Natural Language Processing news and analysis on Web Pulse: 3949 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
Researchers introduced the Onramp-Sequence Cross-Distillation (OSCD), a post-training algorithm that improves native multilingual chain-of-thought reasoning in low-resource Southeast Asian languages by projecting high-re…
Researchers trained a small language model to route queries to specialized retrieval agents, achieving an NDCG@10 of 0.918 on mismatched queries, compared to 0.539 for Amazon Nova Lite and 0.490 for Claude Haiku 4.5, and…
A new arXiv preprint (2608.00585v1) shows that per-chunk verification fails on multi-hop retrieval-augmented generation, with entailment scoring reaching only 0.643, 0.523, and 0.560 AUC on HotpotQA, 2WikiMultihopQA, and…
A new study from arXiv (2608.00205v1) finds that human faithfulness annotations for text summarization exhibit an 'Averaging Bias,' where annotators accept summaries as faithful if most sentences are supported, rather th…
A new arXiv paper (2608.00582v1) introduces BPE-guided insertion, a method to adapt byte-level BPE tokenizers to underrepresented languages without changing the model vocabulary size. On Ukrainian adaptations of Nemotron…
Researchers at Observatorio Lázaro have built a self-populating database that has recorded more than two million English lexical borrowings (anglicisms) in the Spanish digital press, processing 1.88 million articles and …
Researchers introduced DLLM-TTS, a text-to-speech framework using block discrete diffusion over X-Codec2 neural audio codec tokens, achieving a real-time factor of 0.15 with a 0.6B-parameter model trained on 20K hours, a…
Researchers at arXiv introduced a novel pre-training method for logographic character recognition that uses semantic-based contrastive learning to improve visual representations on imbalanced datasets. The method, which …
Researchers introduced DE-NER, a dialogue elicitation framework that improves zero-shot Named Entity Recognition by leveraging the chatting ability of Large Language Models, achieving an average improvement of 3.75% F1 p…
Nous Research shipped Hermes Agent v0.20.0 on August 3rd, adding streaming conversational voice, on-device wake words, and a standard protocol for communicating with other AI agents. The release, signed by co-founder Rya…
Pangram, an AI detection company founded in Brooklyn, NY in 2023 by AI researchers from Tesla and Google, has launched Pangram 4, its most advanced model, which detects AI-generated content with 99.98% accuracy. The tool…
OpenAI engineers Justin Uberti and Zahan Malkani detailed on August 3rd a six-month rebuild of ChatGPT's voice infrastructure that enables GPT-Live to listen and speak simultaneously, with other models handling search, r…
Wispr Flow, the voice dictation product from founders Tanay Kothari and Sahaj Garg, is preparing to launch a meeting notetaker, according to updated terms and privacy documents dated July 25th. The new Notetaker feature …
In a blog post, technology writer and designer Matt Webb proposed a voice-first podcast app that lets listeners talk back to the show to capture notes, citing a need for a minimal Granola-like interface to timestamp mome…
A developer replaced human-written 'gold data' with AI-generated translations to train a specialist small model, boosting Japanese-to-English accuracy from 42% to 84% via distillation. The engineer found that the student…
The Guardian's Football Weekly podcast revealed that artificial intelligence is reshaping football scouting, transfer gossip, and tactical analysis, with clubs increasingly relying on machine learning models for recruitm…
Microsoft has introduced SharePoint Skills, a feature that lets users create reusable, multi-step AI workflows within SharePoint using natural language. The tool, managed via the site's Agent Assets library, automates re…
FunASR, an industrial-grade speech recognition toolkit, released an update in August 2026, achieving up to 340x realtime speed and 26x faster than Whisper, supporting 50+ languages, speaker diarization, emotion detection…
A study on arXiv (2607.28634v1) found that zero-shot GPT-4.1 with a temperature of 0 achieved the highest item difficulty prediction accuracy among large language models, with a quadratic weighted kappa (QWK) of 0.578, b…
A team from ZeroR achieved 2nd place on hate speech detection (F1: 0.797) and 4th place on sentiment analysis (F1: 0.518) in the CHiPSAL 2026 shared task on multimodal hate speech and sentiment detection in Nepali memes,…