Weichuan/TC-DN-MT-Geneevals

该文件夹发布了TC-DN-MT研究的评估统计和可视化就绪结果（CSV格式），涵盖了多种非确定性机器翻译设置下的词汇和语义指标。配套资源包括生成输出、代码库/评估管道和相关论文。内容包括词汇和语义指标的delta百分比统计，以及可直接用于绘图的视觉分析就绪表格。数据集包含多个实验组，如固定温度+不同样本大小、温度消融和多语言方向。数据格式为CSV，每行对应一个模型，列包含各种聚合统计信息。数据集旨在用于重现ND-MT评估分析的结果表和图表，研究在采样大小和温度变化下的指标行为，以及在随机解码下交叉检查词汇与语义指标的一致性。

This folder releases evaluation statistic and visualization-ready results (CSV) for the TC-DN-MT study, covering both lexical and semantic metrics under multiple non-deterministic MT settings. The companion resources include generation outputs, codebase/evaluation pipeline, and the related paper. The content includes delta-percentage statistics for lexical and semantic metrics, and visual-analysis-ready tables prepared for direct plotting. The dataset contains multiple experiment groups, such as fixed temperature + different sample sizes, temperature ablations, and multilingual directions. The data format is CSV, with each row corresponding to one model and columns containing various aggregated statistics. The dataset is intended for reproducing result tables and plots for ND-MT evaluation analysis, studying metric behavior under sampling-size and temperature changes, and cross-checking lexical vs semantic metric consistency under stochastic decoding.