Predicting answer types for question-answering

We tackle the problem of answer type prediction, applying a transition-based neural network parser to understand the syntactic structure of a query and infer the focus word or phrase that contains the correct type. To achive this, we first come up with a framework to combine a dataset of labeled question-answer pairs and a type inventory into a dataset to learn types. Second, we fit a sigmoid prediction layer on top of query words and the inferred syntactic structure of a query to predict the correct answer types. Our model outperforms a logistic regression baseline that does not take into account the syntactic structure of the query.