UCLA

Phonetic variation poses a challenge for language learners tasked with identifying the abstract sound categories (phonemes) and positional allophony of their target language(s). Yet we know relatively little about the actual degree of phonetic variability in IDS and how this variation is structured. In this study, we set out to provide a more holistic understanding of what infants hear by quantifying the extent of variability in the pronunciation of some of the most frequent sound categories of English: coronals (/t/, /d/, /s/, /z/, and /n/). We further examine the degree to which this variation is expected based on English phonotactics. We sampled IDS from the longitudinal Providence Corpus (Demuth et al., 2006) which contains recordings of 5 typically-developing, monolingual, English-speaking 1- to 3-year-olds interacting with their caregivers at home during everyday activities. These utterances were force-aligned (Rosenfelder et al., 2014) according to orthographic transcripts to generate segmental boundaries automatically. We then checked and phonetically annotated 7,000 utterances containing 31,245 coronal segments. We found that overall, canonical variants of /t/ are in the minority (39%) whereas /s/ is overwhelmingly canonical (98%); further, almost every segment had more canonical instances in word-initial compared to word-final position. We also examined the distribution of expected variants based on English phonotactics against the observed variants for /t/ and /d/, two segments that are the most variable. While most variants had high counts of matching observed and expected variants, we also find that unexpected variants are common. The results of the current study help provide an understanding of the full extent of variation in naturalistic IDS. We discuss the implications of these results for theoretical and computational models of morphological and phonological acquisition.