Pivot Machine Translation Using Chinese as Pivot Language

Abstract

Pivoting through a popular language with more parallel corpora available (e.g. English and Chinese) is a common approach to build machine translation (MT) systems for low-resource languages. For example, to build a Russian-to-Spanish MT system, we could build one system using the Russian-Spanish corpus directly. We could also build two systems, Russian-to-English and English-to-Spanish, as the resources of the two language pairs are much larger than the Russian-Spanish pair, and use them cascadingly to translate texts in Russian into Spanish by pivoting through English. There are, however, some confusing results on the Pivot MT approach in the literature. In this paper, we reviewed the performance of Pivot MT with the United Nations Parallel Corpus v1.0 (UN6Way) using both English and Chinese as pivot languages. We also report our system performance on the CWMT 2018 Pivot MT shared task, where Japanese patent sentences are translated into English using Chinese as the pivot language.

Reference Language based Unsupervised Neural Machine Translation

“…Pivot translation has long been studied in statistical machine translation (Wu and Wang, 2007;Utiyama and Isahara, 2007;Paul et al, 2009), supervised NMT (Cheng et al, 2017;Liu et al, 2018;Kim et al, 2019), and UNMT (Leng et al, 2019) as a means of improving the performance of low/zero-resource translations.…”

Simple Automatic Post-editing for Arabic-Japanese Machine Translation

“…A commonly used solution for addressing the lack of parallel data for a certain language pair is to pivot through a third (pivot or bridge) language that has enough shared parallel data with the two languages of interest (Hajič et al, 2000;Utiyama and Isahara, 2007;Wu and Wang, 2007;Bertoldi et al, 2008;Habash and Hu, 2009;Koehn et al, 2009;Liu et al, 2019). Pivoting has been shown useful for closely related languages (Hajič et al, 2000) as well as unrelated languages (Habash and Hu, 2009;Liu et al, 2019).…”

“…A commonly used solution for addressing the lack of parallel data for a certain language pair is to pivot through a third (pivot or bridge) language that has enough shared parallel data with the two languages of interest (Hajič et al, 2000;Utiyama and Isahara, 2007;Wu and Wang, 2007;Bertoldi et al, 2008;Habash and Hu, 2009;Koehn et al, 2009;Liu et al, 2019). Pivoting has been shown useful for closely related languages (Hajič et al, 2000) as well as unrelated languages (Habash and Hu, 2009;Liu et al, 2019). English is the most commonly used pivot language simply because systems/corpora to and from English are typically more available than other possible pivots, although this is not exclusive - (Liu et al, 2019) translated Japanese patent text to English through Chinese as a pivot.…”

“…Pivoting has been shown useful for closely related languages (Hajič et al, 2000) as well as unrelated languages (Habash and Hu, 2009;Liu et al, 2019). English is the most commonly used pivot language simply because systems/corpora to and from English are typically more available than other possible pivots, although this is not exclusive - (Liu et al, 2019) translated Japanese patent text to English through Chinese as a pivot. The simplest pivoting strategy is sentence pivoting, where we build two independent systems and pipeline them.…”

“…A bottleneck for some language pairs is the lack of direct parallel translation data sets. This issue has been addressed through alternative solutions such as zeroshot models (Johnson et al, 2016) or pivoting techniques (Utiyama and Isahara, 2007;Habash and Hu, 2009;Liu et al, 2019). Although such methods can be successful in getting a strong baseline, c 2019 The authors.…”