ASL-SMT

ASL-SML deal with machine translation to sign language. It starts with studying existing systems and issues in order to propose a new model for statistical machine translation from written English text to American Sign Language (English/ASL). The study covers specificity of Sign Language from different communities and a scope of existing tools and solutions. According to the state of the art, the aims of this paper is to propose a new approach aiming to build artificial corpus using grammatical dependencies rules due to the lack of resources for Sign Language. The parallel corpus was the input of our machine translation that has been used to create the statistical memory translation based on the IBM alignment algorithms. These algorithms have been improved and optimized by integrating Jaro-Winkler distances. Then, based on the constructed translation memory, we have implemented a decoder to translate an English text to the American Sign Language using a new transcription system based on gloss annotation. Results had been evaluated by the BLEU evaluation metric.

Live Demo:

Corpus Sample Data:

if you’re writing or working on the corpus, please cite this paper:

Achraf Othman and Zouhour Tmar. “English-ASL Gloss Parallel Corpus 2012: ASLG-PC12, The Second Release”. Fourth International Conference On Information and Communication Technology and Accessibility ICTA’13, Hammamet, Tunisia, October 24-26, 2013.

 

Schema Ressources for Gloss Annotation System (XML-Gloss):

if you’re writing or working on the corpus, please cite this paper:

Achraf Othman, Mohamed Jemni, “An XML-Gloss Annotation System for Sign Language Processing“, 6th International Conference On Information and Communication Technology and Accessibility ICTA’17, Muscat, Oman, December 19-21, 2017.

 

 

 

If you’re just writing about this work, please cite this thesis as follow:

Achraf Othman, “Machine Translation for Sign Language based on Statistical Approach“, PhD Thesis, University of Sfax, Tunisia, (2017).