Natural Language Processing news and analysis on Web Pulse: 5576 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
VEGAS (Video caption Evaluation via GAze Score) introduces a training-free method to align video captions with viewer gaze data, improving caption-to-video retrieval and personalization. The approach uses real-time gaze …
Researchers introduced WCog-VLA, a dual-level World-Cognitive Vision-Language-Action model that enables autonomous vehicles to transition from reactive to proactive navigation by integrating world cognition with generati…
AI Humanizer, a new tool that rewrites AI-generated text into natural copy, launched on Hacker News. The tool offers specialized writing modes for blogs, emails, social media, and creative writing, with features like key…
Researchers introduced LEEVLA, a new architecture for vision-language-action models that uses drift-guided dynamic prioritization to focus on task-relevant evidence, outperforming existing methods on VLA benchmarks. The …
Researchers introduced the PS4 framework, a proxy-supervised training method for target speaker extraction in conversational AI, using a bilingual dataset of 71,771 samples. The framework achieved second place on the REA…
A developer built ShopPilot, an AI shopping assistant for Shopify stores that helps visitors find products and make buying decisions in real time. The project revealed key challenges: product recommendations require nuan…
Researchers introduced LEXIC, a lightweight eye-tracking extension that injects word-level difficulty signals into gaze-only models, achieving AUROC gains of up to +2.9 percentage points on reading comprehension tasks wi…
Researchers propose MASTE, a multi-agent pipeline that decomposes aspect sentiment triplet extraction into four sequential stages handled by specialized agents, achieving zero-shot performance that narrows the gap to sup…
Researchers propose Holographic Neural PCFG (Hol-PCFG), a new unsupervised constituency parsing model that uses holographic embeddings to score grammar rules with a closed-form mathematical structure. Hol-PCFG achieves s…
Researchers found that best-of-N text-to-speech evaluation is confounded by alignment between ASR verifier and evaluator families, with same-family pairs recovering 2-3× more oracle headroom than cross-family pairs. They…
Researchers introduced SQuaD-SQL, a method that uses knowledge distillation from large language models to train small language models for text-to-SQL tasks, achieving 86.9% execution accuracy on WikiSQL with reduced comp…
Researchers developed a modular speaker diarization front end combined with an adapted Qwen3-ASR-1.7B recognizer for multilingual two-speaker conversational speech, achieving a tcpMER of 17.97 on the final evaluation set…
Researchers propose COBART, a method combining prefix control tokens with BART fine-tuning for ad headline generation, achieving a 25.82% improvement in Rouge-L and a 5.82% increase in estimated CTR over previous baselin…
Researchers introduced a human-LLM collaborative annotation framework to create EspanStereo, a Spanish-language stereotype dataset covering multiple Spanish-speaking countries. The dataset captures both well-documented a…
A study comparing LSTM networks with traditional machine learning models for sentiment analysis on Twitter data found that LSTM outperformed logistic regression, random forest, naive Bayes, and gradient boosting, achievi…
Researchers propose Concretized Proposition Prompting (CPP) to resolve the composition-knowledge dichotomy in large language models, enhancing reasoning performance especially in medical benchmarks. CPP bridges compositi…
Researchers developed a two-stage system using large language models to extract fine-grained event tags from SEC 8-K filings, mapping disclosures to a 119-event taxonomy with grounded quotes. Applied to 292,984 filings f…
The ICDAR 2026 HIPE-OCRepair competition evaluated LLM-assisted OCR post-correction for historical documents in English, French, and German. Four teams submitted systems using strategies from zero-shot prompting to fine-…
Researchers propose a multi-cluster boundary learning method for out-of-scope intent detection using MiniLM embedding, achieving state-of-the-art performance on CLINC150, StackOverflow, and Banking77 datasets. The method…
A new survey examines the alignment of large language models with clinical reasoning needs, proposing a five-level competency scheme based on Miller's Pyramid and evaluating 18 state-of-the-art models on a benchmark data…