研究：Bangla习语揭示跨语言概念子空间的读取与擦除并不可逆
What Travels Across Languages? Reading, Transfer and Erasure of Cross-Lingual Concept Subspaces in Bangla Idioms
这篇论文拿10,359个孟加拉语习语当探针，测试6个编码器里跨语言语义共享的真实程度。结论挺反直觉：英文训练的读出器能迁移到Bangla，但LEACE擦除器却几乎擦不动，读取和擦除是两回事。做多语言模型对齐或概念擦除研究的人建议看看。
一篇arXiv论文（编号2611.11386）以10,359个Bangla（孟加拉语）习语词义为测试材料，在6个冻结编码器、8种输入条件下研究跨语言概念子空间中的读取、分类器迁移与擦除操作，并使用 figurative-class（比喻类）和情感标签进行评估。核心发现是“能读取不等于能擦除”：类别均值方向在后验分析中击败随机子空间、置换标签和协方差匹配零假设等全部12项模型-标签测试；用英文方向训练的分类器在12项发现测试中的8项里超过零样本零假设，但性能仅比多数类基线高出3.0-7.3个百分点。
机制层面，论文指出擦除存在明显不对称：英文拟合的LEACE对Bangla比喻类准确率的削减不足1个百分点，而Bangla自身拟合的LEACE可削减4-11个百分点，INLP方法呈现类似格局。作者用模式过滤分析、协方差加权的输入映射和一个二元高斯重叠界解释了这种分歧，并提出多类排序作为经验诊断手段，该诊断在12个单元格中的11个得到观察。
结论支持一种“估计器敏感”的跨语言概念共享观：概念共享程度取决于所用估计器与迁移映射，因此论文主张将读取、分类器迁移、擦除和重叠四类操作分开评估，而非混为一谈。
What Travels Across Languages? Reading, Transfer and Erasure of Cross-Lingual Concept Subspaces in Bangla Idioms
Cross-lingual concept analysis requires distinguishing the information that a subspace retains from the classifiers and erasers that transfer through it. We study these operations in six frozen encoders using 10,359 Bangla (Bengali) idiom senses, eight input conditions, and figurative-class and sentiment labels. Class-mean directions provide the most consistent evidence of shared coarse meaning: in post hoc analyses, they outperform random-subspace, permuted-label and covariance-matched nulls in all 12 model-label tests under balanced accuracy. A prospective internal re-split supports idiom-alone sharing in 10 of 12 tests, spanning five encoders. The original confirmatory logistic-regression family rejects both planned nulls in 4 of 12 tests, where English directions gain only a few accuracy points above the majority-class floor (3.0-7.3). Classifier transfer depends on the input map and regularization: English-trained class-mean readouts beat both zero-shot nulls in 8 of 12 discovery tests, and stronger source-selected penalties improve transfer through logistic and LEACE directions. These readouts use Bangla-validation-selected layers and unlabeled target preprocessing. Input controls show greater coarse-class sharing for explicit glosses than for idiom counterparts. Reading, however, does not imply effective erasure: English-fitted LEACE removes less than one percentage point of Bangla figurative-class accuracy, compared with 4-11 points for Bangla-fitted LEACE; INLP shows a similar asymmetry. Pattern-filter analysis, covariance-weighted input maps and a binary Gaussian overlap bound explain how these operations can diverge. The analogous multiclass ordering is an empirical diagnostic, observed in 11 of 12 cells. The results support an estimator-sensitive account of cross-lingual concept sharing and motivate evaluating reading, classifier transfer, erasure and overlap separately.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include LEACE, INLP, 跨语言对齐, Bangla.
Source attribution: curated from arXiv cs.LG (@Syed Rifat Raiyan, Yousuf Ibne Kamal Niloy).