JSON data block
{
"anthology_id": "2026.loresmt-1.7",
"abstract": "Neural Machine Translation (NMT) models for low-resource languages suffer significant performance degradation under domain shift. We quantify this challenge using <b>Dhao</b>, an indigenous language of Eastern Indonesia with no digital footprint beyond the New Testament (NT). When applied to the unseen Old Testament (OT), a standard NMT model fine-tuned on the NT drops from an in-domain score of 36.17 chrF++ to 27.11 chrF++. To recover this loss, we introduce a <b>hybrid framework</b> where a fine-tuned NMT model generates an initial draft, which is then refined by a Large Language Model (LLM) using Retrieval-Augmented Generation (RAG). The final system achieves 35.21 chrF++ (<tex-math>+8.10</tex-math> recovery), effectively matching the original in-domain quality. Our analysis reveals that this performance is driven primarily by the <b>number of retrieved examples</b> rather than the choice of retrieval algorithm. Qualitative analysis confirms the LLM acts as a robust \"safety net,\" repairing severe failures in zero-shot domains.",
"authors": [
{
"first": "David Samuel",
"last": "Setiawan",
"id": "david-samuel-setiawan/unverified"
},
{
"first": "Raphaël",
"last": "Merx",
"id": "raphael-merx"
},
{
"first": "Jey Han",
"last": "Lau",
"id": "jey-han-lau"
}
],
"authors_old": "David Samuel Setiawan | Raphael Merx | Jey Han Lau",
"authors_new": "David Samuel Setiawan | Raphaël Merx | Jey Han Lau"
}
