Latent question interpretation: parameter adaptation using interpretation policy

Abstract

Majority of question-answering models based on artificial neural networks are limited in their ability to capture natural language variability. To tackle this problem, we derive a model that learns multiple interpretations for a given question, using our ``interpretation policy'' module which automatically adjusts the parameters of a question answering model with respect to a value of a discrete latent variable. Interpretation policy is learned through a semi-supervised variational inference framework. Our framework is capable of combining different learning paradigms such as semi-supervised learning, reinforcement learning, and sample-based variational inference to bootstrap the performance.We analyze the effectiveness of our method through a large panel of experiments. When tested using the Stanford Question Answering Dataset, our model outperforms the baseline methods in finding multiple and diverse answers. Qualitative results underline the ability of the proposed architecture to discover multiple interpretations of a question.