This research project investigates lexical representations in long-term memory and the processes that access those representations. Most theories assume that the speech wave form is converted to a string of phonemes which is then compared to the mental lexicon. When comparison yields a match, the word is """"""""recognized"""""""". This typical view emphasizes the abstract properties of words. A speech signal containing idiosyncratic information (speaker's voice, ambient noise, etc.) is converted to a sequence of ideal, abstract phonemes, which is used to access ideal, abstract words. Few theories consider idiosyncracies of speech as more than """"""""noise"""""""" in the signal. However, multiple-trace memory models posit that details of perceptual experience are not forgotten; instead, they are integral to later perception. The proposed research will test a multiple-trace model of the mental lexicon, using speech perception and memory data (Project 1), and speech production data from a novel shadowing paradigm (Project 2). Preliminary research and pilot simulations of Hintzman's (1986) MINERVA 2 model have already produced encouraging results. The investigators plan to test the predictive limits of the extant model, and to test embellished versions as necessary. In Project 1, perception and memory experiments will test the level of detail and the durability of episodic memory traces produced in spoken word recognition, and will carefully assess the relationship of memory and perception. In Project 2, the model is tested against speech production data from a single-word shadowing task. The typical dependent measure in shadowing is response time; an alternative, rarely- used measure is the acoustic-phonetic content of the Ss' speech responses. Project 2 will examine changes in speech acoustics during single-word shadowing. Stimulus words for Ss to shadow will be produced by multiple speakers, including males and females.
