To correctly use in-context information, language models (LMs) must bind
entities to their attributes. For example, given a context describing a "green
square" and a "blue circle", LMs must bind the shapes to their respective
colors. We analyze LM representations and identify the binding ID mechanism: a
general mechanism for solving the binding problem, which we observe in every
sufficiently large model from the Pythia and LLaMA families. Using causal
interventions, we show that LMs' internal activations represent binding
information by attaching binding ID vectors to corresponding entities and
attributes. We further show that binding ID vectors form a continuous subspace,
in which distances between binding ID vectors reflect their discernability.
Overall, our results uncover interpretable strategies in LMs for representing
symbolic knowledge in-context, providing a step towards understanding general
in-context reasoning in large-scale LMs.

通过分析语言模型表示法，我们识别出绑定 ID 机制：一种在足够大的 Pythia 和 LLaMA 系列模型中观察到的解决绑定问题的通用机制。我们使用因果干预证明语言模型的内部激活通过将绑定 ID 向量附加到相应的实体和属性上来表示绑定信息。我们进一步展示，绑定 ID 向量形成一个连续的子空间，其中绑定 ID 向量之间的距离反映了它们的可辨性。总体而言，我们的结果揭示了语言模型在上下文中表示符号知识的可解释策略，为理解大规模语言模型中通用的上下文推理迈出了一步。