US 11,379,736 B2Granted

Machine comprehension of unstructured text

AssigneeMicrosoft Technology Licensing, LLC

InventorsAdam Trischler, Philip Bachman, Xingdi Yuan, Alessandro Sordoni, Zheng Ye

Life of this patent

Estimating the term…

Loading patent…

Abstract

Described herein are systems and methods for providing a natural language comprehension system that employs a two-stage process for machine comprehension of text. The first stage indicates words in one or more text passages that potentially answer a question. The first stage outputs a set of candidate answers for the question, along with a first probability of correctness for each candidate answer. The second stage forms one or more hypotheses by inserting each candidate answer into the question and determines whether a sematic relationship exists between each hypothesis and each sentence in the text. The second processing circuitry generates a second probability of correctness for each candidate answer and combines the first probability with the second probability to produce a score that is used to rank the candidate answers. The candidate answer with the highest score is selected as a predicted answer.

Claims20

Dependent claims

A system, comprising:

first processing circuitry;
second processing circuitry operably connected to the first processing circuitry;
and
one or more storage devices storing computer executable instructions that when executed by the first and the second processing circuitries, perform a method, the method comprising:
generating, by the first processing circuitry, a question relating to natural-language text and a set of candidate answers to the question based on the natural-language text, wherein the natural-language text includes at least part of each candidate answer;
generating a first probability of a first candidate answer being a correct answer to the question based on a combination of the question and the natural-language text;
generating, by the second processing circuitry, a first hypothesis by inserting the first candidate answer into the question;
generating, based on the first probability and the first hypothesis appearing in the natural-language text, a second probability of the first candidate answer being the correct answer to the question, the second probability being a correction to the first probability of the first candidate answer being the correct answer to the question;
generating, based on the first probability and the second probability, a score associated with the first candidate answer, wherein the score represents a degree of correctness of the first candidate answer of the set of candidate answers as the correct answer to the question;
and
based on the score associated with the first candidate answer, predicting the first candidate answer as the correct answer to the question.

8 more elements

Dependencies are inferred from the claim text.

Facts

Application: 15/598,139Filed 17 May 2017 · us provisional application
Priority: 17 May 2016
Published: US 2017/0337479 A123 Nov 2017
Granted: 5 Jul 2022
Term ends: Not estimatedThe source records don’t support an estimate.

Classification8 codes

G06N 5/04
Computing arrangements using knowledge-based models: inference or reasoning models
G06F 40/279
Natural language analysis: recognition of textual entities
G06F 40/30
Handling natural language data: semantic analysis
G06N 3/0442
Recurrent networks, e.g. Hopfield networks: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]