METIS II: Statistical Machine Translation Using Monolingual Corpora

Start Date: 01/10/2004
End Date: 30/09/2007
Funding: Information Society Technologies (IST)
Project Leader: Markantonatou Stella

METIS II was the continuation of the successful assessment project METIS I (IST-2001-32775). Like METIS I, METIS II also aimed at constructing high quality free text translations by relying on pattern matching/statistical techniques and by retrieving the basic stock for translations from large annotated monolingual corpora. The translation process was based on chunk patterns identified in both the source (SL) and the target (TL) languages.

In particular, the algorithm received as input a sequence of source language chunk patterns, which have been translated employing bilingual lexica and a limited set of rules, and attempted (i) to detect similar patterns within the large TL monolingual corpus and (ii) to collate these patterns to a meaningful sentence.

In METIS II several languages have been studied (Dutch, German, Greek and Spanish), while the consortium also investigated the development of semi-automatic techniques allowing system adaptation to translation needs.