Multilingual Transformer for Dynamic Cricket Commentary Generation

The sport of Cricket is growing in popularity which necessitates the use of technology to adhere to popular game standards. Since this is a data rich and most elaborate game, understanding this for someone who's new to the game had been challenging. We are aiming to solve this unavailability of the game through native commentary to audience of all major languages. To outline our entire research work we ideate a Cricket Game AI through computer vision, use it to produce text commentary of game-play and perform Machine Translation(MT) to produce native commentary for a targeted language, As afirst part of the outlined process in this work we focus on the implementation of the machine translation component of this pipeline, evaluating various methods including Seq2seq and transformer based architectures. Our method of cross feeding Seq2Seq with Transformer model results in Bilingual Evaluation Understudy (BEL U) score on Validation set of 2.08.