A textual question answering and handwritten answer evaluation system for hindi language.

The article focuses on the development of an automated answer evaluation system for reading comprehension questions in the Hindi language, integrating a multilingual textual Question Answering (QA) system and a Handwritten Text Recognition (HTR) model. The QA system employs a fine-tuned RoBERTa-large transformer model trained on a combined multilingual dataset (XQuAD) covering 11 languages, including Hindi, to extract answer spans from given contexts, achieving superior performance compared to other transformer models. The HTR model uses a Convolutional Recurrent Neural Network (CRNN) with Bi-directional Long Short-Term Memory (Bi-LSTM) layers and Connectionist Temporal Classification (CTC) loss to convert handwritten Hindi text images into digital text with an accuracy of 90.37% on the IIIT-HW-Dev dataset. The system compares the predicted answer from the QA model with the recognized handwritten answer using F1-score metrics to assign correctness and marks, demonstrating effectiveness on self-created and existing datasets. Future directions include extending the system to handle multi-word answers, other languages, and different question types, with applications primarily in educational settings.