UCLA

In recent decades, linguists have experimentally demonstrated that the phonetic realization of lexical items and of specific speech sounds within them can be influenced by purely lexical properties such as word frequency (Balota et al 1989; Bybee 1994), contextual predictability (Hawkins and Warren 1994; Lieberman 1963), and, most interestingly, the existence of many phonologically similar words in the lexicon, i.e. lexical neighbors (Wright 1998, 2004; Brown 2001; Scarborough 2004; Baese-Berk and Goldrick 2009).

Several specific phonetic correlates of these lexical factors have been established: the vowel space as a whole is expanded (Wright 1998, 2004); voiceless stops and voiced stops have a larger VOT difference (Goldinger & Summers 1989; Baese-Berk and Goldrick 2009), and the overall amount of coarticulation between local segments is increased (Brown 2001; Scarborough 2004).

The general mechanism that underlies these various effects, however, is not well understood. While they each have the end effect of aiding listener comprehension, and occur under almost precisely the same conditions where word recognition is expected to be more difficult (Luce 1986), there are at least two types of mechanism consistent with this result. The first--the "hyperarticulation" hypothesis--is that speakers diminish processes of reduction, producing realizations of speech sounds which are highly faithful; since the phoneme inventory is generally dispersed, this indirectly facilitates word-recognition. The second--the "dissimilation" hypothesis--is that speakers directly facilitate word recognition by maximizing the perceptual distance between the target word and its lexical competitors, producing realizations of speech sounds which are phonetically distant from competing sounds.

An experiment was devised to distinguish between these two possibilities by using a phonetically medial sound: English /E/ (epsilon), which has the potential for competition from, among other vowels, /æ/ and /I/ (small caps 'i'), which are phonetically similar to /E/ and geometrically opposed in formant space. If the realization of words containing /E/ is influenced not only by the existence of minimal pair neighbors, but by the location in phonetic space of the vowels in such neighbors, the second hypothesis will be strongly supported. The results of the experiment were inconclusive; while some data trended in a direction consistent with the dissimilation hypothesis, no lexical neighborhood effects of any kind reached significance, despite a relatively large sample. This fact weakly supports the hyperarticulation hypothesis, at least with respect to vowels. However, the null result is potentially attributable to a number of factors.