AI-mediated Chinese city publicity translation dataset: Human-coded evaluation of web machine translation and large language model agents

This dataset supports the manuscript Evaluating Machine Translation and Large Language Model Agents for Chinese City Publicity Translation. It contains the anonymised research materials used to evaluate English translations of Chinese city publicity texts under four translation conditions: DeepL web machine translation, Google Translate web machine translation, a GPT-5.5 large language model translation agent, and a DeepSeek V4 Pro large language model translation agent. The corpus contains 90 Chinese city publicity source segments from Qingdao, Xi'an and Hangzhou. Each source segment was translated under four conditions, yielding 360 translation records. Two trained human coders evaluated the translations using five quality dimensions: accuracy, cultural adequacy, pragmatic appropriateness, audience accessibility and city image representation. The dataset also includes a primary issue-code taxonomy, randomisation mapping, formal coding workbooks, statistical analysis outputs, auxiliary LLM pre-coding outputs, CSV exports of key sheets, a Chinese codebook PDF and reproducibility scripts. The dataset is intended to support reproducibility and secondary analysis in specialised translation, machine translation evaluation, AI translation technologies, human-coded translation quality assessment and public-sector intercultural communication. The package does not contain personal data, sensitive private records or human-subject responses. Human coders acted as trained research raters of translation outputs rather than as research participants, and coder identities are not included.