natural-language-processing

The paper introduces L-MAD, a framework for systematically evaluating multi-agent debate structures in legal textual entailment. It finds that increasing agent population improves accuracy but more debate rounds cause over-deliberation, with improvements of up to 8% over single-agent baselines.

This paper presents a two-stage LLM-based system that extracts grounded event tags from SEC 8-K filings using a three-tier taxonomy of 119 event types, with mechanisms for constraint and auditability. The system is evaluated on nearly 300k filings, showing high precision for high-quality tags and enabling event studies that distinguish economically distinct events.

This paper presents a comparative study of LSTM and traditional models for sentiment analysis of public opinion.

Stanford NLP announces that a linguistics-informed NLP paper titled 'The Imperfective Paradox in LLMs' has won a Best Paper Award at ACL 2026, encouraging more detailed linguistic evaluation of large language models.

This paper audits and mitigates dialect bias in large language models, showing they systematically prefer Standard American English over African American English. The authors introduce activation steering, a training-free method that reduces bias significantly while preserving fluency, and release the largest real-AAE parallel corpus to date.

This ICML paper introduces recursive models that recursively invoke themselves to solve subtasks in isolated contexts, proving they can surpass context-bounded autoregressive models for long-horizon reasoning. Experiments on SAT solving and Go game-tree search show improved accuracy with small active contexts.

This paper investigates whether explicit domain adaptation methods are beneficial for sentiment transfer when using frozen pre-trained language model backbones, finding that effectiveness depends on whether the backbone already possesses target-domain knowledge.

This paper presents UCSC NLP's systems for SemEval-2026 Task 10 (PsyCoMark), addressing conspiracy marker extraction using boundary-aware span extraction with RoBERTa, and document-level conspiracy classification with label smoothing. The systems ranked 7th in subtask 1 and 12th in subtask 2.

This study analyzes tweets from self-reported ADHD and ASD users to profile DSM-5 depressive symptoms using NLP, finding population-level differences in symptom expression between the two groups despite limited classification performance.

This paper surveys the challenges and evolution of NLP for Indic languages, emphasizing cultural heritage, and proposes a new research direction called 'Culture Sensing' to address representation gaps and ensure culturally meaningful AI outputs.

This paper introduces the Nimblemind Multi-Agent System (nMAS) for extracting evidence of H. pylori infection from gastric biopsy reports, achieving 98.61% accuracy across 216 feature-case decisions and demonstrating substantial time savings over manual review.

This research examines whether LLMs can generate synthetic consumer data for projective techniques, comparing human and LLM responses on city tourism perceptions and finding substantial overlap but differences in style and diversity.

An explanation of how the Genie Ontology method improves text-to-SQL accuracy by focusing on the underlying mechanism rather than the marketing pitch.

This paper presents the Invariant-Variant Disentangled State-Space Model (IVD-SSM), a submission to SemEval-2026 Task 4, which uses a hybrid Jamba-1.5-Mini backbone and a novel Structurally Gated Alignment head to disentangle structural invariants from lexical variants for narrative similarity assessment.

This paper presents an Attention-Based Multiple Instance Learning (ABMIL) framework that leverages patient-level labels from cancer registries to train deep learning classifiers for tumor group classification without requiring per-report annotations, achieving a macro F1 of 0.83 on tasks at the BC Cancer Registry.

This paper presents CaresAI's approach to the SMM4H-HeaRD 2026 shared task on predicting TNM staging from pathology reports using various embeddings and classifiers, achieving strong results but noting generalization issues.

TACG is a training-free decoder for diffusion language models that uses trajectory-aware signals to decide when to commit tokens, improving accuracy and efficiency on code and math benchmarks.

This paper describes the use of NLP methods including LSTM, BERT, and LLMs for analyzing mental health states from social media posts as part of the CLPsych 2026 shared task, achieving top consistency scores for summarization.

A study analyzing 142K NLP papers from 2010–2026 finds that both established and new NLP authors are increasingly publishing in general ML venues like NeurIPS and ICLR rather than core NLP conferences like ACL, with a significant citation premium favoring ML venues.

BamiBERT is a new BERT-based pre-trained language model for Vietnamese that addresses limitations of PhoBERT, supporting longer context and operating without word segmentation, achieving state-of-the-art results on multiple Vietnamese benchmarks.