What Do We Mean When We Speak About Named Entities?

The concept of Named Entity (NE) has its origin in the Named Entity Recognition and Classification (NERC) tasks, an offspring of Information Retrieval systems, and became one of the main interest points in the Sixth and Seventh Message Understanding Conference (MUC-6, MUC-7) competitions, held back in 1995 and 1998 (Cinchor, 1997; Black, 1998). From then on, most competitions and programs include at least one task related to it. That is the case, for instance, of the Automatic Content Extraction (ACE) program, which began in 1999, and whose aim is to develop technologies in order to automatically infer from human languages the entities being mentioned, their relations and the events in which they participate (Doddington et al., 2004). So, the concept of NE appeared in an environment of NLP applications and is far from being linguistically clear and settled. Although there are some disciplines like linguistics and language analysis which study some particular items that share some characteristics with it, they do not fill in the needs to properly set solid criteria to handle NERC tasks. On the one hand, the classic grammatical approach to Proper Noun (PN) analysis is surely insufficient to deal with the problems NERC poses. In one of the reference contributions on the matter in Spanish (Fernández Leborans, 1999), PNs are defined in their prototypical uses in regard of a series of common but not unique features, such as capitalization, lack of inflection, lack of lexical meaning, lack of determiner, lack of translation, monoreferenciality, or incompatibility with restrictive complementation. As can be seen, that approach cares only about PNs in the strict linguistic sense, so it leaves a vast amount of cases out of consideration (such as alphanumerical expressions or temporal expressions) which are relevant when talking about NEs. This proposal has proven to be too narrow to be used in NLP, especially in the case of weak named entity detection and classification, for weak named entities usually do not even contain a PN. On the other hand, the philosophical approach to language and reference, via the concepts of Singular Terms and Definite Descriptions, is far too wide. Summarizing a lot of bibliography in a few words (Fernández Moreno, 2006), Singular Terms include Proper Names, singular Indexical Terms and singular Definite Descriptions. Definite Descriptions are expressions introduced by a singular definite article which predicate a property possessed by a single individual. Singular Indexical Terms include paradigmatically both pronouns (personal and demonstrative) and some adverbs (place and time deictics). Finally, PNs are defined by two features: their reference does not depend on the context of emission, and it is not dictated by the internal structure of the name itself. This approach deals with a number of linguistic phenomena which should not be part of a linguistically-based approach to NE (at least in most of the cases). In fact, most Reference Theories treat Proper Names, most pronouns and any definite noun phrase as a definite expression, whereas NERC systems care only about a subgroup of