Language comprehension relies on memory. In many cases, we need information scattered throughout an utterance to arrive at its meaning. Consider the sentence: "Mary, who was really, really tired, passes the bag to Suzy." To understand who is doing the passing, we need to retrieve the word "Mary" when reading "passes the bag".
To date, extensive research has investigated memory operations involved in processing syntactic dependencies within a sentence. This dissertation extends that research by focusing on semantic dependencies, which are not constrained by sentence boundary.
The approach taken here is a mix of behavioural experiments and computational modeling. The experiments attempt to establish whether semantic dependencies are subject to distance effects, that is, whether they require more effortful memory operations when the distance between the head and the dependent increases. Having established this, the dissertation turns to modeling to propose a mechanistic computational model explaining this phenomenon. The models, implemented in ACT-R, explore a variety of possible explanations and ultimately suggest that the effect arises as a combination of parallel processing and diminishing activation of the relevant chunk over time.
The dissertation contributes to our understanding of memory systems involved in language comprehension and highlights the difficulties and pitfalls in cognitive modeling.
