Perceptual similarity co-existing with lexical dissimilarity

The extreme case of perceptual similarity is indiscriminability, as when two second-language phonemes map to a single native category. An example is the English had-head vowel contrast for Dutch listeners; Dutch has just one such central vowel, transcribed [E]. We examine whether the failure to discriminate in phonetic categorization implies indiscriminability in otherââ"šÂ¬Ã¢â‚¬Âe.g., lexicalââ"šÂ¬Ã¢â‚¬Âprocessing. Eyetracking experiments show that Dutch-native listeners instructed in English to ``click on the panda'' look (significantly more than native listeners) at a pictured pencil, suggesting that pan- activates their lexical representation of pencil. The reverse, however, is not the case: ``click on the pencil'' does not induce looks to a panda, suggesting that pen- does not activate panda in the lexicon. Thus prelexically undiscriminated second-language distinctions can nevertheless be maintained in stored lexical representations. The problem of mapping a resulting unitary input to two distinct categories in lexical representations is solved by allowing input to activate only one second-language category. For Dutch listeners to English, this is English [E], as a result of which no vowels in the signal ever map to words containing [ae]. We suggest that the choice of category is here motivated by a more abstract, phonemic, metric of similarity.