Multilingual Text-to-Pictogram Translation for Augmentative and Alternative Communication

Augmentative and Alternative Communication (AAC) systems are essential for individuals with language impairments due to genetic conditions, accidents, or strokes. This paper presents two approaches for multilingual text-to-pictogram translation across English, French, and Tamil. The first approach is a hybrid method that combines contextual sentence embeddings extracted using BERT with similarity-based retrieval through an ANN index. The second approach is transformer-based, utilizing contrastive learning for effective mapping of text to AAC pictograms. Both methods aim to enhance communication for cognitively impaired individuals, and their performance is evaluated to determine the most effective solution for AAC applications.