NEUROVIZ and SPECTRANS closing event

15-16 December 2022

rooms 715 (Th morning) and 720

Bâtiment Olympe de Gouges

8 Place Paul Ricoeur

75013 PARIS

No registration fee, but registration is mandatory before December 10. Contact Maud Benard maud.benard@u-paris.fr

This two day workshop is intended as a recap of two Université Paris Cité research projects and a potential platform for exchanges at a European level of comparable research questions. We would like to share our results, but also our doubts (limitations and frustrations) about two funded research projects conceived in an interdisciplinary perspective designed to empower linguists and translators with a better grasp of neural machine translation with strong help from computational scientists, computational linguists and data scientists.

In this spirit, we have put up a programme that welcomes similar European projects that voice comparable research questions carried out in partner universities. For institutional reasons, we have focused on members of the Circle U Alliance.

This event is funded by the Laboratoire de Linguistique formelle (LLF)  and CLILLAC-ARP labs and jointly organised by Nicolas Ballier, Maud Bénard & Guillaume Wiesniewski.

PROGRAMME

DECEMBER 15th

Morning: room 715 

(zoom : https://u-paris.zoom.us/j/85674176101?pwd=TEZYcHo3eEx3dk9JNmVwbXdIL2gyQT09

ID  : 856 7417 6101
secret code : 806693)

9h30 Welcome coffee 

10h Nicolas Ballier (UPCité, CLILLAC-ARP & LLF) Introduction

10h05 François Yvon (Orsay, LISN)  Neural Machine Translation for Linguists of Good Will

11h Guillaume Wisniewski (UPCité, LLF) What can linguists learn from probing neural networks ?

12h Bingzhi Li, Guillaume Wisniewski and Benoit Crabbé  (all UPCité, LLF) What can we learn from neural networks about agreements and coreference?

13h LUNCH BREAK  

Afternoon: room 720 

14h30  NeuroViz team [Nicolas Ballier (UPCité, CLILLAC-ARP & LLF), Lina Conti (UPCité), François Yvon (Orsay, LISN), Guillaume Wisniewski (UPCité, LLF), Lichao Zhu (UPCité, CLILLAC-ARP & LLF)]  Some Lessons from the NeuroViz project. Presentation of the deliverables of the project : papers, scripts, datasets; working with JoeyNMT, a pedagogical toolkit; more to explore on Visualisation : alternate visualising systems

15h10 Discussion (the robustness of translation toolkits)

15h30 Bilal Faye (Université Sorbonne Paris Nord) : Investigating Byte Pair Encoding and linguistic complexity (Some results from the ComplexNeuroViz project)

16h00 Coffee break

16h30 Caroline Rossi (UGA) and Nicolas Ballier (UPCité) the Grenoble Swansea MAKE-NMTViz project

17h Round Table : next steps for Research

François Yvon, Caroline Rossi, Alina Secară (Universität Wien), Dragoș Ciobanu (Universität Wien)

18h End

DECEMBER 16th

All day room 720

(on Zoom https://u-paris.zoom.us/j/88117026423?pwd=MGZDL0hBdTZTK3NIcnpTL0VPbUk5UT09

ID  : 881 1702 6423
secret code : 177146)

9h00  zoom tests

9h15 Nicolas Ballier (UPCité) Introduction : the steep learning curve of Neural Machine Translation (NMT)

9h30 Juan Antonio Pérez-Ortiz (Universitat d’Alacant) the Gourmet Project for NMT for less-resourced languages

10h Marie-Aude Lefer (UCLouvain) Assessing NMTPE quality in translator education: the PostEditMe project

discussant : Natalie Kübler (UPCité) On error typologies

11h00 Coffee break

11h30 Ralph Krüger and Janica Hackenbuchner (Koeln) NMT, MT Literacy and Data Literacy for Specialised translation

discussant : Christoph Declercq (Utrecht), (tbc)

12h30  Lunch break

14h00 SPECTRANS team (UPCité) :  Some lessons from the SPECTRANS project. Presentation of the deliverables of the project: papers, apps, Python notebooks, datasets, How is society impacted with NMT?, A springboard for other projects : Specialised translation and data collection (OCTAVES, ARTES and corpus building).
Alexandra Mestivier & Brice Bricaud Octaves: a platform for collecting specialised student translations and post-edited machine translation output
Mojca Pecman et Brice Bricaud, ARTES et TAN : l’adaptation au domaine.

15h00 Maria Zimina, Nicolas Ballier, Lichao Zhu (UPCité) : PAPTAN : A platform for Neural Machine Translation:  training translation models for research and teaching (Pure Neural Server)

• deep learning for teaching
• partnership for SYSTRAN Pure Neural Server
• WMT22 competitions (biomedical English)

15h30 Coffee break

16h00 Round Table with some lightning presentations from some members of the Circle U Alliance Teaching and Researching NMT 

Dragos Ciobanu & Alina Secara (Universität Wien), Marie-Aude Lefer  &  Anais Tack (UCLouvain, UPCité)

How can we establish collaborations between translators and computational scientists?  

17h00 Round Table Next plans: do we need a COST on NMT? 

Maria Zimina-Poirot (UPCité, CLILLAC-ARP) & Nicolas Ballier The KLILT project, Nicolas Froeliger (UPCité, CILLLAC-ARP), similar initiatives in Europe

The LITHME COST partially deals with digital literacy in humanities. Do we need a COST on Neural Machine Translation?

18h00 Nicolas Ballier Final words