A comparison of Named-Entity Disambiguation and Word Sense Disambiguation

A comparison of Named-Entity Disambiguation and Word Sense Disambiguation Angel X. Chang†, Valentin I. Spitkovsky†, Christopher D. Manning†, Eneko Agirre‡ †Computer Science Department, Stanford University, Stanford, CA, USA ‡IXA Group, University of the Basque Country, Donostia, Basque Country {angelx,vals,manning}@cs.stanford.edu, e.agirre@ehu.eus Abstract Named Entity Disambiguation (NED) is the task of linking a named-entity mention to an instance in a knowledge-base, typically Wikipedia-derived resources like DBpedia. This task is closely related to word-sense disambiguation (WSD), where the mention of an open-class word is linked to a concept in a knowledge-base, typically WordNet. This paper analyzes the relation between two annotated datasets on NED and WSD, highlighting the commonalities and differences. We detail the methods to construct a NED system following the WSD word-expert approach, where we need a dictionary and one classifier is built for each target entity mention string. Constructing a dictionary for NED proved challenging, and although similarity and ambiguity are higher for NED, the results are also higher due to the larger number of training data, and the more crisp and skewed meaning differences. Keywords: Word Sense Disambiguation, Named-Entity Disambiguation 1. Introduction A named-entity mention may refer to multiple entities, and the process of resolving the appropriate meaning in con- text is called entity linking (EL) or named entity disam- biguation (NED). The former terminology stresses the im- portance of linking a mention to an actual instance in the given knowledge-base (McNamee and Dang, 2009). We prefer the latter term (NED), which focuses on the potential ambiguity among several possible instances. It highlights the connection to the closely related problem of word-sense disambiguation (WSD), where the goal is to disambiguate open-class words with respect to a sense inventory, such as WordNet (Fellbaum, 1998). In this paper we compare the closely related worlds of WSD and NED. In WSD, an exhaustive dictionary is provided, while in NED, one has to generate all candidate entities for a target string — a step that has been shown to be critical to success (Hachey et al., 2012), where it is called “candidate generation”. In WSD very few occurrences correspond to senses missing in the dictionary, but in NED this problem is quite prevalent. We show also that ambiguity is larger for NED. On the bright side, there is a lot of potential training data for NED in the form of human-generated anchor-texts, while hand-annotated data for WSD is a scarce resource. Section 2 introduces two popular datasets, TAC-KBP and Senseval-3, which we use to compile statistics on ambigu- ity and synonymy for WSD and NED in Section 3. Sec- tion 4 focuses on the performance of the Most Frequent Sense baseline, a widely used WSD baseline. This section also introduces the method to build dictionaries for NED, which is a requirement for collecting the sense frequency statistics. Section 5 explains how a supervised WSD ar- chitecture (the so-called word-expert approach) is applied to NED with good results. A longer version of this paper is available at http://arxiv.org/abs/1603.04767. 2. Datasets We selected two of …

Content truncated to 500 words