophysiological evidence for inhibition of the native language. Journal of Memory and Language, 67(1), 224-237. Monsell, S., Matthews, G. H., & Miller, D. C. (1992). Repetition of lexicalization across languages: A further test of the locus of priming. The Quarterly Journal of Experimental Psychology, 44(4), 763-783. Moon, J., & Jiang, N. (2012). Non-selective lexical access in different-script bilinguals. Bilingualism: Language and Cognition, 15(01), 173-180. Mosca, M., & Clahsen, H. (2016). Examining language switching in bilinguals: The role of preparation time. Bilingualism: Language and Cognition, 19(02), 415-424. Norman, D., & Shallice, T. (1986). Attention to action: Willed and automatic control of behavior. In R. Davidson, R. G. Schwartz, & D. Shapiro (Eds.), Consciousness and self-regulation: Advances in research and theory (pp. 1–18). New York: Plenum Press. Olson, D. J. (2013). Bilingual language switching and selection at the phonetic level: Asymmetrical transfer in VOT production. Journal of Phonetics, 41, 407–420. Peterson, R. R., & Savoy, P. (1998). Lexical selection and phonological encoding during language production: Evidence for cascaded processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24(3), 539-557.


50

Philipp, A. M., Gade, M., & Koch, I. (2007). Inhibitory processes in language switching: Evidence from switching language-defined response sets. European Journal of Cognitive Psychology, 19(3), 395-416. Poplack, S. (1988). Contrasting patterns of code-switching in two communities. Codeswitching: Anthropological and sociolinguistic perspectives, 215, 44. Poulisse, N., & Bongaerts, T. (1994). First language use in second language production. Applied Linguistics, 15(1), 36–57. R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.Rproject.org/. Roberts, P. M., & Deslauriers, L. (1999). Picture naming of cognate and non-cognate nouns in bilingual aphasia. Journal of Communication Disorders, 32(1), 1-23. Santesteban, M., & Costa A. (2016). Are cognate words "Special"? On the role of cognate words in language switching performance. In J. W. Schwieter (Ed.), Cognitive control and consequences of multilingualism (p. 97 - 126). John Benjamins Publishing. Schwartz, A. I., & Kroll, J. F. (2006). Bilingual lexical activation in sentence context. Journal of Memory and Language, 55(2), 197-212. Schwartz, A. I., Kroll, J. F., & Diaz, M. (2007). Reading words in Spanish and English: Mapping orthography to phonology in two languages. Language and Cognitive Processes, 22, 106129. Strijkers, K., Costa, A., & Thierry, G. (2009). Tracking lexical access in speech production: electrophysiological correlates of word frequency and cognate effects. Cerebral Cortex, 20 (4), 912-928. doi: 10.1093/cercor/bhp153


51

Taylor, I. (1976). Similarity between French and English words—a factor to be considered in bilingual language behavior? Journal of Psycholinguistic Research, 5(1), 85-94. Thomas, M. S., & Allport, A. (2000). Language switching costs in bilingual visual word recognition. Journal of Memory and Language, 43(1), 44-66. Titone, D., Libben, M., Mercier, J., Whitford, V., & Pivneva, I. (2011). Bilingual lexical access during L1 sentence reading: The effects of L2 knowledge, semantic constraint, and L1– L2 intermixing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37(6), 1412. Van Assche, E., Drieghe, D., Duyck, W., Welvaert, M., & Hartsuiker, R. J. (2011). The influence of semantic constraints on bilingual word recognition during sentence reading. Journal of Memory and Language, 64(1), 88-107. Van Hell, J. G., & De Groot, A. M. B. (1998). Conceptual representation in bilingual memory: Effects of concreteness and cognate status in word association. Bilingualism: Language and Cognition,1(3), 193-211. Verhoef, K. M. W., Roelofs, A., & Chwilla, D. J. (2009). Role of inhibition in language switching: Evidence from event-related brain potentials in overt picture naming. Cognition, 110, 84–99. Verhoef, K. M., Roelofs, A., & Chwilla, D. J. (2010). Electrophysiological evidence for endogenous control of attention in switching between languages in overt picture naming. Journal of Cognitive Neuroscience, 22(8), 1832-1843. Waszak, F., Hommel, B., & Allport, A. (2003). Task-switching and long-term priming: Role of episodic stimulus-task bindings in task-shift costs. Cognitive Psychology, 46(4), 361-413.


52


53

Table 1. Means and standard deviations of participant characteristics

Characteristic Age Age of Acquisition of English Age of Acquisition of Spanish Self-rated spoken English proficiencya Self-rated spoken Spanish proficiencya Current percent of English use Percent of English use during childhood Primary caregiver English proficiencya Secondary caregiver English proficiencya Primary caregiver Spanish proficiencya Secondary caregiver Spanish proficiencya Years lived in Spanish-speaking country MINT score in Englishb MINT score in Spanishb

Experiment1 M SD 20.3 2.0 3.3 2.6 0.4 1.0 6.7 0.5 5.9 1.1 78.8 16.8 54.6 15.2 3.7 1.9 3.5 1.8 6.8 0.6 6.9 0.3 0.8 1.8 61.4 2.7 47.0 8.4

Experiment 2 M SD 20.6 2.2 3.2 2.5 0.7 1.3 6.7 0.6 5.9 0.9 79.3 16.5 58.4 15.9 4.0 1.6 3.5 1.9 6.8 0.5 6.8 0.5 1.5 3.1 62.5 2.9 45.4 9.2

Experiment 3 M SD 20.1 2.6 3.4 3.4 0.8 1.6 6.5 0.7 5.8 0.9 87.1 12.4 54.8 20.7 3.6 2.1 3.5 2.2 6.9 0.5 7.0 0 1.7 4.0 61.4 3.3 42.3 8.6

a Proficiency-level self-ratings were obtained using a scale from 1 (almost none) to 7 (like a native speaker). b The maximum possible MINT score is 68 Note. The only difference between Experiments 2 and 3 is that, English was reported to be used more frequently, and second caregivers’ Spanish proficiency was higher than in Experiment 2 (ps ≤ .038).


54

Table 2. The characteristics of the cognate and non-cognate stimuli for Experiments 1 and 2

Frequencya (English name) Frequencyb (Spanish name) Number of Syllables (English Name) Number of Syllables (Spanish Name) Visual Complexityc a

Cognates vs. Non-cognates t-value p-value 0.51 0.62

Cognates M SD 127.04 202.68

Non-cognates M SD 83.16 160.54

131.50

227.84

56.75

71.20

0.94

0.36

1.56

0.72

1.78

0.67

0.68

0.51

2.44

0.88

2.00

0.87

1.08

0.30

1.97

0.84

1.94

1.02

0.08

0.94

The frequency refers to the number per million based on SUBTLEX_US (Brysbaert & New, 2009) b The frequency refers to the number per million based on SUBTLEX_ESP (Vega, Nosti, Gutiérrez, & Brysbaert, 2011) c The visual complexity was rated by 18 native English-speaking college students from 1 (very simple) to 4 (in the middle) to 7 (too complex to recognize the picture).


55

Table 3. The characteristics of the cognate and non-cognate stimuli in Experiment 3

Frequencya (English name) Frequencyb (Spanish name) Number of Syllables (English Name) Number of Syllables (Spanish Name) Visual Complexityc a

Cognates vs. Non-cognates t-value p-value 0.91 0.40

Cognates M SD 37.90 20.30

Non-cognates M SD 178.16 309.26

47.48

38.72

198.70

349.23

0.86

0.42

1.75

0.50

2.00

0.82

0.52

0.62

2.50

0.58

2.75

0.96

0.45

0.67

2.29

0.89

1.86

0.91

0.68

0.52

The frequency refers to the number per million based on SUBTLEX_US (Brysbaert & New2009) b The frequency refers to the number per million based on SUBTLEX_ESP (VegaNostiGutiérrez& Brysbaert2011) c The visual complexity was rated by 18 native English-speaking college students from 1 (very simple) to 7 (too complex to recognize the picture).


56

Table 4. Mean error rates (%) and standard deviation (SD) for each condition (trial type cognate status) and language in Experiment 1 (n = 9 cognates and n = 9 non-cognates were presented in separate blocks) Experiment 2 (n = 9 cognates and n = 9 non-cognates were mixed) and Experiment 3 (n = 4 cognates and n = 4 non-cognates were mixed)

English

Cognates Noncognate

Spanish

Cognates Noncognate

Non-Switch Switch Non-Switch Switch

Experiment 1 M SD 2.55 3.45 4.98 4.85 5.09 6.15 6.71 6.68

Experiment 2 M SD 4.40 5.44 7.06 6.41 5.56 5.40 7.99 6.66

Experiment 3 M SD 2.08 4.23 4.95 5.54 4.43 6.34 5.47 6.21

Non-Switch Switch Non-Switch Switch

1.50 3.47 1.50 3.59

2.31 6.02 2.89 4.63

1.30 4.17 1.82 5.47

2.07 3.87 2.95 3.58

3.61 5.47 4.58 5.64

3.07 4.73 5.07 6.56


57

Figure 1. Mean response time for each condition (trial type, cognate status) and language in Experiment 1 (cognates and non-cognates were in separate blocks), Experiment 2 (cognates and non-cognates were mixed), and Experiment 3 (with fewer items, cognates and non-cognates were mixed). Error bars represent 95% confidence intervals.


58

Figure 2. Mean response time of “first time trials” for each condition (trial type, cognate status) and language in Experiment 1 (cognates and non-cognates were in separate blocks), Experiment 2 (cognates and non-cognates were mixed), and Experiment 3 (with fewer items, cognates and non-cognates were mixed). Error bars represent 95% confidence intervals.


Figure 3. Mean response times of their first third vs. the last third presentation of items in each condition (cognate status and language) on switch trials in Experiment 1 (cognates and noncognates were in separate blocks), Experiment 2 (cognates and non-cognates were mixed), and Experiment 3 (with fewer items, cognates and non-cognates were mixed). Error bars represent 95% confidence intervals.

59


60

Appendix.

Experiments 1 & 2

Experiments 3

Cognates English baby bottle camera circle lemon lion map piano plate

Spanish bebé botella cámara círculo limón león mapa piano plato

Non-cognates English Spanish butterfly mariposa dog perro shoes zapatos king rey leg pierna money dinero pencil lápiz tooth diente rabbit conejo

bottle camera lemon map

botella cámara limón mapa

butterfly money leg pencil

mariposa dinero pierna lápiz