9319368 Dell ABSTRACT Of all the skills that people possess, speaking is the most intricate, the most well developed, and above all the most uniquely human. The present project investigates speech production focussing on a simple, but critical, component-the phonological encoding of single words. Phonological encoding is a step in the sequence of cognitive processes that map from meaning to articulation, in particular the processes involved in retrieving and ordering speech sounds. Current theories of phonological encoding propose that these sounds are encoded by a frame-and-slot mechanism. Individual sounds (phonological segments) are retrieved and then inserted into slots in a separate phonological frame. The frame dictates the order of the sounds and specifies their structure, including how the how the sounds are grouped into syllables and other constituents. Support for the frame-and-slot idea comes from studies of everyday speech errors. In general, this support takes the from of demonstrations that speakers' errors are sensitive to the kind of information that is provided by the phonological frame. For example, speakers will (almost) never mispronounce a word so badly as to violate the basic phonological patterns, or rules, of the language being spoken. These patterns are assumed to reside, for the most part, in the frame. The research contrasts two explicit computational models of phonological encoding, a spreading activation model that directly embodies the frame-and-slot idea, and a controversial alternative, a parallel-distributed processing (PDP) model that attempts to account for the data without using explicit phonological frames. According to the PDP model, frame-like effects result from superimposing the representation of lexical items on to the same network connections. The models offer predictions about how easily certain sequences of syllables can be produced and the proposed experiments test these predictions. The studies employ two methods, a speeded rep etition technique in which subjects attempt to say sequences of 2-4 syllables as many times as possible in a fixed interval, and a rule-learning paradigm in which subjects produce syllables that derive from "local" phonological rules. The studies are aimed at two questions: (1) Are separate representations of structure (frames) required to account for the speaker's sensitivity to the phonological patterns of her language during phonological encoding? (2) To what extent can the production system's sensitivity to these patterns be altered by recent experience? More generally, the experiments address the an issue that lies at the heart of modern psycholinguistics. How are linguistic generalizations-rules, structural frames, and the like-represented and used during language production and comprehension?
