The study investigates how augmenting training data with conceptually related alternatives—beyond nouns to include verbs and adjectives—affects large language models trained with next-token prediction. Results show that while concept augmentation consistently improves perplexity on news data regardless of part of speech, verbs yield the most reliable downstream performance gains. Representational analyses reveal that the conditions producing the largest downstream improvements do not always correspond to the greatest changes in internal representations.
By Laya Iyer, Chen Shani
Large language models are typically trained with next-token prediction (NTP), which rewards predicting specific tokens rather than explicitly accounting for the semantic similarities between different...
The paper critiques a recent NLI benchmark that tests the imperfective paradox, arguing that the benchmark suffers from conceptual and evaluation mis-specifications, notably Aspectual Reduction and a lack of strict NLI standards. The authors re-evaluate the benchmark, identify mis-specifications, and construct lexically matched minimal pairs to control for lexical variation. Their experiments reveal that models often exhibit a Sufficiency Bias, accept simple‑past hypotheses without affirming culmination, and that prompting interventions shift label decisions without improving true semantic understanding, highlighting additional failure modes such as compositional aspectual classification errors and surface‑form attraction.
By Kaiqiao Han, Yizhou Sun
The paper introduces the "Prompt Lottery" phenomenon, where paraphrastic rephrasings of judge prompts in LLM-as-a-Judge (LLMaJ) lead to inconsistent judgments and can even reverse model rankings. A large-scale study across various tasks, judge families, and model backbones demonstrates that these prompt variations significantly alter evaluation outcomes. A user study shows that LLMaJ judges are 6–9 times more sensitive to prompt rephrasings than humans, and that aggregating multiple prompts improves robustness by focusing on consistent judgments.
By Ilija Subasic, Brett Levert, Andrew Rabinovich, Sajjad Pakdamansavoji, Zhao Chen
The Writerslogic team participated in the CLEF 2026 SimpleText shared task, tackling both text simplification (Task 1) and complexity spotting (Task 2). For simplification, they built a multi‑candidate pipeline with GPT‑4o‑mini, selecting the best candidate via a reference‑free heuristic, and their Claude Sonnet 4 submission achieved a SARI of 47.43 and BLEU of 14.21, ranking third overall on the Task 1 leaderboard. For complexity spotting, they fine‑tuned a DeBERTa‑v3‑large NLI model on 350 K labeled pairs, achieving a macro F1 of 0.8081 (0.8085 in an ensemble) on binary over‑generation identification and 0.804 accuracy on multi‑class error classification, placing them second among unique teams.
By David L. Condrey
The study evaluates eleven autoregressive transformer models on English agreement attraction scenarios using a surprisal-based approach. Results show that while transformers match human reading times for prepositional phrase configurations, they perform poorly on object‑extracted relative clauses, with predictions diverging across models and failing to capture human interference patterns. The authors argue that current transformers cannot adequately model human morphosyntactic processing and call for more rigorous, comprehensive testing to avoid misleading conclusions from limited syntactic setups.
By Titus von der Malsburg, Sebastian Pad\'o
arXiv:2604.18835v2 Announce Type: replace-cross
Abstract: We propose a scalable, multifactorial experimental framework that systematically probes LLM sensitivity to subtle semantic changes in pairwis...
By Sinan G. Aksoy, Alexandra A. Sabrio, Erik VonKaenel, Lee Burke
arXiv:2606. 05486v1 Announce Type: cross Abstract: Prompt ambiguity is a common source of failure in large language models, but is difficult to localize because it is a latent property of the prompt, while existing attribution methods are designed to explain observable outputs such as logits or generated tokens.
By Govind Ramesh, Yao Dou, Wei Xu
The paper introduces DescaPE, a decoding framework that uses internal model signals to reduce hallucinations in large language models. By identifying a factual‑salient layer span and training a lightweight probe to approximate its signal, DescaPE penalizes high‑risk continuations and rewards factually grounded ones during inference. Experiments on five factuality benchmarks across three LLMs show that DescaPE improves factuality with only a 1.10× latency overhead.
By Hayeong Ryu, JungMin Yun, Byeonggeuk Lim, Sunhee Jo, YoungBin Kim
arXiv:2603.29396v2 Announce Type: replace
Abstract: Standard evaluations of Large language models (LLMs) focus on task performance, offering limited insight into whether correct behavior reflects app...
By Zo\"e Prins, Samuele Punzo, Frank Wildenburg, Giovanni Cin\`a, Sandro Pezzelle
SWORD is a new benchmark that tests large language models’ ability to reject factually incorrect statements across eight major languages by distorting Wikidata triples. The benchmark reveals that models often perform better on semantically plausible distortions than on random ones, indicating a reliance on distributional familiarity rather than true factual verification. It also shows significant performance drops for East Asian languages, with gaps up to 28 percentage points, highlighting asymmetric multilingual factual reasoning capabilities.
By Sanghyeok Park, Minji Kang, Hosung Kwak, Jinhyuk Yun
arXiv:2609.21227v1 Announce Type: cross
Abstract: Factual hallucination is commonly defined by incorrect factual outputs. We study a paraphrase-induced hallucination setting, where a model answers a...
By Wenhan Yu, Wenxin Wu, Hao Wang, Lei Sha