Infants' On-line Segmentation of Dynamic Human Action

Abstract

Previous research has clarified that infants from 10–11 months segment dynamic human action into units coinciding with actor's goals and intentions (Baldwin, Baird, Saylor, & Clark, 2001). In this study, we explored the scope and robustness of early action segmentation skills by exposing infants to a variety of relatively novel events in the absence of an extensive familiarization period. Infants from 9–11 months viewed two simultaneous displays of live action, each involving a single actor manipulating a different set of objects. While the actors were in continuous motion, tones were presented that corresponded with the completion points of one of the actor's intentions. Infants revealed segmentation skills by looking longer at the action sequence matching the tone presentation than at the other action sequence. Neither repeated presentation nor the salience of the matching actions accounted for the finding. The segmentation skills revealed in this study may lay the foundation for infants’ eventual ability to make sense of others’ actions and intentions.

ACKNOWLEDGMENTS

Jennifer LaBounty is now at the Department of Psychology at the University of Michigan. Jodie A. Baird is now at the Department of Psychology at Villanova University.

Portions of this project served as Jennifer LaBounty's Honors thesis. This study was supported by the National Science Foundation under Grant No. 0214484 and by John Merck Scholars Fund (both awards to Baldwin).

We would like to thank the parents and children who participated. We also thank Lou Moses for assistance with statistical analyses.

Notes

To further explore the possibility that actor-introduced bias drove infants’ looking, a coder judged which action display was the matching action with the sound off (i.e., no tones). Bias on the part of the actors should have lead to correct identification of the matching action. However, the coder showed at chance (55%) choice of the match action (binomial test, ns).

Proportions were used as the primary dependent measure, instead of raw looking times, because trial durations varied slightly as a result of live actors producing the actions. Analyses were conducted after the proportions were submitted to the arcsine transformation to normalize the distributions (X’ = 2∗ arcsine (), with values close to zero or with unity corrected as X’ = 2∗ arcsine () (e.g., Winer, 1977). The same pattern of findings emerged as with the untransformed proportions. Reported means and standard deviations are based on the untransformed proportions.