When exceptions matter: bilinguals regulate their dominant language to exploit structural constraints in sentence production

ABSTRACT

What we say generally follows distributional regularities, such as learning to avoid “the asleep dog” because we hear “the dog that’s asleep” in its place. However, not everyone follows such regularities. We report data on English monolinguals and Spanish-English bilinguals to examine how working memory mediates variation in a-adjective usage (asleep, afraid), which, unlike typical adjectives (sleepy, frightened), tend to resist attributive use. We replicate previous work documenting this tendency in a sentence production task. Critically, for all speakers, the tendency to use a-adjectives attributively or non-attributively was modulated by individual differences in working memory. But for bilinguals, a-adjective use was additionally modulated by an interaction between working memory and category fluency in the dominant language (English), revealing an interactive role of domain-general and language-related mechanisms that enable regulation of competing (i.e. attributive and non-attributive) alternatives. These results show how bilingualism reveals fundamental variation in language use, memory, and attention.

KEYWORDS:

• Sentence production
• bilingualism
• individual differences
• working memory
• category fluency

Acknowledgements

We would like to thank Anne Beatty-Martínez, Adele Goldberg, David W. Green, Jason Gullifer, Holger Hopp, and Manuel Pulido for providing valuable feedback.

Disclosure statement

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

Notes

1 Previous work uses the term “language control” to characterise inhibitory effects when bilinguals switch from one language to another in cued digit or single-word switching paradigms (see Bobb & Wodniecka, 2013; Declerck & Philipp, 2015 for reviews). Language control is often examined correlationally with task switching effects to assess its domain-generality (e.g. Calabria et al., 2012; Declerck et al., 2017). Although we view language control as part of bilingual language regulation, we make the distinction to emphasise the interplay between down-regulatory and up-regulatory dynamics engaged at different levels of language processing (e.g. phonological, lexical-semantic, grammatical, sentential-discourse). As we have stated (see also Kroll et al., 2021; Zirnstein et al., 2018; Zirnstein et al., 2019), such interplay emerges from the interaction between language (e.g. lexical-semantic accessibility) and domain-general cognitive (e.g. attention and working memory) mechanisms, and not from independent or correlational effects.

2 Unlike English, which allows for prenominalisation, attributive adjectives in Spanish typically follow postnominalisation (e.g. el perro grande, “the dog big”) or predicative constructions, with several exceptions that allow for prenominalisation under certain circumstances (Liceras et al., 2002; Varela, 2012). Although it is beyond the focus of the present study to directly examine cross-linguistic interactions, we note that previous work has shown that cross-linguistic variation in adjectival placement can affect children’s production choices (Nicoladis, 2006) and can result in processing interference when there is partial syntactic overlap between the two languages (Luque et al., 2018; Sanoudaki & Thierry, 2014, 2015; Vingron et al., 2021).

3 We also collected response time (RT) data, and conducted a second generalised mixed model with bilinguals that included Spanish fluency and O-span as fixed effects. Given the focus on the construction-based regulation hypothesis, we only report in the main text the model that includes English fluency. Results for the RT and Spanish fluency data can be found in the Appendix.

4 An alternate explanation is that the a-adjective constrain is driven by entrenchment effects (Regier & Gahl, 2004; Stefanowitsch, 2008; Theakston, 2004). Unlike the statistical preemption account, which posits that only frequency between functionally related forms (i.e. “the asleep dog” vs. “The dog that’s asleep”) is relevant, the entrenchment account predicts that the frequency of any non-attributive alternative (e.g. relative clauses such as “The dog that’s alone” and adverbial instances such as “He lives alone”) will block attributive use. While the present study was not designed to test both accounts, we conducted an exploratory analysis comparing real a-adjectives that can also be used as verbals (i.e. asleep and awake can be used as verbs, and alone is also an adverb) against the remaining real a-adjectives (alive, afraid, afloat, aware). An entrenchment account would predict a greater tendency to use verbals non-attributively because these items provide greater instances of non-attributive usage. However, an exploratory analysis yielded no differences in attributive rates between the two subsets of a-adjectives across conditions (all ps > .400). We also point to a third experiment in Boyd and Goldberg (2011), in which the exposure block included examples such as “the dog that’s ablim and proud of itself”. The entrenchment account predicts that these instances should also block attributive use with novel a-adjectives. But since complex adjectival phrases (i.e. proud of itself) never occur attributively, the preemption account predicts that these instances should not block attributive usage. Indeed, the authors found that in such pseudo-preemptive contexts, novel a-adjectives were overwhelmingly used attributively, suggesting that the constraint emerges only in contexts with functionally related alternatives.

5 A correlational analysis for monolinguals indicated a positive association between raw O-span recall scores and raw English fluency scores, r(54) = .29, p = .026, 95% CI [0.04, 0.52]. To diagnose potential multicollinearity, we calculated the Variance Inflation Factor (VIF) and a kappa value from the model estimates. All VIF values were below 1.30. Likewise, the kappa value was 1.99, indicating minimal-to-no collinearity between the two predictors and their interaction terms. To further examine this issue, we reran the model twice, each time excluding one of the two predictor variables. The pattern of results remained the same: when the model included O-span without fluency, O-span significantly interacted with adjective type, β = 0.39, SE = 0.14, z = 2.73, p = .006. Conversely, when the model included fluency without O-span, no significant effects or interactions with fluency emerged.

6 Unlike monolinguals, the correlational analysis in bilinguals indicated no significant pattern of association between O-span recall scores and fluency in English, r(53) = .25, p = .068, or Spanish, r(51) = .19, p = .180.

7 The interaction between O-span and English fluency in bilinguals may also account for exposure discrepancies between bilinguals and monolinguals (see Figure 3). Following the construction-based regulation hypothesis, bilinguals with low English fluency but high O-span may be more likely to extend forms regardless of their semantic associations. These individuals may also be more likely to disprefer attributive use without taking its competing alternative into account, which may explain why some L2 speakers show reduced sensitivity to novel sentences with competing alternatives (Robenalt & Goldberg, 2016; Tachihara & Goldberg, 2020). An interesting prediction is that these individuals would adopt an entrenchment (see Footnote 4), rather than preemption, strategy. Although we are unable to directly test this prediction here, our results suggest that taking into account speakers’ regulatory profile will be important to determine whether they are more likely to adopt specific usage strategies.

Additional information

Funding

This work was supported in part by the National Science Foundation [Award number DGE-1255832 and Grant number BCS-1946051 to Christian A. Navarro-Torres; Grant numbers BCS-1535124, OISE-1545900 to Judith F. Kroll] and the National Institute of Health [Grant number HD098783 to Christian A. Navarro-Torres; Grant number HD082796 to Judith F. Kroll; Grant number HD071758 to Paola E. Dussias].