Seminar by Amit Dubey
The Influence of Discourse Information on Human Syntactic Ambiguity Resolution
Amit Dubey
Division of Informatics, University of Edindurgh
Date: Monday, October 12, 2009
Time: 4:00 PM
Venue: CS101. Abstract:
Humans combine information from a variety of sources while understanding language. This includes several different kinds of linguistic information sources such as syntax and discourse, as well as non-linguistic sources such as world knowledge. In the first half of this talk, I will revisit an old debate in psycholinguistics concerning how quickly information from various information sources are integrated while understanding language. Although recent work has shown these various information sources work interactively to determine the meaning of an utterance, I will demonstrate that, in certain circumstances, it is possible to make discourse information "hard" to use. This results in syntax-first behaviour in reading, whereby a (possibly incorrect) structure of a temporarily ambiguous sentence is first suggested by the syntactic module, and later updated by semantic and discourse information.
Theories of language understanding have often been driven by computational models of language from the Artificial Intelligence community. Early syntax-first theories were influenced by rule-based AI systems, whereas later models hypothesizing various forms of interaction drew their inspiration from statistical AI, including artificial neural networks. In the second half of the talk, which will be of interest to Computer Scientists, I will show how recent hybrids of rule-based and statistical AI systems provide a sound basis for a computational model of our experimental results.
About The Speaker:
Dr. Amit Dubey has done a Ph.D. in Computational Linguistics from Saarland University in 2006.