- 24 views
- 39 downloads
Toward a Concept-Based Theory of Lexical Semantics
-
- Author / Creator
- Hauer, Bradley
-
This work aims to address the lack of clear theoretical foundations in computational lexical semantics, the sub-field of natural language processing pertaining to computing with the meaning of words. Semantic tasks are of interest for end-user applications (e.g. contextual translation), downstream tasks (e.g. semantic parsing), and evaluating language models and contextualized representations. Nevertheless, the linguistic phenomena on which semantic methods depend – such as senses, synonymy, and translation – lack a clear, coherent theory, consisting of explicitly stated assumptions, definitions, and proven theorems. Further, there is a deficiency of prior work empirically assessing the utility of such theoretical developments.
This thesis, in short, argues for a theory of lexical semantics grounded in universal lexical concepts, and demonstrates, via experimental evidence, that such a theory is important for developing novel, useful, interpretable methods and resources. The thesis begins with a novel theoretical model for wordnets, a class of resources commonly used in lexical semantics. Key definitions are grounded in lexical concepts, culminating in an empirically validated set of best practices for wordnet construction. Next is an investigation of word senses and translations, beginning at the level of lexemes, the most basic level of semantic distinction in a lexicon, and proceeding to more fine-grained sense distinctions. This ultimately yields a novel method semantic tagging method, with applications to word sense disambiguation. The thesis concludes with a novel analysis of semantic tasks themselves, borrowing from theoretical computing science the notion of reducibility, and finally proposing the first taxonomy of semantic tasks. Taken together, these contributions represent substantial progress toward the development of a theory of semantics, and for the development of interpretable methods and resources for semantic tasks.
-
- Subjects / Keywords
-
- Graduation date
- Spring 2024
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.