A model termed the 'Focusing Hypothesis' is presented. It is proposed that language processing is shared by an analytic and a holistic system, according to a task specific balance of demand and efficiency. The analytic system could function alone, but it is more economical, in normal communication, for holistic processing to operate up to clausal level and analysis to deal with the evaluation of propositions. The severe limitations on the abilities of the holistic system originate from its use of formulae to recognise familiar words in familiar structures. Where problems arise, the analytic system 'trouble-shoots', by focusing attention onto the language, at the expense of propositional focus. The relative involvement of the two systems is variable, according to the strategy selected from a task specific strategy option range; the strategy option range and preferences within it are built up as a response to the environmental requirements placed on the individual. Apparent evidence for left hemisphere lateralised language is re-examined in the light of this hypothesis, which proposes that the test environment of most psycholinguistic and clinical assessments induces a language-focusing strategy and thus deactivates the right hemisphere (holistic) mechanisms. It is predicted that careful modifications to the methods of test administration could reveal right hemisphere activity by permitting it to occur. Support for the hypothesis is drawn from the literature relating to neurophysiological (dynamic) studies and from the reported symptoms of left and right hemisphere damaged patients. Accounts of polyglot (bilingual) acquisition and storage and of differential language loss in polyglot aphasia are also examined. Output processing is examined with reference to one specific hypothesis (Pawley & Syder 1983) which closely aligns with the one for input presented by the Focusing Hypothesis. Two experiments attempt to examine contrasts in strategy as a function of age (Experiment I) and stimulus type (Experiment II). Neither displays strong patterns of the kind predicted to be associated with contrasts in hemispheric superiority according to strategy choice, and it is suggested that, despite the attempt, the experimental designs failed to enable consistent access to the proposition-focused strategies held to be operational in normal communication, that is, those involving holistic processing.

No description available.

FrameNet is used to enhance the performance of semantic QA systems. FrameNet is a linguistic resource that encapsulates Frame Semantics and provides scenario-based generalizations over lexical items that share similar semantic backgrounds. / Doctor of Philosophy

Representation of sentences that captures semantics is an essential part of natural language processing systems, such as information retrieval or machine translation. The representation of a sentence is commonly built by combining the representations of the words that the sentence consists of. Similarity between words is widely used as a proxy to evaluate semantic representations. Word similarity models are well-studied and are shown to positively correlate with human similarity judgements. Current evaluation of models of sentential similarity builds on the results obtained in lexical experiments. The main focus is how the lexical representations are used, rather than what they should be. It is often assumed that the optimal representations for word similarity are also optimal for sentence similarity. This work discards this assumption and systematically looks for lexical representations that are optimal for similarity measurement between sentences. We find that the best representation for word similarity is not always the best for sentence similarity and vice versa. The best models in word similarity tasks perform best with additive composition. However, the best result on compositional tasks is achieved with Kroneckerbased composition. There are representations that are equally good in both tasks when used with multiplicative composition. The systematic study of the parameters of similarity models reveals that the more information lexical representations contain, the more attention should be paid to noise. In particular, the word vectors in models with the feature size at the magnitude of the vocabulary size should be sparse, but if a small number of context features is used then the vectors should be dense. Given the right lexical representations, compositional operators achieve state-of-the-art performance, improving over models that use neural-word embeddings. To avoid overfitting, either several test datasets should be used or parameter selection should be based on parameters' average behaviours.