Variation in English infant-directed speech

In adult-directed speech (ADS), words are rarely produced canonically. Infant-directed speech (IDS) has been argued to contain more canonical productions. However, recent analyses show that IDS is as variable as ADS. Then, how could infants learn to privilege canonical forms, as has been shown for adult listeners? Previous research on variation in IDS has focused on word-final productions. In this study, we investigate whether the extent of variation in IDS differs by the position of a segment in a word. We sampled IDS to 6 infants between 16 and 24-mo-old from the Providence corpus. Utterances with /t/, /d/, /n/, /s/ and /z/ were identified orthographically, forced-aligned, corrected, and transcribed by 3 phonetically-trained native speakers of English. This yielded 28,775 segment tokens in word initial, medial and final position. Results confirmed that IDS is at least as variable as ADS (canonical pronunciations < 50%). However, variation was limited to coda positions; on average, over 90% of onsets were produced canonically compared to just 60% of codas. This positional difference could benefit category learning. Word-initial segments would bolster acquisition of canonical forms, and possibly support word segmentation, while word-final variation would encourage learning of phonetic variants resulting from processes in connected speech.