Retrieval-based fact-checking systems ground verification in large document collections, creating a critical vulnerability: adversarial evidence contamination. We demonstrate that factually inconsistent but contextually coherent evidence generated to satisfy both retrieval-similarity and linguistic-fluency constraints can systematically mislead verification models. Contaminating just 3.1-7.4% of training corpora with five label-specific strategies causes adversarial content to penetrate up to 35.2% of top-k retrievals, degrading verification accuracy by up to 46.9% relative F1. Existing retrieval-stage defenses, validated against gradient-optimized or syntactically anomalous attacks, transfer poorly to this threat model: detection rates collapse and F1 recovery exhibits high variance across seeds.

We propose GLAD (Geometric-Lexical Adversarial Defense), which fuses dense embedding peer-coherence with lexical distributional isolation. Across two benchmarks, two embedding models, and five random seeds, GLAD achieves the highest detection rates (70-86%) and the most consistent F1 recovery, reclaiming 73-80% of contamination-induced degradation on LIAR-RAW while approaching clean baselines on RAWFC. Crucially, unlike defenses that trade clean performance for robustness, GLAD imposes negligible cost on uncontaminated corpora.

The proliferation of misinformation necessitates robust yet computationally efficient fact verification systems. While current state-of-the-art approaches leverage Large Language Models (LLMs) for generating explanatory rationales, these methods face significant computational barriers and hallucination risks in real-world deployments. We present DeReC (Dense Retrieval Classification), a lightweight framework that demonstrates how general-purpose text embeddings can effectively replace autoregressive LLM-based approaches in fact verification tasks. By combining dense retrieval with specialized classification, our system achieves better accuracy while being significantly more efficient. DeReC outperforms explanation-generating LLMs in efficiency, reducing runtime by 95% on RAWFC (23 minutes 36 seconds compared to 454 minutes 12 seconds) and by 92% on LIAR-RAW (134 minutes 14 seconds compared to 1692 minutes 23 seconds), showcasing its effectiveness across varying dataset sizes. On the RAWFC dataset, DeReC achieves an F1 score of 65.58%, surpassing the state-of-the-art method L-Defense (61.20%). Our results demonstrate that carefully engineered retrieval-based systems can match or exceed LLM performance in specialized tasks while being significantly more practical for real-world deployment.

The increasing size and diversity of corpora in natural language processing requires highly efficient processing frameworks. Building on the universal corpus format, Teanga, we present Cuaċ, a format for the compact representation of corpora. We describe this methodology based on short-string compression and indexing techniques and show that the files created with this methodology are similar to compressed human-readable serializations and can be further compressed using lossless compression. We also show that this introduces no computational penalty on the time to process files. This methodology aims to speed up natural language processing pipelines and is the basis for a fast database system for corpora.

This eye-tracking study links language-model surprisal and contextual entropy to how 23 non-expert adults read French health texts. Participants read seven texts (clinical case, medical, general), each available in an Original and Simplified version. Surprisal and entropy were computed with eight autoregressive models (82M–8B parameters), and four complementary eye-tracking measures were analyzed. Surprisal correlates positively with early reading measures, peaking in the smallest GPT-2 models (r ≈ 0.26) and weakening with model size. Entropy shows the opposite pattern, with negative correlations strongest in the 7B-8B models (r ≈ −0.13), consistent with a skim-when-uncertain strategy. Surprisal effects are largest in Clinical Original passages and drop by ∼20% after simplification, whereas entropy effects are stable across domain and version. These findings expose a scaling paradox – where different model sizes are optimal for different cognitive signals – and suggest that French plain-language editing should focus on rewriting high-surprisal passages to reduce processing difficulty, and on avoiding high-entropy contexts for critical information.