Benchmarking Mental State Representations in Language Models
Matteo Bortoletto,
Constantin Ruhdorfer,
Lei Shi,
Andreas Bulling
Proc. ICML 2024 Workshop on Mechanistic Interpretability,
pp. 1–21,
2024.
Abstract
Links
BibTeX
Project
While numerous works have assessed the generative performance of language models (LMs) on tasks requiring Theory of Mind reasoning, research into the models’ internal representation of mental states remains limited. Recent work has used probing to demonstrate that LMs can represent beliefs of themselves and others. However, these claims are accompanied by limited evaluation, making it difficult to assess how mental state representations are affected by model design and training choices. We report an extensive benchmark with various LM types with different model sizes, fine-tuning approaches, and prompt designs to study the robustness of mental state representations and memorisation issues within the probes. Our results show that the quality of models’ internal representations of the beliefs of others increases with model size and, more crucially, with fine-tuning. We are the first to study how prompt variations impact probing performance on theory of mind tasks. We demonstrate that models’ representations are sensitive to prompt variations, even when such variations should be beneficial. Finally, we complement previous activation editing experiments on Theory of Mind tasks and show that it is possible to improve models’ reasoning performance by steering their activations without the need to train any probe.
@inproceedings{bortoletto24_icmlw,
author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Shi, Lei and Bulling, Andreas},
title = {Benchmarking Mental State Representations in Language Models},
booktitle = {Proc. ICML 2024 Workshop on Mechanistic Interpretability},
year = {2024},
pages = {1--21},
doi = {},
url = {https://openreview.net/forum?id=yEwEVoH9Be}
}
