Evaluating Chinese Noun Compound Interpretation in Natural Language Inference

Noun compound interpretation, generally framed as the generation of a paraphrase, is important for many natural language processing tasks. In our study, we focus on Chinese noun compound interpretation in the context of natural language inference, by presenting a new dataset designed to study whether language models identify the inferential relations between noun compounds and
their paraphrases.
We compare the performance of models fine-tuned on natural language inference data and Large Language Models (LLMs) with prompting, and find that they have a decent ability of drawing the correct inferences, close to or sometimes better than fine-tuned models.