Garden-Path Traversal in GPT-2

In recent years, large-scale transformer decoders such as the GPT-x family of models have become increasingly popular. Studies examining the behavior of these models tend to focus only on the output of the language modeling head and avoid analysis of the internal states of the transformer decoder. In this study, we present a collection of methods to analyze the hidden states of GPT-2 and use the model's navigation of garden path sentences as a case study. To enable this, we compile the largest currently available dataset of garden path sentences. We show that Manhattan distances and cosine similarities provide more reliable insights compared to established surprisal methods that analyze next-token probabilities computed by a language modeling head. Using these methods, we find that negating tokens have minimal impacts on the model's representations for unambiguous forms of sentences with ambiguity solely over what the object of a verb is, but have a more substantial impact of representations for unambiguous sentences Figure 1: Hidden state relations (Top: cosine similarity, whose ambiguity would stem from the voice Middle: Manhattan distance, Bottom: surprisal difference) of a verb. Further, we find that analyzing the between negated and non-negated forms of garden decoder model's hidden states reveals periods path and unambiguous sentences. The ambiguous of ambiguity that might conclude in a garden verb "walked" primes the effect later in the sentence, path effect but happen not to, whereas surprisal while the unambiguous "taken" avoids it. The verb "lit" analyses routinely miss this detail.