Refine
Document Type
- Conference Proceeding (6) (remove)
Has Fulltext
- yes (6)
Is part of the Bibliography
- no (6)
Keywords
Institute
- Informatik (6)
In the last years, much effort went into the design of robust anaphor resolution algorithms. Many algorithms are based on antecedent filtering and preference strategies that are manually designed. Along a different line of research, corpus-based approaches have been investigated that employ machine-learning techniques for deriving strategies automatically. Since the knowledge-engineering effort for designing and optimizing the strategies is reduced, the latter approaches are considered particularly attractive. Since, however, the hand-coding of robust antecedent filtering strategies such as syntactic disjoint reference and agreement in person, number, and gender constitutes a once-for-all effort, the question arises whether at all they should be derived automatically. In this paper, it is investigated what might be gained by combining the best of two worlds: designing the universally valid antecedent filtering strategies manually, in a once-for-all fashion, and deriving the (potentially genre-specific) antecedent selection strategies automatically by applying machine-learning techniques. An anaphor resolution system ROSANA-ML, which follows this paradigm, is designed and implemented. Through a series of formal evaluations, it is shown that, while exhibiting additional advantages, ROSANAML reaches a performance level that compares with the performance of its manually designed ancestor ROSANA.
In the last decade, much effort went into the design of robust third-person pronominal anaphor resolution algorithms. Typical approaches are reported to achieve an accuracy of 60-85%. Recent research addresses the question of how to deal with the remaining difficult-toresolve anaphors. Lappin (2004) proposes a sequenced model of anaphor resolution according to which a cascade of processing modules employing knowledge and inferencing techniques of increasing complexity should be applied. The individual modules should only deal with and, hence, recognize the subset of anaphors for which they are competent. It will be shown that the problem of focusing on the competence cases is equivalent to the problem of giving precision precedence over recall. Three systems for high precision robust knowledge-poor anaphor resolution will be designed and compared: a ruleset-based approach, a salience threshold approach, and a machine-learning-based approach. According to corpus-based evaluation, there is no unique best approach. Which approach scores highest depends upon type of pronominal anaphor as well as upon text genre.