Metrical segmentation across dialects

Abstract

Norris, McQueen & Cutler (1995) tested the Metrical Segmentation Strategy (MSS; Cutler & Norris, 1988) as part of the spoken-word recognition model Shortlist, using British English stimuli and listeners. We replicate their study using American English listeners, who we exposed to one of two sets of stimuli. One group heard a new set of stimuli recorded in American English, while the other was exposed to the original British English recordings. Norris et al. used a word-spotting task: listeners had to spot words within speech (e.g. “stamp” in [stæmpɪdʒ]). Target words were CVCC (like “champ") or CVC (like “done"), and were followed by a full vowel (e.g. /tʃæmpoʊʃ/) or a reduced vowel (e.g. /tʃæmpəʃ/). The original study found different behavior for CVCC versus CVC targets, with the results suggesting that listeners hypothesize a word onset at the start of a full-vowel strong syllable (the MSS). The results for the current study with American stimuli partially replicate the original findings, showing even more consistent support for the MSS. The results with British stimuli also support the MSS, but with higher error rates. The results indicate that the MSS has a very strong effect even in the difficult setting of cross-dialectal perception.