2024/12/17 by Rikke Louise Bundgaard-Nielsen, Robert Mailhammer, Brett Baker +3 · 1 voice
Computer Science · #Library Science and Information Systems
paper · doi:10.16995/labphon.15347
openalex publication_date 2024/12/17 · openalex created_date 2024/12/18 · openalex updated_date 2026/06/11
Many models of word recognition assume that spoken words are faithful to their phonological shape in the lexicon and that word recognition begins with the first incoming segment and proceeds linearly. Some languages, however, including Mawng and Iwaidja (Australia), exhibit alternations in word-initial segments, rendering these segments potentially unreliable. We tested the effect of word-initial segmental variability in Mawng and Iwaidja in a Two-Alternate Forced Choice experiment which paired canonical productions of nouns with forms beginning with both attested and non-attested variant onsets. All participants preferred canonical forms, but Mawng speakers were tolerant of /ɡ/-initial lenition. Results demonstrate that speakers prefer input consistent with the lexical specifications. Variance is only tolerated when phonetic/phonological deviance does not compromise native phonological and phonetic boundaries. The results highlight the importance of language-specific lexicon-phonology-phonetics interfaces in word-recognition and may guide developments in models of continuous parsing regarding the question of the nature of the input.