Ecological validity and bilingual language control: voluntary language switching between sentences

ABSTRACT

Few language switching studies have found conditions in which there is no significant cost to switching languages. Since language-switch costs are a measure of language control, this could be seen as evidence for the ubiquity of this process in bilingual language production. However, one claim is that ecologically valid bilingual contexts lead to small or even absent switch costs. To further investigate this, we examined voluntary language switching between sentences. This ecologically more valid setup (compared to the more prominent involuntary language switching setup with single word production) resulted in switch costs for sentences produced in the second language, but no significant switch costs for sentences produced in the first language, whereas involuntary language switching between sentences resulted in substantial switch costs across both languages. These results indicate that more ecologically valid contexts can lead to circumstances that might require little to no language control.

KEYWORDS:

• Bilingual language production
• voluntary language switching
• cued language switching
• sentence production

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We also ran an analysis based on error rates. However, before delving into this analysis, it should be noted that participants were told that in the voluntary language switching task they could use any language that they wanted. So, participants could actually use another language in a sentence for one (or several) word(s), and it would not be problematic according to the task at hand. While this only happened very infrequently, and we took these sentences out of the analyses, this means that the most informative error type in language switching studies, namely language errors, could not be taken into account.

The main effect of Language was not significant, b = 0.176, SE = 0.135, z = 1.306, p = .191. Trial type was significant, b = 0.653, SE = 0.126, z = 5.197, p < .001, with more errors during switch trials (13.8%) than during repetition trials (7.1%). Language switching variant was also significant, b = 0.755, SE = 0.190, z = 3.982, p < .001, with more errors during cued (13.6%) than during voluntary (6.4%) language switching. The interaction between Language and Trial type was significant, b = 0.634, SE = 0.248, z = 2.551, p = .011, with larger L2 (8.2%) than L1 (4.9%) switch costs. The interaction between Language and Language switching variant was not significant, b = 0.072, SE = 0.256, z = 0.281, p = .779. The interaction between Trial type and Language switching variant was significant, b = 1.294, SE = 0.251, z = 5.158, p < .001, with larger switch costs in cued (14.4%) than in voluntary language switching (0.1%). Finally, also the three-way interaction was significant, b = 1.048, SE = 0.496, z = 2.115, p = .035, with similar L1 (14.9%) and L2 (13.8%) switch costs during cued language switching and larger L2 (3.1%) than L1 (-3.8%) switch costs during voluntary language switching.

2 The following strategy was used in case of an issue with the fully randomised model (cf. Barr et al., 2013; Matuschek et al., 2017): First, random effects for the item-specific random slopes were excluded, starting with the higher-order interactions. If the issue was not resolved, the higher-order interactions of the participant-specific random slopes were excluded. If this does not resolve the issue, the lower-order terms were removed, again starting with the item-specific random slopes before moving on to the participant-specific random slopes.

3 To exclude the possibility that any switch cost differences between voluntary and cued language switching were due to a difference in switch rate, we conducted an analysis with the addition of switch rate as a continuous (centered) variable. The results showed no three-way interaction between Trial type, Language switching variant, and switch rate, b = 0.001, SE = 0.016, z = 0.618, p = .537. Furthermore, the four-way interaction between Language, Trial type, Language switching variant, and switch rate was also not significant, b = 0.034, SE = 0.030, z = 1.135, p = .256. So, it seems unlikely that switch rate caused any switch cost differences between voluntary and cued language switching.

4 It should be noted that we are not the first voluntary language switching paper that has a switch rate around 47%. Jevtović et al. (2020) found a 50.8% switch rate in Spanish and 40.7% in Basque, whereas de Bruin et al. (2018) observed a 49.8% switch rate in Spanish and 38.0% in Basque.

5 In an exploratory analysis, along the lines of Kleinman and Gollan (2016), we also included the following continuous variable: Block type order (Voluntary first = −0.5; Cued first = 0.5). The results showed no significant four-way interaction between Language, Trial type, Language switching variant and Block type order, b = 0.352, SE = 0.886, z = 0.397, p = .691. However, there was a trend towards larger cued switch costs (7.1%) than voluntary switch costs (5.3%) when performing cued language switching first, whereas smaller cued switch costs (0%) than voluntary switch costs (4.7%) were observed when performing voluntary language switching first, b = 0.814, SE = 0.449, z = 1.814, p = .070.