How Verbs are Represented in Caregiver-infant Interactions

• In Event: Lexical Development: environment and mechanisms

Abstract

How do infants learn language through everyday interactions? Contexts matter: the linguistic contexts in which a given word tends to occur are far from random, and these non-random co-occurrences of words could help children induce word meaning and assign them to syntactic categories (e.g., Mintz, 2003). Moreover, research on nouns suggest that the embodied contexts of caregivers’ and infants’ attention or action when caregivers produce utterances are also important (Yu & Smith, 2012). To understand how infants might acquire statistical information about language through input, previous research has focused on words with concrete referents: primarily object-labeling nouns. However, little is known about distributional information in social interactions for words with more dynamic or less perceptible meanings, like verbs (Maguire, Hirsh-Pasek & Golinkoff, 2006). To investigate how verbs are represented in infant everyday interactions, we 1) annotated the linguistic and embodied contexts in caregiver-infant interactions, 2) calculated the co-occurrences of different verb usages and linguistic as well embodied contexts, and 3) applied hierarchical clustering to verbs based on the co-occurrences.

Forty-two caregivers and their 12-month-old infants were video-recorded for 15-min free play sessions. We included the 51 most frequent verbs, classified according to meanings and usage: movement, object-oriented action, perception/cognition, volition, transitive and auxiliary. For contexts, we included 17 frequent pronouns, infants’ looking and holding targets, infant locomotion, and caregivers’ holding target (Table 1). To calculate co-occurrence, we compare the frequency ratio of a subtype of verbs (SV) in all verbs (V): SV/V, and the frequency ratio of SV co-occurring with a certain contexts (C) compared to all V happening with C: (SV|C)/ (V|C). The closer (SV/V)/ [ (SV|C)/ (V|C)] is to 1, the more likely there is no significantly higher or lower co-occurrence than baseline.

We found that movement verbs were characterized by low co-occurrence with deictic pronouns and with infants' object handling, indicating that movement was negatively related to object play. Moreover, the play context also afforded separability of mental verbs: verbs of perception and cognition were more associated with first-person pronouns (primarily used with the caregiver as subject, or in the imperative) whereas verbs of volition or intention were much more likely to co-occur with second-person and deictic pronouns, which is consistent with Laakso and Smith’s (2006) findings. Further analyses will incorporate utterance length and word frequency in additional significance tests. We applied hierarchical clustering to verbs based on their co-occurrences with pronouns, gaze and hands data (Figure 1). Qualitatively the clusters are consistent with verb usages.

Our data show that verbs are predicted by statistical regularities in caregiver speech input as well as caregiver and infant embodied contexts. Without providing any systematic linguistic knowledge, we found clusters similar to English verb categories based on co-occurrences. This suggests that although verbs are more abstract and less frequent than nouns, their meanings and usages may still be learned through linguistic and embodied contexts. Our data contribute to the statistical learning literature, and reveal new statistical regularities of under-studied English verb categories in caregiver-infant interactions.

Authors

• Hanwen Vivian Vivian Zhang, Cornell University
  Presenting Author

• Lucas Chang, University of California, San Diego
  Non-Presenting Author

• Gedeon O Deak, University of California - San Diego
  Non-Presenting Author