Multilingual NLP roadmap: researcher maps open problems from data selection to LID across 350+ languages
StellaLisy · x · 2026-10-15
Researcher Niyati Bafna published a curated list of open questions in multilingual NLP ahead of EMNLP, aimed at people hunting for research projects.
- Data: how to select and curate training/eval data under tight budgets in low-resource settings, and whether filtering noisy samples helps multilingual pretraining
- Language identification: beyond a few hundred high-resource languages, LID becomes ill-defined due to closely related languages, sparse data, and noisy web text — hierarchical models like LIMIT (350+ languages) are one response
- Data mixtures and sampling: handling severe language imbalance in training
She notes the list is LLM-focused, skips (computational) linguistics, and reflects personal taste rather than a canonical survey; she invites discussion at EMNLP.
More from Research
- Colossal Uses AI to Rebuild Extinct Genomes and Track Endangered Species — IgorCarron · 2026-10-15
- Thinking Machines ships Dimuon optimizer for LoRA fine-tuning, beats AdamW in fewer steps — simonguozirui · 2026-10-15
- NVIDIA report: agentic setup lets GPT-6 Astra hit 74.8% on RoboLab-120 robot manipulation — YuXiang_IRVL · 2026-10-15
- Bittensor Subnet 55 completes 10M synthetic CRISPR experiments with 248 competing miners — markjeffrey · 2026-10-15
- Logit lens shows Qwen3-1.7B's refusal to be shut down depends on how many layers you train — Sauers_ · 2026-10-15
- DeepSeek's first LLM paper was literally titled "Scaling Open-Source Language Models with Longtermism" — teortaxesTex · 2026-10-15