Machine reading comprehension system for answering queries related to a document

A method, comprising:

• in response to a received query, dividing a document into multiple sections that each comprise a portion of the document;
• processing each section separately to determine a first score that indicates the section provides an answer to the received query;
• processing each section separately to determine a second score that indicates the section provides an answer to the received query;
• processing each section in a subset of the processed sections separately to determine a third score that indicates whether the section provides an answer to the received query, wherein each section in the subset of the processed sections is included in the subset of the processed sections based on a ranking of a combination of the first score and the second score, wherein the processing of each section further comprises:
• separating a section into sentences;
• encoding each sentence in the section and the received query;
• processing the encoded sentences to determine one or more relevant sentences in the section that are relevant to the encoded query;
• for each of the one or more relevant sentences, comparing a relevant sentence to the encoded query to produce the third score;
• and
• determining a predicted answer to the received query based on each third score.

A method, comprising:

• in response to a received query, dividing a document into multiple sections that each comprise a portion of the document;
• processing, by processing circuitry, each section separately to determine a first score that indicates the section provides an answer to the received query;
• processing, by the processing circuitry, each section separately to determine a second score that indicates the section provides an answer to the received query;
• processing each section in a subset of sections separately to determine whether a section provides an answer to the received query, wherein each section in the subset of sections is included in the subset of sections based on a ranking of a combination of the first score and the second score, wherein the processing of each section further comprising:
• separating a section into sentences;
• encoding, by the processing circuitry, each sentence in the section and the received query;
• processing, by the processing circuitry, the encoded sentences to determine one or more relevant sentences in the section that are relevant to the encoded query;
• for each of the one or more relevant sentences, comparing a relevant sentence to the encoded query to produce a third score;
• and
• selecting a predicted answer to the received query based on each third score.