A Theory of Timing Effects in a Self-Organizing Model of Sentence Processing

Many leading theories of human sentence processing assume that language compre- hension and production take place under the strict control of a symbolic grammar. For example, in sentence comprehension, reading or hearing a word triggers the application of a grammar rule that incorporates the word into the existing sentence structure so that the resulting structure is consistent with the all of the rules of the grammar. These theories have had wide success in explaining important timing effects, e.g., predicting speed-ups or slowdowns while reading a sentence. A number of phenomena have been identified, though, that challenge these grammar-controlled theories and motivate the development of an alternative theory. In local coherenceeffects, people seem to entertain syntactic structures that are compatible with a subset of the words in a sentence but ungrammatical in the context of the rest of the sentence.Agreement attraction occurs when the verb of a sentence agrees in number with a noun other than the subject, in violation of the rules of a grammar. The existence of these phenomena, which grammar-controlled theories struggle to account for, motivatesself-organizing sentence processing-treelet harmony (SOSP-TH), the focus of this dissertation. Instead of being strictly controlled by a symbolic grammar, lexically anchored syntactic treelets in SOSP-TH self-organize into larger structures via local interactions that try to maximize the well-formedness (harmony) of the resulting structure. Importantly, SOSP-TH includes less-than-perfect syntactic structures, which allows it to account for local coherence and agreement attraction effects as a natural by-product of its strongly bottom-up syntactic processing. In contrast to many previous self-organizing models, the mathematical formulation of SOSP-TH allows us to make precise predictions about processing times, which we test in three experiments on interference effects in subject-verb agreement. Overall, the model provides a good fit to the human data and provides a parsimonious explanation for the semantic interference effects tested in the experiments. SOSP-TH does face some challenges in accounting for certain data points, but those, combined with a set of new predictions, make it a promising theory for future research at the intersection of theoretical linguistics, dynamical systems modeling, and experimental psycholinguistics