• Noun Detection & Lemmatization

  • Weighted Common Sense cues of the word "Orange" with the meaning "Color"

  • Best overlapping meanings of the words "Orange" and "Color"

  • Relative overlap of the enforced & and wrong meanings "Fruit" of "Orange" and "Vividness" of "Color"

  • Evaluation of the Word Sense Disambiguation step on SemCor Corpus

In this paper I outline an approach, which can be used to disambiguate nouns in English texts by their local context. A formal notion of relation-free context models for noun lemmata is introduced as well as an algorithm that describes how to build these models from common sense with the aid of lexical databases.

The approach was implemented in JAVA, utilizing WordNet & ConceptNet as knowledge resources; spot-check evaluation was done with the SemCor corpus.

read paper (German; ∼265 KB)

"Relationsfreie Kontextmodelle auf der Basis von common sense als integrierter Ansatz zur word sense disambiguation und Bildung lexikalischer Ketten"