Material suplementario de «Itinerarios semánticos multilingües en ESMAS-ES+: del diseño digital a la justicia semántica. Estudio de caso sobre el léxico cultural del paisaje marítimo atlántico gallego»

This dataset contains the supplementary material associated with the communication “Itinerarios semánticos multilingües en ESMAS-ES+: del diseño digital a la justicia semántica. Estudio de caso sobre el léxico cultural del paisaje marítimo atlántico gallego”, presented at the XI Congreso Internacional de Lexicografía Hispánica: Tres siglos del Diccionario de Autoridades, Universidad de Cádiz, 3–5 June 2026. The dataset documents a pilot analysis of the Galician lexical unit batea within a Galician-Italian lexicographic workflow. It includes the exact prompt used for the consultation, the JSON outputs generated by five large language models (ChatGPT, Gemini, Claude, DeepSeek and Qwen), a synthetic results table in CSV and Markdown formats, a qualitative analysis note, a codebook and a file manifest. The consultation was designed as a zero-shot, single-turn, domain-framed prompting strategy with structured JSON output. The purpose was not to compare or rank model performance, but to collect convergences and divergences useful for human lexicographic post-editing. The model outputs are therefore treated as exploratory material, not as automatically validated lexicographic data. The dataset supports transparency, traceability and reuse of the methodological procedure. It is intended to document how generative AI outputs can assist, but not replace, expert lexicographic decision-making in the representation of culturally situated lexical units.本数据集包含与提交至第11届国际西班牙语词典学大会：《权威词典》三百年（加的斯大学，2026年6月3日至5日）的学术报告《ESMAS-ES+中的多语言语义路径：从数字设计到语义公正——加利西亚大西洋海岸景观文化词汇的案例研究》（原标题：*Itinerarios semánticos multilingües en ESMAS-ES+: del diseño digital a la justicia semántica. Estudio de caso sobre el léxico cultural del paisaje marítimo atlántico gallego*）相关的补充材料。 本数据集记录了加利西亚语-意大利语词典学工作流中针对加利西亚语词汇单元batea的试点分析，涵盖本次咨询所用的精准提示词、由ChatGPT、Gemini、Claude、DeepSeek及Qwen这5款大语言模型生成的JSON格式输出结果、CSV与Markdown格式的综合结果统计表、质性分析说明、编码手册以及文件清单。 本次咨询采用零样本（Zero-shot）单轮领域限定提示策略，并要求输出结构化JSON格式内容。其设计初衷并非对比或排序模型性能，而是收集有助于人工词典学后期编辑的词汇趋同与差异信息。因此，模型输出仅作为探索性材料，而非经过自动验证的词典学数据。 本数据集旨在提升方法流程的透明度、可追溯性与可复用性，用于阐释生成式AI（Generative AI）输出如何辅助而非替代专家在文化语境词汇单元表征中的词典学决策工作。