Video-based datasets for Continuous Sign Language are scarce due to the challenging task of recording videos from native signers and the reduced number of people who can annotate sign language. COVID-19 has evidenced the key role of sign language interpreters in delivering nationwide health messages to deaf communities. In this paper, we present a framework for creating a multi-modal sign language interpretation dataset based on videos and we use it to create the first dataset for Peruvian Sign Language (LSP) interpretation annotated by hearing volunteers who have intermediate knowledge of PSL guided by the video audio. We rely on hearing people to produce a first version of the annotations, which should be reviewed by native signers in the future. Our contributions: i) we design a framework to annotate a sign Language dataset; ii) we release the first annotated LSP multi-modal interpretation dataset (AEC); iii) we evaluate the annotation done by hearing people by training a sign language recognition model. Our model reaches up to 80.3% of accuracy among a minimum of five classes (signs) AEC dataset, and 52.4% in a second dataset. Nevertheless, analysis by subject in the second dataset show variations worth to discuss.

Due to the lack of more variate, native and continuous datasets, sign languages are low-resources languages that can benefit from multilingualism in machine translation. In order to analyze the benefits of approaches like multilingualism, finding the similarity between sign languages can guide better matches and contributions between languages. However, calculating the similarity between sign languages again implies a laborious work to measure how close or distant signs are and their respective contexts. For that reason, we propose to support the similarity measurement between sign languages through a video-segmentation-based machine learning model that will quantify this match among signs of different countries’ sign languages. Using a machine learning approach the similarity measurement process can run more smoothly, compared to a more manual approach. We use a pre-trained temporal segmentation model for British Sign Language (BSL). We test it on three datasets, an American Sign Language (ASL) dataset, an Indian Sign Language (ISL), and an Australian Sign Language (AUSLAN) dataset. We hypothesize that the percentage of segmented and recognized signs by this machine learning model can represent the percentage of overlap or similarity between British and the other three sign languages. In our ongoing work, we evaluate three metrics considering Swadesh’s and Woodward’s list and their synonyms. We found that our intermediate-strict metric coincides with a more classical analysis of the similarity between British and American Sign Language, as well as with the classical low measurement between Indian and British sign languages. On the other hand, our similarity measurement between British and Australian Sign language just holds for part of the Australian Sign Language and not the whole data sample.

We present a study in which we train neural models that approximate rules that assess the quality of English sentences. We modeled five rules using deep LSTMs trained over a dataset of sentences whose quality is evaluated under such rules. Preliminary results suggest the neural architecture can model such rules to high accuracy.