Corelli Project


Project Description

The goal of the Corelli project was to develop an architecture and a tool-set for rapid deployment of multilingual Machine-Translation systems, with an emphasis on machine-translation for assimilation purposes and on languages for which electronic or human resources are scarce or difficult to obtain. Building on previous achievements from the Temple project, the Corelli project concentrated on three major areas:

1. A Component-Based MT Architecture

The public open-source MEAT Language Engineering Toolset was developed within the Corelli project to support the rapid deployment of multilingual Machine-Translation systems. The toolset is implemented in C++ and TCL and is available for Unix and Windows. It is structured around:

  1. An open library of generic NLP algorithms. Includes an extension mechanism to enrich the core library by third parties or for specific applications. Each component can be used as a stand-alone tool or can be combined with other components to build more complex systems.
  2. A linguistic knowledge representation language based on charts and typed feature structures.
  3. An integration architecture which supports information interchange between modules based on the `plug-and-play' concept without constraining the control model.
  4. A set of development tools to help the computational linguist to develop and test a complete system.

2. Development of new MT systems

Two new machine-translation systems have been developed using essentially the same approach as in the Temple project (`glossary-based MT'):

3. Porting and improvement of existing MT systems

Language components developed in the Temple project are enhanced and ported to the new architecture:

The Corelli architecture also supports the Shiraz Persian-English Machine-Translation system, the Expedition Turkish-English MT system, and is used as the MT engine for the Expedition Boas system.

Project Members

The Corelli project was funded by DoD, Maryland Procurement Office, MDA904-96-C-1040.