arXiv Research
Abstract:LargeLanguage…▽ MoreLargeLanguageModels(LLMs) have been progressively exhibiting there capabilities in various areas of research. The performance of the LLMs in acute maternal healthcare area, predominantly in low resourcelanguageslike Telugu, Hindi, Tamil,Urduetc are still unstudied. This study presents how ChatGPT-4o, GeminiAI, and Perplexity AI respond to pregnancy related questions as
Open Full News
arXiv Research
Abstract:…yet it remains underrepresented in multimodal evaluation. To address this gap, we introduce BanglaVerse, a culturally grounded benchmark for evaluating multilingual vision-language…▽ MoreBangla culture is richly expressed through region, dialect, history, food, politics, media, and everyday visual life, yet it remains underrepresented in multimodal evaluation. To address this gap, we int
Open Full News
arXiv Research
Abstract:In this article, we present a gold-standard benchmark dataset for BiomedicalUrduNamed Entity Recognition (BioUNER), developed by crawling health-related articles from online…▽ MoreIn this article, we present a gold-standard benchmark dataset for BiomedicalUrduNamed Entity Recognition (BioUNER), developed by crawling health-related articles from onlineUrdunews portals, medical prescription
Open Full News
arXiv Research
Abstract:…and inadequate formal curricula, leaving women with few trusted resources to lean on. In response to these challenges, we introduce a WhatsApp-based chatbot powered by a largelanguage…▽ MoreMenstrual health education (MHE) in Pakistan is constrained by cultural taboos and inadequate formal curricula, leaving women with few trusted resources to lean on. In response to these challenges, we
Open Full News
arXiv Research
Abstract:Pashto is absent from Whisper's pre-training corpus despite being one of CommonVoice's largestlanguagecollections, leaving off-the-shelf…▽ MorePashto is absent from Whisper's pre-training corpus despite being one of CommonVoice's largestlanguagecollections, leaving off-the-shelfmodelsunusable: all Whisper sizes output Arabic, Dari, orUrduscript on Pashto audio, achieving word error rates
Open Full News
arXiv Research
Abstract:Word error rate (WER) is the dominant metric for automatic speech recognition, yet it cannot detect a systematic failure mode:modelsthat produce fluent output in the wrong writing system. We define Script Fidelity Rate (SFR), the fraction of hypothesis characters in the target script block, computable without reference transcriptions, and report the first…▽ MoreWord error rate (WER) is th
Open Full News
arXiv Research
Abstract:Across multiplelanguagepairings (English $\to$ \{Spanish, French, German, Mandarin, Japanese,…▽ MoreAcross multiplelanguagepairings (English $\to$ \{Spanish, French, German, Mandarin, Japanese,Urdu, Cantonese\}), we find reasoning errors in translation. To quantify how often these reasoning errors occur, we leverage an automated annotation protocol for reasoning evaluation wherein the goa
Open Full News
arXiv Research
Abstract:Emotion classification in multilingual settings remains constrained by the scarcity of annotated data: existing corpora are predominantly English, single-label, and cover fewlanguages. We address this gap by constructing a large-scale synthetic training corpus of over 1M multi-label samples (50k per…▽ MoreEmotion classification in multilingual settings remains constrained by the scarcity
Open Full News
arXiv Research
Abstract:Many benchmarks show that largelanguage…▽ MoreMany benchmarks show that largelanguagemodelscan answer direct questions about culture. We study a different question: do they also change how they speak when culture is only implied by the situation? We evaluate 60 culturally grounded conversational scenarios across fivelanguagesin three conditions: a neutral baseline (Prompt A), an explicit
Open Full News
arXiv Research
Abstract:Facilitating cross-lingual transfer in multilinguallanguage…▽ MoreFacilitating cross-lingual transfer in multilinguallanguagemodelsremains a critical challenge. Towards this goal, we propose an embedding-based data augmentation technique called XITE. We start with unlabeled text from a low-resource targetlanguage, identify an English counterpart in a task-specific training corpus using em
Open Full News
Web
Pakistan to Develop Urdu LLM for Generative AI
Open Full News
Research
Assessing the Feasibility of Lightweight Whisper Models for Low ...
Open Full News