Interactive Analysis of LLMs using Meaningful Counterfactuals

Counterfactual examples are useful for exploring the decision boundaries of machine learning models and determining feature attributions. How can we apply counterfactual-based methods to analyze and explain LLMs? We identify the following key challenges. First, the generated textual counterfactuals should be meaningful and readable to users and thus can be mentally compared to draw conclusions. Second, to make the solution scalable to long-form text, users should be equipped with tools to create batches of counterfactuals from perturbations at various granularity levels and interactively analyze the results. In this paper, we tackle the above challenges and contribute 1) a novel algorithm for generating batches of complete and meaningful textual counterfactuals by removing and replacing text segments in different granularities, and 2) LLM Analyzer, an interactive visualization tool to help users understand an LLM's behaviors by interactively inspecting and aggregating meaningful counterfactuals. We evaluate the proposed algorithm by the grammatical correctness of its generated counterfactuals using 1,000 samples from medical, legal, finance, education, and news datasets. In our experiments, 97.2% of the counterfactuals are grammatically correct. Through a use case, user studies, and feedback from experts, we demonstrate the usefulness and usability of the proposed interactive visualization tool.

A Theoretical Result on the Inductive Bias of RNN Language Models

Recent work by Hewitt et al. (2020) provides a possible interpretation of the empirical success of recurrent neural networks (RNNs) as language models (LMs). It shows that RNNs can efficiently represent bounded hierarchical structures that are prevalent in human language. This suggests that RNNs' success might be linked to their ability to model hierarchy. However, a closer inspection of Hewitt et al.'s (2020) construction shows that it is not limited to hierarchical LMs, posing the question of what \emph{other classes} of LMs can be efficiently represented by RNNs. To this end, we generalize their construction to show that RNNs can efficiently represent a larger class of LMs: Those that can be represented by a pushdown automaton with a bounded stack and a generalized stack update function. This is analogous to an automaton that keeps a memory of a fixed number of symbols and updates the memory with a simple update mechanism. Altogether, the efficiency in representing a diverse class of non-hierarchical LMs posits a lack of concrete cognitive and human-language-centered inductive biases in RNNs.