Social sciences · OECD FoS 6.2 · China MoE 050102 语言学及应用语言学
Linguistics
语言学
L2Local breakthroughs → Steady Score 1.90 · Last reviewed 2026-10-22 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2024-06-05 AI · majorField · major
NLLB: scaling neural MT to 200 languages (Nature) ↗
200 languages in one model, +44% for low-resource
Why this level
NLLB extended neural MT to 200 languages (Nature 2024), helping low-resource documentation; LLMs are now evidence in debates on acquisition and syntax.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
Low-resource MTLanguage documentationLLMs & linguistic theory
Papers
- ELARSpeech: A 30K-Hour Dataset Across 675 Underrepresented Languages for Long-Tail Multilingual Speech Technologies
AI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)
2026-10-14 · arXiv
- AI-generated language remains systematically distinguishable from human communication
AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation
2026-10-12 · Proceedings of the National Academy of Sciences
- The shrinking landscape of linguistic diversity in the age of large language models
AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation
2026-08-24 · Nature Human Behaviour
- Scaling neural machine translation to 200 languages
200 languages in one model, +44% for low-resource
2024-06-05 · Nature
- Dissociating language and thought in large language models
Formal vs functional competence distinction
2024-06 · Trends in Cognitive Sciences
- A Benchmark for Learning to Translate a New Language from One Grammar Book
Turns learning from a descriptive grammar into a measurable task
2023-09-28 · arXiv
- Large Language Models Demonstrate the Potential of Statistical Learning in Language
LLMs enter the poverty-of-the-stimulus debate
2023-02-25 · Cognitive Science
- Robust Speech Recognition via Large-Scale Weak Supervision
Open multilingual ASR cuts transcription cost for fieldwork and corpora
2022-12-06 · arXiv
- Scaling Multilingual Pretraining of Pixel Language Models
Scales multilingual pretraining of pixel language models across languages, scripts and capacity, evaluated on 83 languages.
2026-10-21 · arXiv · Ilker Kesen, Jenny Kunz, Kasra Malihi, Desmond Elliott
- Benchmarking Arabic Dialects for Math Problems
DAR-MGSM, the first dialectal Arabic math benchmark; models drop on culturally localised versions.
2026-10-20 · arXiv · Noor Abo Mokh, Teresa Lynn, Nizar Habash, Ahmed Elshabrawy
- Global Security Blind Spots: Translation as a Safety Bypass in Low-Resource Languages
Across 78 low-resource languages, translation bypasses English safety alignment 21.9–44.4% of the time.
2026-10-19 · arXiv · Jalil Alva, David Harper, Abigail E. Thornton, Rafael S. Villalobos Alva
- DeepSEA: A Multilingual Evaluation Benchmark for Southeast Asia
A benchmark professionally translated into 12 Southeast Asian languages.
2026-10-19 · arXiv · Jiuzhou Han, Abhishek Bijay Mishra, Trevor Cohn, Shubham Mittal
- The World as Viewed by 136 Countries: An Open, Curated, Annotated, and Multilingual Dataset of 118 Million State Media Articles
WorldStateMedia: 118M state-media articles from 136 countries in 114 languages, extracted from pretraining corpora.
2026-10-18 · arXiv · Bryan Li, Samar Haider, Grace Kim, Fiona Luo
- MAS-Bai: Linguistically Informed Adaptive Multi-Agent Translation for Bowen-Script Bai
A linguistically informed multi-agent system for Bowen-script Bai–Chinese translation.
2026-10-16 · arXiv · Jingting Zheng, Hong Ren, Yuehan Cui, Shaolin Zhu
- From Triage to Impact: A Multi-Hop Reasoning Benchmark for Explainable Persian Hate Speech
A multi-hop benchmark for explainable Persian hate-speech analysis.
2026-10-14 · arXiv · Ghazal Kalhor, Sima Kamandloo, Mehrsa Shamsian, Behnam Bahrak
- Do Syntax Heads Influence Model Behavior?
Tests whether specialised syntax attention heads causally drive syntactic predictions; evidence is limited.
2026-10-13 · arXiv · William Timkey, Tal Linzen
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
Milestones 9
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Dataset
ELARSpeech: a 30K-hour speech dataset across 675 underrepresented languages from the Endangered Languages ArchiveAI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)
arXivPreprint -
Paper
AI-generated language remains systematically distinguishable from human communication (PNAS)AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation
PNAS -
Paper
The shrinking landscape of linguistic diversity in the age of large language models (Nat. Hum. Behav.)AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation
Nature Human Behaviour -
Paper
Evaluating frontier LLM translation for Lakota (AmericasNLP 2026)Frontier LLMs still unreliable on endangered languages
ACL AnthologyTo be verified -
Paper
NLLB: scaling neural MT to 200 languages (Nature)200 languages in one model, +44% for low-resource
Nature -
Paper
Dissociating language and thought in LLMs (Trends Cogn. Sci.)Formal vs functional competence distinction
Trends in Cognitive SciencesTo be verified -
Benchmark
MTOB: learning to translate Kalamang from a single grammar book (arXiv 2309.16575)Turns learning from a descriptive grammar into a measurable task
arXivPreprint -
Paper
LLMs demonstrate the potential of statistical learning in language (Cognitive Science)LLMs enter the poverty-of-the-stimulus debate
Cognitive Science -
Model
Whisper: robust multilingual speech recognition via large-scale weak supervision (arXiv 2212.04356)Open multilingual ASR cuts transcription cost for fieldwork and corpora
arXivPreprintVendor claim
No milestones match this filter.
More papers & repos 8
| Type | Title | Date |
|---|---|---|
| Paper | AI-generated language remains systematically distinguishable from human communication | 2026-10-12 |
| Paper | The shrinking landscape of linguistic diversity in the age of large language models | 2026-08-24 |
| Paper | Scaling neural machine translation to 200 languages | 2024-06-05 |
| Paper | Dissociating language and thought in large language models To be verified | 2024-06 |
| Paper | A Benchmark for Learning to Translate a New Language from One Grammar Book Preprint | 2023-09-28 |
| Paper | Large Language Models Demonstrate the Potential of Statistical Learning in Language | 2023-02-25 |
| Paper | Robust Speech Recognition via Large-Scale Weak Supervision Preprint | 2022-12-06 |
| Repo | facebookresearch/fairseq (NLLB branch) To be verified | — |
Latest news 5
- Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine translationarXiv cs.CLenPaperPreprint
- Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both ChallengedApple Machine Learning Researchen
- 斩获融资的这8家语言学习平台，都在用AI做什么？芥末堆zh