Wait long and prosper! Delaying production alleviates its detrimental effect on word learning

ABSTRACT

Recent work by Baese-Berk and Samuel (Baese-Berk, M. M., & Samuel, A. G. (2022). Just give it time: Differential effects of disruption and delay on perceptual learning. Attention, Perception, & Psychophysics, 84(3), 960–980.) suggests that immediate – but not delayed – production has a detrimental effect on learning a non-native speech sound contrast. We tested whether this pattern is also found for word learning. Each participant learned 12 new words in one of four training conditions: Perception-Only, Immediate-Production, 2-seconds-Delayed-Production, and 4-seconds-Delayed-Production. At test, we assessed how well new words were embedded into the mental lexicon by measuring the degree to which they could drive phonemic recalibration (also called “perceptual learning”). Training and testing were repeated on the next day along with a word recognition task assessing lexical configuration. Replicating previous findings, Day 1 results showed that repeating a new word immediately after hearing it disrupted learning compared to just hearing it. Critically, in line with our prediction, this negative effect disappeared when a 4-second pause was inserted between hearing and producing each word.

KEYWORDS:

• Word learning
• production
• spoken word recognition
• mental lexicon

Acknowledgements

The authors thank Alexandra Kypta-Vivanco for help with data collection. Support for this project was provided by the Spanish Ministry of Science and Innovation through Grant # PSI2017-82563-P, awarded to AGS, and Grant # PID2020-113348GB-I00, awarded to AGS and ECK, and by the Spanish Ministry of Economy and Competitiveness through the Juan de la Cierva-Formación fellowship, # FJCI-2016- 28019, awarded to ECK.

Disclosure statement

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

Notes

1 Recent work has cast doubt on whether acoustic information can be maintained beyond two seconds (Caplan et al., 2021). However, this issue is still very much a matter of debate. For example, Sarrett et al. (2020) showed that listeners retain subphonemic information for approximately ∼900 ms even when they do not expect further disambiguating information. Regardless of how this debate is resolved, the delays that we used here produced interpretable and interesting results.

2 We added background noise in the stimuli used in this task in case the versions without noise produced ceiling effects.

3 Training and testing stimuli were spoken by two speakers of different sex. This was done because pilot data showed that when the same voice was used for training and testing, no phonemic recalibration was detected. This drop in the effect likely reflects the fact that exposure to unambiguous tokens of critical speech sounds can interfere with phonemic recalibration (Kraljic et al., 2008). In addition, using speakers of different genders in training and testing allowed us to assess the establishment of lexical representations independently of voice-specific information. We come back to these issues in the General Discussion.

4 Given the availability of different statistical approaches, as well as the rising popularity of mixed effects models, it is worth explaining the rationale of our analytical approach. We used ANOVAs instead of a mixed effects model (1) to make the results more directly comparable to previous work (e.g., Leach and Samuel) and (2) because in this case using mixed effects would not likely have an advantage over ANOVA. This is because (a) there was only one item (ufi/usi), so there was no item-driven variability to account for, and (b) any subject-driven effects would be confounded with condition-driven effects, given that our critical manipulation was between-subjects (that is, each subject was exposed to either ambiguous /s/s or ambiguous /f/s). Moreover, our goal was to test whether each of the three types of production (Immediate/2-secs-Delayed/4-secs-Delayed) led to different degrees of lexical integration when compared to the baseline (Perception-Only) condition. This is why we opted for an analytical approach that would test these specific questions (i.e., three ANOVAs, each testing the corresponding question).

Additional information

Funding

This work was supported by Basque Government: [Grant Number BERC 2018-2021,BERC 2022-2025]; H2020 Marie Sklodowska Curie Actions: [Grant Number 793919]; Spanish Ministry of Science and Innovation: [Grant Number PID2020-113348GB-I00,PSI2017-82563-P]; Spanish Ministry of Economy and Competitiveness: [Grant Number FJCI-2016- 28019]; Spanish State Research Agency: [Grant Number SEV-2015-0490,CEX2020-001010-S].