Poster #55 - Dynamics of infant vocal trajectories in acoustic space

• In Event: Poster Session 06

Abstract

Previous studies have used a foraging theory framework to study non-spatial cognitive behaviors, establishing similarities with how animals forage in physical space for food, mates, etc. (e.g., Hills et al., 2015; Rhodes & Turvey, 2007). Following this approach, it has been proposed that infant vocal behaviors can be characterized as foraging in acoustic feature space (Fig. 1a), motivated by both social and intrinsic rewards. Past work focused on infant vocalization-to-vocalization pitch and amplitude differences and inter-vocalisation intervals (Fig. 1a). A limitation has been reliance on the LENA system’s automated labels, and limited exploration of different ways of defining the acoustic space.

Here, we expand on that prior work in two ways. First, we consider utterance duration as an additional acoustic feature (Fig. 1b). Second, we use LENA-annotated and human-listener annotated 5-minute intervals to assess whether the patterns are robust to data quantity and annotation method.

We analyzed 204 daylong child-centered audio recordings from 54 infants aged 2 to 18 months. We used Praat to estimate the mean pitch and intensity of infant vocalizations automatically identified by LENA’s software. In addition, trained human listeners annotated up to three five-minute sections from a subset of the recordings, corresponding to high infant volubility times of the day according to LENA. These five-minute sections were then labeled by listeners in ELAN using a protocol that included labeling the onsets and offsets of all target infant non-vegetative vocalizations and categorizing each as speech/speech-related or cry/laugh. This provides the opportunity to compare foraging patterns between the day-long and 5-minute timescales, and between LENA labeling and human listener annotation.

A two-dimensional (2D) acoustic space positioned each infant utterance based on its mean pitch and amplitude. Expanding on this, a three-dimensional (3D) acoustic space added utterance duration as a feature (Fig 1). We then tracked utterance-to-utterance acoustic distances within each space, as well as corresponding inter-vocalization intervals.

Linear mixed effects regression (Table 1) indicated that the step size between successive infant speech/speech-related vocalizations in the 2- and 3-D acoustic spaces was weakly, but significantly, positively correlated with the inter-vocalization interval. This association was found across all acoustic space and annotation method combinations, indicating that a previously reported positive association between acoustic difference and inter-vocalization interval generalizes across annotation methods. It also generalizes to a somewhat more comprehensive acoustic space that incorporates utterance duration (which may relate to utterance complexity). We also observed a small but consistent trend for step sizes in acoustic space to decrease with infant age, a developmental trend we plan to explore in more detail as part of the broader project.

Overall, these results support a parallel between infant vocalizations in natural settings and foraging phenomena in other domains, where the physical or psychological distance traversed is positively correlated with the time taken to cover the distance.

Authors

• Ritwika VPS, University of California - Los Angeles
  Presenting Author

• Sara Schneider, University of California

• Lukas D Lopez, University of Utah

• Jeffrey Mai, UCLA

• Christopher T Kello, University of California

• Anne Warlaumont, University of California