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

• URL: http://arxiv.org/abs/2606.30815v1
• Date: Mon, 29 Jun 2026 18:42:03 GMT
• Title: When transformers learn "impossible" languages, what do they learn?
• Abstract summary: We evaluate two theoretically motivated linking hypotheses: impossibility arising from deficiencies in grammatical sensitivity or generative production.<n>Using GPT-2 style models trained on perturbed "impossible" variants of English, we measure sensitivity to grammaticality.<n>We find that model performance exhibits only gradual degradation, mediated by the language's information locality.
• Score: 9.276738475766512
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• 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.

Related papers

• A Unified Assessment of the Poverty of the Stimulus Argument for Neural Language Models [10.792630632317653]
  We introduce poshbench, a training-and-evaluation suite targeting question formation, islands to movement, and other English phenomena.<n>We find indications of generalization on all phenomena even without direct positive evidence.<n>Our findings challenge the claim that innate syntax is the only possible route to generalization.
  arXiv  Detail & Related papers  (2026-02-10T17:16:29Z)
• Parallel Universes, Parallel Languages: A Comprehensive Study on LLM-based Multilingual Counterfactual Example Generation [49.2073409243885]
  Large language models (LLMs) excel at generating English counterfactuals and demonstrate multilingual proficiency.<n>We conduct automatic evaluations on both directly generated counterfactuals in the target languages and those derived via English translation across six languages.<n>We identify and categorize four main types of errors that consistently appear in the generated counterfactuals across languages.
  arXiv  Detail & Related papers  (2026-01-01T08:53:49Z)
• Biasless Language Models Learn Unnaturally: How LLMs Fail to Distinguish the Possible from the Impossible [4.7831562043724665]
  We show that GPT-2 learns each language and its impossible counterpart equally easily.<n>By considering cross-linguistic variance in various metrics computed on the perplexity curves, we show that GPT-2 provides no systematic separation between the possible and the impossible.
  arXiv  Detail & Related papers  (2025-10-08T16:17:13Z)
• Anything Goes? A Crosslinguistic Study of (Im)possible Language Learning in LMs [14.78046527879077]
  We train language models to model impossible and typologically unattested languages.<n>Our results show that while GPT-2 small can largely distinguish attested languages, it does not achieve perfect separation between all the attested languages and all the impossible ones.<n>These findings suggest that LMs exhibit some human-like inductive biases, though these biases are weaker than those found in human learners.
  arXiv  Detail & Related papers  (2025-02-26T04:01:36Z)
• Can Language Models Learn Typologically Implausible Languages? [62.823015163987996]
  Grammatical features across human languages show intriguing correlations often attributed to learning biases in humans.<n>We discuss how language models (LMs) allow us to better determine the role of domain-general learning biases in language universals.<n>We test LMs on an array of highly naturalistic but counterfactual versions of the English (head-initial) and Japanese (head-final) languages.
  arXiv  Detail & Related papers  (2025-02-17T20:40:01Z)
• Understanding and Mitigating Language Confusion in LLMs [76.96033035093204]
  We evaluate 15 typologically diverse languages with existing and newly-created English and multilingual prompts.<n>We find that Llama Instruct and Mistral models exhibit high degrees of language confusion.<n>We find that language confusion can be partially mitigated via few-shot prompting, multilingual SFT and preference tuning.
  arXiv  Detail & Related papers  (2024-06-28T17:03:51Z)
• Injecting structural hints: Using language models to study inductive biases in language learning [40.8902073270634]
  We inject inductive bias into language models by pretraining on formally-structured data. We then evaluate the biased learners' ability to learn typologically-diverse natural languages. We show that non-context-free relationships form the best inductive biases.
  arXiv  Detail & Related papers  (2023-04-25T18:00:08Z)
• Discovering Latent Knowledge in Language Models Without Supervision [72.95136739040676]
  Existing techniques for training language models can be misaligned with the truth. We propose directly finding latent knowledge inside the internal activations of a language model in a purely unsupervised way. We show that despite using no supervision and no model outputs, our method can recover diverse knowledge represented in large language models.
  arXiv  Detail & Related papers  (2022-12-07T18:17:56Z)
• Do Multilingual Language Models Capture Differing Moral Norms? [71.52261949766101]
  Massively multilingual sentence representations are trained on large corpora of uncurated data. This may cause the models to grasp cultural values including moral judgments from the high-resource languages. The lack of data in certain languages can also lead to developing random and thus potentially harmful beliefs.
  arXiv  Detail & Related papers  (2022-03-18T12:26:37Z)
• Recurrent Neural Network Language Models Always Learn English-Like Relative Clause Attachment [17.995905582226463]
  We compare model performance in English and Spanish to show that non-linguistic biases in RNN LMs advantageously overlap with syntactic structure in English but not Spanish. English models may appear to acquire human-like syntactic preferences, while models trained on Spanish fail to acquire comparable human-like preferences.
  arXiv  Detail & Related papers  (2020-05-01T01:21:47Z)
• Limits of Detecting Text Generated by Large-Scale Language Models [65.46403462928319]
  Some consider large-scale language models that can generate long and coherent pieces of text as dangerous, since they may be used in misinformation campaigns. Here we formulate large-scale language model output detection as a hypothesis testing problem to classify text as genuine or generated.
  arXiv  Detail & Related papers  (2020-02-09T19:53:23Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.