Chinese-Portuguese Machine Translation: A Study on Building Parallel Corpora from Comparable Texts

Chinese–Portuguese Machine Translation: A Study on Building Parallel Corpora from Comparable Texts Siyou Liu♥ Longyue Wang♠ Chao-Hong Liu♠ ♥ School of Languages and Translation, Macao Polytechnic Institute, Macao S.A.R., China ♠ ADAPT Centre, School of Computing, Dublin City University, Ireland violetal@ipm.edu.mo {longyue.wang, chaohong.liu}@adaptcentre.ie Abstract Although there are increasing and significant ties between China and Portuguese-speaking countries, there is not much parallel corpora in the Chinese–Portuguese language pair. Both languages are very populous, with 1.2 billion native Chinese speakers and 279 million native Portuguese speakers, the language pair, however, could be considered as low-resource in terms of available parallel corpora. In this paper, we describe our methods to curate Chinese–Portuguese parallel corpora and evaluate their quality. We extracted bilingual data from Macao government websites and proposed a hierarchical strategy to build a large parallel corpus. Experiments are conducted on existing and our corpora using both Phrased-Based Machine Translation (PBMT) and the state-of-the-art Neural Machine Translation (NMT) models. The results of this work can be used as a benchmark for future Chinese–Portuguese MT systems. The approach we used in this paper also shows a good example on how to boost performance of MT systems for low-resource language pairs. Keywords: Chinese–Portuguese, Low-Resource, Statistical Machine Translation, Neural Machine Translation, Parallel Corpus 1. Introduction Chinese and Portuguese are widely used by a large amount of people in the world. With the development of eco- nomic globalization, communications between Chinese and Portuguese-speaking countries are increasing in a fast path. Translation services between these two languages is be- coming more and more demanding. However, Chinese and Portuguese belong to distinct language families (Sino- Tibetan and Romance, respectively) and only a relative much smaller proportion of people have bilingual profi- ciency of the language pair. Therefore, the use of Chinese– Portuguese MT systems to provide auxiliary translation ser- vices between the two sides is highly demanded. Pivot-based machine translation is a commonly used method when large quantities of parallel data are not read- ily available for some language pairs. Utiyama and Isahara (2007), Wu and Wang (2007), Bertoldi et al. (2008) inves- tigated phrase-level, sentence-level and system-level pivot strategies for low resource translation in SMT. A pivot lan- guage, which is usually English, can bridge the source and target languages and make translation possible. However, the domains of these two are often different and thus results in low performance and even ambiguities. A few researchers have investigated how to improve the Chinese–Portuguese MT by incorporating linguistic knowl- edge into the systems. For instance, Wong and Chao (2010) proposed a hybrid MT system combining rule-based and example-based components. Oliveira et al. (2010) explored Constraint Synchronous Grammar parsing for SMT. Lu et al. (2014) and Liu and Leal (2016) focused on specific lin- guistic phenomena (i.e. present articles and temporal adver- bials) in translation. Although NMT has been rapidly de- veloped in recent years (Kalchbrenner and Blunsom, 2013; Sutskever et al., 2014; Bahdanau et al., 2015; Tu et al., 2016), Chinese–Portuguese MT has not received much at- tention using NMT …

Content truncated to 500 words