Marathi Text to ISL Gloss Translation Using LSTM, GRU and Transformer Neural Models

In the Deaf and hard-of-hearing groups, sign languages are crucial for efficient communication. However, there is a significant language barrier between sign languages and spoken/written languages. The challenging work of translating written information into glosses in sign language is essential to improving communication for these people. In the context of text-to-sign language gloss translation, this article compares three neural architectures: Gated Recurrent Units (GRU), Attention-based Neural Machine Translation (NMT), and Long Short-Term Memory (LSTM). A custom dataset that includes paired text and sign gloss sequences is used to assess the models, with an emphasis on accuracy, fluency, and contextual relevance.