Chess as a Testbed for Language Model State Tracking
The DOI matches its Crossref record; 19 of the 19 references it cites were found in a registry or an index.
Abstract
Transformer language models have made tremendous strides in natural language understanding tasks. However, the complexity of natural language makes it challenging to ascertain how accurately these models are tracking the world state underlying the text. Motivated by this issue, we consider the task of language modeling for the game of chess. Unlike natural language, chess notations describe a simple, constrained, and deterministic domain. Moreover, we observe that the appropriate choice of chess notation allows for directly probing the world state, without requiring any additional probing-related machinery. We find that: (a) With enough training data, transformer language models can learn to track pieces and predict legal moves with high accuracy when trained solely on move sequences. (b) For small training sets providing access to board state information during training can yield significant improvements. (c) The success of transformer language models is dependent on access to the entire game history i.e. “full attention”. Approximating this full attention results in a significant performance drop. We propose this testbed as a benchmark for future work on the development and analysis of transformer language models.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The clearest evidence that such models are learnable from next-token prediction comes from GPT-style models trained on chess (Toshniwal et al., 2022) and Othello games (Li et al., 2022) which were shown to have explicit representations of the board and game state, with further work showing these representations are linear (Nanda et al., 2023).”
From Language Models Represent Space and Time (2023) · cites this paper as Toshniwal et al., 2022
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 1×
“Recently, transformer-based language models have stretched notions of what is possible with the simple self-supervised objective of language modeling, becoming a fixture in state of the art language technologies (Vaswani et al. 2017; Devlin et al. 2019; Brown et al. 2020).”
BackgroundLanguage Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
Scaling Laws for Neural Language Models
Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., & Amodei, D. (2020) · arXiv · doi:10.48550/arxiv.2001.08361 · Cited by 9,550 · cited 2×
“These results are along the expected lines of larger training sets alleviating the overfitting problem with larger models (Kaplan et al. 2020).”
BackgroundResultA general reinforcement learning algorithm that masters chess, shogi, and Go through self-play
Silver, D., Hubert, T., Schrittwieser, J., Antonoglou, I., Lai, M., Guez, A., Lanctot, M., Sifre, L., Kumaran, D., Graepel, T., et al. (2018) · Science · doi:10.1126/science.aar6404 · Cited by 3,562 · cited 1×
“For example, using only self-play and the rules of chess, AlphaZero achieves superhuman performance starting from random play (Silver et al. 2018).”
MethodLanguage Models as Knowledge Bases?
Petroni, F., Rocktäschel, T., Riedel, S., Lewis, P., Bakhtin, A., Wu, Y., & Miller, A. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1250 · Cited by 1,834
Reformer: The Efficient Transformer
Kitaev, N., Kaiser, Ł., & Levskaya, A. (2020) · arXiv · doi:10.48550/arxiv.2001.04451 · Cited by 1,435 · cited 7×
“Reformer replaces the “full attention” with attention based on locality-sensitive hashing, while Performer approximates the “full attention” with random features.”
BackgroundMethodInfluentialMicikevicius, P., Narang, S., Alben, J., Diamos, G., Elsen, E., Garcia, D., Ginsburg, B., Houston, M., Kuchaiev, O., Venkatesh, G., et al. (2017) · arXiv · doi:10.48550/arxiv.1710.03740 · Cited by 1,179 · cited 1×
“To accelerate training, we use mixed precision training (Micikevicius et al. 2018).”
MethodTransformers are RNNs: Fast Autoregressive Transformers with Linear Attention
Katharopoulos, A., Vyas, A., Pappas, N., & Fleuret, F. (2020) · arXiv · doi:10.48550/arxiv.2006.16236 · Cited by 827 · cited 1×
“…of board state tracking in chess can serve as an excellent testbed for recently proposed transformer variants (Kitaev, Kaiser, and Levskaya 2020; Katharopoulos et al. 2020; Choromanski et al. 2021, inter alia ) that attempt to make use of long histories or contexts, but without incurring a…”
BackgroundRethinking Attention with Performers
Choromanski, K., Likhosherstov, V., Dohan, D., Song, X., Gane, A., Sarlos, T., Hawkins, P., Davis, J., Mohiuddin, A., Kaiser, L., et al. (2020) · arXiv · doi:10.48550/arxiv.2009.14794 · Cited by 695 · cited 5×
“Among the transformer models, we experiment with two “approximate” attention models (i.e., models which approximate the full attention of vanilla transformer models), namely, Reformer (Kitaev, Kaiser, and Levskaya 2020) and Performer (Choromanski et al. 2021).”
BackgroundMethodInfluentialTowards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks
Weston, J., Bordes, A., Chopra, S., Rush, A. M., van Merriënboer, B., Joulin, A., & Mikolov, T. (2015) · arXiv · doi:10.48550/arxiv.1502.05698 · Cited by 670 · cited 1×
“The bAbI framework simulates a world modeled via templates to generate question answering tasks (Weston et al. 2015).”
MethodA Corpus and Cloze Evaluation for Deeper Understanding of Commonsense Stories
Mostafazadeh, N., Chambers, N., He, X., Parikh, D., Batra, D., Vanderwende, L., Kohli, P., & Allen, J. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1098 · Cited by 598
Understanding intermediate layers using linear classifier probes
Alain, G., & Bengio, Y. (2016) · arXiv · doi:10.48550/arxiv.1610.01644 · Cited by 581 · cited 2×
“A typical solution used by prior work is to train a probing model on top of a pretrained model (Ettinger, Elgohary, and Resnik 2016; Alain and Bengio 2017; Adi et al. 2017; Tenney et al. 2019; Hewitt and Liang 2019).”
MethodWhat BERT Is Not: Lessons from a New Suite of Psycholinguistic Diagnostics for Language Models
Ettinger, A. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00298 · Cited by 511 · cited 1×
“…tasks range from testing general text understanding (Paperno et al. 2016) to targeting particular aspects of natural language, such as commonsense/pragmatics (Mostafazadeh et al. 2016; Ettinger 2020), narrative understanding (Mostafazadeh et al. 2017), and factual knowledge (Petroni et al. 2019).”
BackgroundDesigning and Interpreting Probes with Control Tasks
Hewitt, J., & Liang, P. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1275 · Cited by 393 · cited 1×
“A typical so-lution used by prior work is to train a probing model on top of a pretrained model (Ettinger, Elgohary, and Resnik 2016; Alain and Bengio 2017; Adi et al. 2017; Tenney et al. 2019; Hewitt and Liang 2019).”
MethodThe LAMBADA dataset: Word prediction requiring a broad discourse context
Paperno, D., Kruszewski, G., Lazaridou, A., Pham, N. Q., Bernardi, R., Pezzelle, S., Baroni, M., Boleda, G., & Fernandez, R. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p16-1144 · Cited by 87 · cited 1×
“These tasks range from testing general text understanding (Paperno et al. 2016) to targeting particular aspects of natural language, such as commonsense/pragmatics (Mostafazadeh et al. 2016; Ettinger 2020), narrative understanding (Mostafazadeh et al. 2017), and factual knowledge (Petroni et al.…”
BackgroundDeepChess: End-to-End Deep Neural Network for Automatic Learning in Chess
David, O. E., Netanyahu, N. S., & Wolf, L. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-44781-0_11 · Cited by 76
Cited by
Language Models Represent Space and Time
Gurnee, W., & Tegmark, M. (2023) · arXiv · doi:10.48550/arxiv.2310.02207 · Cited by 46
“The clearest evidence that such models are learnable from next-token prediction comes from GPT-style models trained on chess (Toshniwal et al., 2022) and Othello games (Li et al., 2022) which were shown to have explicit representations of the board and game state, with further work showing these…”
BackgroundOpening the AI Black Box: Distilling Machine-Learned Algorithms into Code
Michaud, E. J., Liao, I., Lad, V., Liu, Z., Mudide, A., Loughridge, C., Guo, Z. C., Kheirkhah, T. R., Vukelić, M., & Tegmark, M. (2024) · Entropy · doi:10.3390/e26121046 · Cited by 30
Evaluating the World Model Implicit in a Generative Model
Vafa, K., Chen, J., Rambachan, A., Kleinberg, J., & Mullainathan, S. (2024) · Advances in Neural Information Processing Systems 37 · doi:10.52202/079017-0846 · Cited by 22
What Do Language Models Hear? Probing for Auditory Representations in Language Models
Ngo, J., & Kim, Y. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.297 · Cited by 5
Metadata
- Authors
- Toshniwal, S., Wiseman, S., Livescu, K., & Gimpel, K.
- Venue
- Proceedings of the AAAI Conference on Artificial Intelligence
- Published
- 2022-06-28
- Volume
- 36
- Issue
- 10
- Pages
- 11385-11393
- ISSN
- 2374-3468
- 2159-5399
- Cited by
- 9
- References
- 19
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.