Testing the limits of natural language models for predicting human language judgments

TL;DR

This study evaluates how well various neural network language models, including GPT-2, predict human language judgments using controversial sentence pairs, revealing strengths and limitations in model-human alignment.

Contribution

Introduces a novel experimental method using controversial sentence pairs to assess model-human consistency across diverse language models.

Findings

GPT-2 showed the highest alignment with human judgments

Controversial sentence pairs effectively reveal model failures

All models exhibited notable shortcomings in matching human perception

Abstract

Neural network language models can serve as computational hypotheses about how humans process language. We compared the model-human consistency of diverse language models using a novel experimental approach: controversial sentence pairs. For each controversial sentence pair, two language models disagree about which sentence is more likely to occur in natural text. Considering nine language models (including n-gram, recurrent neural networks, and transformer models), we created hundreds of such controversial sentence pairs by either selecting sentences from a corpus or synthetically optimizing sentence pairs to be highly controversial. Human subjects then provided judgments indicating for each pair which of the two sentences is more likely. Controversial sentence pairs proved highly effective at revealing model failures and identifying models that aligned most closely with human judgments. The most human-consistent model tested was GPT-2, although experiments also revealed significant shortcomings of its alignment with human perception.