Resumen
Today?s natural language processing research frequently addresses the issue of content semantization (including the semantization of unstructured texts such as electronic news) by means of semantic annotation or its special case, ontology-based and domain-oriented conceptual annotation. Conceptual annotation is often complicated by conceptual ambiguity manifested in one-to-many mappings between lexical items and ontology concepts. This paper examines the causes of conceptual ambiguity in restricted domain texts, with the case study of English-language electronic news on terror attacks. Four causes of conceptual ambiguity are revealed: part-of speech homonymy, lexical ambiguity, the plurality of conceptual meanings (the most productive), and the extralinguistic context (the least productive, but the hardest to resolve). Three quantitative disambiguation methods are studied: a) tag ranking, b) a bigram-model-based contextual method, and c) a positional method. All the methods are found useful for computer-aided conceptual disambiguation, yet it is pointed out that these quantitative methods are not quite accurate when used alone and rule-based methods would be a good addition.