One of our senior PhD students, Kyle Richardson, and a team of colleagues including Aleksei Mokeev (project lead) and Natalia Mokeeva placed first (out of 162 teams) in the Russian Text Normalization Challenge hosted by Kaggle.com and Google’s Text Normalization Research group, and 7th place (out of 260 teams) in the English version of the same competition.
Text normalization is the task of converting written expressions (e.g., time and date expressions in numerical short hand) to their “spoken form”, and is a primary component in automatic text-to-speech systems and other related technologies. Using the competition’s provided training data consisting of example unnormalized/normalized sentence pairs, their solution learns normalization patterns using state-of-the-art techniques from the emerging field of neural machine translation. For more information about their solution, please see https://github.com/amokeev/ru_norm_kaggle