Prosodic Variation in Infant-Directed Speech

• In Event: 3-193 - The Moment-to-Moment Dynamics of Infant-Directed Speech: What Infants Hear and How They Learn

Integrative Statement

Infant-directed speech (IDS) is a dynamic signal. It is characterized by repetition and exaggerated prosody, suggesting that the same word may occur with different pitch patterns across tokens. This variation may help infants solve a key learning problem in acquisition—determining which acoustic variants in speech are meaningfully relevant for learning and recognizing words. Hearing rich prosodic variation may help infants learning non-tonal languages recognize that a word has the same referent regardless of pitch contour. The present research takes two approaches to understanding the role of prosodic variation in learning. First, we collected samples of IDS and adult-directed speech (ADS) to examine variation in pitch contours. Second, we tested infants’ ability to exploit experience with variation to shape their interpretation of prosody.

We collected IDS and ADS from 10 parent-infant dyads (infant ages 10 to 18 months). Each parent performed tasks with their infant (object labeling, object sorting, wordless picture book description) for approximately 25 minutes. Parents also performed parallel tasks with an experimenter (object labeling, object locating game, book description). All tasks shared a set of 12 target words (e.g., baby, dog, car, carrot). We adopted Quam’s (2008) procedure for identifying pitch contours based on areas of local fundamental frequency maxima and minima. Each target token fit one of 6 contour types: rise, fall, rise-fall, fall-rise, flat, and complex. We predicted that IDS would have a more even distribution of contours than ADS, indicating greater variation in the contours used. As shown in Figure 1, our prediction was supported. We calculated skewness, a measure of the evenness of distributions, over the proportion of the 6 contours produced in each register. A linear mixed-effects regression model fitted to these values, with a main predictor of register and by-speaker random intercepts, indicated that IDS (.31) had a more even distribution of pitch contours than ADS (.96) (p=.01), which had a preponderance of tokens with flat contours.

In our experimental test of the effects of variation, we built on evidence that 14-month-olds learning a non-tonal language misinterpret pitch in new words. Hay et al. (2015) found that they treat pitch as a contrastive dimension and accept object labels that differ only in pitch contour (i.e., the syllable ku with a rising versus falling pitch). We tested whether infants’ learning of pitch contour contrasts could be affected by concentrated prior experience with pitch variation. Before seeing the novel objects, the infants (n=24) heard a set of familiar objects labeled with variable pitch contours (rise-fall, fall-rise, flat). They then heard labels for novel objects that different only in pitch contour (ku rising vs. falling). Exposure to pitch variation effectively inhibited infants’ learning of labels that differed only in contour, allowing them to show more sophisticated prosodic interpretation. We propose that this experiment demonstrates a concentrated version of a naturally-occurring mechanism whereby experience hearing variation helps infants learn how to interpret it. Our IDS analyses and experimental results provide evidence that variation in prosody offers crucial information infants need to interpret native language prosody.

Authors

• Katharine Graf Estes, University of California, Davis
  Presenting Author

• Georgia Zellou, University of California, Davis
  Non-Presenting Author