🤖 AI Summary
Arabic–English parallel corpora are scarce, especially high-quality benchmarks covering challenging domains such as law and literature and supporting many-to-many sentence alignment. To address this gap, we propose the first generative sentence alignment framework tailored to low-resource language pairs: it leverages domain-adapted large language models (LLMs), introduces a novel benchmark—ArabAlign—with a “Hard” subset featuring complex nested structures, ellipsis, and cross-sentence coreference, and defines a fine-grained evaluation protocol. Our method achieves an F1 score of 85.5% on the hard test set, outperforming the prior state of the art by 9 percentage points. Crucially, it is the first approach to robustly model non-one-to-one alignments (e.g., one-to-many, many-to-one) in legal and literary texts. We publicly release a high-quality dataset, open-source implementation, and standardized evaluation tools—filling a critical gap in fine-grained alignment research for low-resource languages.
📝 Abstract
High-quality parallel corpora are essential for Machine Translation (MT) research and translation teaching. However, Arabic-English resources remain scarce and existing datasets mainly consist of simple one-to-one mappings. In this paper, we present AlignAR, a generative sentence alignment method, and a new Arabic-English dataset comprising complex legal and literary texts. Our evaluation demonstrates that "Easy" datasets lack the discriminatory power to fully assess alignment methods. By reducing one-to-one mappings in our "Hard" subset, we exposed the limitations of traditional alignment methods. In contrast, LLM-based approaches demonstrated superior robustness, achieving an overall F1-score of 85.5%, a 9% improvement over previous methods. Our datasets and codes are open-sourced at https://github.com/XXX.