Mechanistic Interpretability of GPT-2: Lexical and Contextual Layers in Sentiment Analysis
This provides causal evidence for mechanistic interpretability in large language models, addressing a foundational problem for AI researchers, though it is incremental in refining existing hypotheses.
The study tackled the problem of understanding how GPT-2 processes sentiment information across its layers, finding that early layers detect lexical sentiment independently of context, while contextual integration occurs primarily in late layers through a unified mechanism, falsifying hypothesized mid-layer specialization.
We present a mechanistic interpretability study of GPT-2 that causally examines how sentiment information is processed across its transformer layers. Using systematic activation patching across all 12 layers, we test the hypothesized two-stage sentiment architecture comprising early lexical detection and mid-layer contextual integration. Our experiments confirm that early layers (0-3) act as lexical sentiment detectors, encoding stable, position specific polarity signals that are largely independent of context. However, all three contextual integration hypotheses: Middle Layer Concentration, Phenomenon Specificity, and Distributed Processing are falsified. Instead of mid-layer specialization, we find that contextual phenomena such as negation, sarcasm, domain shifts etc. are integrated primarily in late layers (8-11) through a unified, non-modular mechanism. These experimental findings provide causal evidence that GPT-2's sentiment computation differs from the predicted hierarchical pattern, highlighting the need for further empirical characterization of contextual integration in large language models.