Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions
This work provides insights into the internal mechanisms of models for MCQA, which is incremental as it builds on existing methods to analyze model behavior without introducing new performance gains.
The study investigated how transformer language models perform multiple-choice question answering (MCQA) by using vocabulary projection and activation patching to identify key hidden states, finding that middle layers and specific attention heads causally encode answer predictions and that logit differences between answer tokens increase during training.
Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slightly (such as by shuffling answer choice order). In this work we ask: how do successful models perform formatted MCQA? We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information for predicting the correct answer. We find that the prediction of a specific answer symbol is causally attributed to a few middle layers, and specifically their multi-head self-attention mechanisms. We show that subsequent layers increase the probability of the predicted answer symbol in vocabulary space, and that this probability increase is associated with a sparse set of attention heads with unique roles. We additionally uncover differences in how different models adjust to alternative symbols. Finally, we demonstrate that a synthetic task can disentangle sources of model error to pinpoint when a model has learned formatted MCQA, and show that logit differences between answer choice tokens continue to grow over the course of training.