Effects of Parsing Errors on Pre-Reordering Performance for Chinese-to-Japanese SMT

Effects of Parsing Errors on Pre-reordering Performance for Chinese-to-Japanese SMT Dan Han1,2 Pascual Mart´ınez-G´omez2,3 Yusuke Miyao1,2 Katsuhito Sudoh4 Masaaki Nagata4 1The Graduate University For Advanced Studies 2National Institute of Informatics, 3The University of Tokyo 4NTT Communication Science Laboratories, NTT Corporation {handan,pascual,yusuke}@nii.ac.jp {sudoh.katsuhito,nagata.masaaki}@lab.ntt.co.jp Abstract Linguistically motivated reordering meth- ods have been developed to improve word alignment especially for Statistical Ma- chine Translation (SMT) on long dis- tance language pairs. However, since they highly rely on the parsing accuracy, it is useful to explore the relationship between parsing and reordering. For Chinese-to- Japanese SMT, we carry out a three-stage incremental comparative analysis to ob- serve the effects of different parsing errors on reordering performance by combining empirical and descriptive approaches. For the empirical approach, we quantify the distribution of general parsing errors along with reordering qualities whereas for the descriptive approach, we extract seven in- fluential error patterns and examine their correlation with reordering errors. 1 Introduction Statistical machine translation is a challenging and well established task in the community of compu- tational linguistics. One of the key components of statistical machine translation systems are word alignment techniques, where the words from sen- tences in a source language are mapped to words from sentences in a target language. When esti- mating the most appropriate word alignments, it is unfeasible to explore every possible word cor- respondence due to the combinatorial complexity. Considering local permutations of words might be effective to translate languages with a similar sen- tence structure, but these methods have a limited performance when translating sentences from lan- guages with different syntactical structures. An effective technique to translate sentences between distant language pairs is pre-reordering, where words in sentences from the source lan- guage are re-arranged with the objective to resem- ble the word order of the target language. Re- arranging rules are automatically extracted (Xia and McCord, 2004; Genzel, 2010), or linguisti- cally motivated (Xu et al., 2009; Isozaki et al., 2010; Han et al., 2012; Han et al., 2013). We work following the latter strategy, where the source sen- tence is parsed to find its syntactical structure, and linguistically-motivated rules are used in combi- nation with the structure of the sentence to guide the word reordering. The language pair under con- sideration is Chinese-to-Japanese, which despite their common roots, it is a well known language pair for their different sentence structure. However, syntax-based pre-reordering tech- niques are sensitive to parsing errors, but insight into their relationship has been elusive. The con- tribution of this work is two fold. First, we provide an empirical analysis where we quantify the aggre- gated impact of parsing errors on pre-reordering performance. Second, we define seven patterns of the most common and influential parsing errors and we carry out a descriptive analysis to exam- ine their relationship with reordering errors. We combine an empirical and descriptive approach to present a three-stage incremental comparative analysis to observe the effect of different parsing errors on reordering performance. In Section 2, after a brief description on …

Content truncated to 500 words