There is ample evidence in the experimental as well as theoretical literature that phonetic variation, measurable in the speech signal, is not entirely random. Articulatory gestures can be controlled to express specific phonetic distinctions. On the other hand, these distinctions do not usually pose considerable problems for speech perception. Listeners successfully map the highly variably speech signal onto discrete lexical representations. How can this efficient and adaptive interface between continuous and discrete speech characteristics be best described?

The present approach espouses a model which resolves phonetic variation directly in the mental lexicon via underspecification of phonological representations. A very detailed phonetic input is assumed to be mapped onto abstract lexical forms, of which the underspecification of particular phonological features interacts with principles of language change and morphological categorization. For both points, evidence is drawn from experimental findings.

In relation to language change, a dialectal "variation" between American and New Zealand English exemplifies the advantage of underspecified lexical segments in order to capture perceptual asymmetries based on vocalic tongue height features.

(1) [bɛt]
American English: bet
New Zealand English: bat

The experimental results suggest that phonetically identical vowels are interpreted differentially by American and New Zealand English speakers.

As for morphological categorization, data supporting asymmetric perception and underspecified recognition are drawn from German umlaut where two distinct phonetic segments refer to one underlying representation.

(2) [ʃtœkə]~[ʃtɔk] 'stick' plural~singular

A formerly purely phonetic process acquired morphological significance by grammaticalization. The phenomenon of German umlaut shows that phonetic differences not only have phonological but also morphological and syntactic repercussions. The strongest evidence for this claim stems from a neurolinguistic pilot study, again highlighting asymmetries in the mapping of phonetic to phonological information.