Automatic Contexonym Organizing Model (ACOM)

Normal language user’s word-association intuition (e.g. drunken – stagger) raises questions about the mental lexicon organization and its application for natural language processing tasks. We present an automatic contextually related words (contexonym) organizing model (ACOM) that reflects this intuition, giving one of the possible answers to this question. Trained on large corpora, the model (1) selects contexonyms for a given word and (2) classifies these groups of related words on a geometric representation. Some near-synonyms discussed in Near-Synonymy and Lexical Choice (Edmonds and Hirst, 2002) were chosen to test the model. The results showed that our model provides valuable contexonyms that reflect different usage and nuance of each word. Furthermore, the test on polysemous words showed that the model can classify contexonyms by grouping the different senses of a target word. The model can can be used as both theoretical lexicon-related research and practical natural language processing (NLP) research as well as an interactive reference.