Design of Anaphora Resolution for Afaan Oromo Personal Pronoun

Abstract

Anaphora is defined as the linguistic phenomenon of pointing back to a previously stated item in the text. The pointing back word or phrase is called an anaphor and the entity to which it refers or for which it stands is its antecedent. Anaphora resolution is the process of determining the antecedent of anaphor. The scope of this resolution can be Intersentential or Intrasentential. The implementation of Anaphora resolution improves most of NLP applications such as machine translation, question answering, and text summarization and information extraction. Most of Anaphora resolutions are studied for the English language. Nowadays research on anaphora resolution has been studied for other languages, such as Norwegian, Estonian, Spanish, Arabic, Turkish andAmharic. As anaphora resolution, systemfor one language is notdirectlyadapted to another language, because it requires specific design for Afan Oromo based on the grammatical behavior of the language. This study presented a model for resolving anaphora occurrences in Afan Oromo text using knowledge poor approach. The approach is implemented without any sophistication of linguistic knowledge and its core method is a list of multilingual antecedent indicators like a subject place, recency, frequency and constraints rules like gender, person and number agreement. The proposed model focuses mainly on pronominal anaphora types and specifically on third personal pronouns. The models deal with Intrasentential and Intersentential types of anaphor. These personal pronouns can be hidden anaphor that resides in verbs and independent anaphor that occurs as personal pronouns. The proposedmodel follow different sub tasks, These are: preprocessing text which includes POS tagging, locating independent anaphor in a sentence in the text, extracting hidden personal pronouns, identifying possible antecedent candidates in defined range of preceding sentences, application of eliminative rule – constraint rules and optional rule - preferential rules, and selection of the candidate with the highest aggregate score. Data used as datasets for our experiment were collected from Afan Oromo Holly Bible and Fiction. The evaluation of the prototype is performed on 330 sentences and conducted for two different scenarios. First, the hidden intrasentential anaphor algorithm scored a success rate of 57.84% and for independent intrasentential, anaphor the algorithm scored 47.51% success rate. For both Intrasentential anaphor, the algorithm scored a success rate of 55.20% On the other scenario, the algorithm scored success rate of 98.28% for hidden intersentential anaphor algorithm and 98.85% for independent intersentential algorithm. For both Intersentential anaphor, the algorithm scored success rate of 98.43%. The challenging tasks in the study are extracting hidden anaphor from the verb word class because there are ambiguity of words in the language that are extracted based on the meaning of the sentence. Therefore, further works focusing on the knowledge of pragmatic is the major direction with regard to Afan Oromo anaphora resolution.