Reconsidering SMT Over NMT for Closely Related Languages: A Case Study of Persian-Hindi Pair
Waisullah Yousofi, Pushpak Bhattacharyya
TL;DR
This study investigates whether phrase-based SMT can outperform Transformer-based NMT for closely related languages using Persian–Hindi as a case study. Leveraging a moderate-sized parallel corpus assembled from multiple sources and filtered with LABSE, the authors compare PBSMT (Moses) against a Transformer NMT (OpenNMT). The results show SMT achieving a BLEU of 66.32, significantly higher than the NMT score of 53.7, with additional experiments revealing that romanization degrades performance and reversing script direction harms translation. The work highlights that SMT can be a competitive, environmentally friendlier option for certain language pairs and data regimes, and it outlines future directions such as cross-lingual embeddings and pivot-based NMT approaches.
Abstract
This paper demonstrates that Phrase-Based Statistical Machine Translation (PBSMT) can outperform Transformer-based Neural Machine Translation (NMT) in moderate-resource scenarios, specifically for structurally similar languages, like the Persian-Hindi pair. Despite the Transformer architecture's typical preference for large parallel corpora, our results show that PBSMT achieves a BLEU score of 66.32, significantly exceeding the Transformer-NMT score of 53.7 on the same dataset. Additionally, we explore variations of the SMT architecture, including training on Romanized text and modifying the word order of Persian sentences to match the left-to-right (LTR) structure of Hindi. Our findings highlight the importance of choosing the right architecture based on language pair characteristics and advocate for SMT as a high-performing alternative, even in contexts commonly dominated by NMT.