UC Santa Cruz

This dissertation addresses the role of memory processes in real-time language comprehension. A rapidly growing body of research indicates that the memory access required for incremental sentence comprehension utilizes a content-addressable architecture that gives rise to similarity-based interference effects, thereby unifying psycholinguistic research with independently-motivated principles of memory and cognitive models (McElreeet al., 2003; Van Dyke, 2003). I investigate the nature of the information encoded in linguistic representations, using interference effects to diagnose the properties that are used to retrieve constituents from memory. Empirically, the focus is on the retrieval of subject constituents in two situations: a subject separated from its verb by an intervening relative clause, and so-called control dependencies ('Mary promised John to leave') where the interpretation of an unexpressed infinitive subject depends on a preceding overt subject argument. Experimental manipulations vary the similarity between the target subject and a grammatically-illicit subject along various linguistic dimensions, and the morpho-syntactic complexity of these constituents. I tie the results to computational simulations implemented in the ACT-R theory of cognition (Anderson & Lebiere, 1998; Lewis et al., 2005), investigating different retrieval structures and targets. I conclude that subject retrieval is interference-prone, supporting content-addressable architectures, and that subject encodings are retrieved based on highly abstract syntactic properties. The results also indicate that constituent complexity modulates processing difficulty at the retrieval site (Hofmeister, 2011), which I argue to be an elaboration effect.