대규모 말뭉치에서의 ‘N+N’형 임시어 추출과 그 특성

초록

This study constructs a large-scale corpus of 2.5 billion Korean tokens, balanced and refined for linguistic analysis, and employs SoyNLP’s LRNounExtractor_v2 to extract candidate N+N nonce words. The resulting dataset of 8,165 items, filtered by morpho-lexical criteria, has been released as the Korean Nonce Word Database (KNWD). Drawing on this lexical resource, we track nonce-word usage trends across time-series corpora from 2009 to 2023 and classify items into temporary and persistent types (rising, declining, or fluctuating). We then analyze the quantitative properties of these categories and explore socio-cognitive factors that appear to shape their distribution and use.

키워드