Game theory and discourse anaphora
Journal of Logic, Language and Information
Dependency-Based Construction of Semantic Space Models
Computational Linguistics
Parenthetical constructions: an argument against modularity
GEAF '09 Proceedings of the 2009 Workshop on Grammar Engineering Across Frameworks
Combining relations for information extraction from free text
ACM Transactions on Information Systems (TOIS)
DAARC'11 Proceedings of the 8th international conference on Anaphora Processing and Applications
| Hi-index | 0.02 |
The central claim of this thesis is that, unlike main clauses, adjunct subordinate clauses do not form independent processing units in the computation of entity-based topic continuity (attention structure) in discourse. This claim has two primary consequences. First, discourse entities in adjunct subordinate clauses are assigned lower salience than main clauses entities, especially subjects. Second, the process that selects antecedents for pronouns in main clauses is qualitatively different from the process of anaphoric interpretation in subordinate clauses. The former is affected by the mechanism responsible for directing attention in discourse. The latter depends heavily on verbs semantics and the effect of connectives. The claims of this thesis are empirically tested for English and Greek. Primary evidence for the low salience of entities in adjunct subordinate clauses comes from corpus studies, which show that entities in adjunct subordinate clauses (a) make poor competitors in the selection of antecedents for subject pronouns in main clauses, and (b) are unlikely to be referred to in the subsequent discourse with a pronoun (unless they are already old). Primary evidence for the two level anaphora resolution mechanism comes from psycholinguistic experiments designed to test if there is a consistent difference in the way we interpret pronouns in main and adjunct subordinate clauses. These findings form the basis for the specification of two new NLP models, a system for the automated evaluation of coherence in student essays and a two level anaphora resolution algorithm. They also make two significant contributions to the Centering Model, (a) Centering's “utterance” is formally defined on the basis of empirical evidence, and (b) Centering's Rough-Shift transition is for the first time validated as a reliable estimator of poor coherence, empirically tested on an operable essay scoring system.