Survey data underlying the MSc thesis: "Generative AI: Investigating Consistency and Neutrality in Multilingual Outputs"

This dataset contains responses from an online survey designed to evaluate how consistently and neutrally ChatGPT’s English and Arabic answers align across ten prompts (seven politically sensitive, three non-sensitive). Each row captures one participant’s ratings of sentiment and factual consistency between the two language outputs, neutrality scores for each response and the prompt itself, and optional comments. The data were collected via Qualtrics from English- and Arabic-fluent respondents who compared side-by-side model answers, providing quantitative Likert-scale ratings to assess multilingual consistency and neutrality of Generative AI output in a human evaluation study.

本数据集包含一项在线调研的反馈数据，该调研旨在评估ChatGPT在10个提示词（其中7个涉及政治敏感议题、3个为非敏感议题）下生成的英文与阿拉伯语回答的一致性与中立对齐程度。每一行数据对应一名参与者的评分内容：涵盖两种语言输出的情感倾向与事实一致性评分、两份模型回答及提示词本身的中立性得分，以及可选的评论内容。本数据集通过Qualtrics平台收集，招募了精通英语与阿拉伯语的受访者，让其对比并排展示的模型生成回答，并通过李克特量表（Likert-scale）给出量化评分，以此在本次人类评估研究中评估生成式AI（Generative AI）输出的多语言一致性与中立性。