Can Language Models Handle Recursively Nested Grammatical Structures? A Case Study on Comparing Models and Humans
Open accessThe DOI matches its Crossref record; 103 of the 103 references it cites were found in a registry or an index.
Abstract
How should we compare the capabilities of language models (LMs) and humans? In this article, I draw inspiration from comparative psychology to highlight challenges in these comparisons. I focus on a case study: processing of recursively nested grammatical structures. Prior work suggests that LMs cannot process these structures as reliably as humans can. However, the humans were provided with instructions and substantial training, while the LMs were evaluated zero-shot. I therefore match the evaluation more closely. Providing large LMs with a simple prompt—with substantially less content than the human training—allows the LMs to consistently outperform the human results, even in more deeply nested conditions than were tested with humans. Furthermore, the effects of prompting are robust to the particular structures and vocabulary used in the prompt. Finally, reanalyzing the existing human data suggests that the humans may not perform above chance at the difficult structures initially. Thus, large LMs may indeed process recursively nested grammatical structures as reliably as humans, when evaluated comparably. This case study highlights how discrepancies in the evaluation methods can confound comparisons of language models and humans. I conclude by reflecting on the broader challenge of comparing human and model capabilities, and highlight an important difference between evaluating cognitive models and foundation models.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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
“Without a fair comparison (59–63) that evaluates humans in the same settings, we cannot make strong claims about whether the properties listed in Table 1 make LLMs qualitatively unlike humans.”
From Embers of autoregression show how large language models are shaped by the problem they are trained to solve (2024) · cites this paper as [62]
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 9×
“3 The models are decoder-only Transformers (Vaswani et al., 2017), using SentencePiece (Kudo and Richardson, 2018) with 32000 tokens; they are trained to predict the next token in context on a large text corpus.”
InfluentialExploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
Language 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 · cited 10×
“This model is 10 × smaller than Chinchilla 70B, and 25 × smaller than GPT-3.”
InfluentialNeural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 1×
“However, the most broadly successful neuralnetwork language-processing systems have not incorporated explicit recursion; instead they have used recurrence, and more recently, attention (Bahdanau et al., 2014; Vaswani et al., 2017).”
Kaugeranna, R., Kaugeranna, E., & Aion, (. S. 4. 6. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.giscience.2023.43 · Cited by 13,690
The weirdest people in the world?
Henrich, J., Heine, S. J., & Norenzayan, A. (2010) · Behavioral and Brain Sciences · doi:10.1017/s0140525x0999152x · Cited by 12,641
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929 · cited 1×
“…particularly effective at increasing base model performance (Webson and Pavlick 2021) and should probably be replaced with examples if possible—though models trained or tuned to follow instructions (e.g., Raffel et al. 2020; Ouyang et al. 2022; Chung et al. 2024) would perhaps show larger bene-fits.”
Chomsky, N. (2014) · The MIT Press eBooks · doi:10.7551/mitpress/9780262527347.001.0001 · Cited by 8,244
Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank
Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C. D., Ng, A., & Potts, C. (2013) · Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d13-1170 · Cited by 6,878
Chomsky, N. (2001) · Ken Hale · doi:10.7551/mitpress/4056.003.0004 · Cited by 5,105 · cited 4×
“Merge is generally described as a classical symbolic operation over structures—one that is “indispensable” (Chomsky 1999).”
InfluentialChain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022) · arXiv · doi:10.48550/arxiv.2201.11903 · Cited by 4,263 · cited 4×
“…attention-based (Transformer) models are trained simply to predict the next token on large corpora of human-generated text, they can exhibit emergent capabilities (e.g., Wei et al. 2022a), such as answering college questions in subjects like medicine with some accuracy (e.g., Hoffmann et al. 2022).”
InfluentialTraining language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208 · cited 2×
“…particularly effective at increasing base model performance (Webson and Pavlick 2021) and should probably be replaced with examples if possible—though models trained or tuned to follow instructions (e.g., Raffel et al. 2020; Ouyang et al. 2022; Chung et al. 2024) would perhaps show larger bene-fits.”
Training Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117 · cited 2×
“…particularly effective at increasing base model performance (Webson and Pavlick 2021) and should probably be replaced with examples if possible—though models trained or tuned to follow instructions (e.g., Raffel et al. 2020; Ouyang et al. 2022; Chung et al. 2024) would perhaps show larger bene-fits.”
Kudo, T., & Richardson, J. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations · doi:10.18653/v1/d18-2012 · Cited by 2,869 · cited 1×
“3 The models are decoder-only Transformers (Vaswani et al. 2017), using SentencePiece (Kudo and Richardson 2018) with 32,000 tokens; they are trained to predict the next token in context on a large text corpus.”
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676 · cited 4×
“…Models vs. Foundation Models: I believe that a key aspect of model evaluation has been under-emphasized: the difference between cognitive models designed to explore a capability in a narrow, restricted domain, and foundation models (Bommasani et al. 2021) that are trained to perform many tasks.”
InfluentialThe lexical nature of syntactic ambiguity resolution
MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994) · Psychological Review · doi:10.1037/0033-295x.101.4.676 · Cited by 2,259 · cited 2×
“…human language learning, or whether domain-general processes could explain the acquisition of syntactic structures (Elman, Bates, and Johnson 1996; MacDonald, Pearlmutter, and Seidenberg 1994; Bates and Goodman 2013)— particularly recursive ones (Christiansen and Chater 1999; Christiansen and…”
Large language models are zero-shot reasoners
Kojima, T., Gu, S. (., Reid, M., Matsuo, Y., & Iwasawa, Y. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1613 · Cited by 2,013 · cited 2×
“Evaluation strategies that allow models to produce a reasoning trace before an answer will likely improve performance (Nye et al., 2021; Wei et al., 2022b; Kojima et al., 2022; Zhou et al., 2022; Suzgun et al., 2022), and may allow more fair comparisons to humans given time to think.”
Rethinking Innateness: A Connectionist Perspective on Development
Harris, M. J., Karmiloff‐Smith, A., Parisi, D., & Plunkett, K. (1996) · Medical Entomology and Zoology · Cited by 1,990
Learning and development in neural networks: the importance of starting small
Elman, J. L. (1993) · Cognition · doi:10.1016/0010-0277(93)90058-4 · Cited by 1,988 · cited 1×
“Early work showed that simple recurrent networks could learn simple grammars (Elman 1991, 1993).”
Psychology of Reasoning: Structure and Content
Wason, P. C., & Johnson-Laird, P. N. (1972) · Medical Entomology and Zoology · Cited by 1,784
Rethinking Innateness: A Connectionist Perspective on Development
Elman, J., Bates, E., Johnson, M. H., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996) · doi:10.7551/mitpress/5929.001.0001 · Cited by 1,511
Scaling Instruction-Finetuned Language Models
Chung, H. W., Hou, L., Longpre, S., Zoph, B., Tay, Y., Fedus, W., Li, Y., Wang, X., Dehghani, M., Brahma, S., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.11416 · Cited by 1,441
Distributed representations, simple recurrent networks, and grammatical structure
Elman, J. L. (1991) · Machine Learning · doi:10.1007/bf00114844 · Cited by 1,354
Emergent Abilities of Large Language Models
Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., et al. (2022) · arXiv · doi:10.48550/arxiv.2206.07682 · Cited by 1,264 · cited 3×
“…attention-based (Transformer) models are trained simply to predict the next token on large corpora of human-generated text, they can exhibit emergent capabilities (e.g., Wei et al. 2022a), such as answering college questions in subjects like medicine with some accuracy (e.g., Hoffmann et al. 2022).”
Structural priming: A critical review.
Pickering, M. J., & Ferreira, V. S. (2008) · Psychological Bulletin · doi:10.1037/0033-2909.134.3.427 · Cited by 1,047 · cited 3×
“Humans exhibit structural priming (e.g. Pickering and Branigan, 1999; Pickering and Ferreira, 2008)— a tendency to reproduce syntactic structures that they have recently encountered.”
Bias in Human Reasoning: Causes and Consequences
Evans, J. S. B. T. (1990) · Medical Entomology and Zoology · Cited by 1,033
Chang, F., Dell, G. S., & Bock, K. (2006) · Psychological Review · doi:10.1037/0033-295x.113.2.234 · Cited by 984 · cited 1×
“Connectionist models of these phenomena offer the potential to explain human errors, developmental processes, and interactions between syntax and semantics (e.g., Chang, Dell, and Bock 2006).”
Training Compute-Optimal Large Language Models
Hoffmann, J., Borgeaud, S., Mensch, A., Buchatskaya, E., Cai, T., Rutherford, E., Casas, D. D. L., Hendricks, L. A., Welbl, J., Clark, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.15556 · Cited by 929 · cited 22×
“Even prompts with substantially less content than the human training allow state-of-the-art LMs (Hoffmann et al. 2022) to perform far better than human subjects, even in the hardest conditions.”
InfluentialYarkoni, T. (2022) · Behavioral and Brain Sciences · doi:10.1017/s0140525x20001685 · Cited by 701
Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity
Lu, Y., Bartolo, M., Moore, A., Riedel, S., & Stenetorp, P. (2021) · arXiv · doi:10.48550/arxiv.2104.08786 · Cited by 615 · cited 1×
“Fewshot prompts are more effective when examples are selected or ordered by validation performance (Perez et al., 2021; Lu et al., 2021)—this is somewhat analogous to how experimenters might choose illustrative examples to introduce an experimental paradigm to participants.”
Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity
Lu, Y., Bartolo, M., Moore, A., Riedel, S., & Stenetorp, P. (2022) · Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2022.acl-long.556 · Cited by 610 · cited 1×
“Fewshot prompts are more effective when examples are selected or ordered by validation performance (Perez et al., 2021; Lu et al., 2021)—this is somewhat analogous to how experimenters might choose illustrative examples to introduce an experimental paradigm to participants.”
Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
Srivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., Brown, A. R., Santoro, A., Gupta, A., Garriga-Alonso, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2206.04615 · Cited by 569
Self-citationUsing cognitive psychology to understand GPT-3
Binz, M., & Schulz, E. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2218523120 · Cited by 559 · cited 2×
“For example, a recent survey finding successes and failures of language models in cognitive tasks (Binz and Schulz, 2022) considered model evaluation without context, while comparing to prior human studies in extended experimental contexts.”
The neural architecture of language: Integrative modeling converges on predictive processing
Schrimpf, M., Blank, I. A., Tuckute, G., Kauf, C., Hosseini, E. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2021) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2105646118 · Cited by 517
Recurrent Neural Network Grammars
Dyer, C., Kuncoro, A., Ballesteros, M., & Smith, N. A. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1024 · Cited by 517 · cited 2×
“Other researchers have attempted to build neural networks with explicit recursive structure for language processing (e.g. Socher et al., 2013; Dyer et al., 2016).”
Toward a connectionist model of recursion in human linguistic performance
Christiansen, M. (1999) · Cognitive Science · doi:10.1016/s0364-0213(99)00003-8 · Cited by 406 · cited 3×
“Elman (1991) showed that simple recurrent networks can learn recursive compositional syntactic structures in a simplified language, which initiated a line of research (e.g., Christiansen and Chater 1999; Christiansen and MacDonald 2009).”
Rapid Expectation Adaptation during Syntactic Comprehension
Fine, A. B., Jaeger, T. F., Farmer, T. A., & Qian, T. (2013) · PLoS ONE · doi:10.1371/journal.pone.0077661 · Cited by 397 · cited 2×
“…is important, because humans exhibit structural priming effects (e.g., Pickering and Branigan 1999; Pickering and Ferreira 2008)—they are more likely to produce a structure they have recently encountered—and this adaptation occurs even in comprehension (Fine et al. 2013; Fine and Jaeger 2016).”
Least-to-Most Prompting Enables Complex Reasoning in Large Language Models
Zhou, D., Schärli, N., Hou, L., Wei, J., Scales, N., Wang, X., Schuurmans, D., Cui, C., Bousquet, O., Le, Q., et al. (2022) · arXiv · doi:10.48550/arxiv.2205.10625 · Cited by 380 · cited 1×
“Evaluation strategies that allow models to produce a reasoning trace before an answer will likely improve performance (Nye et al., 2021; Wei et al., 2022b; Kojima et al., 2022; Zhou et al., 2022; Suzgun et al., 2022), and may allow more fair comparisons to humans given time to think.”
Syntactic priming in language production
Pickering, M. (1999) · Trends in Cognitive Sciences · doi:10.1016/s1364-6613(99)01293-0 · Cited by 343 · cited 2×
“This mixing of structures is important, because humans exhibit structural priming effects (e.g., Pickering and Branigan 1999; Pickering and Ferreira 2008)—they are more likely to produce a structure they have recently encountered—and this adaptation occurs even in comprehension (Fine et al. 2013;…”
Emergent linguistic structure in artificial neural networks trained by self-supervision
Manning, C. D., Clark, K., Hewitt, J., Khandelwal, U., & Levy, O. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1907367117 · Cited by 339 · cited 1×
“In this context, these experiments contribute to the growing evidence (e.g., Manning et al. 2020) that modern LMs can learn to process complex syntactic structures, even ones that challenge humans, with a relatively domain-general architecture and learning process applied to imperfect natural data.”
Letting structure emerge: connectionist and dynamical systems approaches to cognition
McClelland, J. L., Botvinick, M. M., Noelle, D. C., Plaut, D. C., Rogers, T. T., Seidenberg, M. S., & Smith, L. B. (2010) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2010.06.002 · Cited by 305
Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
Suzgun, M., Scales, N., Schärli, N., Gehrmann, S., Tay, Y., Chung, H. W., Chowdhery, A., Le, Q., Chi, E., Zhou, D., et al. (2023) · Findings of the Association for Computational Linguistics: ACL 2023 · doi:10.18653/v1/2023.findings-acl.824 · Cited by 253
Yarkoni, T. (2019) · doi:10.31234/osf.io/jqw35 · Cited by 244
Scaling Language Models: Methods, Analysis & Insights from Training Gopher
Rae, J. W., Borgeaud, S., Cai, T., Millican, K., Hoffmann, J., Song, F., Aslanides, J., Henderson, S., Ring, R., Young, S., et al. (2021) · arXiv · doi:10.48550/arxiv.2112.11446 · Cited by 235 · cited 1×
“Fortunately, the Chinchilla models evaluated in this work were trained on data collected in 2020 (see Rae et al. 2021), which is before the BIG-Bench challenge containing these tasks (Srivastava et al. 2023) was released, which reduces the potential for contamination.”
True Few-Shot Learning with Language Models
Perez, E., Kiela, D., & Cho, K. (2021) · arXiv · doi:10.48550/arxiv.2105.11447 · Cited by 226 · cited 2×
“Fewshot prompts are more effective when examples are selected or ordered by validation performance (Perez et al., 2021; Lu et al., 2021)—this is somewhat analogous to how experimenters might choose illustrative examples to introduce an experimental paradigm to participants.”
Constraints on multiple center-embedding of clauses
KARLSSON, F. (2007) · Journal of Linguistics · doi:10.1017/s0022226707004616 · Cited by 223 · cited 1×
“(Example based on the dataset of Lakretz et al. 2022, as released in Srivastava et al. 2023.) of embedding—and almost absent in speech (Karlsson 2007).”
Do Prompt-Based Models Really Understand the Meaning of Their Prompts?
Webson, A., & Pavlick, E. (2022) · Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/2022.naacl-main.167 · Cited by 203
Are Emergent Abilities of Large Language Models a Mirage?
Schaeffer, R., Miranda, B., & Koyejo, S. (2023) · arXiv · doi:10.48550/arxiv.2304.15004 · Cited by 200 · cited 1×
“Few-shot prompting effects are known to depend on model scale (Brown et al. 2020; Wei et al. 2022a; Schaeffer, Miranda, and Koyejo 2023; Lu et al. 2023).”
(2023) · doi:10.52202/075280-2425 · Cited by 200
Compositionality Decomposed: How do Neural Networks Generalise?
Hupkes, D., Dankers, V., Mul, M., & Bruni, E. (2020) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.11674 · Cited by 194 · cited 2×
“…of generalization in more controlled settings, where we know that even the structures in the test set do not appear in training (cf. Linzen 2020; Hupkes et al. 2020)—and indeed, several more controlled studies show that Transformer models trained on simple, synthetic hierarchical structure…”
On the Emergence of Grammar From the Lexicon
(2013) · The Emergence of Language · doi:10.4324/9781410602367-7 · Cited by 191 · cited 2×
“…learning, or whether domain-general processes could explain the acquisition of syntactic structures (Elman, Bates, and Johnson 1996; MacDonald, Pearlmutter, and Seidenberg 1994; Bates and Goodman 2013)— particularly recursive ones (Christiansen and Chater 1999; Christiansen and MacDonald 2009).”
A Usage‐Based Approach to Recursion in Sentence Processing
Christiansen, M. H., & MacDonald, M. C. (2009) · Language Learning · doi:10.1111/j.1467-9922.2009.00538.x · Cited by 168 · cited 4×
“Elman (1991) showed that simple recurrent networks can learn recursive compositional syntactic structures in a simplified language, which initiated a line of research (e.g., Christiansen and Chater 1999; Christiansen and MacDonald 2009).”
InfluentialDistributed Representations, Simple Recurrent Networks, and Grammatical Structure
Elman, J. L. (1991) · Connectionist Approaches to Language Learning · doi:10.1007/978-1-4615-4008-3_5 · Cited by 159
Performance vs. competence in human–machine comparisons
Firestone, C. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1905334117 · Cited by 150 · cited 5×
“More fundamentally, they illustrate the challenges of fairly comparing models and humans (cf. Firestone 2020).”
InfluentialHow Can We Accelerate Progress Towards Human-like Linguistic Generalization?
Linzen, T. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.465 · Cited by 148
Neural language models as psycholinguistic subjects: Representations of syntactic state
Futrell, R., Wilcox, E., Morita, T., Qian, P., Ballesteros, M., & Levy, R. (2019) · Proceedings of the 2019 Conference of the North · doi:10.18653/v1/n19-1004 · Cited by 144 · cited 1×
“Some recent work has explored whether networks trained on natural language can learn syntactic structures (e.g., Futrell et al. 2019; Hu et al. 2020; Wilcox et al. 2020), and in particular recursive dependencies (Futrell et al. 2018; Wilcox, Levy, and Futrell 2019; Lakretz et al. 2021)—and even…”
Show Your Work: Scratchpads for Intermediate Computation with Language Models
Nye, M., Andreassen, A. J., Gur-Ari, G., Michalewski, H., Austin, J., Bieber, D., Dohan, D., Lewkowycz, A., Bosma, M., Luan, D., et al. (2021) · arXiv · doi:10.48550/arxiv.2112.00114 · Cited by 144 · cited 2×
“Evaluation strategies that allow models to produce a reasoning trace before an answer will likely improve performance (Nye et al., 2021; Wei et al., 2022b; Kojima et al., 2022; Zhou et al., 2022; Suzgun et al., 2022), and may allow more fair comparisons to humans given time to think.”
Can language models learn from explanations in context?
Lampinen, A., Dasgupta, I., Chan, S., Mathewson, K., Tessler, M., Creswell, A., McClelland, J., Wang, J., & Hill, F. (2022) · Findings of the Association for Computational Linguistics: EMNLP 2022 · doi:10.18653/v1/2022.findings-emnlp.38 · Cited by 142
Self-citationA Systematic Assessment of Syntactic Generalization in Neural Language Models
Hu, J., Gauthier, J., Qian, P., Wilcox, E., & Levy, R. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.158 · Cited by 134
NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark
Sainz, O., Campos, J., García-Ferrero, I., Etxaniz, J., de Lacalle, O. L., & Agirre, E. (2023) · Findings of the Association for Computational Linguistics: EMNLP 2023 · doi:10.18653/v1/2023.findings-emnlp.722 · Cited by 129
Surface Form Competition: Why the Highest Probability Answer Isn’t Always Right
Holtzman, A., West, P., Shwartz, V., Choi, Y., & Zettlemoyer, L. (2021) · Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2021.emnlp-main.564 · Cited by 127 · cited 3×
“However, multiple choice evaluation, by forcing the model to choose among a small set of answers, effectively hides any possibilities—such as alternative phrasings of the correct answer—that the model would strongly prefer (cf. Holtzman et al. 2021).”
Assessing the belief bias effect with ROCs: It's a response bias effect.
Dube, C., Rotello, C. M., & Heit, E. (2010) · Psychological Review · doi:10.1037/a0019634 · Cited by 116 · cited 1×
“(Cf . related cognitive arguments, e.g., that belief bias interaction effects are simply response bias effects with nonlinearity; Dube, Rotello, and Heit 2010.)”
On the Predictive Power of Neural Language Models for Human Real-Time Comprehension Behavior
Wilcox, E., Gauthier, J., Hu, J., Qian, P., & Lévy, R. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.01912 · Cited by 105 · cited 2×
“Some recent work has explored whether networks trained on natural language can learn syntactic structures (e.g. Futrell et al., 2019; Wilcox et al., 2020), and in particular recursive dependencies (Wilcox et al.”
McClelland, J. L., Hill, F., Rudolph, M., Baldridge, J., & Schütze, H. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1910416117 · Cited by 91 · cited 1×
“However, it remains to be seen whether similar models could learn and generalize complex structures from a human-like amount of language experience (cf. Linzen 2020; Hosseini et al. 2022); learning language within an embodied setting might help (cf. McClelland et al. 2020).”
The role of verb repetition in cumulative structural priming in comprehension.
Fine, A. B., & Jaeger, T. (2016) · Journal of Experimental Psychology. Learning, Memory and Cognition · doi:10.1037/xlm0000236 · Cited by 70 · cited 2×
“…is important, because humans exhibit structural priming effects (e.g., Pickering and Branigan 1999; Pickering and Ferreira 2008)—they are more likely to produce a structure they have recently encountered—and this adaptation occurs even in comprehension (Fine et al. 2013; Fine and Jaeger 2016).”
Valmeekam, K., Marquez, M., Olmo, A., Sreedharan, S., & Kambhampati, S. (2022) · arXiv · doi:10.48550/arxiv.2206.10498 · Cited by 70
Language models, like humans, show content effects on reasoning tasks
Lampinen, A. K., Dasgupta, I., Chan, S. C. Y., Sheahan, H. R., Creswell, A., Kumaran, D., McClelland, J. L., & Hill, F. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae233 · Cited by 69
Self-citationA resource-rational model of human processing of recursive linguistic structure
Hahn, M., Futrell, R., Levy, R., & Gibson, E. (2022) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2122602119 · Cited by 67 · cited 4×
“Indeed, recent results using different behavioral measures (reading time) suggest that language models may need to be impaired with lossy memories to accurately model human responses to complex nested structures (Hahn et al., 2022); similarly, I find that large models given a few sentences of context exhibit far fewer…”
InfluentialHosseini, E. A., Schrimpf, M., Zhang, Y., Bowman, S., Zaslavsky, N., & Fedorenko, E. (2024) · Neurobiology of Language · doi:10.1162/nol_a_00137 · Cited by 63
Using cognitive psychology to understand GPT-3Preprint
Binz, M., & Schulz, E. (2022) · doi:10.31234/osf.io/6dfgk · Cited by 62
Prompting GPT-3 To Be Reliable
Si, C., Gan, Z., Yang, Z., Wang, S., Wang, J., Boyd-Graber, J. L., & Wang, L. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.09150 · Cited by 62 · cited 1×
“, 2022), how to prompt the model for reliable performance (Si et al., 2022), and how to choose the model’s answer (Holtzman et al.”
Mechanisms for handling nested dependencies in neural-network language models and humans
Lakretz, Y., Hupkes, D., Vergallito, A., Marelli, M., Baroni, M., & Dehaene, S. (2021) · Cognition · doi:10.1016/j.cognition.2021.104699 · Cited by 60 · cited 16×
“, 2020), and in particular recursive dependencies (Wilcox et al., 2019; Lakretz et al., 2021; Hahn et al., 2022).”
InfluentialPrompting is not a substitute for probability measurements in large language models
Hu, J., & Levy, R. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.306 · Cited by 60
Language models show human-like content effects on reasoning tasks
Dasgupta, I., Lampinen, A. K., Chan, S. C. Y., Sheahan, H. R., Creswell, A., Kumaran, D., McClelland, J. L., & Hill, F. (2022) · arXiv · doi:10.48550/arxiv.2207.07051 · Cited by 53 · cited 4×
“There is an increasing interest in comparing the capabilities of large language models to human capabilities (e.g. Lakretz et al., 2021a; Binz and Schulz, 2022; Misra et al., 2022; Dasgupta et al., 2022).”
InfluentialSelf-citationReplies to commentaries on the generalizability crisis
Yarkoni, T. (2022) · Behavioral and Brain Sciences · doi:10.1017/s0140525x21001758 · Cited by 51
Sinclair, A., Jumelet, J., Zuidema, W., & Fernández, R. (2022) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00504 · Cited by 48 · cited 3×
“Similarly, LMs exhibit some analogous structural priming effects (Prasad, van Schijndel, and Linzen 2019; Sinclair et al. 2022), and might therefore likewise benefit from exposure to related structures before testing.”
Recursion: what is it, who has it, and how did it evolve?
Coolidge, F. L., Overmann, K. A., & Wynn, T. (2010) · Wiley Interdisciplinary Reviews Cognitive Science · doi:10.1002/wcs.131 · Cited by 47 · cited 2×
“In the minimalist program (Chomsky, 2014b), the ability to recursively Merge smaller constituents into a single larger syntactic structure is considered a key, evolved component (Chomsky, 1999; Coolidge et al., 2011; Berwick and Chomsky, 2019).”
All or nothing: No half-Merge and the evolution of syntax
Berwick, R. C., & Chomsky, N. (2019) · PLoS Biology · doi:10.1371/journal.pbio.3000539 · Cited by 46 · cited 2×
“In the minimalist program (Chomsky, 2014b), the ability to recursively Merge smaller constituents into a single larger syntactic structure is considered a key, evolved component (Chomsky, 1999; Coolidge et al., 2011; Berwick and Chomsky, 2019).”
Experimental psychology: An introduction.
Underwood, B. J. (1949) · Appleton-Century-Crofts eBooks · doi:10.1037/11294-000 · Cited by 45
Investigating Data Contamination in Modern Benchmarks for Large Language Models
Deng, C., Zhao, Y., Tang, X., Gerstein, M., & Cohan, A. (2024) · Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) · doi:10.18653/v1/2024.naacl-long.482 · Cited by 45
The Dangers of Underclaiming: Reasons for Caution When Reporting How NLP Systems Fail
Bowman, S. (2022) · Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2022.acl-long.516 · Cited by 44 · cited 1×
“We should avoid underclaiming, by reporting not only the limitations of the model evaluated (Bowman, 2022), but also the limitations of the evaluation paradigm used—and whether human evaluation paradigms had similar limitations (Firestone, 2020).”
Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
Suzgun, M., Scales, N., Schärli, N., Gehrmann, S., Tay, Y., Chung, H. W., Chowdhery, A., Le, Q. V., Chi, E. H., Zhou, D., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.09261 · Cited by 44 · cited 1×
“Evaluation strategies that allow models to produce a reasoning trace before an answer will likely improve performance (Nye et al. 2021; Wei et al. 2022b; Kojima et al. 2022; Zhou et al. 2023; Suzgun et al. 2023), and may allow more fair comparisons to humans given time to think.”
Using Priming to Uncover the Organization of Syntactic Representations in Neural Language Models
Prasad, G., van Schijndel, M., & Linzen, T. (2019) · Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL) · doi:10.18653/v1/k19-1007 · Cited by 37 · cited 2×
“Similarly, LMs exhibit some analogous structural priming effects (Prasad, van Schijndel, and Linzen 2019; Sinclair et al. 2022), and might therefore likewise benefit from exposure to related structures before testing.”
Context Limitations Make Neural Language Models More Human-Like
Kuribayashi, T., Oseki, Y., Brassard, A., & Inui, K. (2022) · Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2022.emnlp-main.712 · Cited by 37
RNNs as psycholinguistic subjects: Syntactic state and grammatical dependency
Futrell, R., Wilcox, E., Morita, T., & Lévy, R. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1809.01329 · Cited by 36 · cited 1×
“…language can learn syntactic structures (e.g., Futrell et al. 2019; Hu et al. 2020; Wilcox et al. 2020), and in particular recursive dependencies (Futrell et al. 2018; Wilcox, Levy, and Futrell 2019; Lakretz et al. 2021)—and even whether they can be used to help model human processing of these…”
Are Emergent Abilities in Large Language Models just In-Context Learning?
Lu, S., Bigoulaeva, I., Sachdeva, R., Tayyar Madabushi, H., & Gurevych, I. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.279 · Cited by 35
The Importance of a Truly Comparative Methodology for Comparative Psychology
Smith, M. F., Watzek, J., & Brosnan, S. F. (2018) · International Journal of Comparative Psychology · doi:10.46867/ijcp.2018.31.01.12 · Cited by 33
Experimental Psychology; An Introduction
B., M., & Underwood, B. J. (1950) · The American Journal of Psychology · doi:10.2307/1418453 · Cited by 28
Transcending Scaling Laws with 0.1% Extra Compute
Tay, Y., Wei, J., Chung, H., Tran, V., So, D., Shakeri, S., Garcia, X., Zheng, S., Rao, J., Chowdhery, A., et al. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.91 · Cited by 26
What a performance! Some problems with the competence‐performance distinction
Milroy, L. (1985) · Australian Journal of Linguistics · doi:10.1080/07268608508599333 · Cited by 25
Hierarchical Representation in Neural Language Models: Suppression and Recovery of Expectations
Wilcox, E., Levy, R., & Futrell, R. (2019) · Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP · doi:10.18653/v1/w19-4819 · Cited by 21 · cited 2×
“…syntactic structures (e.g., Futrell et al. 2019; Hu et al. 2020; Wilcox et al. 2020), and in particular recursive dependencies (Futrell et al. 2018; Wilcox, Levy, and Futrell 2019; Lakretz et al. 2021)—and even whether they can be used to help model human processing of these structures (Hahn et…”
Interpretability Illusions in the Generalization of Simplified ModelsPreprint
Friedman, D., Lampinen, A. K., Dixon, L., Chen, D., & Ghandeharioun, A. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2312.03656 · Cited by 19
Self-citationBeliefs and Bayesian reasoning
Cohen, A. L., Sidlowski, S., & Staub, A. (2016) · Psychonomic Bulletin & Review · doi:10.3758/s13423-016-1161-z · Cited by 18 · cited 2×
“(2022) note, dismissals of language model reasoning capabilities often ignore the extensive literature on content effects and biases in human reasoning (e.g. Wason and Johnson-Laird, 1972; Evans, 1989; Cohen et al., 2017)—in fact, language models reproduce several classic human effects.”
Schuster, S., & Linzen, T. (2022) · Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/2022.naacl-main.71 · Cited by 15
Misra, K., Rayz, J., & Ettinger, A. (2023) · Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.213 · Cited by 15
(2023) · doi:10.18653/v1/2023.acl-short.38 · Cited by 12
Are Language Models Worse than Humans at Following Prompts? It’s Complicated
Webson, A., Loo, A., Yu, Q., & Pavlick, E. (2023) · Findings of the Association for Computational Linguistics: EMNLP 2023 · doi:10.18653/v1/2023.findings-emnlp.514 · Cited by 12
Stay on topic with Classifier-Free Guidance
Sanchez, G., Fan, H., Spangher, A., Levi, E., Ammanamanchi, P. S., & Biderman, S. (2023) · arXiv · doi:10.48550/arxiv.2306.17806 · Cited by 8
Investigating Data Contamination in Modern Benchmarks for Large Language Models
Deng, C., Zhao, Y., Tang, X., Gerstein, M., & Cohan, A. (2023) · arXiv · doi:10.48550/arxiv.2311.09783 · Cited by 8
Auxiliary task demands mask the capabilities of smaller language models
Hu, J., & Frank, M. C. (2024) · arXiv · doi:10.48550/arxiv.2404.02418 · Cited by 8 · cited 1×
“…what architecture and training is most useful for zero-shot evaluation (Wang et al. 2022; Tay et al. 2023), how to prompt the model or frame the question for reliable performance (Si et al. 2023; Hu and Levy 2023; Hu and Frank 2024), and how to choose the model’s answer (Holtzman et al. 2021).”
Assessing the belief bias effect with ROCs: It's a response bias effect
Dubé, C., Rotello, C. M., & Heit, E. (2009) · PsycEXTRA Dataset · doi:10.1037/e520562012-013 · Cited by 5
Misra, K., Ettinger, A., & Mahowald, K. (2024) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2024.emnlp-main.687 · Cited by 5
Research works printed but not linked: 6
- Chomsky, 2014, Aspects of the Theory of Syntax
- Frank, 2024, Experimentology: An Open Science Approach to Experimental Psychology Methods
- Lakretz, 2022, Can transformers process recursive nested constructions, like humans?, 3226
- Marcus, 2020, GPT-3, Bloviator: OpenAI’s language generator has no idea what it’s talking about, Technology Review
- Radford, 2019, Language models are unsupervised multitask learners, OpenAI blog, 1, 9
- Wang, 2022, What language model architecture and pretraining objective works best for zero-shot generalization?, 22964
Cited by
Dissociating language and thought in large language models
Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2024) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2024.01.011 · Cited by 345
“Life is what we make of it, and it’s up to each of us to find our own meaning in life.”
Niu, Q., Liu, J., Bi, Z., Feng, P., Peng, B., Chen, K., Li, M., Yan, L. K. Q., Zhang, Y., Yin, C. H., et al. (2026) · BIO Integration · doi:10.15212/bioi-2025-0199 · Cited by 121
Language Model Behavior: A Comprehensive Survey
Chang, T. A., & Bergen, B. K. (2024) · Computational Linguistics · doi:10.1162/coli_a_00492 · Cited by 105
McCoy, R. T., Yao, S., Friedman, D., Hardy, M. D., & Griffiths, T. L. (2024) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2322420121 · Cited by 96
(Ir)rationality and cognitive biases in large language models
Macmillan-Scott, O., & Musolesi, M. (2024) · Royal Society Open Science · doi:10.1098/rsos.240255 · Cited by 47
“Lampinen[15] discusses this problem when it comes to language models in particular, highlighting that different approaches must be taking to evaluate cognitive and foundation models.”
Background
Metadata
- Authors
- Lampinen, A.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Corresponding author
- Lampinen, Andrew
- ORCID
- Lampinen, Andrew
- Venue
- Computational Linguistics
- Publisher
- MIT Press
- Published
- 2024
- Volume
- 50
- Issue
- 4
- Pages
- 1441-1476
- ISSN
- 0891-2017 (ISSN-L)
- 1530-9312
- OpenAlex
- W4401123261
- Cited by
- 32
- References
- 93
- Citations by year
- 2026: 11 · 2025: 7 · 2024: 6
- FWCI
- 3.54
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Computational Linguistics (publishedVersion, cc-by-nc-nd)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Social Sciences › Arts and Humanities › Language and Linguistics › Syntax, Semantics, Linguistic Variation
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- language models, human language processing, model evaluation, zero-shot evaluation, prompting, comparative psychology, cognitive models, foundation models
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.