EUVIMEDEUVIMED entdecken
Crossref

Do Large Language Models Reflect the Definiteness Constraint in Chinese? : The Definiteness Constraint on NP₂ in Existential Sentences

Nahee Kim, Sukyong Yu

Korea Association of Chinese Linguistics · 30.04.2026 · 123 · 177-210

DOI: 10.38068/kjcl.123.6 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Korea Association of Chinese Linguistics, online erschienen am 30.04.2026. Verfasst von Nahee Kim und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study quantitatively investigates how the definiteness restriction on “NP₂” in the Chinese existential construction (存现句) “NP₁+VP+NP₂” varies depending on discourse context, using probability-based metrics derived from large language models (LLMs). Previous studies have shown that indefinite noun phrases are generally preferred in the “NP₂” position, while definite noun phrases tend to be judged as less natural. Although information-structural accounts suggest that this restriction may be weakened when the referent is activated through prior discourse, quantitative evidence remains limited. To address this issue, this study employed a 2×2 experimental design crossing [±context] and [±definiteness], based on 120 minimal pairs extracted from the CCL and BCC corpora. Three pretrained LLMs—DeepSeek-LLM-7B-base, Qwen2.5-7B-base, and GLM-4-9B (base)—were used to compute cross-entropy values measured in bits per token (bits/t), where lower values indicate greater predictability and naturalness. The results showed that, without contextual information, all three models assigned lower bits/t values to sentences containing indefinite “NP₂,” confirming the probabilistic tendency of the definiteness restriction. When contextual information was provided, the bits/t values of sentences with definite “NP₂” consistently decreased, indicating that prior discourse weakens the restriction by activating the referent and establishing a discourse anchor. This pattern was consistently observed across all three models, highlighting the important role of contextual information in probabilistic sentence processing. These findings suggest that the definiteness restriction on “NP₂” is not an absolute syntactic constraint, but a probabilistic constraint sensitive to discourse and information structure. Furthermore, this study demonstrates that LLM-based probabilistic metrics provide an effective quantitative approach to investigating the interaction between discourse and grammatical constraints, offering empirical support for information structure theory and expanding methodological approaches to grammar research using large language models.

Zitieren

APA

Kim, N., & Yu, S. (2026). Do Large Language Models Reflect the Definiteness Constraint in Chinese? : The Definiteness Constraint on NP₂ in Existential Sentences. Korea Association of Chinese Linguistics, 123, 177-210. https://doi.org/10.38068/kjcl.123.6

Vancouver

Kim N, Yu S. Do Large Language Models Reflect the Definiteness Constraint in Chinese? : The Definiteness Constraint on NP₂ in Existential Sentences. Korea Association of Chinese Linguistics. 2026;123:177-210. doi:10.38068/kjcl.123.6

BibTeX
@article{Kim2026,
  author = {Kim, Nahee and Yu, Sukyong},
  title = {Do Large Language Models Reflect the Definiteness Constraint in Chinese? : The Definiteness Constraint on NP₂ in Existential Sentences},
  journal = {Korea Association of Chinese Linguistics},
  year = {2026},
  volume = {123},
  pages = {177--210},
  doi = {10.38068/kjcl.123.6}
}
Kostenlos starten →