Natural Language Processing news and analysis on Web Pulse: 5477 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
IBM researchers released Granite Speech 5.0 Turbo CTC, two 470-million-parameter English speech-recognition models that remove the language-model decoder to achieve over 20 times the throughput of earlier Granite Speech …
Oro announced a $3 million strategic funding round co-led by MH Ventures and Mapleblock Capital, bringing its total raised to $4 million, to advance its natural-language DeFi platform that converts user goals into multi-…
IBM released two compact 470M-parameter English speech recognition models, granite-speech-5.0-470m-turboctc and granite-speech-5.0-470m-turboctc-nc, achieving over 12,600 RTFx on an NVIDIA H200 GPU and transcribing more …
De-stink, a free browser tool, lets users paste prose to detect and remove AI writing tells without using LLMs, relying instead on mechanical rule-based fixes. The tool, created by an unnamed developer, offers a 'stink s…
Oro, a Dubai-based financial AI platform, raised $3 million in funding led by MH Ventures and MapleBlock Capital, bringing its cumulative funding to $4 million. The platform, which has over 350,000 active users across 80…
A new accelerator program, the Conversational Data Intelligence Accelerator, aims to give business users an Enterprise AI Data Assistant that converts natural-language questions into validated SQL and decision-ready visu…
Voicecard, a new command-line tool by an independent developer, resolves contradictions in rambling voice-note transcripts, achieving 15 of 18 on its adversarial test suite with the local qwen2.5:7b model. The tool, inst…
Telnyx has released an example application that turns live phone calls into actionable data using its Edge Compute platform. The edge-call-transcription-agent answers inbound calls, streams real-time transcription, store…
Vladimir Besk released the complete personal archive of Konstantin Tsiolkovsky as a dataset on Hugging Face, comprising 51,008 sheets across 2,019 archival files under CC0. The archive includes 1,759 manuscript-typescrip…
Researchers introduced Semantic Compression Trees (SCT), a hierarchical index storing only semantic residuals, and found it matches dense retrieval on answer quality (0.274 vs. 0.277 F1, p = 0.37) using 30% fewer context…
A new study from the KSE-Web project finds that BM25 retrieval achieves the strongest performance for Khmer semantic search, reaching 0.943 Recall and 0.876 nDCG, outperforming hybrid and dense retrieval methods. The res…
A new arXiv study (2608.21384v1) finds that Ukrainian and other Cyrillic-script languages face 68-121% token overhead on modern tokenizers and 220% on the older cl100k, compared to English, across nine production tokeniz…
Researchers introduced Wazobia Eval, a benchmark for evaluating Nigerian Pidgin emotion understanding, sarcasm detection, and cultural reasoning, built on a manually annotated dataset of over 550 examples with a 16-categ…
Researchers propose SCoRe, an LLM-based framework for Knowledge-based Visual Question Answering that improves reasoning by structuring context into explicit entity-relation triplets. On OK-VQA and A-OKVQA benchmarks, SCo…
A new arXiv study (2608.21408v1) comparing hate speech classification techniques for Roman Urdu found that parameter-efficient fine-tuning (PEFT) with LoRA and prompt engineering methods offer computationally efficient s…
A new arXiv paper (2608.21364v1) distinguishes two update operators in incremental narrative interpretation: revision-driven update, which retracts prior structure non-monotonically, and delayed elaboration, which monoto…
Researchers propose KBMR, the first MLLM-based embedding retriever for Knowledge-Based Visual Question Answering (KB-VQA), which outperforms CLIP baselines by up to 14.7% in retrieval Recall@1 and 9.4% in end-to-end VQA …
A new arXiv preprint (2608.21794v1) finds that GUI grounding evaluations using sentence embeddings often conflate visible-label recovery with semantic grounding, as lexical baselines remain competitive at top-1 across th…
A new arXiv paper (2608.21376v1) investigates how human judges and four open-source LLMs evaluate citations in preference data for scientific question answering, finding that humans prefer more diverse citations but fewe…
Researchers introduced L"etzCross, a benchmark for cross-lingual page-level retrieval over Luxembourgish PDF documents, and found that ColPali-style page-image retrievers outperform OCR-based text-only retrievers across …