UK/Nigeria driver-coaching NLG study wins Best Evaluation award at INLG 2026
EhudReiter · x · 2026-10-22
Two former PhD students of NLG researcher Ehud Reiter, Iniakpokeikiye Thompson and Jawwad Baig, won the Best Evaluation award at INLG-2026 for their paper on localising data-to-text driver coaching for the UK and Nigeria.
- The paper studies content-level localisation in NLG, beyond language translation: generated feedback must fit local data availability and culture.
- Content depends on data: the UK system leverages reliable per-road speed limit data (from Damoov) to flag speeding; the Nigerian system cannot, due to missing data.
- Conversely, the Nigerian system checks drink-driving using a model trained on self-reported data, considered reasonably accurate since the behaviour carries less stigma locally.
Reiter argues this is an underexplored topic in NLG. Both systems' evaluations showed reductions in unsafe driving.
More from Research
- Yi Ma's new course "Principles of Deep Representation Learning" is on YouTube — YiMaTweets · 2026-10-22
- Tsinghua's SRHarness and MechBench push AI from curve-fitting to mechanism discovery — jiqizhixin · 2026-10-22
- New paper compares 11 rewards for training budget-aware multi-turn agents — adam_golinski · 2026-10-22
- 130K-sample RAG claim verification dataset lands on Hugging Face — MaziyarPanahi · 2026-10-22
- Culture is the moat: Scobleizer on cultural data for humanoid robot learning — Scobleizer · 2026-10-22
- Tencent Hunyuan's WorkForge builds 16.7K verifiable real-world environments for work agents — TencentHunyuan · 2026-10-22