The TransBank Aligner: Cross-Sentence Alignment with Deep Neural Networks

Abstract

Sentence-aligned parallel bilingual corpora are the main and sometimes the only required resource for training Statistical and Neural Machine Translation systems. We propose an end-to-end deep neural architecture for sentence alignment. In addition to one-to-one alignment, our aligner can perform cross alignment as well. We used three language pairs from Europarl corpus and an English-Persian corpus to generate an alignment dataset. Using this dataset, we tested our system both in isolation and in an SMT system. In both settings, we obtained significantly better results compared to two competitive baselines.