Alkhder team fine-tunes NileChat-3B for Arabic dialect translation
Team Alkhder presented a system for the AlexandriaX-2026 shared task, focusing on translating English dialogue into thirteen distinct Arabic dialects. The approach involves fine-tuning the NileChat-3B-Base model using QLoRA on the Alexandria parallel dialogue corpus. A single set of adapter weights was trained to handle all dialects, with translations conditioned on specific factors such as the…
Key points
- Team Alkhder fine-tuned NileChat-3B-Base with QLoRA for thirteen Arabic dialects.
- The system uses one adapter set for all dialects, conditioned on context and roles.
- It achieved a spBLEU of 20.36 and chrF++ of 36.64 on the AlexandriaX-2026 test set.
During inference, the system translates conversations autoregressively, feeding its own previous outputs back as context without relying on gold target-side test information. The training process is efficient, taking about one hour, while translating the private test set takes under two hours. The system achieved a country-macro-averaged spBLEU score of 20.36 and a chrF++ score of 36.64 on both tracks of the competition.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic says small AI may self‑replicate by year‑end after experiment with GLM‑5.2 · 1 src
- Anthropic co-founder Jack Clark says AI could boost US GDP by nearly 50% by 2030 · 13 src
- Hybrid Reasoning Agents improve LLM performance by 10.4% and reduce costs up to 3.7x · 1 src
- Chinese team proves 3D Ghys conjecture using Intern‑S2 AI model · 2 src
- StarSkirmish Bench v0.2 ranks GPT-6 Astra as top StarCraft bot writer · 2 src
Comments
via GitHub Discussions