Skip to content
PetterHaugereid edited this page Jul 26, 2012 · 27 revisions

Japanese English Machine Translation

This page describes the Jaen MT system, an MT system based on the LOGON architecture, using Jacy for the source language analysis and the ERG for the target language generation. In order to degrade gracefully in the presence of input we cannot yet translate, we are also working on a backup Statistical MT system based on Moses.

Jaen is the elder sibling of Noja, the Norwegian-Japanese MT system.

Most of the rules of the Jaen MT system are extracted automatically from parallel corpora. The procedure for automatic rule extraction is described here: MT rule extraction

Some results can be found here: MtJaenTanaka, MtJaenFeedbackCleaning.

There are some note on how to set up an MT system with the DELPH-In tools in the MT tutorial.

A list of things that need to be fixed in Jaen can be found here: Todo list for Jaen

References

Clone this wiki locally