Faithfulness and Perception in Loanword Adaptation: A Case Study from Hawaiian Numerous cross-linguistic studies

Numerous cross-linguistic studies suggest that the adaptation of loanwords proceeds in such a way as to maximize perceptual similarity between input and output (Fleischacker 2000, Kang 2002, Kenstowicz 2003). Other authors have characterized the relationship between input and output sounds in borrowings as being based on phonological closeness rather than perceptual similarity (Paradis' TCRS loanword model: 1996; with LaCharité 1997). Within the speech perception literature, there are also models of nonnative speech perception that is grounded in articulatory similarity (e.g., Best's Perceptual Assimilation Model: 1994, 1995, et seq.). The Hawaiian language, which has a small, uniquely structured phoneme inventory and strict phonotactic constraints, can serve as a useful testing ground for these approaches. Among other restrictions, the language has no oral coronal stops and no oral fricatives, and it disallows all codas and consonant clusters. When borrowing words from English, Hawaiians are faced with a number of sounds and sound combinations which are not present in their language. Speakers must make the choice to either delete or adapt illicit consonants, and either insert vowels or delete consonants to repair phonotactic violations. I present a study of novel (non-lexicalized) adaptations of English words into the Hawaiian language, based on data from two Hawaiian-English bilingual speakers. We find that, of all the logically possible feature changes which would change an impossible segment into a native Hawaiian phoneme, only a subset are actually used (see chart). In contrast to previous findings in the perceptual literature, it seems that high paradigmatic (segmental or 'inherent') salience does not necessarily lead to preservation of segments in less salient positions. I will show that the variation observed in adaptation and deletion is often expected under Best's Perceptual Assimilation Model (PAM). PAM provides a framework for categorizing nonnative phones into classes according to the ease with which they can be assimilated into a listener's existing phoneme inventory. Both the PAM and TCRS approaches to nonnative sound adaptation suggest that segmental mappings will be based on a judgment of closeness or similarity of some sort. In Best's model, the relevant factor is similarity between articulatory gestures, assuming the articulatory phonology of Browman and Goldstein (1986 et seq.). This accords with the data I will discuss, but the model as it stands is not able to account for all of the observed mappings. One difficulty encountered is the lack of an explicit formulation of how to determine which repair among …