Experimental approaches to sentence processing focus on localizing areas responsible for language comprehension in the brain oftentimes disregarding the role of time in both the construction and...Show moreExperimental approaches to sentence processing focus on localizing areas responsible for language comprehension in the brain oftentimes disregarding the role of time in both the construction and deconstruction of linguistic structure. Inspired by predictive coding and cue integration, this thesis proposes a theoretical framework for sentence processing where the hierarchical structure of language and its evolution over time profoundly influences its processing leading to time-contingent weighted integration of information. Essential to this theory is the assumption that the reliability of the internal representations generated by each level of linguistic analysis influences the gain of the predictions formulated by the other levels. Multivariate Pattern Analysis was used to compare the gain of semantic and phonological processing at two different timepoints in a sentence. Experiment 1 was the design of an EEG Multivariate Pattern Classification experiment where the classification accuracy of a phonological and semantic classifier for words in early and late positions in a sentence would be compared. We expected classification accuracy of the phonological classifier to be constant regardless of word position and a higher classification accuracy for the semantic classifier at later time points relative to the phonological classifier. Experiment 2 was a Representational Similarity Analysis of nouns in early and late positions from MEG audiobook data. When correlating Phonological and Semantic models with the data, no significant time windows were found. However, the presence of uncorrected clusters suggests the implementation of nested timescales as variations in temporal integration frequency.Show less
This thesis consists of two parts. In the first part, I go over some of the difficulties in representing semantic structures, including a discussion of the characteristics of syntactic and semantic...Show moreThis thesis consists of two parts. In the first part, I go over some of the difficulties in representing semantic structures, including a discussion of the characteristics of syntactic and semantic structures, syntax-semantics correspondence, and two types of semantic underspecification. In the second part, I offer a design for the visualization of semantic structure as derived by the Delilah parser, as well as a software tool for drawing these structures automatically.Show less
Artificial Grammar Learning (AGL) is a powerful experimental paradigm for testing specific hypotheses about language acquisition but is limited because of its reliance on meaningless grammatical...Show moreArtificial Grammar Learning (AGL) is a powerful experimental paradigm for testing specific hypotheses about language acquisition but is limited because of its reliance on meaningless grammatical structures. Meanwhile, formal and computational semantics provide rigorous ways to define and calculate meanings for formal languages, but are typically only used to describe or simulate the linguistic competence of adult speakers. This thesis attempts to connect these two fields by proposing a new type of AGL experiment that uses a language with both a context-free syntax and a formally defined semantics which can be used to express spatial relationships between objects. Moreover, using a computer simulation in which an Intelligent Agent (IA) acquires such a language, it shows how this new paradigm can be used to test psycholinguistic hypotheses about the acquisition of both syntax and semantics.Show less
