This paper describes a method for the automatic inference of structuraltransfer rules to be used in a shallow-transfer machine translation (MT) systemfrom small parallel corpora. The structural transfer rules are based onalignment templates, like those used in statistical MT. Alignment templates areextracted from sentence-aligned parallel corpora and extended with a set ofrestrictions which are derived from the bilingual dictionary of the MT systemand control their application as transfer rules. The experiments conductedusing three different language pairs in the free/open-source MT platformApertium show that translation quality is improved as compared to word-for-wordtranslation (when no transfer rules are used), and that the resultingtranslation quality is close to that obtained using hand-coded transfer rules.The method we present is entirely unsupervised and benefits from information inthe rest of modules of the MT system in which the inferred rules are applied.
展开▼
