Word-sense disambiguation using statistical methods

Abstract

We describe a statistical technique for assigning senses to words. An instance of a word is assigned a sense by asking a question about the context in which the word appears. The question is constructed to have high mutual information with the translation of that instance in another language. When we incorporated this method of assigning senses into our statistical machine translation system, the error rate of the system decreased by thirteen percent.

Disambiguating proteins, genes, and RNA in text: a machine learning approach

“…Earlier work on this problem with general language terms (e.g., disambiguating between the use of the word bank as a financial institution and as the slope next to a river) proposed modeling the context of each ambiguous word as a vector of neighboring words (Brown et al, 1991;Gale et al, 1992); for example, words such as "teller" and "money" would likely indicate that a nearby ambiguous bank referred to the financial institution sense. These methods obtain classification accuracies (percentage of correctly disambiguated cases) of 65-92% depending on the word being disambiguated and the alternative senses (choosing between bank/institution and bank/river is much easier than choosing between bank/institution and bank/building (Buitelaar, 1998)).…”

Hybrid data-driven models of machine translation

“…Without any notion of syntax, such as phrases, non-local dependencies are extremely difficult to capture and without any morphological analysis, related words are treated as completely separate types. Huge improvements were seen even when simple morphological analysis was added to the translation models (Brown et al, 1991).…”

Analysis of Statistical Question Classification for Fact-Based Questions

“…Although WordNet expansion introduces a certain amount of noise into our data, it does improve classification as we showed in Section 5.2. The use of word sense disambiguation techniques [2] could be useful to overcome such issues. Furthermore, other semantic ontologies such as the Suggested Upper Merged Ontology (SUMO) [20] have been created based on WordNet and may be provide useful tools for extracting better semantic meaning from questions.…”