| 1 |
Towards resolving ambiguity in semantic parsing with hints
|
|
|
|
Abstract:
Thesis (Ph. D.)--University of Rochester. Department of Computer Science, 2020. ; The first step towards any approach to Natural Language Understanding pipelines is to transform the incoming utterance into a form that can be reasoned on. The task of Semantic Parsing creates a graphical meaning representation of utterances by identifying and labeling predicate-argument structures. Several paradigms of meaning representation have been proposed to encode the complexity of human natural language. In this dissertation we examine and compare the structure of two such meaning representations, as well as the surrounding framework of semantic and lexical resources and parsers, in order to identify methods in which we can improve the accuracy of the semantic parsers. We propose an extension to the evaluation metrics to score similarity between instances of a representation. We observe that issues surrounding high levels of lexical ambiguity can create cascading errors in the parsing process and propose two systems to improve the parses by reducing the level of ambiguity.
|
|
Keyword:
Artificial intelligence; Natural language understanding; Semantic parsing
|
|
URL: http://hdl.handle.net/1802/36166
|
|
BASE
|
|
Hide details
|
|
|
|
|
