Hindi
PulseAugur coverage of Hindi — every cluster mentioning Hindi across labs, papers, and developer communities, ranked by signal.
15 day(s) with sentiment data
-
New dataset tackles Hindi media bias in Indian political news
Researchers have developed a new dataset focused on media bias in Hindi political news articles from India. This dataset includes 448 articles from four leading Hindi news agencies, annotated by 14 experts for four type…
-
New multi-word tokenizer Havon v4 outperforms GPT-4o on 204 languages
Researchers have developed a new multi-word tokenizer called Havon v4, which demonstrates improved efficiency over GPT-4o's tokenizer across 204 languages. Havon v4, particularly when trained with a larger vocabulary, r…
-
Indian languages face 'token tax' with AI models, research finds
A new research paper, "The Indic Token Tax in 2026," highlights significant disparities in how different language models tokenize Indian languages compared to English. The study found that tokenizers for models like GPT…
-
Developer builds 40M-parameter multilingual AI for delivery riders in 28 minutes
A developer has created a small, multilingual language model designed to assist delivery riders, particularly in areas with poor or no network connectivity. This 40-million-parameter model, trained in under 30 minutes o…
-
Otter.ai expands multilingual transcription to 31 languages
Otter.ai, an AI-powered note-taking service, has significantly expanded its multilingual transcription capabilities. The company is introducing a new feature that can follow conversations even when speakers switch langu…
-
AI metrics fail romanized Indic languages, new study finds
A new research paper highlights a significant gap in evaluating AI models for romanized Indic languages, where automatic metrics fail to accurately assess generation quality. The study found that metrics like BGE-M3 and…
-
GPT-2 self-training alters word dominance and entropy profiles
Researchers have explored the effects of recursive self-training on language models, specifically fine-tuning GPT-2 on its own output over multiple generations. They observed that as the training temperature increases, …
-
New transliteration models aim to simplify mixed-script typing
Developers have created UnimeType, a new input method and browser converter, to address the challenges of mixed-script typing. They trained two specialized transliteration models, Transliteration 2.0 (0.8B and 4B), base…
-
Familiar scripts, not related languages, unlock AI understanding of unseen languages
Researchers have demonstrated that the familiarity of a script, rather than linguistic relatedness, is key to enabling AI models to understand unseen languages. In experiments with a model called GlyphMind, which proces…
-
Anthropic's Claude AI shows language-dependent response variations
Anthropic's Claude AI model exhibits distinct response patterns when queried in different languages, showing a more encouraging tone when responding in Hindi compared to English. This linguistic variability suggests tha…
-
New TRIDENT framework and SHAKTI benchmark tackle Hindi AI-text detection
Researchers have developed TRIDENT, a novel framework for detecting AI-generated text in Hindi, a language previously underserved by such tools. TRIDENT utilizes a round-trip translation approach, analyzing semantic con…
-
New Bhasha model identifies languages word-by-word, struggles with Tamil
A new open-source model named Bhasha has been developed to identify the language of individual words within code-mixed Indian text, supporting languages like Hinglish, Tanglish, and Kanglish. While Bhasha performs well …
-
Bengali coreference resolution research questions multilingual model assumptions
A new research paper on arXiv challenges common assumptions about multilingual transformer models, particularly for coreference resolution in low-resource languages like Bengali. The study found that a small amount of h…
-
Maya Research releases multilingual streaming TTS model Maya 2 Native
Maya Research Program has released Maya 2 Native, a streaming text-to-speech model supporting 11 Indic languages and English. The model generates 24 kHz speech with 20 built-in voices and features zero-shot voice clonin…
-
New meta-learning framework enhances code-switched ASR accuracy
Researchers have developed SyRA-CS, a novel meta-learning framework designed to improve automatic speech recognition (ASR) for code-switched speech. This approach uses synthetic code-switched data for meta-training, fol…
-
Amazon Polly upgrades Generative engine with new voices and languages
Amazon Polly has released a significant upgrade to its Generative engine, enhancing speech output to be more natural, expressive, and accurate. This update introduces 14 new voices across 9 locales, including the first-…
-
AI tutor Acharya 2.0 aids Hindi students with RAG
A new AI tutor named Acharya 2.0 has been developed to assist Hindi-speaking students with self-learning subjects. This application utilizes retrieval-augmented generation (RAG) to enhance large language models (LLMs) b…
-
Gemini 4 Argon tokenizer shifts token costs across languages
Google's Gemini 4 Argon model features a new tokenizer that alters token counts for various languages. While Korean and Japanese texts require fewer tokens (8% reduction), Indian languages like Hindi, Tamil, and Kannada…
-
New LoWAgent benchmark reveals performance gaps in AI web agents for low-resource languages
A new benchmark called LoWAgent has been introduced to evaluate the performance of web agents across low-resource languages, addressing a gap in current testing methodologies. The benchmark includes six localized websit…
-
LLM pipelines show broad cross-lingual reusability, but domain-specific validation is key
A new arXiv paper investigates the cross-lingual reusability of large language model (LLM) pipelines, specifically focusing on financial extraction, multilingual reasoning, and semantic screening. The study found that p…