Speakers
Description
Aligning a target word in a sentence with a dictionary sense is a challenging task, given the dynamicity of meaning (in context) and the fixed discrete senses in dictionaries, among other reasons. However, it is still a practical necessity for a lexicographer to select example sentences that best illustrate dictionary senses and, hence, find a way to address this challenge. Further challenges appeared after the established use of large corpora as sources of lexicographic evidence and the need to update dictionary entries with additional illustrative examples. The present study tested the usability of an LLM as a facilitator in sense-aligning and example-selecting tasks. The role of the model was limited to assigning a score that reflects the match between (corpus-based and dictionary-cited) examples and existing dictionary senses under several experimental conditions. Quantitative and qualitative analyses showed the usefulness of the model in selecting candidate senses that best represent a dictionary sense, detecting misalignment cases in dictionaries, excluding less representative examples despite their high GDEX scores and identifying highly overlapping senses (especially in the category of adjectives) that are likely to be puzzling to the users.