Statistical Learning in Languages with Different Rhythm and Prosody

• In Event: Poster Session 05
  In Poster Session: PS 05 Topic - Language, Communication

Abstract

By their first birthday, infants demonstrate sensitivity to language-general (e.g., transitional probability (TP) between syllables) and language-specific cues (e.g., stress pattern of their native language) during word segmentation. As infants gain more experience with their native language, they come to rely on language-specific cues to identify word boundaries (e.g., Thiessen & Saffran, 2003). However, in natural languages, language-general and language-specific cues often provide redundant information regarding word boundaries. For example, in English, disyllabic words tend to have both strong internal co-occurrence statistics and a trochaic (strong/weak) stress pattern. Infants may benefit from these redundant cues during speech segmentation. Interestingly, much of the recent work on tracking TP information in natural language has been conducted using Italian (e.g., Pelucchi, Hay, & Saffran, 2009a), which like English, puts stress on the initial syllable of disyllabic words. While the two languages have similar lexical stress patterns, they have different rhythmic patterns (English is stress-timed and Italian is syllable-timed). Although previous work has demonstrated that 8-month-old English-learning infants readily use TP information to extract trochaic words with high transitional probability (HTP) from a fluent Italian corpus and seem to be able to segment a language with a different rhythm, we do not know whether infants are able to track TP in natural language input if redundant language-specific cues to word boundaries are not present.

To address this gap in knowledge, we tested 8-month-old English-learning infants’ ability to statistically segment either a natural Italian (Experiment 1) or a natural Farsi (Experiment 2) familiarization corpus. Like Italian, Farsi is a syllable-timed language, but unlike Italian (and English) Farsi disyllabic words are predominantly iambic (i.e., weak/strong). In both experiments infants were familiarized with a naturally produced corpus (either Italian or Farsi) that contained two embedded target words. The words were high TP because their syllables never occurred anywhere else in the corpus (HTP; TP =1.0). Importantly, the target words were trochaic in the Italian corpus and iambic in the Farsi corpus. Following familiarization, infants were tested using the head-turn preference procedure on their ability to discriminate HTP words from Novel words that had never occurred in the corpus. Infants showed a familiarity preference for HTP words vs Novel words, t(23)= 3.192, p= 0.004, d=0.648, after familiarization with the Italian corpus (Experiment 1, n=24), but they failed to extract iambs from the Farsi natural speech stream, t(21)=.634, p=.53, d =.14, (Experiment 2, n = 22).

The results show that while 8-month-old monolingual English-learning infants can statistically segment Italian trochees, replicating previous findings, they failed to segment Farsi iambs. These findings suggest that while infants readily track statistical regularities in a rhythmically unfamiliar language, non-native stress patterns appear to impair statistical learning in English-learning infants. Thus, infants appear to only benefit from redundant cues to word boundaries if those cues are consistent with the properties of their native language.

Authors

• Sara Parvanezadeh Esfahani, University of Tennessee, Knoxville
  Presenting Author

• Jessica Sari Fleming Hay, University of Tennessee - Knoxville
  Non-Presenting Author