Poster #121 - What gets Predicted? Evaluating the Specificity of Infants' Verbal Predictions

• In Event: 2-110 - Poster Session 07
  In Poster Session: PS 07 Section - Language, Communication

Integrative Statement

Prediction is a potential language learning mechanism (Dell & Chang, 2014). Prior studies have demonstrated prediction among infants (Reuter et al., 2018), but prediction is measured indirectly. For example, if an infant looks towards a cookie while hearing “eat” these looks are operationalized as a pre-activation of the representation “cookie.” However, for prediction to be a viable learning mechanism, predictions must be specific – including semantic, syntactic, and phonological detail (Rabagliati, Gambi & Pickering, 2016). We hypothesized that infants’ predictions do include representational detail. If so, then an accurate prediction of a word should be behaviorally identical to the perception of that word.
To evaluate this hypothesis, we investigated verbal prediction among infants 14-19 months old (n=16). In an eye-tracking task, infants viewed pairs of images (e.g., a cookie and a shoe) and heard three types of auditory stimuli: Prediction trials included semantically-related words that infants could use to predict the upcoming noun (e.g., Let’s go eat! Oh yum yum! Open your mouth!). Repetition trials included repetitions of the upcoming noun (e.g., Look at the cookie! There’s the cookie! Do you see the cookie?). Neutral trials did not include any information about the upcoming noun (e.g., Look at that! There it is! Do you see it?). All trials ended with the same phrases (e.g., Where’s the cookie? Find the cookie?).
We analyzed infants’ proportion of looks to the target (e.g., cookie) during two a priori defined time windows: the prediction time window (i.e., looks before target noun onset) and the comprehension time window (i.e., looks after target noun onset). If infants comprehend target nouns, looks should be above chance during the comprehension time window. If infants predict target nouns, looks should be greater for prediction trials than for neutral trials during the prediction time window. Critically, if infants’ predictions are specific, then looks should be equal for prediction trials and repetition trials during the prediction time window (Figure 1).
Preliminary findings revealed robust comprehension and prediction of target nouns: Target looks were above chance during the comprehension time window for all trial types (ps<0.001), and target looks were greater for prediction trials than for neutral trials during the prediction time window (t(15)=3.02, p=0.004). Critically, findings revealed no significant difference among prediction trials and repetition trials during the prediction time window (t(15)=-0.91, p=0.38). We also analyzed looks within 100-ms time-bins during the prediction time window, using paired sample t-tests, and using FDR to correct all p-values for multiple comparisons (Benjamini & Hochberg, 1995). We again found no significant differences among prediction trials and repetition trials (ps>0.65).
Thus, findings indicate that infants robustly comprehend target nouns and can use semantically-related words to predict the upcoming noun. Importantly, infants’ looking behavior indicates that their predictions are specific: We observed no differences between prediction trials and repetition trials during the prediction time window. This pattern of results suggests that infants’ predictions are representationally detailed (Rabagliati, Gambi & Pickering, 2016) and supports the claim that prediction is a language learning mechanism (Dell & Chang, 2014).

Authors

• Tracy Reuter, Princeton University
  Presenting Author

• Lauren Emberson, Princeton University
  Non-Presenting Author