发现语言模型对词段概率计算不一致，尤其自回归模型更明显。
CONTESTS: a Framework for Consistency Testing of Span Probabilities in Language Models
- 设计一致性测试框架ConTestS，检验不同生成顺序下的概率稳定性。
- 自回归模型比MLM概率不一致更严重，大MLM更稳定，大自回归模型反而更差。
- 预测熵可反映真实概率，指导解码策略选择。
尽管语言模型的得分常被视为概率，但其作为概率估计器的可靠性主要通过校准研究，忽略了其他方面。特别是，语言模型对词段联合概率的不同赋值方式是否产生相同结果尚不明确。本文提出新框架ConTestS（Span一致性测试），通过统计检验评估不同补全与条件顺序下的得分一致性。我们在发布后的真实与合成数据上进行实验，排除训练影响。结果表明，掩码语言模型（MLMs）和自回归模型均存在预测不一致，其中自回归模型差异更大。更大的MLM产生更一致的预测，而自回归模型则相反。此外，两种模型的预测熵可揭示真实词段概率，有助于选择最优解码策略。这些不一致性及其与熵、模型类型的关系，可为未来改进提供指引。
原文摘要 · Abstract (English)
Although language model scores are often treated as probabilities, their reliability as probability estimators has mainly been studied through calibration, overlooking other aspects. In particular, it is unclear whether language models produce the same value for different ways of assigning joint probabilities to word spans. Our work introduces a novel framework, ConTestS (Consistency Testing over Spans), involving statistical tests to assess score consistency across interchangeable completion and conditioning orders. We conduct experiments on post-release real and synthetic data to eliminate training effects. Our findings reveal that both Masked Language Models (MLMs) and autoregressive models exhibit inconsistent predictions, with autoregressive models showing larger discrepancies. Larger MLMs tend to produce more consistent predictions, while autoregressive models show the opposite trend. Moreover, for both model types, prediction entropies offer insights into the true word span likelihood and therefore can aid in selecting optimal decoding strategies. The inconsistencies revealed by our analysis, as well their connection to prediction entropies and differences between model types, can serve as useful guides for future research on addressing these limitations.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。