Learning to identify animate references

Learning to identify animate references Constantin Or˘asan School of Humanities, Languages and Social Sciences University of Wolverhampton C.Orasan@wlv.ac.uk Richard Evans School of Humanities, Languages and Social Sciences University of Wolverhampton R.J.Evans@wlv.ac.uk Abstract Information about the animacy of nouns is important for a wide range of tasks in NLP. In this paper, we present a method for determining the animacy of English nouns using WordNet and machine learning techniques. Our method firstly categorises the senses from WordNet using an annotated corpus and then uses this information in order to classify nouns for which the sense is not known. Our evaluation results show that the accuracy of the classification of a noun is around 97% and that animate entities are more difficult to identify than inanimate ones. 1 Introduction Information on the gender of noun phrase (NP) referents can be exploited in a range of NLP tasks including anaphora resolution and the applications that can benefit from it such as coreference resolution, information retrieval, information extraction, machine translation, etc. The gender of NP referents is explicitly realised morphologically in languages such as Romanian, French, Russian, etc. in which the head of the NP or the NP’s determiner undergoes predictable morphological transformation or affixation to reflect its referent’s gender. In the English language, the gender of NPs’ referents is not predictable from the surface morphology. Moreover, in (Evans and Or˘asan, 2000) it was argued that it is not always desirable to obtain information concerning the specific gender of a NP’s referent in English. Instead, it is more effective to obtain the animacy of each NP. We define animacy as the property of a NP whereby its referent, in singular rather than plural number, can be referred to using a pronoun in the set fhe, him, his, himself, she, her, hers, herself During the course of this paper, we will discussg. animate and inanimate senses of nouns and verbs. We use these expressions to denote the senses of nouns that are the heads of NPs referring to animate/inanimate entities and the senses of verbs whose agents are typically animate/inanimate entities. In our previous work, we investigated the use of WordNet in order to determine the animacy of entities in discourse. There, we used the fact that each noun and verb sense is derived from unique classes called unique beginners. We classified each unique beginner as being a hypernym of a set of senses that were for the most part either animate or inanimate (in the case of nouns) or indicative of animacy/inanimacy in their subjects (in the case of verbs). In classifying a noun, the number of its senses that belong to an animate class is compared with the number belonging to an inanimate class, and this information is used to make the final classification. In addition, if the noun is the head of a subject, the same information is computed for the verb. Our assumption was that a noun with many animate senses is likely to be used to refer to an animate …

Content truncated to 500 words