Not quite Sherlock Holmes: Language model predictions do not reliably differentiate impossible from improbable events (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing work has shown that language models can select the most likely or plausible of a set of possible events, but they are far from robust. |
| Approach: | They focus on whether language models can select the most likely or plausible of a set of possibilities and compare them to a broader behavior that humans exhibit largely unconsciously. |
| Outcome: | The proposed models perform worse than expected under certain conditions, compared with Llama 3, Gemma 2, and Mistral NeMo, and they are significantly more sensible than leaves. |
Similar Papers
Biasless Language Models Learn Unnaturally: How LLMs Fail to Distinguish the Possible from the Impossible (2026.eacl-long)
Copied to clipboard
| Challenge: | linguists have discovered patterns which hold across virtually all known natural languages . lingulists are able to learn languages by comparing their learning curves to those of humans . |
| Approach: | They compare LLM learning curves on existing and "impossible" datasets . they find that GPT-2 learns each language and its impossible counterpart equally easily . |
| Outcome: | The proposed model learns each language and its impossible counterpart equally easily, the study shows . the study also shows that the proposed model does not provide any kind of separation between the possible and the impossible . |
How Reliable are Model Diagnostics? (2021.findings-acl)
Copied to clipboard
| Challenge: | Contemporary statistical models trade off interpretability and simplicity for powerful parameterizations and inductive biases, enabling impressive performance. |
| Approach: | They examine three recent models and find they are not yet reliable . they also formulate recommendations for practitioners and researchers . |
| Outcome: | The proposed models are not as reliable as previously assumed, the authors argue . their findings suggest that they are needed for improving models and training setups . |
Methods for Estimating and Improving Robustness of Language Models (2022.naacl-srw)
Copied to clipboard
| Challenge: | Large language models suffer from weak generalisation ability due to shallow textual relations over full semantic complexity of the problem. |
| Approach: | They propose to incorporate some of these measures into training objectives to enhance distributional robustness of LLMs. |
| Outcome: | The proposed models outperform human models on complex tasks and outperformed other models on deep networks. |
Predicting Reference: What do Language Models Learn about Discourse Models? (2020.emnlp-main)
Copied to clipboard
| Challenge: | a growing literature that probes neural language models to assess their latent acquisition of grammatical knowledge has not investigated their acquisition of discourse modeling ability. |
| Approach: | They draw on a psycholinguistic literature that has established how different contexts affect referential biases concerning who is likely to be referred to next. |
| Outcome: | The proposed models do not resemble human language users, the authors show . their models capture the linguistic knowledge required to perform discourse modeling . |
Can Large Language Models Always Solve Easy Problems if They Can Solve Harder Ones? (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) have impressive capabilities, but still suffer from inconsistency issues. |
| Approach: | They develop a ConsisEval benchmark to evaluate LLMs' inconsistency . they find that LLM models can paradoxically fail at easier problems . |
| Outcome: | The proposed model achieves highest consistency score but inconsistent to specific questions due to distraction by redundant information, misinterpretation of questions, etc. |
What Kind of Language Is Hard to Language-Model? (P19-1)
Copied to clipboard
| Challenge: | a recent study suggests that language models perform poorly across languages. |
| Approach: | They propose a model that fits a paired-sample multiplicative mixed-effects model to obtain language difficulty coefficients from at least-pairwise parallel corpora. |
| Outcome: | The proposed model is able to handle missing data and is aware of inter-sentence variation. |
Do Neural Language Models Overcome Reporting Bias? (2020.coling-main)
Copied to clipboard
| Challenge: | Recent studies show that pre-trained language models can overcome reporting bias by estimating the plausibility of rare but unspoken facts. |
| Approach: | They revisit the experiments conducted by Gordon and Van Durme (2013) . they find that pre-trained language models overestimate the very rare . |
| Outcome: | The proposed approach overestimates the rare at the expense of the rare, while minimizing reporting bias. |
Can Language Models Learn Typologically Implausible Languages? (2026.tacl-1)
Copied to clipboard
| Challenge: | Language models provide a naturalistic framework for studying artificial language learning . authors: typological universals and tendencies are thought to be caused by a learning bias . |
| Approach: | They propose to train LMs on highly naturalistic counterfactual versions of English and Japanese . they show that LM learn subtly implausible languages more slowly . |
| Outcome: | The proposed language models learn subtly implausible languages more slowly compared to human models . the findings suggest that LMs exhibit typologically aligned learning preferences . |
Language Model Probabilities are Not Calibrated in Numeric Contexts (2025.acl-long)
Copied to clipboard
Charles Lovering, Michael Krumdick, Viet Dac Lai, Varshini Reddy, Seth Ebner, Nilesh Kumar, Rik Koncel-Kedziorski, Chris Tanner
| Challenge: | Using language model outputs, we find that even in simple settings, the best LMs (1) are poorly calibrated and (2) have systematic biases. |
| Approach: | They argue that language model outputs should capture natural distributions over multiple options within their textual contexts. |
| Outcome: | The proposed model outputs are calibrated to the numeric content of their contexts. |
Overestimation of Syntactic Representation in Neural Language Models (2020.acl-main)
Copied to clipboard
| Challenge: | Several testing methodologies have been developed to probe models’ syntactic representations. |
| Approach: | They propose a method to determine syntactic structure by training a model on strings generated according to a template and testing its ability to distinguish between similar ones with different syntax. |
| Outcome: | The proposed method reproduces positive results with two non-syntactic baseline language models: an n-gram model and an LSTM model trained on scrambled inputs. |