Composition of semantic relations and semantic representation of negation.

摘要

Understanding the meaning of text is key to several natural language processing applications. In this dissertation, we focus on two topics: composition of semantic relations and semantic representation of negation. We present a model to compose relations applicable to any relation inventory and introduce a novel representation for negated statements that reveals implicit positive meaning.;The framework to compose semantic relations automatically obtains inference axioms in an unsupervised manner. In contrast to approaches that extract relations from text, axioms take as their input relations and output a new relation previously ignored. The key to the model is an extended definition for semantic relations including semantic primitives coupled with domain and range restrictions.;Composition of semantic relations manipulates relations by using this extended definition. Composing relations is reduced to composing their domains, ranges and primitives, three analytic tasks that can be automated. Primitives are composed using a manually defined algebra. In this dissertation, we present and evaluate inference axioms obtained over Prop- Bank and a richer set of 26 relations.;Thoroughly representing the meaning of negated statements has been avoided in the past. Negation is a complex phenomenon that greatly influences the semantics of text. Going beyond scope detection, we present a novel representation that reveals implicit positive meaning from negated statements.;The key to the representation is detecting the focus of negation. We target verbal, analytic and clausal negation. New annotation over PropBank is presented and simple models for focus detection are depicted. We further lay out and evaluate how to compose relations when an argument is negated.