BEAMetrics

BEAMetrics (Benchmark to Evaluate Automatic Metrics) 是用于研究用于评估生成语言的新指标更容易评估的资源。 BEAMetrics 用户可以在各种任务、质量维度（流畅性、连贯性和信息量等）和语言中快速将现有和新指标与人类判断进行比较。BEAMetrics (Benchmark to Evaluate Automatic Metrics) is a specialized resource that facilitates research on automatic metrics for evaluating generated language, streamlining the evaluation process for both existing and novel metrics. Users of BEAMetrics can rapidly compare existing and novel metrics against human judgments across diverse tasks, quality dimensions (e.g., fluency, coherence, informativeness, etc.) and languages.

搜集汇总

背景与挑战

背景概述

BEAMetrics是一个用于评估生成语言自动指标的基准资源，支持在多种任务、质量维度和语言中快速比较新指标与人类判断。该数据集由DeepMind和索邦大学于2021年发布，旨在促进语言生成评估研究。