Are our preferences and evaluations conditioned by the language context?

ABSTRACT

Our preferences and evaluations are often affected by contextual factors. One unavoidable context is language. We used an evaluative conditioning (EC) paradigm (pairing neutral stimuli with emotional or neutral stimuli) to investigate whether our evaluations are equally conditioned in a first (L1) and a second language (L2). An EC effect was observed in both languages, however, if in L1 it occurred independently of recollection of the pairing of the stimuli, in second language memory seemed to play a larger role. These results were confirmed using a more implicit measure (memory confusion paradigm). Overall, the results suggest that conditioning occurs both in L1 and L2 but is weaker and more sensitive to memory of the emotional stimuli in L2. The study is the first demonstration that EC is modulated by language and converges with recent findings showing that linguistic context can modulate our behaviours.

KEYWORDS:

• Foreign language
• evaluative conditioning
• emotion

Acknowledgements

This study was funded by the Agencia Estatal de Investigación (AEI, National Research Agency) and Fondo Europeo de Desarrollo Regional (FEDER, European Regional Development Fund) under projects PSI2017-84539-P, the Catalan Government (2017 SGR 268), and was carried out within the framework of the research project IN.MIGRA3-CM (ref. H2019/HUM5772), funded by the Community of Madrid and co-funded by the European Social Fund. We are grateful to Dr. Olivier Corneille for the fruitful discussions. Albert Costa passed away in December 2018.

Disclosure statement

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

Notes

1 We performed a statistical power analysis to estimate the required sample size for Task 3, based on effect size on EC effects (ηp² = 0.11) reported by Pleyers and colleagues (Pleyers et al. 2009), and the correlation between repeated measures (r = 0.35) reported in a meta-analysis from Hoffmann and colleagues (Hofmann et al. 2010). With 3 groups (positive, negative and neutral) and 3 repetitions per group, using a two-tailed ANOVA test with an alpha = .05 and power = 0.90, the estimated sample size was approximately N = 24. However, we anticipated that a number of participants would have to be excluded based on their linguistic proficiency (evaluated on the day of the experiments with the language proficiency test and translation), which is why more participants than estimated by the power analysis were tested. Note, however, that all the participants were tested before data analysis. We also performed a sensitivity power analysis testing our main hypotheses for each task of the study. For Task 2, we performed an analysis on the comparison of the distributions of performance on the Old/New memory task (Experiments 1a and 1b). Between 2 groups, two-tailed comparisons between two paired groups with alpha = .05, power = .80 and sample size N = 67 (Wilcoxon signed-rank test), dz = 0.356. Between 3 groups, required effect size f (‘as in SPSS’) with alpha = .05, power = .80 and sample size N = 67, f(U) = 0.273. For Task 3 (Experiment 1a), the sensitivity power analysis for the effect of valence on ratings showed a required effect size f (‘as in SPSS’) with alpha = .05, power = .80 and sample size N = 26 f(U) = 0.452. Our results show significant effects of valence on the ratings in both language groups and effect sizes of 0.659 in L1 and 0.506 in L2 (Figure 4). For Tasks 4 and 5 (Experiments 1a and 1b), we performed an analysis for the comparisons of the distributions of proportions of errors of correct valence, two-tailed comparisons between two paired groups with alpha = .05, power = .80 and sample size N = 67, d = 0.347. Our results show that the proportions of errors of correct valence were significantly higher in L1 than in L2 with an effect size of d = .32 (Figure 8). In Experiment 1b, we performed an analysis for the comparisons of the distributions of proportions of errors of correct valence within each contingency awareness group, two-tailed comparisons between two paired groups with alpha = .05, power = .80 and sample size N = 41, d = 0.395. Our results show that the proportions of errors of correct valence within in the presence of contingency awareness is significantly higher in L1 than in L2 with an effect size of d = .55 (Figure 8).

2 Although we cannot guarantee that participants were not able to recall the US associated with the sign, putting participants under time-pressure would definitely impair their memory, as access to mnemonic information about source (the context in which an item was presented) is reduced under time-pressure (Benjamin and Craik 2001).

Additional information

Funding

This study was funded by the Agencia Estatal de Investigación (AEI, National Research Agency) and Fondo Europeo de Desarrollo Regional (FEDER, European Regional Development Fund) under projects [grant number PSI2017-84539-P], the Catalan Government [grant number 2017 SGR 268], and was carried out within the framework of the research project IN.MIGRA3-CM [ref. H2019/HUM5772], funded by the Community of Madrid and co-funded by the European Social Fund; Comunidad de Madrid.