The EMNLP 2024 Conference

The EMNLP 2024 conference, hosted by the Association for Computational Linguistics (ACL), took place from 12 to 16 Nov 2024, in Miami, Florida, USA. This flagship event in the field of Natural Language Processing (NLP) focused on driving innovation within the discipline, with a special emphasis on exploring large language models in areas such as training and inference efficiency, interpretability analysis, and applications downstream. The conference attracted over 4,000 researchers from around the globe.

Mr. Zexuan Qiu, a Ph.D. student from AIFT, presented his research paper titled “CLongEval: A Chinese Benchmark for Evaluating Long-Context Large Language Models.” In the real world, many documents, such as financial reports and legal texts, are extensive, making the development of robust long-context capabilities a central focus of recent research. His paper introduces ClongEval, a benchmark designed for the Chinese long-text domain. This benchmark is characterized by three key features: (1) a large data volume, including 7 distinct tasks and 7,267 examples; (2) broad applicability, suitable for models with context window sizes ranging from 1K to 100K; (3) high quality, with more than 2,000 manually annotated question-answer pairs in addition to automatically constructed labels. Utilizing this benchmark, the paper conducts comprehensive evaluations of various language models and provides in-depth analysis, highlighting the critical challenges faced in long-context scenarios.