SBIR Phase I: Multilingual Adaptive School for Youth (MASY)

Award ID: 2431924

The broader/commercial impact of this SBIR Phase I project addresses significant educational disparities faced by over 200 million young people with no access to schooling and an additional 600 million struggling with basic literacy and numeracy skills despite being enrolled. The United Nations projects that the number of out-of-school children will increase by 84 million by 2030. This project aims to solve this problem by developing a universal platform for delivering educational content in multiple languages and cultural contexts. Education and language translation technologies rarely engage in cross-field research, yet they are deeply interconnected. Unlike existing digital education programs, this innovation is designed specifically for low-resourced language communities, where digital education is most needed. Low-resourced languages typically lack extensive digital data, comprehensive dictionaries, and detailed linguistic analysis, making it impossible for current natural language processing (NLP) models. Consequently, no comprehensive or high-quality educational content exists in these low-resource languages. This project aims to merge these fields by developing a global comprehensive PreK-grade 12 school platform that employs advanced language translation in its educational content delivery system, to exponentially increase access to high-quality education across linguistic barriers in regions where education in local languages is severely limited or completely absent. This Small Business Innovation Research (SBIR) Phase I project aims to develop a core universal natural language processing (NLP) model specifically designed for low-resourced languages. The primary technical objective is to create a model capable of handling the unique linguistic features of low-resourced languages, leveraging limited data, and incorporating cultural and contextual nuances to reduce language barriers as well as increase access to high-quality education globally while simultaneously extending the current ability of extant NLP models. The research will involve selecting and fine-tuning pre-trained models and adapting them to low-resourced languages through transfer learning, cross-linguistic techniques, and generating synthetic data to enrich training datasets. The project will also develop specialized layers for morphological analysis, tonal recognition, and flexible syntax parsing. Data collection and preprocessing pipelines will be established to ensure high-quality training data, while bias detection and mitigation techniques will be integrated to promote fairness and accuracy. The anticipated technical results include a highly adaptable NLP model that can deliver educational content in multiple low-resourced languages with high accuracy and cultural relevance. This project has the potential to significantly advance the field of NLP while providing a scalable and effective solution for the delivery of high-quality educational content globally. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

NSF Program Director: Lindsay Portnoy