Handling Conjunctions in Named Entities

Named entity recognition consists of identifying ‘mentions’ — strings in a text that correspond to named entities — and then classifying each such mention as corresponding to a specific type of named entity, with typical categories being Company, Person and Location. The full range of named entity categories to be identified is usually application dependent. Introduced for the first time as a separately evaluated task at the Sixth Message Understanding Conference in 1995 (see, for example, [Grishman and Sundheim, 1995; 1996]), named entity recognition has attracted a considerable amount of research effort. Initially handled with hand crafted rules (as, for example, in many of the participating systems in MUC-6 and MUC-7) and later by means of statistical approaches (see, for example, [Sang, 2002; Sang and Meulder, 2003]), the state-of-the-art provides high performance for named entity identification and classification both for specific domains and for languageand domain-independent systems. However, our experience with existing software tells us that there are still some categories of named entities that remain problematic. In particular, very little work has explored the problem of the potential ambiguity of conjunctions appearing in named entity strings. Consider, for example, a string like the following: