It seems to be assumed that phonological representation in the mental lexicon and lexical effects on phonological representations of speech input share the burden of overcoming variations in natural pronunciation, producing a percept of roughly equal quality whatever the input. This paper summarizes work which shows why that outcome would be counterproductive.
In fact, naturally occurring variations in word pronunciation are not noise, but information [2, 3]. They indicate, for example, whether a word stands alone or forms part of an utterance and whether a nominal refers to a Given or a New entity. Three studies of spontaneous production show that, claims to the contrary notwithstanding, natural variation in pronunciation is not adjusted to sources of knowledge which the listener might apply simply. First, the degree to which words display phonological reductions and consequent intelligibility loss when heard in isolation is not constrained by lexical competition. Second, the differences between introductory and second mentions are not simply a binary categorial contrast between accented New items and deaccented Given items. Third, changes in pronunciation are not attuned to the listener's current knowledge of the discourse and its domain, but to the speaker's [1]. In effect, the listener has a delicate perceptual task for which the key is the speaker's internal representation of the discourse. Since grasping this is the listener's real goal, good design would be served by inducing the listener to find the key information rather than by managing all of lexical access without it.
To serve the discourse understanding system, however, lexical access must take the hit: it must be affected but not blocked by degraded word tokens. A series of perception experiments show that variations in the clarity of spoken words both hamper lexical access and enhance access to a record of the discourse. First, tokens with naturally occurring phonological reductions and assimilations yield substantial but substantially less cross-modal identity priming than more canonical tokens. Second, when listeners perform a primed probe task after hearing a portion of a monologue containing a single token of the prime in the same sentence as the probe, both natural and artificial decrements in intelligibility enhance the priming effect. Because the effect is restricted to those monologues where the prime word will ultimately be mentioned again, perceptual difficulty appears to be inducing selective rapid access to important features of the discourse.
References
[1] Bard, E. G., Anderson, A. H., Sotillo, C., Aylett, M., Doherty-Sneddon, G., and Newlands, A. (2000). Controlling the intelligibility of referring expressions in dialogue. Journal of Memory and Language, 42, 1-22.
[2] Fowler, C., & Housum, J. (1987). Talkers' signalling of `new' and `old' words in speech and listeners' perception and use of the distinction. Journal of Memory and Language, 26, 489-504.
[3] Lindblom, B. (1990). Explaining variation: a sketch of the H and H theory. In W. Hardcastle & A. Marchal (Eds.), Speech production and speech modelling, (pp. 403-439). Dordrecht, Netherlands: Kluwer Academic Publishers.