Back LaSTUS papers for the SignON project
LaSTUS papers for the SignON project
Members of LaSTUS lab have recently been leading and contributing to research as part of the SignON project.
This week, on the 6th September, Santiago Egea Gómez presented his work titled "Syntax-aware Transformers for Neural Machine Translation: The Case of Text to Sign Gloss Translation" during the 14th Workshop on Building Comparable Corpora (RANLP 2021). In this paper, text to sign language gloss translation is improved (5.7 BLEU) by adding syntactic dependency information to the embeddings used in state of the art transformer-based models. Our lab will continue research in this area, adding different language combinations and linguistic features to the experiments. You can find the paper freely available.
Another paper, "SignON: Bridging the gap between Sign and Spoken Languages", which introduces the project and all the areas of research that will be involved - including computer vision, machine translation, natural language processing and 3D graphics. The project will be highly multidisciplinary incorporating 17 European universities and institutions, user-centric, and will led by the deaf community whose input is crucial in building the SignON service. This paper will be presented at the forthcoming SEPLN2021 conference, beginning 22nd September 2021.