揭示大模型如何内部编码词义和词形变化，发现词形信息贯穿深层，词义信息随深度减弱。
Model Internal Sleuthing: Finding Lexical Identity and Inflectional Features in Modern Language Models
- 通过逐层探测25个模型，分析词义与词形特征的表示规律。
- 词形特征在各层线性可解，词义信息越往深层越弱，尤其在压缩层失效。
- 适合关注模型内部机制、语言表征演化的研究者阅读。
大型基于Transformer的语言模型主导现代自然语言处理，但对其语言信息编码的理解仍主要基于早期模型（如BERT和GPT-2）。本文系统探测了从BERT Base到Qwen2.5-7B共25个模型，聚焦六种不同语言中的词义身份与词形特征。结果表明：词形特征在整个模型层级中均可线性解码，而词义身份在早期显著，随深度增加逐渐减弱。进一步分析显示，具有强烈中层维度压缩的模型，尽管探测准确率高，但在该层的可控性（steering）效果下降。预训练分析表明，词形结构在早期即稳定，而词义表示持续演化。综合来看，变压器模型在各层维持词形特征，同时以牺牲词义信息换取更紧凑、更具预测性的表示。代码已开源：https://github.com/ml5885/model_internal_sleuthing。
原文摘要 · Abstract (English)
Large transformer-based language models dominate modern NLP, yet our understanding of how they encode linguistic information relies primarily on studies of early models like BERT and GPT-2. We systematically probe 25 models from BERT Base to Qwen2.5-7B focusing on two linguistic properties: lexical identity and inflectional features across 6 diverse languages. We find a consistent pattern: inflectional features are linearly decodable throughout the model, while lexical identity is prominent early but increasingly weakens with depth. Further analysis of the representation geometry reveals that models with aggressive mid-layer dimensionality compression show reduced steering effectiveness in those layers, despite probe accuracy remaining high. Pretraining analysis shows that inflectional structure stabilizes early while lexical identity representations continue evolving. Taken together, our findings suggest that transformers maintain inflectional features across layers, while trading off lexical identity for compact, predictive representations. Our code is available at https://github.com/ml5885/model_internal_sleuthing
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。