Inferring Shallow-Transfer Machine Translation Rules from Small Parallel Corpora

Abstract

This paper describes a method for the automatic inference of structural transfer rules to be used in a shallow-transfer machine translation (MT) system from small parallel corpora. The structural transfer rules are based on alignment templates, like those used in statistical MT. Alignment templates are extracted from sentence-aligned parallel corpora and extended with a set of restrictions which are derived from the bilingual dictionary of the MT system and control their application as transfer rules. The experiments conducted using three different language pairs in the free/open-source MT platform Apertium show that translation quality is improved as compared to word-for-word translation (when no transfer rules are used), and that the resulting translation quality is close to that obtained using hand-coded transfer rules. The method we present is entirely unsupervised and benefits from information in the rest of modules of the MT system in which the inferred rules are applied.1. Small compared to the size of corpora commonly used to build corpus-based MT systems (Och, 2005).

A generalised alignment template formalism and its application to the inference of shallow-transfer machine translation rules from scarce bilingual corpora

“…Experiments conducted using five different language pairs with the free/open-source rule-based MT platform Apertium show that translation quality significantly improves when compared to the method proposed by Sánchez-Martínez and Forcada (2009), and is close to that obtained using handcrafted rules. For some language pairs, our approach is even able to outperform them.…”

“…Unlike previous approaches in literature, our formalism does not require linguistic knowledge about the languages involved in the translation. Moreover, it is the first time that conflicts between rules are resolved by choosing the most appropriate ones according to a global minimisation function rather than proceeding in a pairwise greedy fashion.Experiments conducted using five different language pairs with the free/open-source rule-based MT platform Apertium show that translation quality significantly improves when compared to the method proposed by Sánchez-Martínez and Forcada (2009), and is close to that obtained using handcrafted rules. For some language pairs, our approach is even able to outperform them.…”

“…The approach by Sánchez-Martínez and Forcada (2009) is based on the alignment template (AT) approach (Och, 2002;Och and Ney, 2004) initially proposed in the context of SMT. An AT performs a generalisation of bilingual phrase 4 pairs (pairs of segments which are mutual translations) by using word classes rather than the words themselves.…”

“…An AT performs a generalisation of bilingual phrase 4 pairs (pairs of segments which are mutual translations) by using word classes rather than the words themselves. Sánchez-Martínez and Forcada (2009) adapted the AT approach for their application in RBMT by extending the ATs with a set of restrictions in order to control their application as shallowtransfer rules. An extended AT (henceforth, EAT) is defined as a tuple z = (S, T, A, R), consisting of a sequence S of SL word classes, the corresponding sequence T of TL word classes, a set A of pairs of word class indexes (i, j) with the alignment information between the word classes in the two sequences, and a set R of restrictions over the TL inflection information that the words to translate need to meet.…”

“…The following section presents a brief description of the approach by Sánchez-Martínez and Forcada (2009), stressing its main limitations and summarising how they are overcome in our approach. Sections 3 and 4, respectively, introduce the AT formalism used in our approach and the method employed to extract ATs using this formalism.…”