State-of-the-Art Translation of Text-to-Gloss using mBART : A case study of Bangla
This addresses the understudied problem of Bangla Sign Language translation for the deaf and dumb population, though it is incremental as it applies an existing method to a new domain.
The study tackled the lack of Bangla text-to-gloss translation by creating a rule-based synthetic dataset and fine-tuning mBART-50, achieving state-of-the-art performance with ScareBLEU scores of 79.53 on Bangla and 63.89 on the PHOENIX-14T benchmark.
Despite a large deaf and dumb population of 1.7 million, Bangla Sign Language (BdSL) remains a understudied domain. Specifically, there are no works on Bangla text-to-gloss translation task. To address this gap, we begin by addressing the dataset problem. We take inspiration from grammatical rule based gloss generation used in Germany and American sign langauage (ASL) and adapt it for BdSL. We also leverage LLM to generate synthetic data and use back-translation, text generation for data augmentation. With dataset prepared, we started experimentation. We fine-tuned pretrained mBART-50 and mBERT-multiclass-uncased model on our dataset. We also trained GRU, RNN and a novel seq-to-seq model with multi-head attention. We observe significant high performance (ScareBLEU=79.53) with fine-tuning pretrained mBART-50 multilingual model from Facebook. We then explored why we observe such high performance with mBART. We soon notice an interesting property of mBART -- it was trained on shuffled and masked text data. And as we know, gloss form has shuffling property. So we hypothesize that mBART is inherently good at text-to-gloss tasks. To find support against this hypothesis, we trained mBART-50 on PHOENIX-14T benchmark and evaluated it with existing literature. Our mBART-50 finetune demonstrated State-of-the-Art performance on PHOENIX-14T benchmark, far outperforming existing models in all 6 metrics (ScareBLEU = 63.89, BLEU-1 = 55.14, BLEU-2 = 38.07, BLEU-3 = 27.13, BLEU-4 = 20.68, COMET = 0.624). Based on the results, this study proposes a new paradigm for text-to-gloss task using mBART models. Additionally, our results show that BdSL text-to-gloss task can greatly benefit from rule-based synthetic dataset.