Measuring Inductive Biases of In-Context Learning with Underspecified Demonstrations
This work addresses the problem of understanding and controlling generalization in in-context learning for researchers and practitioners, though it is incremental as it builds on existing studies of model biases.
The study investigated the inductive biases of in-context learning in large language models by analyzing which features they prioritize when given underspecified demonstrations, finding that models strongly favor sentiment over shallow lexical features like punctuation. It also showed that interventions like natural language instructions can influence feature preferences but often fail to overcome strong prior biases.
In-context learning (ICL) is an important paradigm for adapting large language models (LLMs) to new tasks, but the generalization behavior of ICL remains poorly understood. We investigate the inductive biases of ICL from the perspective of feature bias: which feature ICL is more likely to use given a set of underspecified demonstrations in which two features are equally predictive of the labels. First, we characterize the feature biases of GPT-3 models by constructing underspecified demonstrations from a range of NLP datasets and feature combinations. We find that LLMs exhibit clear feature biases - for example, demonstrating a strong bias to predict labels according to sentiment rather than shallow lexical features, like punctuation. Second, we evaluate the effect of different interventions that are designed to impose an inductive bias in favor of a particular feature, such as adding a natural language instruction or using semantically relevant label words. We find that, while many interventions can influence the learner to prefer a particular feature, it can be difficult to overcome strong prior biases. Overall, our results provide a broader picture of the types of features that ICL may be more likely to exploit and how to impose inductive biases that are better aligned with the intended task.