Natural Language Processing news and analysis on Web Pulse: 5573 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
A new graph-based framework from arXiv research (2607.11894v1) detects disinformation narratives in Telegram ecosystems by combining weak supervision with propagation graph analysis, enabling identification of coordinate…
A new study from arXiv reveals that state-of-the-art large language models (LLMs) fail at bidirectional Korean-Braille translation, producing poor and unstable outputs with substantial disagreement from human judgments, …
Researchers introduced a multi-feature fusion framework for non-invasive semantic reconstruction from brain recordings, achieving state-of-the-art performance by combining static lexical representations (W2V) with dynami…
Researchers fine-tuned LLaMA 3 (8B) as a drop-in reranker for RAG pipelines, achieving gains of 14% in answer relevancy, 16% in context precision, 19% in answer similarity, and 21% in answer correctness over a cross-enco…
Researchers introduced CANDI-QA, a novel dataset for evaluating large language models (LLMs) on contextual alignment in specialized domains such as medical diagnostics and financial advisory. The dataset features expert-…
A new study from researchers publishing on arXiv extends noisy-channel inference models to explain intercomprehension, the zero-shot understanding of a related unfamiliar language (L2) by a speaker of a related language …
AgentCall, a new tool from Pattern AI, lets any AI agent join Google Meet, Zoom, and Microsoft Teams as a live participant with voice, video, and screenshare capabilities. The agent runs locally on the user's machine, wi…
A developer released GAP (Graph-Anchor Pyramid), a hierarchical context architecture that combines graph retrieval, topology-aware leaf grouping, and prompt-level semantic anchoring to solve the multi-hop relational blin…
Researchers introduced MET, a framework for multilingual moral reasoning in language models that uses culture-aware, theory-grounded prompting. The approach improved macro-F1 scores by up to 12.94 points on the MCLASH be…
Spotify launched a beta AI feature called Talk to Spotify, a conversational assistant that lets users search music, create playlists, manage queues, and ask about listening habits via text or voice. The feature is initia…
Hume AI released Real World VoiceEQ, a benchmark evaluating the human quality of voice interaction across more than 40 voice models using over 1 million human ratings, finding that no single model ranks among the top fiv…
Granola offers a device-capture transcription tool for Google Meet that does not join the call as a participant, avoiding the awkwardness of bots or the visible note-taking indicator of Gemini. The app works on Mac and W…
Researchers at Apple propose CLaRa (Continuous Latent Reasoning), a unified framework that bridges retrieval and generation in retrieval-augmented generation (RAG) by performing embedding-based compression and joint opti…
Granola offers iOS and Android apps that capture in-person conversations through a phone's microphone, producing a transcript, summary, and action items. The tool addresses the limitation of bot-based notetakers, which r…
Granola, a note-taking tool, captures client-call audio on a user's device without joining as a visible participant, preserving conversational candor. The tool provides transcripts, summaries, and action items after call…
Apple has released the iOS 27 public beta, introducing a new Siri AI that can access data across devices, answer complex questions, and perform onscreen awareness tasks. The update also includes AI-powered Spatial Refram…
Kyle launched Xtraktr, an iOS app that uses an on-device AI model to help users clarify their thoughts through a question-and-answer process, priced at $5.83 per month. The app runs entirely on the iPhone's chip to prote…
BotCritic audited an e-commerce support bot, applied its own suggested fixes to the system prompt, and re-tested it, raising the score from 86 to 91. The biggest improvement was in robustness, as the bot now handles ambi…
A developer built a private, local voice assistant using Python and OpenAI's Whisper model, combining speech-to-text with offline text-to-speech to avoid cloud dependencies. The assistant runs entirely on the user's mach…
Databricks released Genie One mobile apps for iOS and Android, bringing its data-smart AI coworker to business users on the go. The app lets users ask natural-language questions of their data, view dashboards, and access…