LLM 기반 현대 중국어 동결식 추출 및 언어적 특징 분석

초록

This study reexamines the linguistic characteristics of Verb-Resultative Complement Compounds (VRCs) in Modern Chinese based on authentic language use. A specific corpus of approximately 26,000 VRC sentences was constructed by collecting and refining raw data using LLM-based generative AI, and a comprehensive quantitative analysis was conducted using #LancsBox X. The findings are as follows. First, the perfective aspect marker le (了) , exhibited the highest frequency in the corpus, confirming a dominant co-occurrence with VRCs. Co-occurrence with guo (過) was restricted; co-occurrence with zhe (著) was limited but possible, whereas with zai (在) it was unattested. Second, the high frequency of de (的) revealed notable VRC usage in shi-de (是-的) and hui-de (會-的) constructions—noteworthy given the potential tension between focus-marking and the resultative semantics of VRCs. Third, collocation analysis (L1–L5, R1–R3) identified strong collocates including ba (把), bei (被), dou (都), jiu (就), and shi (是), suggesting an association between VRCs and emphatic sentence functions. The strongest rightward collocate was le (了), reaffirming “VRC-le (了)” as the canonical pattern. Finally, although ba (把) constructions accounted for only one-third of object-taking VRCs in raw frequency, they encompassed 76% of all attested complement types, confirming the ba (把)-construction as the primary syntactic environment for VRC combinatorial diversity. This study demonstrates that LLM-based generative AI enables the efficient construction of a linguistically specific corpus at scale. However, the risk of AI misclassification—as seen in the underextraction of negative constructions—underscores the need for conservative rule-setting and ongoing researcher oversight.

키워드