Pronominal and Sortal Anaphora Resolution for Biomedical Literature

Pronominal and Sortal Anaphora Resolution for Biomedical Literature Yu-Hsiang Lin and Tyne Liang Department of Computer and Information Science National Chiao Tung University Hsinchu, Taiwan Email: gis91534@cis.nctu.edu.tw; tliang@cis.nctu.edu.tw; Abstract. Anaphora resolution is one of essential tasks in message understanding. In this paper resolution for pronominal and sortal anaphora, which are common in biomedical texts, is addressed. The resolution was achieved by employing UMLS ontology and SA/AO (subject-action/action-object) patterns mined from biomedical corpus. On the other hand, sortal anaphora for unknown words was tackled by using the headword collected from UMLS and the patterns mined from PubMed. The final set of antecedents finding was decided with a salience grading mechanism, which was tuned by a genetic algorithm at its best-input feature selection. Compared to previous approach on the same MEDLINE abstracts, the presented resolution was promising for its 92% F-Score in pronominal anaphora and 78% F-Score in sortal anaphora. 1 Introduction Anaphora resolution is one of essential tasks in message understanding as well as knowledge discovering. For example recognizing biomedical relations among biomedical entities from research literature like MEDLINE database requires anaphora resolution for those mentioned entities from texts. There are different types of anaphora to be solved like pronominal, sortal (definite), zero, event, and coreference anaphora. In biomedical literature, pronominal anaphora and sortal anaphora are the two common anaphora phenomena. Pronominal anaphora is that mentioned entity is substituted by the pronoun. Sortal (definite) anaphora occurs in the situation that a noun phrase is referred by its general concept entity. Definite noun phrases are noun phrases stating with demonstrative articles, such as those, this, both, each and these or starting with a definite article. Generally identifying antecedents of an anaphor can be handled by using syntactic, semantic or pragmatic clues. In past literature, syntax-oriented approaches for general texts can be found in [Hobbs, 76; Lappin and Leass 94; Kennedy and Boguraev 96] in which syntactic representations like grammatical role of noun phrases were used. On the other hand more information other than syntactic information like co-occurring patterns obtained from the corpus was employed during antecedent finding in [Dagan and Itai, 90]. Information with limited knowledge and linguistic resources for resolving pronouns were found in [Baldwin, 97]. In [Denber, 98, Mitkov, 02], more knowledge from the outer resource like WordNet was employed in solving anaphora. Similarly WordNet together with additional heuristic rules were applied for resolving pronominal anaphora in [Liang and Wu, 04] which animacy information is obtained by analyzing the hierarchical relation of nouns and verbs in the surrounding context learned from WordNet. In biomedical literature, it was found that sortal anaphors are prevalent in the texts like MEDLINE abstracts [Castaño et al., 02]. To deal this type of anaphora, Castaño et al. [02] used UMLS (Unified Medical Language System) as ontology to tag semantic type for each noun phrase and used some significant verbs in biomedical domain to extract most frequent semantic types associated to agent (subject) and patient (object) role of SA/AO-patterns. The result showed SA/AO-pattern could gain increase …

Content truncated to 500 words