pith. sign in

arxiv: 2606.30815 · v1 · pith:T4KJ7WHQnew · submitted 2026-06-29 · 💻 cs.CL · cs.AI

When transformers learn "impossible" languages, what do they learn?

Ram Janarthan , Coleman Haley , Sharon Goldwater This is my paper

Pith reviewed 2026-07-01 02:10 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords transformer language modelsimpossible languagesgrammatical sensitivitygenerative productionBLiMPnon-attestationlanguage acquisitioninformation locality
0
0 comments X

The pith

Transformers show only gradual loss of grammatical sensitivity on impossible languages but severe failures generating long sentences from them.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests why language models appear biased against impossible languages by comparing two possible reasons: weak detection of grammatical violations versus inability to produce coherent output. Models trained on English variants altered to be unlearnable by humans still respond to grammar tests with only gradual drops that track how local the information is. The same models however generate far fewer high-quality sentences once length increases. This pattern points to production and transmission problems rather than recognition problems as the more plausible link to why impossible languages remain unattested. A reader would care because the result reframes model bias as a concrete capacity limit instead of a general preference for natural structure.

Core claim

Using GPT-2 style models trained on perturbed impossible variants of English, grammatical sensitivity measured by BLiMP minimal pairs exhibits only gradual degradation mediated by the language's information locality, whereas the models exhibit pronounced failures in generation by producing substantially fewer high-quality sentences at longer lengths; together these results support generative deficiency and transmission failures as a plausible linking hypothesis for non-attestation.

What carries the argument

The contrast between BLiMP minimal-pair grammaticality judgments and direct measurement of generation quality across sentence lengths on information-locally perturbed English variants.

If this is right

  • Grammatical sensitivity alone does not explain model bias against impossible languages.
  • Generation quality declines sharply with length under the perturbed conditions.
  • Transmission failures become a candidate mechanism connecting model behavior to human non-attestation.
  • Information locality modulates how quickly grammatical performance degrades.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If production limits are central, similar length-dependent failures might appear in other sequence models trained on the same variants.
  • The result suggests testing whether human learners also show production bottlenecks before recognition ones on these constructed languages.
  • The mediation by information locality implies that locality-preserving perturbations might produce different patterns than locality-breaking ones.

Load-bearing premise

The specific changes made to English preserve the property that the resulting languages remain unacquirable by humans while still permitting fair comparison to natural English.

What would settle it

An experiment that measures whether the same models generate high-quality long sentences from the perturbed variants at rates comparable to natural English, or a human learning study showing that people acquire the perturbed languages at rates predicted by the model results.

Figures

Figures reproduced from arXiv: 2606.30815 by Coleman Haley, Ram Janarthan, Sharon Goldwater.

Figure 1
Figure 1. Figure 1: BLiMP task accuracy of models trained on English and impossible languages, correlated with 4-local [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Evaluation of generations from the impossible language models. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: BLiMP task accuracy of models trained on English and impossible languages, correlated with test-set [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
read the original abstract

Recent work suggests that transformer language models show a bias towards human languages over unnatural ("impossible") languages argued to be unacquirable by humans. However, this literature has largely based these claims on differences in sample efficiency and test-set perplexity, rather than on direct evaluations of the linguistic capacities that could plausibly explain non-attestation in human languages. We evaluate two theoretically motivated linking hypotheses: impossibility arising from deficiencies in grammatical sensitivity or generative production. Using GPT-2 style models trained on perturbed "impossible" variants of English, we measure sensitivity to grammaticality using BLiMP minimal pairs, finding that model performance exhibits only gradual degradation, mediated by the language's information locality. In contrast, these models exhibited pronounced failures in generation, producing substantially fewer high-quality sentences at longer lengths. Together, these results suggest generative deficiency and transmission failures as a plausible linking hypothesis between language model behaviour and non-attestation of impossible languages.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that GPT-2-style transformers trained on perturbed 'impossible' variants of English exhibit only gradual degradation in grammatical sensitivity on BLiMP minimal pairs (mediated by information locality), but show pronounced failures in generation by producing substantially fewer high-quality sentences at longer lengths. These results are used to argue that generative deficiency and transmission failures provide a plausible linking hypothesis for the non-attestation of impossible languages, moving beyond prior work focused on sample efficiency and perplexity.

Significance. If the results hold, the work strengthens the literature by shifting from indirect metrics to direct tests of grammatical sensitivity and generative capacity on held-out sets. The empirical contrast between BLiMP and generation tasks offers a falsifiable basis for connecting model behavior to linguistic non-attestation, with potential implications for both acquisition theory and model evaluation.

major comments (2)
  1. [language construction section] Language construction section: The perturbations used to create the impossible variants must be explicitly justified against documented linguistic universals or acquisition constraints to ensure the languages remain unacquirable by humans while preserving structural comparability to English; absent this grounding, the observed gradual BLiMP degradation versus generation failures cannot securely support the transmission-failure hypothesis.
  2. [generation evaluation] Generation evaluation (results section): The definition and measurement of 'high-quality sentences' at longer lengths, including any statistical controls or inter-annotator details, is load-bearing for the central claim of pronounced generative deficiency; without these, the contrast with BLiMP cannot be assessed as robust.
minor comments (2)
  1. [Abstract] Abstract: Include the number of models, exact data splits, and perturbation types to allow readers to evaluate the directional results without needing the full methods.
  2. [Methods] Methods: Ensure all baseline comparisons and information-locality mediation analyses are described with sufficient detail for replication.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major point below and will incorporate revisions to strengthen the presentation of our methods and results.

read point-by-point responses

  1. Referee: [language construction section] Language construction section: The perturbations used to create the impossible variants must be explicitly justified against documented linguistic universals or acquisition constraints to ensure the languages remain unacquirable by humans while preserving structural comparability to English; absent this grounding, the observed gradual BLiMP degradation versus generation failures cannot securely support the transmission-failure hypothesis.

    Authors: We agree that more explicit grounding is needed. The language construction section motivates the perturbations from theoretical proposals in the linguistics literature on impossible languages and information locality, but we will add a dedicated subsection that directly links each perturbation to specific documented universals and acquisition constraints (with additional citations). This will clarify how the variants are positioned as unacquirable by humans while preserving structural comparability to English, thereby reinforcing the transmission-failure hypothesis. revision: yes

  2. Referee: [generation evaluation] Generation evaluation (results section): The definition and measurement of 'high-quality sentences' at longer lengths, including any statistical controls or inter-annotator details, is load-bearing for the central claim of pronounced generative deficiency; without these, the contrast with BLiMP cannot be assessed as robust.

    Authors: We concur that greater methodological detail is required here. In the revised manuscript we will expand the generation evaluation section to provide an explicit definition of high-quality sentences (based on grammaticality, fluency, and coherence criteria), report inter-annotator agreement metrics, and include statistical controls such as length-matched comparisons and significance testing. These additions will allow a clearer assessment of the contrast with BLiMP results. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical results on held-out data

full rationale

The paper's claims rest on direct experimental measurements: models are trained on perturbed language variants, then evaluated for grammatical sensitivity via BLiMP minimal pairs and for generation quality at varying lengths. These outcomes are reported as observed performance differences on test sets, not as quantities derived from the paper's own equations, fitted parameters renamed as predictions, or self-referential definitions. No load-bearing self-citations, uniqueness theorems, or ansatzes smuggled via prior work are described in the abstract or methodology summary that would reduce the central linking hypothesis to an input by construction. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work is empirical and relies on standard transformer training plus existing benchmarks; no free parameters, ad-hoc axioms, or new invented entities are introduced in the abstract.

pith-pipeline@v0.9.1-grok · 5688 in / 1126 out tokens · 32530 ms · 2026-07-01T02:10:57.417657+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

298 extracted references · 171 canonical work pages · 3 internal anchors

  1. [1]

    , month = sep, year =

    Smith, Kenny and Kirby, Simon and Guo, Shangmin and Griffiths, Thomas L. , month = sep, year =. Nature , publisher =. doi:10.1038/d41586-024-03023-y , language =

    work page doi:10.1038/d41586-024-03023-y

  2. [2]

    Cognition , author =

    Compression and communication in the cultural evolution of linguistic structure , volume =. Cognition , author =. 2015 , keywords =. doi:10.1016/j.cognition.2015.03.016 , abstract =

    work page doi:10.1016/j.cognition.2015.03.016 2015

  3. [3]

    AI models collapse when trained on recursively generated data.Nature, 631(8022):755–759, 2024

    Shumailov, Ilia and Shumaylov, Zakhar and Zhao, Yiren and Papernot, Nicolas and Anderson, Ross and Gal, Yarin , month = jul, year =. Nature , publisher =. doi:10.1038/s41586-024-07566-y , abstract =

    work page doi:10.1038/s41586-024-07566-y

  4. [4]

    Biasless

    Ziv, Imry and Lan, Nur and Chemla, Emmanuel and Katzir, Roni , month = oct, year =. Biasless. doi:10.48550/arXiv.2510.07178 , abstract =

    work page doi:10.48550/arxiv.2510.07178

  5. [5]

    Xu, Tianyang and Kuribayashi, Tatsuki and Oseki, Yohei and Cotterell, Ryan and Warstadt, Alex , month = feb, year =. Can. doi:10.48550/arXiv.2502.12317 , abstract =

    work page doi:10.48550/arxiv.2502.12317

  6. [6]

    and Giulianelli, Mario and Cotterell, Ryan , editor =

    Someya, Taiga and Svete, Anej and DuSell, Brian and O'Donnell, Timothy J. and Giulianelli, Mario and Cotterell, Ryan , editor =. Information. Proceedings of the 63rd. 2025 , pages =. doi:10.18653/v1/2025.acl-long.1357 , abstract =

    work page doi:10.18653/v1/2025.acl-long.1357 2025

  7. [7]

    Transactions of the Association for Computational Linguistics , author =

    How. Transactions of the Association for Computational Linguistics , author =. 2023 , pages =. doi:10.1162/tacl_a_00567 , abstract =

    work page doi:10.1162/tacl_a_00567 2023

  8. [8]

    Syntactic

    Linzen, Tal and Baroni, Marco , month = jan, year =. Syntactic. Annual Review of Linguistics , publisher =. doi:10.1146/annurev-linguistics-032020-051035 , abstract =

    work page doi:10.1146/annurev-linguistics-032020-051035

  9. [9]

    , year =

    LeBrun, Benjamin and Sordoni, Alessandro and O'Donnell, Timothy J. , year =. Evaluating distributional distortion in neural language modeling , url =. The tenth international conference on learning representations (

  10. [10]

    Transactions of the Association for Computational Linguistics , author =

    What. Transactions of the Association for Computational Linguistics , author =. 2026 , pages =. doi:10.1162/TACL.a.611 , abstract =

    work page doi:10.1162/tacl.a.611 2026

  11. [11]

    , month = oct, year =

    Hu, Jennifer and Wilcox, Ethan Gotlieb and Song, Siyuan and Mahowald, Kyle and Levy, Roger P. , month = oct, year =. What. doi:10.48550/arXiv.2510.16227 , abstract =

    work page doi:10.48550/arxiv.2510.16227

  12. [12]

    Understanding emergent abilities of language models from the loss perspective , url =

    Du, Zhengxiao and Zeng, Aohan and Dong, Yuxiao and Tang, Jie , editor =. Understanding emergent abilities of language models from the loss perspective , url =. Advances in neural information processing systems 38:

  13. [13]

    Chomsky, Noam , year =. On the. Language and

  14. [14]

    Guo, Yanzhu and Shang, Guokan and Vazirgiannis, Michalis and Clavel, Chloé , editor =. The. Findings of the. 2024 , pages =. doi:10.18653/v1/2024.findings-naacl.228 , abstract =

    work page doi:10.18653/v1/2024.findings-naacl.228 2024

  15. [15]

    Proceedings of the

    Heafield, Kenneth , editor =. Proceedings of the. 2011 , pages =

    2011

  16. [16]

    Priorless

    Mitchell, Jeff and Bowers, Jeffrey , editor =. Priorless. Proceedings of the 28th. 2020 , pages =. doi:10.18653/v1/2020.coling-main.451 , abstract =

    work page doi:10.18653/v1/2020.coling-main.451 2020

  17. [17]

    Behavioral and Brain Sciences , author =

    Rules and representations , volume =. Behavioral and Brain Sciences , author =. 1980 , keywords =. doi:10.1017/S0140525X00001515 , abstract =

    work page doi:10.1017/s0140525x00001515 1980

  18. [19]

    Futrell, Richard and Mahowald, Kyle , pages =. How. doi:10.1017/s0140525x2510112x , journal =

    work page doi:10.1017/s0140525x2510112x

  19. [20]

    why is my face purple -

  20. [21]

    Transformer-

    Oh, Byung-Doh and Schuler, William , editor =. Transformer-. Findings of the. 2023 , pages =. doi:10.18653/v1/2023.findings-emnlp.128 , abstract =

    work page doi:10.18653/v1/2023.findings-emnlp.128 2023

  21. [22]

    Hu, Jennifer and Gauthier, Jon and Qian, Peng and Wilcox, Ethan and Levy, Roger , editor =. A. Proceedings of the 58th. 2020 , pages =. doi:10.18653/v1/2020.acl-main.158 , abstract =

    work page doi:10.18653/v1/2020.acl-main.158 2020

  22. [23]

    Proceedings of the National Academy of Sciences , author =

    Cumulative cultural evolution in the laboratory:. Proceedings of the National Academy of Sciences , author =. 2008 , pages =. doi:10.1073/pnas.0707835105 , abstract =

    work page doi:10.1073/pnas.0707835105 2008

  23. [24]

    Language models as tools for investigating the distinction between possible and impossible natural languages , url =

    Kallini, Julie and Potts, Christopher , year =. Language models as tools for investigating the distinction between possible and impossible natural languages , url =

  24. [25]

    arXiv.org , author =

    Language models as tools for investigating the distinction between possible and impossible natural languages , url =. arXiv.org , author =

  25. [26]

    Prompting is not a substitute for probability measurements in large language models , url =

    Hu, Jennifer and Levy, Roger , editor =. Prompting is not a substitute for probability measurements in large language models , url =. Proceedings of the 2023. 2023 , pages =. doi:10.18653/v1/2023.emnlp-main.306 , abstract =

    work page doi:10.18653/v1/2023.emnlp-main.306 2023

  26. [27]

    Cortex , author =

    Large languages, impossible languages and human brains , volume =. Cortex , author =. 2023 , keywords =. doi:10.1016/j.cortex.2023.07.003 , abstract =

    work page doi:10.1016/j.cortex.2023.07.003 2023

  27. [28]

    Form and meaning in intrinsic multilingual evaluations , url =

    Poelman, Wessel and de Lhoneux, Miryam , year =. Form and meaning in intrinsic multilingual evaluations , url =

  28. [29]

    Thomas and Min, Junghyun and Linzen, Tal , editor =

    McCoy, R. Thomas and Min, Junghyun and Linzen, Tal , editor =. Proceedings of the. 2020 , pages =. doi:10.18653/v1/2020.blackboxnlp-1.21 , abstract =

    work page doi:10.18653/v1/2020.blackboxnlp-1.21 2020

  29. [30]

    arXiv.org , author =

    Can. arXiv.org , author =

  30. [31]

    Anything

    Yang, Xiulin and Aoyama, Tatsuya and Yao, Yuekun and Wilcox, Ethan , editor =. Anything. Proceedings of the 63rd. 2025 , pages =. doi:10.18653/v1/2025.acl-long.1264 , abstract =

    work page doi:10.18653/v1/2025.acl-long.1264 2025

  31. [32]

    Explanatory

    Chomsky, Noam , month = jan, year =. Explanatory. Studies in. doi:10.1016/S0049-237X(09)70617-2 , pages =

    work page doi:10.1016/s0049-237x(09)70617-2

  32. [33]

    and Dautriche, Isabelle and Mahowald, Kyle and Bergen, Leon and Levy, Roger , month = may, year =

    Gibson, Edward and Futrell, Richard and Piantadosi, Steven P. and Dautriche, Isabelle and Mahowald, Kyle and Bergen, Leon and Levy, Roger , month = may, year =. How. Trends in Cognitive Sciences , publisher =. doi:10.1016/j.tics.2019.02.003 , language =

    work page doi:10.1016/j.tics.2019.02.003 2019

  33. [34]

    Simplicity and

    Culbertson, Jennifer and Kirby, Simon , month = jan, year =. Simplicity and. Frontiers in Psychology , publisher =. doi:10.3389/fpsyg.2015.01964 , abstract =

    work page doi:10.3389/fpsyg.2015.01964 2015

  34. [35]

    Moro, Andrea , month = may, year =. The. doi:10.7551/mitpress/9780262134989.001.0001 , abstract =

    work page doi:10.7551/mitpress/9780262134989.001.0001

  35. [36]

    The New York Times , author =

    Opinion. The New York Times , author =. 2023 , keywords =

    2023

  36. [38]

    and Bovy, Jo , month = jan, year =

    Leung, Henry W. and Bovy, Jo , month = jan, year =. Towards an astronomical foundation model for stars with a transformer-based model , volume =. Monthly Notices of the Royal Astronomical Society , publisher =. doi:10.1093/mnras/stad3015 , abstract =

    work page doi:10.1093/mnras/stad3015

  37. [39]

    and Sirelkhatim, Hassan and Richard, Guillaume and Skwark, Marcin and Beguir, Karim and Lopez, Marie and Pierrot, Thomas , month = feb, year =

    Dalla-Torre, Hugo and Gonzalez, Liam and Mendoza-Revilla, Javier and Lopez Carranza, Nicolas and Grzywaczewski, Adam Henryk and Oteri, Francesco and Dallago, Christian and Trop, Evan and de Almeida, Bernardo P. and Sirelkhatim, Hassan and Richard, Guillaume and Skwark, Marcin and Beguir, Karim and Lopez, Marie and Pierrot, Thomas , month = feb, year =. Nu...

    work page doi:10.1038/s41592-024-02523-z

  38. [40]

    An image is worth 16x16 words:

    Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil , year =. An image is worth 16x16 words:. 9th international conference on learning representations,

  39. [41]

    Linguistic Inquiry , author =

    Using. Linguistic Inquiry , author =. 2024 , pages =. doi:10.1162/ling_a_00491 , abstract =

    work page doi:10.1162/ling_a_00491 2024

  40. [42]

    Trends in Cognitive Sciences , author =

    Dissociating language and thought in large language models , volume =. Trends in Cognitive Sciences , author =. 2024 , keywords =. doi:10.1016/j.tics.2024.01.011 , abstract =

    work page doi:10.1016/j.tics.2024.01.011 2024

  41. [43]

    Sentence-

    Kann, Katharina and Rothe, Sascha and Filippova, Katja , editor =. Sentence-. Proceedings of the 22nd. 2018 , pages =. doi:10.18653/v1/K18-1031 , abstract =

    work page doi:10.18653/v1/k18-1031 2018

  42. [44]

    Measuring

    Padmakumar, Vishakh and Yueh-Han, Chen and Pan, Jane and Chen, Valerie and He, He , month = oct, year =. Measuring. doi:10.48550/arXiv.2504.09389 , abstract =

    work page doi:10.48550/arxiv.2504.09389

  43. [45]

    Death of the

    Saakyan, Arkadiy and Kim, Najoung and Muresan, Smaranda and Chakrabarty, Tuhin , year =. Death of the. doi:10.48550/ARXIV.2509.22641 , abstract =

    work page doi:10.48550/arxiv.2509.22641

  44. [46]

    Guo, Yanzhu and Shang, Guokan and Vazirgiannis, Michalis and Clavel, Chloé , year =. The. doi:10.48550/ARXIV.2311.09807 , abstract =

    work page doi:10.48550/arxiv.2311.09807

  45. [47]

    Thomas and Smolensky, Paul and Linzen, Tal and Gao, Jianfeng and Celikyilmaz, Asli , month = nov, year =

    McCoy, R. Thomas and Smolensky, Paul and Linzen, Tal and Gao, Jianfeng and Celikyilmaz, Asli , month = nov, year =. How much do language models copy from their training data?. doi:10.48550/arXiv.2111.09509 , abstract =

    work page doi:10.48550/arxiv.2111.09509

  46. [48]

    , editor =

    Warstadt, Alex and Parrish, Alicia and Liu, Haokun and Mohananey, Anhad and Peng, Wei and Wang, Sheng-Fu and Bowman, Samuel R. , editor =. Transactions of the Association for Computational Linguistics , publisher =. 2020 , pages =. doi:10.1162/tacl_a_00321 , abstract =

    work page doi:10.1162/tacl_a_00321 2020

  47. [49]

    and Blank, Idan A

    Mahowald, Kyle and Ivanova, Anna A. and Blank, Idan A. and Kanwisher, Nancy and Tenenbaum, Joshua B. and Fedorenko, Evelina , month = mar, year =. Dissociating language and thought in large language models , url =. doi:10.48550/arXiv.2301.06627 , abstract =

    work page doi:10.48550/arxiv.2301.06627

  48. [50]

    Schaeffer, Rylan and Miranda, Brando and Koyejo, Sanmi , month = may, year =. Are. doi:10.48550/arXiv.2304.15004 , abstract =

    work page doi:10.48550/arxiv.2304.15004

  49. [51]

    Emergent

    Berti, Leonardo and Giorgi, Flavio and Kasneci, Gjergji , month = mar, year =. Emergent. doi:10.48550/arXiv.2503.05788 , abstract =

    work page doi:10.48550/arxiv.2503.05788

  50. [52]

    Understanding

    Du, Zhengxiao and Zeng, Aohan and Dong, Yuxiao and Tang, Jie , month = jan, year =. Understanding. doi:10.48550/arXiv.2403.15796 , abstract =

    work page doi:10.48550/arxiv.2403.15796

  51. [53]

    Emergent Abilities of Large Language Models

    Wei, Jason and Tay, Yi and Bommasani, Rishi and Raffel, Colin and Zoph, Barret and Borgeaud, Sebastian and Yogatama, Dani and Bosma, Maarten and Zhou, Denny and Metzler, Donald and Chi, Ed H. and Hashimoto, Tatsunori and Vinyals, Oriol and Liang, Percy and Dean, Jeff and Fedus, William , month = oct, year =. Emergent. doi:10.48550/arXiv.2206.07682 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2206.07682

  52. [54]

    , month = jan, year =

    Newmeyer, Frederick J. , month = jan, year =. The. The. doi:10.1163/9781849500098_009 , language =

    work page doi:10.1163/9781849500098_009

  53. [55]

    Comparing

    Hartmann, Mareike and Kementchedjhieva, Yova and Søgaard, Anders , editor =. Comparing. Advances in. 2019 , pages =

    2019

  54. [56]

    Theoretical stances shape empirical generalizations on inflection vs

    Kyjánek, Lukáš and Bonami, Olivier , year =. Theoretical stances shape empirical generalizations on inflection vs. derivation: quantitative evidence from. Word-

  55. [57]

    Marked nominative in africa , volume =

    König, Christa , year =. Marked nominative in africa , volume =. Studies in Language , publisher =

  56. [58]

    Lindsey, Jack and Gurnee, Wes and Ameisen, Emmanuel and Chen, Brian and Pearce, Adam and Turner, Nicholas L. and Citro, Craig and Abrahams, David and Carter, Shan and Hosmer, Basil and Marcus, Jonathan and Sklar, Michael and Templeton, Adly and Bricken, Trenton and McDougall◊, Callum and Cunningham, Hoagy and Henighan, Thomas and Jermyn, Adam and Jones, A...

  57. [59]

    Givón, Talmy , year =. On

  58. [60]

    Givón, Talmy , month = mar, year =. On. doi:10.1075/z.213 , abstract =

    work page doi:10.1075/z.213

  59. [61]

    Syntactic categories and grammatical relations:

    Croft, William , year =. Syntactic categories and grammatical relations:

  60. [62]

    Distill , author =

    Multimodal. Distill , author =. 2021 , pages =. doi:10.23915/distill.00030 , abstract =

    work page doi:10.23915/distill.00030 2021

  61. [63]

    Applying

    Vasselli, Justin and Martínez Peguero, Arturo and Sung, Junehwan and Watanabe, Taro , editor =. Applying. Proceedings of the 4th. 2024 , pages =. doi:10.18653/v1/2024.americasnlp-1.24 , abstract =

    work page doi:10.18653/v1/2024.americasnlp-1.24 2024

  62. [64]

    Tanzer, Garrett and Suzgun, Mirac and Visser, Eline and Jurafsky, Dan and Melas-Kyriazi, Luke , year =. A. The

  63. [65]

    Composable

    Ansell, Alan and Ponti, Edoardo and Korhonen, Anna and Vulić, Ivan , editor =. Composable. Proceedings of the 60th. 2022 , pages =. doi:10.18653/v1/2022.acl-long.125 , abstract =

    work page doi:10.18653/v1/2022.acl-long.125 2022

  64. [66]

    Multilingual

    Papadimitriou, Isabel and Lopez, Kezia and Jurafsky, Dan , editor =. Multilingual. Findings of the. 2023 , pages =. doi:10.18653/v1/2023.findings-eacl.89 , abstract =

    work page doi:10.18653/v1/2023.findings-eacl.89 2023

  65. [67]

    Language

    Misra, Kanishka and Mahowald, Kyle , editor =. Language. Proceedings of the 2024. 2024 , pages =. doi:10.18653/v1/2024.emnlp-main.53 , abstract =

    work page doi:10.18653/v1/2024.emnlp-main.53 2024

  66. [68]

    Hu, Michael Y. and Mueller, Aaron and Ross, Candace and Williams, Adina and Linzen, Tal and Zhuang, Chengxu and Cotterell, Ryan and Choshen, Leshem and Warstadt, Alex and Wilcox, Ethan Gotlieb , editor =. Findings of the. The 2nd. 2024 , pages =

    2024

  67. [69]

    Findings of the

    Warstadt, Alex and Mueller, Aaron and Choshen, Leshem and Wilcox, Ethan and Zhuang, Chengxu and Ciro, Juan and Mosquera, Rafael and Paranjabe, Bhargavi and Williams, Adina and Linzen, Tal and Cotterell, Ryan , editor =. Findings of the. Proceedings of the. 2023 , pages =. doi:10.18653/v1/2023.conll-babylm.1 , urldate =

    work page doi:10.18653/v1/2023.conll-babylm.1 2023

  68. [70]

    Charpentier, Lucas and Choshen, Leshem and Cotterell, Ryan and Gul, Mustafa Omer and Hu, Michael Y. and Liu, Jing and Jumelet, Jaap and Linzen, Tal and Mueller, Aaron and Ross, Candance and Shah, Raj Sanjay and Warstadt, Alex and Wilcox, Ethan Gotlieb and Williams, Adina , editor =. Findings of the. Proceedings of the. 2025 , pages =

    2025

  69. [71]

    , editor =

    Woodard, Roger D. , editor =. Greek. The. 2023 , keywords =. doi:10.1017/9780511842788.008 , abstract =

    work page doi:10.1017/9780511842788.008 2023

  70. [72]

    Peyraube, Alain and Chappell, Hilary and Vovin, Alexander , editor =. East. The. 2023 , keywords =. doi:10.1017/9780511842788.006 , abstract =

    work page doi:10.1017/9780511842788.006 2023

  71. [73]

    Amsler, Mark , editor =. The. The. 2023 , keywords =. doi:10.1017/9780511842788.004 , abstract =

    work page doi:10.1017/9780511842788.004 2023

  72. [74]

    doi:10.1017/9780511842788 , abstract =

    The. doi:10.1017/9780511842788 , abstract =

    work page doi:10.1017/9780511842788

  73. [75]

    Unlocking finite-state morphological transducers:

    Haley, Coleman , month = jun, year =. Unlocking finite-state morphological transducers:. Society for Computation in Linguistics , publisher =. doi:10.7275/scil.3172 , abstract =

    work page doi:10.7275/scil.3172

  74. [76]

    Computational Linguistics , author =

    A. Computational Linguistics , author =. 2025 , pages =. doi:10.1162/COLI.a.577 , abstract =

    work page doi:10.1162/coli.a.577 2025

  75. [77]

    What is “

    Ploeger, Esther and Poelman, Wessel and de Lhoneux, Miryam and Bjerva, Johannes , editor =. What is “. Proceedings of the 2024. 2024 , pages =. doi:10.18653/v1/2024.emnlp-main.326 , abstract =

    work page doi:10.18653/v1/2024.emnlp-main.326 2024

  76. [78]

    Universals in

    Lieber, Rochelle and Štekauer, Pavol , editor =. Universals in. The. 2014 , pages =. doi:10.1093/oxfordhb/9780199641642.013.0041 , abstract =

    work page doi:10.1093/oxfordhb/9780199641642.013.0041 2014

  77. [79]

    Sampling for variety , volume =

    Miestamo, Matti and Bakker, Dik and Arppe, Antti , month = oct, year =. Sampling for variety , volume =. Linguistic Typology , publisher =. doi:10.1515/lingty-2016-0006 , abstract =

    work page doi:10.1515/lingty-2016-0006 2016

  78. [80]

    Foundations of Language

    Dryer, Matthew S. , month = jan, year =. Large. Studies in Language. International Journal sponsored by the Foundation “Foundations of Language” , publisher =. doi:10.1075/sl.13.2.03dry , abstract =

    work page doi:10.1075/sl.13.2.03dry

  79. [81]

    Lexical-semantic versus syntactic disorders in aphasia: the processing of prepositions , author =

  80. [82]

    Cerebral Cortex , author =

    Fractionating. Cerebral Cortex , author =. 1992 , pages =. doi:10.1093/cercor/2.3.244 , abstract =

    work page doi:10.1093/cercor/2.3.244 1992

Showing first 80 references.