ايمان براتي

In recent years, significant advancements have been made in the field of natural language processing, particularly in question-answering systems. However, these systems often face limitations in responding to complex multi-hop questions in multilingual environments, especially for low-resource languages. One of the fundamental limitations of current models is their inability to effectively combine information extracted from different languages. In this research, by introducing a new task called cross-lingual multi-hop question answering, we have addressed these challenges. For this purpose, an artificial dataset named XhotpotQA has been created by translating the HotpotQA dataset into 23 different languages. This dataset enables the eva‎luation of systems in their ability to integrate an‎d infer information from multiple languages. To solve this complex task, we propose a method based on large language models an‎d a two-stage selec‎tor-reader architecture. In this approach, the problem is broken down into two separate sub-tasks, an‎d each part is independently processed using large language models. Additionally, by employing techniques such as chain-of-thought an‎d question decomposition, the performance of the models has been improved. Experimental results show that the proposed method not only can solve the multi-hop question-answering problem in English with high accuracy an‎d efficiency but also can be effectively generalized to the cross-lingual setting. These results indicate the high capacity of large language models in solving complex an‎d multilingual problems. This research, by providing an innovative approach, paves the way for future studies in the field of multilingual understan‎ding an‎d reasoning.

سامانه پرسش و پاسخ , پرسش و پاسخ بين زباني , پرسش پيچيده چندگامي , مدلهاي زباني بزرگ , درك زبان طبيعي

Question Answering System , Cross-Lingual Question Answering , Complex Multi-Hop Question , Large Language Models , Natural Language Understanding

Eimani

Dr. Bidgoli