Word2Dict – Lemma Selection and Dictionary Editing Assisted by Word Embeddings

Page 819-826
Author Nicolai Hartvig Sørensen, Sanni Nimb
Title Word2Dict – Lemma Selection and Dictionary Editing Assisted by Word Embeddings
Abstract We describe the use of a tool that assists lexicographers with extending the lexical coverage of an online Danish dictionary. The tool is based on a word embedding model (word2vec) trained on a large Danish corpus, and it presents semantically related lemmas already included in the dictionary and, importantly, their definitions. Furthermore, lemma candidates, i.e. words from the corpus which are not included in the dictionary, are presented in the tool, supplemented by information on corpus frequency. The tool thereby facilitates the lemma selection as well as the process of writing consistent definitions across synonyms and near synonyms. We discuss the shortcomings of the tool and the semantic model when it comes to identifying words similar in meaning from different genres and registers. We also look closer into whether it does in fact benefit the dictionary-making process or not by studying a number of previously edited words, including their synonyms, and comparing them with the output data from the tool.
Session Poster Presentations
Keywords lemma selection, word embedding, word2vec, semantic similarity, dictionary-making process
author={Nicolai Hartvig Sørensen, Sanni Nimb},
title={Word2Dict – Lemma Selection and Dictionary Editing Assisted by Word Embeddings},
booktitle={Proceedings of the XVIII EURALEX International Congress: Lexicography in Global Contexts},
address={Ljubljana, Slovenia},
editor={Jaka Čibej, Vojko Gorjanc, Iztok Kosem, Simon Krek},
publisher={Ljubljana University Press, Faculty of Arts},
isbn={978-961-06-0097-8}, }