用定义学习概念嵌入，让冻结大模型可控地调整行为。
Concept Tokens: Learning Behavioral Embeddings Through Concept Definitions
- 新增特殊标记符，仅训练其嵌入向量，保持原模型不变。
- 否定该标记可减少幻觉，但增加不回答；肯定则加剧幻觉并降低准确率。
- 比直接输入定义更有效保留其他指令的遵守性，适合教学与控制场景。
我们提出概念标记（Concept Tokens），一种轻量级方法：在预训练大语言模型中添加一个特殊标记，仅从多个自然语言定义中学习该目标概念的嵌入向量，并将原文中概念出现位置替换为该标记。模型参数保持冻结，通过标准语言建模范式优化嵌入。我们在三个场景中评估该方法：首先在HotpotQA闭卷问答中研究幻觉现象，发现方向性效应——否定概念标记主要通过增加弃答来减少幻觉，而肯定它则提升幻觉并降低精度；其次，在第二语言教学中引入重构反馈策略，同样观察到相同方向效应；相比直接以上下文方式提供完整定义语料，概念标记更能保持对其他指令（如追问）的遵守性；最后，通过艾菲尔铁塔与虚构的‘澳大利亚塔’进行定性分析，揭示所学嵌入捕捉的信息及其局限。总体而言，概念标记提供了一种由定义学习的紧凑控制信号，可在冻结大模型中引导行为。
原文摘要 · Abstract (English)
We propose Concept Tokens, a lightweight method that adds a new special token to a pretrained LLM and learns only its embedding from multiple natural language definitions of a target concept, where occurrences of the concept are replaced by the new token. The LLM is kept frozen and the embedding is optimized with the standard language-modeling objective. We evaluate Concept Tokens in three settings. First, we study hallucinations in closed-book question answering on HotpotQA and find a directional effect: negating the hallucination token reduces hallucinated answers mainly by increasing abstentions, whereas asserting it increases hallucinations and lowers precision. Second, we induce recasting, a pedagogical feedback strategy for second language teaching, and observe the same directional effect. Moreover, compared to providing the full definitional corpus in-context, concept tokens better preserve compliance with other instructions (e.g., asking follow-up questions). Finally, we include a qualitative study with the Eiffel Tower and a fictional "Austral Tower" to illustrate what information the learned embeddings capture and where their limitations emerge. Overall, Concept Tokens provide a compact control signal learned from definitions that can steer behavior in frozen LLMs.
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。