Relying on discourse analysis to answer complex questions by neural machine reading comprehension

A computing device, comprising:

• one or more processors;
• and
• one or more memories storing computer-executable instructions for generating an answer to a complex question from unstructured text, that, when executed by the one or more processors, cause the computing device to:
• receive a user query comprising the complex question;
• obtain, from a corpus of unstructured texts, an answer candidate text corresponding to the user query and comprising text from which the answer is subsequently generated;
• generate first linguistic data corresponding to the user query and second linguistic data corresponding to the answer candidate text, the first linguistic data comprising first syntactic data, first semantic data, and first discourse data generated from the user query, the second linguistic data comprising second syntactic data, second semantic data, and second discourse data generated from the unstructured text;
• provide the first linguistic data, the second linguistic data, and the answer candidate text to a self-attention neural network comprising at least a syntax self-attention layer, a semantic self-attention layer, and a discourse self-attention layer, the self-attention neural network being previously trained to generate a corresponding answer from an instance of unstructured text based at least in part on the complex question, the self-attention neural network being configured to receive input comprising linguistic data associated with a question and separate linguistic data associated with the instance of unstructured text;
• receive the answer generated by the self-attention neural network;
• and
• provide the answer in response to the complex question of the user query.

A non-transitory computer readable medium storing instructions for identifying an answer to a complex question from unstructured text, that, when executed by one or more processors of a computing device, cause the computing device to:

• receive a user query comprising the complex question;
• obtain, from a corpus of unstructured texts, an answer candidate text corresponding to the user query and comprising text from which the answer is subsequently identified;
• generate first linguistic data corresponding to the user query and second linguistic data corresponding to the answer candidate text, the first linguistic data comprising first syntactic data, first semantic data, and first discourse data generated from the user query, the second linguistic data comprising second syntactic data, second semantic data, and second discourse data generated from the answer candidate text;
• provide the first linguistic data, the second linguistic data, and the answer candidate text to a self-attention neural network comprising at least a syntax self-attention layer, a semantic self-attention layer, and a discourse self-attention layer, the self-attention neural network being previously trained to generate a corresponding answer from an instance of unstructured text based at least in part on the complex question, the self-attention neural network being configured to receive input comprising linguistic data associated with a question and separate linguistic data associated with the instance of unstructured text;
• receive the answer generated by the self-attention neural network;
• and
• provide the answer in response to the complex question of the user query.