Specificity of Infant Statistical Word Segmentation

By their first birthday, infants develop sensitivity to language-general (e.g., transitional probability (TP) between syllables) and language-specific (e.g., lexical stress pattern) cues to identify word boundaries. Across 5 experiments, I investigated two questions: 1) how do English learning 8-month-old infants segment high TP (TP=1; HTP) words in non-native languages that have a different rhythm and prosody? 2) how do English learning 8-month-old infants represent nascent word representations across indexical (i.e., talker’s voice), segmental (i.e., onset consonant) and suprasegmental (i.e., stress pattern) information? To that end, in each experiment, using HTPP, I familiarized infants with a familiarization corpus with two embedded HTP words and at test, infants were evaluated on their ability to discriminate between the target words versus foils (either low TP or Novel words). I found that while the redundant language-specific cues can facilitate infants’ ability to identify word boundaries, a different language-specific cue can create a competition in infants’ attention to statistically segmented words versus words with familiar language-specific cues. For the second question, I found that English learning infants can generalize nascent word representation across indexical and suprasegmental information but not across segmental information that is essential to word meaning. Together, the findings suggest that while English-learning infants may need redundant information from stress cues to segment nonnative speech, they do not necessarily encode stress information in nascent word representations as they tend to only store information that is more readily relevant to word meaning (i.e., segmental information).