MMNLG 2023 : Workshop on Multimodal, Multilingual Natural Language Generation In conjunction with INLG/SIGDIAL 2023
Call For Papers
Workshop on Multimodal, Multilingual Natural Language Generation
In conjunction with INLG/SIGDIAL 2023
Prague, 12 September, 2023
We invite the submission of long and short papers for the first Workshop on Muiltimodal, Multilingual NLG (MM-NLG), which will be held in Prague, in conjunction with the joint meetings of the 16th International Conference on Natural Language Generation (INLG 2023) and the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDial 2023).
Workshop goals and topics
This event aims to bring together researchers working on text generation from multimodal input data. The workshop also emphasises multilinguality as an ongoing, open challenge for text generation methods, especially for languages which are relatively under-resourced.
We therefore invite papers on all topics related to text generation from multimodal inputs, multilingual text generation, or a combination of the two. We welcome submissions which focus on multimodal and/or multilingual generation in both dialogue and non-interactive settings.
NLG and multimodal inputs
By Multimodal NLG, we intend to capture a broad variety of input data types and formats from which text can be generated using neural, statistical or rule-based methods. For example, while several contemporary NLG models generate based on textual prompts or prefixes, others rely on structured inputs which can take the form of `flat' semantic representations, RDF triples, etc. In a different vein, vision-to-text models generate captions, paragraphs or short narratives from visual inputs such as images or video. Finally, there is a long tradition in data-to-text NLG which seeks to generate text from numerical or other, less structured inputs. The sheer diversity is also reflected in the broad range of datasets available for training and evaluating NLG models.
This workshop will provide a forum to discuss NLG research based on any input modality, fostering a debate on the directions in which the field has developed, and especially the relationship between different NLG tasks, as characterised by the variety of possible inputs, among others.
NLG and multilingual outputs
As the field has become increasingly dominated by large, pretrained language models, it has become increasingly evident that not all languages are on a level playing field. For example, when training data is opportunistically sourced from the web, data for certain languages is often very limited, and highly noisy. On the other hand, developing curated multilingual data for under-represented languages is very challenging, as some recent efforts (for example, the BLOOM model) have shown.
This workshop will provide an opportunity for researchers to discuss challenges and report on recent work targeting NLG in multiple languages, including, but not limited to, data-lean scenarios, where transfer learning, few-shot and zero-shot approaches would be expected to play an important role.
This one-day workshop will consist of an oral and a poster session, together with a special session The oral session will feature talks by two invited speakers, as well as regular paper presentations.
The workshop will be hybrid. We encourage all participants to be present, but will provide online access for those who are unable, or prefer not to travel.
Special session: WebNLG Challenge on Under-Resourced Languages
In line with the goals of MM-NLG, the workshop will include a special session dedicated to the recently launched, ongoing WebNLG 2023 Challenge, which focuses on generation for under-resourced languages in few-shot and zero-shot settings.
More info here: https://synalp.gitlabpages.inria.fr/webnlg-challenge/challenge_2023/
We solicit two kinds of papers:
- Long papers must not exceed eight (8) pages of content, plus unlimited pages of ethical considerations, supplementary material statements, and references.
- Short papers must not exceed four (4) pages, plus unlimited pages of ethical considerations, supplementary material statements, and references.
Submissions should follow ACL Author Guidelines and policies for submission, review and citation, and be anonymised for double blind reviewing. See https://www.aclweb.org/adminwiki/index.php?title=ACL_Author_Guidelines
Please use ACL 2023 style files; LaTeX style files and Microsoft Word templates are available at https://2023.aclweb.org/calls/style_and_formatting
Authors must honour the ethical code set out in the ACL Code of Ethics, available at https://www.aclweb.org/portal/content/acl-code-ethics
If your work raises any ethical issues, you should include an explicit discussion of those issues. This will also be taken into account in the review process. You may find this checklist of use: https://aclrollingreview.org/responsibleNLPresearch/
Authors are strongly encouraged to ensure that their work is reproducible; see, e.g., the reproducibility checklist at https://2021.aclweb.org/calls/reproducibility-checklist/
Papers involving any kind of experimental results (human judgments, system outputs, etc) should incorporate a data availability statement into their paper. Authors are asked to indicate whether the data is made publicly available. If the data is not made available, authors should provide a brief explanation why. (E.g. because the data contains proprietary information.) A statement guide is available on the INLG 2023 website.
The workshop will only accept direct submissions. Submissions can be made to the MM-NLG START website: https://softconf.com/n/mmnlg2023/
Accepted papers will be published in the Workshop proceedings on the ACL Anthology.
- Deadline for long and short papers: 16 July, 2023
- Notification of acceptance: 6 August, 2023
- Deadline for camera-ready papers: 14 August, 2023
- MM-NLG Workshop: 12 September, 2023
Anya Belz, ADAPT, Dublin City University, Ireland
Claudia Borg, University of Malta, Malta
Liam Cripwell, CNRS/LORIA and Lorraine University, France
Aykut Erdem, Koc University, Turkey
Erkut Erdem, Hacettepe University, Turkey
Claire Gardent, CNRS/LORIA, France
Albert Gatt, Utrecht University, The Netherlands
John Judge, ADAPT, Dublin City University, Ireland
William Soto-Martinez, CNRS/LORIA and Lorraine University, France
Support and acknowledgements
This workshop is a joint initiative which has received the support of the following projects:
- LT-Bridge funded by the EU Horizon 2020 Work Programme Spreading Excellence and Widening Participation (WIDESPREAD) 2018-2020 Grant No. 952194 https://lt-bridge.eu/
- The xNLG AI Chair on Multilingual, Multi-Source Text Generation funded by the French National Research Agency (Gardent; ANR-20-CHIA-0003), Meta and the Region Grand Est https://members.loria.fr/CGardent/xnlg.html
- Multi3Generation: Multimodal, Multi-task, Multi-Lingual Natural Language Generation COST Action CA18231 https://multi3generation.eu/