Scaling up a Hybrid MT System: From low to full resources

Vincent Vandeghinste

doi:10.52034/lanstts.v8i.245

Abstract

This article describes a hybrid approach to machine translation (MT) that is inspired by the rule-based, statistical, example-based, and other hybrid machine translation approaches currently used or described in academic literature. It describes how the approach was implemented for language pairs using only limited monolingual resources and hardly any parallel resources (the METIS-II system), and how it is currently implemented with rich resources on both the source and target side as well as rich parallel data (the PaCo-MT system). We aim to illustrate that a similar paradigm can be used, irrespectively of the resources available, but of course with an impact on translation quality.

Highlights

There are myriad approaches to machine translation, but none have shown acceptable levels of translation quality from an end-user’s perspective
machine translation (MT) systems that exist today reach at best a level of translation quality that might speed up the work of a human translator
In this paper we described a hybrid approach towards machine translation, seeking to combine the strengths and avoid the weaknesses of the classic approaches towards MT

Summary

Introduction

There are myriad approaches to machine translation, but none have shown acceptable levels of translation quality from an end-user’s perspective. MT systems that exist today reach at best a level of translation quality that might speed up the work of a human translator. The most widespread use of MT systems are online translation services, which are available through many Web sites and provide a gist translation of the source language text. MT systems in limited domains are occasionally sufficiently accurate to be useful for real translation tasks. In rule-based machine translation, the development of a new language pair, especially with so-called ‘smaller’ languages, is rather rare due to high costs and long development times. These expenses depend on the availability of parallel corpora containing aligned sentences in both the source and target language

Objectives

Results

Conclusion