Swansea-Grenoble Joint Workshop

Swansea-Grenoble Joint Workshop

MAKE NMT-Viz: Explainability for Neural Machine Translation (NMT)

Workshop Overview

We are delighted to invite you to participate in this one-day workshop, designed to help translators visualise neural machine translation (NMT) outputs using explainable AI visualisation tools. This workshop will introduce the NMT-Viz toolkit, which promotes literacy and explainability in the translation studies community. It will feature keynote presentations from leading experts in the field, along with hands-on tutorials. The workshop aims to provide both theoretical insights and practical skills, focusing on literacy, analysis, and evaluation techniques for more transparent and explainable NMT.

This work is supported by the Grenoble-Swansea Joint Centre in Human-Centred AI and Data Systems, and the ANR CREMA Project.

Workshop Date: 4th November 2024

Workshop Programme Schedule

09:45 - 10:00 | Welcome and Introductions

-----------------------------------------------------------------

10:00 - 11:00 | Keynote Presentation 1:

“Explainability for NMT (with) users?”

Keynote Speakers:

 Professor Caroline Rossi (Université Grenoble Alpes)

Professor Nicolas Ballier (Université Paris Cité)

-----------------------------------------------------------------

11:00 - 12:00 | Keynote Presentation 2:

“Neural Machine Translation: An Overview”

Keynote Speaker:

Dr Marco Dinarelli (Université Grenoble Alpes)

-----------------------------------------------------------------

12:00 – 13:30 | Lunch Break & Networking

-----------------------------------------------------------------

13:30 - 15:30 | Hands-on Tutorials

This session will consist of four tutorials focused on different aspects of NMT explainability, giving participants the opportunity to engage directly with the NMT-Viz toolkit.

·         Tutorial 1: Literacy - Learning NMT Working Mechanisms

This tutorial will explain the architecture and working of NMT systems. Participants will learn about key components like encoders, decoders, and attention mechanisms, with a hands-on exploration of how NMT models generate translations.

·         Tutorial 2: Analyse - Reasons for Poor Translations

In this tutorial, participants will learn how to diagnose poor translations and understand the root causes of errors using explainability techniques. Tools and strategies for identifying common issues in NMT outputs will be covered.

·         Tutorial 3: Compare, Trust and Select – Evaluate the performance of NMT systems

Participants will be guided through methods for comparing different NMT systems. The session will cover how to evaluate the output of NMT models and how to select the best system based on explainability metrics.

·         Tutorial 4: Evaluate - In-Context Translation Results & Exploring Alternatives

This session will focus on evaluating NMT translations within specific contexts and exploring alternative translations.

-----------------------------------------------------------------

15:45 - 16:30 | Participants Discussions & Feedback

-----------------------------------------------------------------

16:30 - 16:45 | Closing Remarks

-----------------------------------------------------------------

Registration link to follow.

Workshop Organizing Committee:

Dr Jun Yang, Dr Sui He, Dr James Robert Turner, Swansea University

Dr Gabriela Gonzalez-Saez, Mr Fabien Lopez, Université Grenoble Alpes