A hybrid machine translation architecture guided by syntax

Abstract

This article presents a hybrid architecture which combines rule-based machine translation (RBMT) with phrase-based statistical machine translation (SMT). The hybrid translation system is guided by the rule-based engine. Before the transfer step, a varied set of partial candidate translations is calculated with the SMT system and used to enrich the tree-based representation with more translation alternatives. The final translation is constructed by choosing the most probable combination among the available fragments using monotone statistical decoding following the order provided by the rule-based system. We apply the hybrid model to a pair of distantly related languages, Spanish and Basque, and perform extensive experimentation on two different corpora. According to our empirical evaluation, the hybrid approach outperforms the best individual system across a varied set of automatic translation evaluation metrics. Following some output analysis to better understand the behaviour of the hybrid system, we explore the possibility of adding alternative parse trees and extra features to the hybrid decoder. Finally, we present a twofold manual evaluation of the translation systems studied in this paper, consisting of (i) a pairwise output comparison and (ii) a individual task-oriented evaluation using HTER. Interestingly, the manual evaluation shows some contradictory results with respect to the automatic evaluation; humans tend to prefer the translations from the RBMT system over the statistical and hybrid translations.Peer ReviewedPostprint (author’s final draft

What Do Professional Translators Do when Post-Editing for the First Time? First Insight into the Spanish-Basque Language Pair

“…Despite the inability of the two metrics to establish a specific level of quality, we have compared the scores with those reported in other experiments for Basque (Díaz de Ilarraza et al 2008, España-Bonet et al 2011, Labaka et al 2014, 2016 and confirmed that they are in the same range. Given the impact MT quality has on post-editing opportunities, we hoped to provide workshop participants with MT quality similar to that attainable by state-of-the-art generic systems for the Spanish-Basque language direction such as that reported in Labaka et al (2014).…”

“…Given the impact MT quality has on post-editing opportunities, we hoped to provide workshop participants with MT quality similar to that attainable by state-of-the-art generic systems for the Spanish-Basque language direction such as that reported in Labaka et al (2014). In terms of system differences, we observed that in week 6, when Google Translate was used, scores were substantially lower, confirming the experience of the IXA research group and the perception of workshop participants, who complained about the decrease in quality after completing the tasks for week 6.…”

Sustainable language technology for African languages

“…Also rule-based systems can be enhanced with statistical components (Habash and Monz 2009). It is claimed that better translation results can be achieved using a hybrid combination of these approaches (Labaka et al 2014). Nevertheless, the quality of translation continues to be relatively low and research continues for finding better quality testing methods (Felice and Specia 2013;Birch et al 2010).…”

A Hybrid Approach Using Phrases and Rules for Hindi to English Machine Translation

“…Apart from the WMT shared task translation systems [18] presented a hybrid machine translation architecture guided by syntax. In this work, the authors develop a hybrid translation system guided by the rule based machine translation engine and, before transference, few partial candidate translations given by the statistical translation schemes.…”