In this work we try to apply Large Language Models (LLMs) to reframe the Question Answering task as Programming (QAaP). Due to the inherent dynamic nature of the real world, factual questions frequently involve a symbolic constraint: time, solving these questions necessitates not only extensive world knowledge, but also advanced reasoning ability to satisfy the temporal constraints. Despite the remarkable intelligence exhibited by LLMs in various NLP tasks, our experiments reveal that the aforementioned problems continue to pose a significant challenge to existing LLMs. To solve these time-sensitive factual questions, considering that modern LLMs possess superior ability in both natural language understanding and programming,we endeavor to leverage LLMs to represent diversely expressed text as well-structured code, and thereby grasp the desired knowledge along with the underlying symbolic constraint.

翻译：在本工作中，我们尝试利用大型语言模型（LLMs）将问答任务重构为编程式问答（QAaP）。由于现实世界固有的动态特性，事实性问题通常涉及一种符号约束：时间。解决这些问题不仅需要广泛的世界知识，还需要具备高级推理能力以满足时间约束。尽管LLMs在各种自然语言处理任务中展现出卓越的智能，但我们的实验表明，上述问题对现有LLMs仍构成显著挑战。为解决这些时间敏感的事实性问题，考虑到现代LLMs在自然语言理解与编程方面均具备卓越能力，我们致力于利用LLMs将多样化的文本表述转化为结构化良好的代码，从而同时捕获所需知识及其隐含的符号约束。