Joint Video and Text Parsing for Understanding Events and Answering Queries