Make NMT-Viz:

Make NMT-Viz is an on-going project developed by the Swansea-Grenoble AI Centre, with funding from Grenoble Alpes University's Multidisciplinary Institue in Artificial Intelligence (MIAI) and involves four research laboratories: Laboratoire d’Informatique de Grenoble (LIG),[1] ILCEA4,[2] CLILLAC-ARP,[3] and the Swansea Translation and Interpreting Group (STING).[4]

The project explores explainability for neural machine translations (NMT), the linguistic properties of neural networks for machine translation, and uses knowledge from Translation Studies to qualitatively analyse the effects of linguistic structures and neural representations on the quality of automatic translations.

As existing NMT systems are too complex to be installed and too experimental to really reach the corpus linguistics and translation studies communities, we plan to develop a visualisation and explainability toolkit featuring a greater range of accessibility and functionality tools, and which is ultimately more user-orientated than previous systems. The project is expected to explore the knowledge limit between the input and the output of NMT systems which humans often fail to visualise. Thus, the development of an NMT-Viz toolkit will provide visual evidence of the workings of the different steps involved within NMT. During the final stage of the project, Swansea University will host a workshop aimed at professional translators and students alike to disseminate the final toolkit (expected June 2024). This project will promote literacy around NMT for linguists, translation researchers, and translation trainees. The final stage of the project will involve conducting a reception-based study that will explore the ways to extend potential applications of the toolkit and inform the direction for our future research.

The project aims to provide translators with a tool that visualises several aspects of the neural machine translation process.

Funding Secured:

·         Welsh Government: Llywio’r Byd Ymchwil (Navigating the World of Research) - TAITH: £2,475.00 awarded April 2024.

·         Welsh Government: Llywio’r Byd Ymchwil (Navigating the World of Research) - TAITH: £1,390.00 awarded July 2023.

·         Grenoble-Swansea AI Centre on Amplifying Human Abilities – €68,374 awarded September 2022.

Published Conference Proceedings:

Gonzalez-Saez, G., et. al. (2024). Exploring NMT Explainability for Translators Using NMT Visualising Tools. In Proceedings of the 25th Annual Conference of the European Association for Machine Translation, Sheffield, United Kingdom. Vol.1 (pp. 396-410).

Gonzalez-Saez, G., et. al. (2024). The MAKE-NMTViz Project: Meaningful, Accurate and Knowledge-limited Explanations of NMT Systems for Translators. In Proceedings of the 25th Annual Conference of the European Association for Machine Translation, Sheffield, United Kingdom. Vol.2 (pp. 12-13).

Lopez, F. et al. (2023). The MAKE-NMTVIZ System Description for the WMT23 Literary Task. In Proceedings of the Eighth Conference on Machine Translation, Singapore: Association for Computational Linguistics. (pp. 287–295).

[1] Didier Schwab, Emmanuelle Esperança-Rodier, Fabien Lopez, Gabriela Gonzalez-Saez, Marco Dinarelli, Mariam Nakhle

[2] Caroline Rossi

[3] Nicolas Ballier

[4] Jun Yang, Sui He, James Robert Turner