Function Words Constrain On-Line Recognition of Verbs and Nouns in French 18-Month-Olds

Abstract

In this experiment using the conditioned head-turn procedure, 18-month-old French-learning toddlers were trained to respond to either a target noun (“la balle”/the ball) or a target verb (“je mange”/I eat). They were then tested on target word recognition in two syntactic contexts: the target word was preceded either by a correct function word (“une balle”/a ball or “on mange”/they eat), or by an incorrect function word, signaling a word from the other category (*“on balle”/they ball or *“une mange”/a eat). We showed that 18-month-olds exploit the syntactic context on-line to recognize the target word: verbs were recognized when preceded by a personal pronoun but not when preceded by a determiner and vice-versa for nouns. These results suggest that 18-month-olds already know noun and verb contexts. As a result, they might be able to exploit them to categorize unknown words and constrain their possible meaning (nouns typically refer to objects whereas verbs typically refer to actions).

Acknowledgments

The work reported in this paper was supported by a doctoral fellowship from Ecole Doctorale “Cerveau Comportement Cognition” from Université Pierre et Marie Curie Paris to Elodie Cauvet, as well as by a grant from the French Agence Nationale pour la Recherche (Bootstrapping language acquisition: an integrative approach) to Anne Christophe and by a Marie Curie Intra-European Fellowship within the 6th European Community Framework Programme (No. 024843) to Séverine Millotte.

Notes

¹Because some sentences were dropped from the analysis, not all children had 12 sentences per condition: the number of sentences per condition varied between 12 and 9 (for the more restless kids). Since 6 head-turns out of 12 possible responses is a smaller response rate than 6 out of 9 possible sentences, we present percentages instead of raw numbers.

³We thank an anonymous reviewer for suggesting this additional analysis.

⁴This window was computed for each item separately, because the target words varied in duration. Since mean target duration was 430.5ms, the window for fast responses lasted 1030.5ms on average.

⁵To test for this learning effect, we analyzed separately the first four trials and the last four trials of the experiment for each condition. We conducted an ANOVA with two within-subject factors: Condition (with three levels: Correct, Incorrect, and Distractor) and First-Last (with two levels: First four trials vs. Last four trials). The ANOVA revealed a significant effect of Condition (F₁(2,70) = 20; p < 10⁻⁷), reflecting the fact that toddlers responded more often for Correct sentences than for either Incorrect or Distractor sentences, as well as a significant First-Last effect (F₁(1,35) = 11; p < 0,002), that was due to the fact that toddlers responded less often overall towards the end of the experiment (probably because they got tired). Crucially, there was no interaction between these 2 factors, indicating that the effect of Condition was not significantly different in the First vs. Last trials (F₁(2,70) = 2.1, p = 0.13). If anything, the difference between Correct and Incorrect sentences was larger at the beginning of the experiment (First four trials: Correct 63.9%, Incorrect 42.4%, Distractor 41.7%) that at the end (Last four trials: Correct 46.5%, Incorrect 37.5%, Distractor 25.7%), probably because toddlers got more tired and distracted towards the end of the experiment. We thank an anonymous reviewer for this suggestion.

⁶This corpus is a transcription of mother-infant interactions from two families. Both children (Timothée and Marie) were recorded between the age of 1.5 and 2.5 years. The corpus contains around 200 000 tokens for around 26,500 utterances.