TextQ-German
收藏资源简介:
TextQ-German是一个面向德语自然语言生成的体验质量评估数据集,由柏林工业大学和德国人工智能研究中心联合创建。该数据集涵盖自动文本摘要和机器翻译两大任务,包含六个子集:原始语料、大语言模型扩展集及最终验证集,共计收集了人类对多种感知质量维度的评分。数据通过众包实验获取,参与者使用语义差分法对输出文本的多项感知属性进行评价,并据此识别出任务相关的质量维度。该数据集旨在为人本评估提供公开资源,推动预测模型开发,使自然语言生成系统更好地与人类质量感知对齐。
TextQ-German is a quality of experience (QoE) evaluation dataset for German natural language generation, jointly developed by Technische Universität Berlin and the German Research Center for Artificial Intelligence (DFKI). This dataset encompasses two core tasks: automatic text summarization and machine translation, and comprises six subsets, including the original corpus, large language model (LLM) extended set, and final validation set among others. In total, this dataset includes human ratings across multiple perceptual quality dimensions. The data is collected via crowdsourcing experiments, where participants evaluate various perceptual attributes of the generated text using the semantic differential method, and identify task-relevant quality dimensions based on these assessments. This dataset aims to serve as an open resource for human-centered evaluation, advance the development of predictive models, and enable natural language generation systems to better align with human quality perceptions.

- 1Assessing Quality of Experience in Natural Language Generation of German Text柏林工业大学; 德国人工智能研究中心 · 2026年



