First Workshop on Multimodal Machine Translation for Low Resource Languages (MMTLRL-2021)
Varna, Bulgaria
September 7, 2021
AN ONLINE EVENT
Language does not exist in a vacuum. Yet, for a long time, large parts of NLP have focused on text- (or speech-) only scenarios: most work on machine translation (MT) e.g. is on text-to-text MT. In principle, the inclusion of additional context in the form of other modalities offers the promise of improving a translation. In practice, this is often hard (Lala et al. 2017, Elliott 2018). In this workshop, we would like to combine two strands of research that are hitherto not well connected: research on low-resource MT and research on multi-modal MT (MMMT). The reason why we would like to explore the connection is the following: while there has been important progress on both sides, including unsupervised (Artetxe et al. 2018, Lample et al. 2018) and self-supervised MT (Ruiter et al. 2019), and neural-network based modality combinations in MMMT (Çağlayan et al. 2019), the potential of mustering information in other modalities (such as images, videos and spoken language) to complement the text signal in low-resource MT has not yet been explored extensively. However, a combination may hold promise: a richer multimodal signal may help address some of the challenges that come with low-resource scenarios. Of course, there are no guarantees: a richer multimodal signal and with it an increase in the dimensionality of the data may make the problem worse.
CALL FOR PAPERS
We invite original contributions describing the latest trends, developments and solutions. Topics of interest include, but are not limited to:
Data harvesting and preparation for low resource MMMT
Multimodality and its impact on machine translation
MMMT for low resource scenarios
Neural approaches towards multimodal machine translations
Speech and Visual modalities for MMMT
Multimodal quality estimation
Multilingual MMMT
Multimodal attention
Evaluation of MMMT
SUBMISSION GUIDELINES
All papers must be submitted in PDF format through the conference management system at https://www.softconf.com/ranlp2021/MMTLRL2021/. The papers should follow the format of the main conference, described at the main RANLP website, Submission Guidelines Section.
Full Papers must describe original unpublished work in any topic area of the workshop. Full papers are limited to 8 pages for content, with 2 additional pages for references.
Short Papers may describe either work in progress or a research proposal. They may also be in the style of a position paper that surveys and criticizes existing literature. Short papers must include clear directions for future research. Submissions of this type are limited to 6 pages for content, with 2 additional pages for references.
DOUBLE SUBMISSION
Authors may submit the same paper at several conferences. In this case, they must notify the organizers by filling in the corresponding information in the submission form, as well as notifying the contact organizer by e-mail.
REGISTRATION
At least one author of each accepted submission must register in order to present the paper. Participants can register here, click HERE
IMPORTANT DATES
Workshop paper submission deadline:
30 June 202123 July 2021Workshop paper acceptance notification:
31 July 202115 August 2021Workshop paper camera-ready versions:
31 August 2021Workshop video presentations due:
31 August 2021Workshop camera-ready proceedings ready: 5 September 2021
Workshops: 7 September 2021
PROGRAM SCHEDULE