Synthetic Data Augmentation for Zero-Shot Cross-Lingual Question Answering

Coupled with the availability of large scale datasets, deep learning\narchitectures have enabled rapid progress on the Question Answering task.\nHowever, most of those datasets are in English, and the performances of\nstate-of-the-art multilingual models are significantly lower when evaluated on\nnon-English data. Due to high data collection costs, it is not realistic to\nobtain annotated data for each language one desires to support.\n We propose a method to improve the Cross-lingual Question Answering\nperformance without requiring additional annotated data, leveraging Question\nGeneration models to produce synthetic samples in a cross-lingual fashion. We\nshow that the proposed method allows to significantly outperform the baselines\ntrained on English data only. We report a new state-of-the-art on four\nmultilingual datasets: MLQA, XQuAD, SQuAD-it and PIAF (fr).\n