Year: 2007
Type: dissertation
Abstract: Semantic parsing involves deep semantic analysis that maps natural \nlanguage sentences to their formal executable meaning representations. This \nis a challenging problem and is critical for developing computing systems that \nunderstand natural language input. This thesis presents a new machine learning \napproach for semantic parsing based on string-kernel-based classification. \nIt takes natural language sentences paired with their formal meaning representations \nas training data. For every production in the formal language grammar, \na Support-Vector Machine (SVM) classifier is trained using string similarity \nas the kernel. Meaning representations for novel natural language sentences \nare obtained by finding the most probable semantic parse using these classi- \nfiers. This method does not use any hard-matching rules and unlike previous \nand other recent methods, does not use grammar rules for natural language, \nprobabilistic or otherwise, which makes it more robust to noisy input. \nBesides being robust, this approach is also flexible and able to learn \nunder a wide range of supervision, from extra to weaker forms of supervision. \nIt can easily utilize extra supervision given in the form of syntactic parse trees \nfor natural language sentences by using a syntactic tree kernel instead of a \nstring kernel. Its learning algorithm can also take advantage of detailed supervision \nprovided in the form of semantically augmented parse trees. A simple \nextension using transductive SVMs enables the system to do semi-supervised \nlearning and improve its performance utilizing unannotated sentences which \nare usually easily available. Another extension involving EM-like retraining \nmakes the system capable of learning under ambiguous supervision in which \nthe correct meaning representation for each sentence is not explicitly given, \nbut instead a set of possible meaning representations is given. This weaker and \nmore general form of supervision is better representative of a natural training \nenvironment for a language-learning system requiring minimal human supervision. \nFor a semantic parser to work well, conformity between natural language \nand meaning representation grammar is necessary. However meaning \nrepresentation grammars are typically designed to best suit the application \nwhich will use the meaning representations with little consideration for how \nwell they correspond to natural language semantics. We present approaches \nto automatically transform meaning representation grammars to make them \nmore compatible with natural language semantics and hence more suitable for \nlearning semantic parsers. Finally, we also show that ensembles of different \nsemantic parser learning systems can obtain the best overall performance.
Authors: Raymond J. Mooney, Rohit J. Kate
Institution: The University of Texas at Austin
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W113528495