How Accurately Do Infants Represent Lexical Stress Information in Recently Segmented Words?

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

Integrative Statement

Finding words in continuous speech is one of the many challenges young infants face when learning their native language. Infants appear to overcome this problem by relying on both language-general cues (e.g., transitional probabilities (TP) between syllables) as well as language-specific cues (e.g., stress pattern of infants' native language) to identify words boundaries. Findings from previous research demonstrate that by eight months of age monolingual English learning infants are able to pull out words with strong internal co-occurrence statistics from continuous speech in both artificial (e.g., Saffran, Aslin, & Newport,1996) and natural languages (Pelucchi, Hay, & Saffran, 2009). Although these findings have been replicated numerous times, we still know very little about how these newly extracted words are represented. For example, if infants use TP information to segment a word with a trochaic (strong/weak) stress pattern in speech, will they recognize the same word if it is presented with an iambic (weak/strong) stress pattern?

Here, we build on work by Pelucchi et al. (2009), but specifically test how well 8-month-old infants represent the stress pattern of newly segmented words. Infants were familiarized with Italian sentences that had two embedded high transitional probability (HTP; TP=1.0) trochaic target words (e.g., FUga & MElo) - their syllables never occurred anywhere else in the corpus (HTP; TP =1.0). Following familiarization, infants were tested using the head-turn preference procedure on their ability to discriminate HTP words from two novel words (e.g., PAne & TEma) that had never occurred in the corpus. In a counterbalanced language the HTP and novel words were switched. In Experiment 1 (n = 23), the trochaic stress pattern of the target words was consistent across familiarization and test, while in Experiment 2 (n = 22), the stress pattern of the words was changed between familiarization and test, such that if the HTP words in the corpus were trochaic (e.g., FUga and MElo), infants were tested on their ability to discriminate the iambic version of the target words (e.g. fuGA & meLO) from novel iambic words (e.g., paNE & teMA) (for acoustic characteristics of test tokens see Table 1).

In Experiment 1 we successfully replicated the familiarity preference for HTP words found by Pelucchi et al (2009) and Karaman & Hay (2017, Exp3), t(22) = 2.84, p = .009, d = .57 (see Figure 1). Interestingly, in Experiment 2 infants also showed a familiarity preference for the modified iambic HTP words, t(21) = 2.23, p = .036, d = .47. Thus, English learning 8-month-olds appear to recognize recently segmented words even when they are presented with a different stress pattern. These findings suggest that segmental information may override suprasegmental information at this age, and further that representations of suprasegmental information may be weak or underspecified early in development. We will discuss upcoming work on the robustness of segmental and suprasegmental representations of newly segmented words across development.

Authors

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

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