PaperScope
LIVE · 2026-09-29 05:40 UTC

Reference-Grounded Data Curation for Instruction-Following Thai-English Machine Translation

Thodsaporn Chay-intr, Krittapad Harnchang, Mahannop, Thabua, Kobkrit Viriyayudhakorn, Thanaruk Theeramunkong

Latestcs.CLcs.LGcs.AIcs.CV
arXiv ID
2609.34770 v1
Category
Submitted
2026-09-28

Abstract

Instruction-following machine translation (IF-MT) requires respecting prompt-level rules on terminology, formatting, and register. Rule compliance typically trades off against translation quality, a tension that general-purpose IF data augmentation methods do not address. We propose Reference-Grounded Data Curation, a two-phase pipeline that extracts every supervised constraint from a reference translation that already satisfies it, ensuring feasibility by construction. Phase 1 applies Instruction-Following Difficulty (IFD) scoring to retain the hardest-but-learnable instances from an English-Thai parallel pool. Phase 2 extracts constraints from each reference target and keeps only generations satisfying every constraint, yielding the 1.97M-record Grounded dataset. We fine-tune open-weight bases on Grounded to produce ChindaMT, a Thai-English translation family at 4B, 2B, and 0.8B parameters. Under length-controlled pairwise judging, ChindaMT outperforms or matches every same-size baseline at every tier on both plain translation and under explicit rules, reaching up to a 68.4% win rate against the strongest baseline. The recipe transfers cleanly across Qwen generations. We release model weights, the Grounded dataset, and evaluation suites.

Comment: Accepted at AACL-IJCNLP 2026 (Main Conference)

arXiv abs page · PDF · same-day batch