BEAMetrics

BEAMetrics数据集由索邦大学和DeepMind的研究人员创建，旨在为自然语言生成（NLG）系统的评估提供一个统一的平台。该数据集包含11个不同的子数据集，覆盖了从机器翻译到文本摘要等多个NLG任务，支持多种语言，并针对流畅性、连贯性、信息性等多个评价维度进行评估。BEAMetrics的创建旨在解决现有自动评价指标在不同任务和语言上的局限性，通过提供一个多任务、多语言和多维度的评估框架，促进NLG领域评价指标的发展和优化。

The BEAMetrics dataset was created by researchers from Sorbonne University and DeepMind, aiming to provide a unified platform for the evaluation of natural language generation (NLG) systems. This dataset includes 11 distinct sub-datasets, covering multiple NLG tasks ranging from machine translation to text summarization, supporting multiple languages, and enabling evaluation across multiple dimensions such as fluency, coherence and informativeness. The creation of BEAMetrics is intended to address the limitations of existing automatic evaluation metrics across different tasks and languages. By providing a multi-task, multi-lingual and multi-dimensional evaluation framework, it promotes the development and optimization of evaluation metrics in the NLG field.