Extraction of Syllable Order During Word Segmentation

• In Event: 2-041 - Specificity of Infant Statistical Learning

Integrative Statement

We investigate the specificity of infant's lexical representations after word segmentation. In particular, we assess whether infants encode and retain the correct ordering of syllables when they segment a string of speech into single words. In addressing this question we focus on two accounts of statistical learning: models that propose that infants track transitional probabilities (TPs) between syllables and those proposing that infants chunk syllables according to co-occurrence frequency. Although Aslin et al. (1998) and studies since contain controls to rule out learning based on co-occurrence frequency, a consequence of those controls is that some syllables are more frequent in the corpus than others. In the present study, we are able to control the frequency of individual syllables and instead focus on determining whether infants retain sequential order information. These two accounts (TPs vs. co-occurrence frequency) differ on their predictions regarding the encoding of syllable order during word segmentation. By definition, TPs reflect the order of syllables within larger units. For example, for syllable pair AB, the probability that A will follow B in a speech corpus does not equal the probability that B will follow A in the same corpus. Therefore, if infants segment and encode words based on TPs, they should retain the order of syllables within words. Alternatively, infants could disregard TPs and form candidate words by encoding high-frequency strings of syllables, regardless of the TPs between syllables. If infants only encode co-occurrence frequency, then they should not track syllable order. In the present study, we investigate the specificity of infants' encoding during word segmentation to ask whether infants retain order of syllables within word candidates or chunk bigrams without regard to syllable order. We familiarized 23 13-month-old infants to a language containing four disyllabic words. As in other word segmentation studies, the words occurred within a fluent string of speech containing no pauses, stress patterns, or intonation contours. Therefore, there were no acoustic cues to indicate word boundaries. After the 13-minute familiarization, we tested infants for their retention of order using the Head Turn Preference Procedure. During "forwards-order" trials infants heard legal words from the language, with their syllables in the correct order (e.g. tani and budo). On "backwards-order trials," infants instead heard words with the syllable order reversed (e.g. if rodi was a word in the language, diro would be its "backwards-order" counterpart). A paired-samples, two-tailed t-test revealed that overall, infants looked significantly longer on the forwards-order trials (M = 8.66 sec) in comparison to the backwards-order trials (M = 7.31 sec), t = 2.45, p = .023). This pattern of results supports the notion that infants are monitoring TPs within syllables, and retain order-related information after the familiarization phase.

Authors

• Lucia Sweeney, University of Arizona
  Presenting Author

• Rebecca Gómez, University of Arizona
  Non-Presenting Author