A Discrete Hard EM Approach for Weakly Supervised Question Answering

Benchmark Model Rank Results
question-answering-on-narrativeqaBERT-QA with Hard EM objective#1Rouge-L: 58.8