ABSTRACT The goal is to investigate the architecture of the language processing system via the study of how the system copes with ambiguity. Two classes of models are contrasted: serial models, in which only one interpretation of an ambiguity is considered, and constrained parallel models, in which several interpretations may be briefly considered in parallel. A second factor is whether probabilistic cues may influence the interpretation(s) that are generated for an ambiguity. Two different ambiguities are investigated: syntactic ambiguities and pronominal reference ambiguities. The specific aims are to investigate 1) the interaction of cues that are encountered before an ambiguity and ones encountered after an ambiguity, 2) the interaction of cues from different levels of linguistics representations, such as syntactic and discourse-level cues, and 3) individual differences in ambiguity resolution as a function of working memory capacity. Earlier research has relied heavily on one dependent measure, reading time at the disambiguation. Six supplementary measures are proposed: 1) examination of reading time in other regions, 2) studies in which ambiguities are resolved with a highly favored interpretation, 3) examination of cues that follow the ambiguity, 4) norming of stimuli to manipulate cue strength on a continuum, 5) the use of examination of individual differences. It is expected that the addition of these methods will distinguish the opposing models more precisely than in much previous work.
