Competence-Based Analysis of Language Models

Despite the recent success of large, pretrained neural LLMs on a variety of prompting tasks, these models can be alarmingly brittle to small changes in inputs or application contexts. To better understand such behavior and motivate the design of more robust LLMs, we provide a causal formulation of linguistic competence in the context of LLMs and propose a general framework to study and measure LLM competence. Our framework, CALM (Competence-based Analysis of Language Models), establishes the first quantitative measure of LLM competence, which we study by damaging models' internal representations of various linguistic properties in the course of performing various tasks using causal probing and evaluating models' alignment under these interventions with a given causal model. We also develop a novel approach for performing causal probing interventions using gradient-based adversarial attacks, which can target a broader range of properties and representations than existing techniques. We carry out a case study of CALM using these interventions to analyze BERT and RoBERTa's competence across a variety of lexical inference tasks, showing that the CALM framework and competence metric can be valuable tools for explaining and predicting their behavior across these tasks.