CCL23-Eval 任务2系统报告:WestlakeNLP,基于生成式大语言模型的中文抽象语义表示解析

This paper presents our participating system in the Chinese Abstract Meaning Representation Parsing Evaluation Task at the 22nd China National Conference on Computational Linguistics. Chinese Abstract Meaning Representation (CAMR) not only captures sentence semantics through graphical representation but also ensures the alignment of concepts and relations. Recently, generative large language models have demonstrated exceptional abilities in generation and generalization across various natural language processing tasks. Motivated by these advancements, we fine-tune the Baichuan-7B model to directly generate serialized CAMR from the provided text in an end-to-end manner. Experimental results demonstrate that our system achieves comparable performance to existing methods, eliminating the need for part-of-speech, dependency syntax, and complex rules.