Evidence-Based Natural Language Input Recognition

Patent metadata

Field	Value
Publication number	US-2016259779-A1
Application number	US-201514640833-A
Country	US
Kind code	A1
Filing date	Mar 6, 2015
Priority date	Mar 6, 2015
Publication date	Sep 8, 2016
Grant date	—

Abstract

Methods, devices, and computer program products for recognizing and responding to natural language input are described herein. Natural language input is received at a natural language input interface of a computing device and transformed into computer-usable text. A natural language input recognizer obtains evidence from one or more evidence source and generates an evidence graph based on the evidence obtained. Evidence may be obtained asynchronously, and the natural language input recognizer may update the evidence graph upon receipt of additional evidence. The natural language input recognizer generates a set of recognition hypotheses based on the evidence graph and selects one of the recognition hypotheses as a recognition result for the natural language input. Semantic models, evidence models, and response models may be employed to generate the evidence graph and respond to the recognition result selected for the natural language input.

First claim

What is claimed is:

1 . A method of recognizing natural language input comprising:
receiving natural language input at a natural language input interface of a computing device comprising a natural language input recognizer; obtaining, by the natural language input recognizer, evidence from an evidence source; generating, by the natural language input recognizer, an evidence graph based, at least in part, on the evidence obtained; generating a set of recognition hypotheses based, at least in part, on the evidence graph; and selecting one of the recognition hypotheses as a recognition result for the natural language input.

2 . The method of claim 1 , wherein:
obtaining evidence from the evidence source comprises asynchronously obtaining evidence from a plurality of evidence sources; and generating the set of recognition hypotheses comprises i) generating a first set of recognition hypotheses based on evidence obtained during a first timeframe, and ii) generating a second set of recognition hypotheses based on evidence obtained during a second timeframe subsequent to the first timeframe.

3 . The method of claim 2 , wherein:
generating the set of recognition hypotheses further comprises iii) traversing a path of the evidence graph, and iv) generating a recognition hypothesis based on the path traversed.

4 . The method of claim 3 , further comprising:
generating a score for each one of the recognition hypotheses; wherein the recognition hypothesis selected as the recognition result for the natural language input is selected based on its score.

5 . The method of claim 4 , further comprising:
comparing the score for each one of the recognition hypotheses to a score threshold; wherein the score of the recognition hypothesis selected as the recognition result for the natural language input exceeds the score threshold.

6 . The method of claim 1 , wherein:
the evidence graph comprises a set of evidence nodes wherein each evidence node in the set of evidence nodes is generated based on evidence obtained by the natural language input recognizer.

7 . The method of claim 6 , wherein:
the set of evidence nodes comprises a set of intent evidence nodes wherein each intent evidence node comprises an intent hypothesis and an intent score for the intent hypothesis; the set of evidence nodes further comprises a set of phrase evidence nodes wherein each phrase evidence node comprises a slot hypothesis and a slot score for the slot hypothesis; and each phrase evidence node is connected to at least one of the intent evidence nodes in the evidence graph.

8 . The method of claim 7 wherein:
the set of phrase evidence nodes is a first set of phrase evidence nodes; and
the set of evidence nodes further comprises a second set of phrase evidence nodes wherein each phrase evidence node in the second set of phrase evidence nodes is connected to at least one of the phrase evidence nodes in the first set of phrase evidence nodes or connected to at least one of the intent evidence nodes.

9 . The method of claim 6 , wherein:
generating set of evidence nodes comprises comparing evidence obtained by the natural language input recognizer to one or more evidence definitions defined in an evidence model; and determining whether the evidence holds for the one or more evidence definitions.

10 . The method of claim 1 , wherein:
the recognition result selected for the natural language input is a preliminary recognition result that is selected before obtaining additional evidence from one of the evidence sources.

11 . The method of claim 1 , wherein:
the recognition result selected for the natural language input is a final recognition result selected for the natural language input received.

12 . The method of claim 1 , further comprising:
triggering, by the natural language input recognizer, a response associated with the recognition result selected for the natural language input.

13 . The method of claim 12 , wherein:
triggering the response comprises comparing a response definition defined in a response model to the recognition result selected for the natural language input.

14 . The method of claim 13 , wherein:
triggering the response further comprises determining that an evidence definition defined in the response model is satisfied by the evidence obtained.

15 . The method of claim 1 , wherein obtaining the evidence from the evidence source comprises receiving, by the natural language input recognizer, a notification comprising the evidence from the evidence source.

16 . The method of claim 1 , wherein obtaining the evidence from the evidence source comprises querying, by the natural language input recognizer, the evidence source and receiving, by the natural language input recognizer, a response comprising the evidence from the evidence source.

17 . The method of claim 1 , wherein the natural language input interface is one of a speech input interface that receives the natural language input as speech input, a text input interface that receives the natural language input as text input, or a handwriting input interface that receives the natural language input as handwriting input.

18 . The method of claim 1 , wherein the evidence comprises at least one recognition hypothesis provided by an automatic speech recognition (ASR) engine.

19 . The method of claim 18 , wherein the ASR engine comprises at least one of a local ASR engine that resides locally at the computing device and a remote ASR engine that is located remotely relative to the computing device and is in signal communication with the computing device via a network.

20 . The method of claim 1 , wherein the evidence comprises at least one hypothesis provided by a natural language understanding (NLU) engine.

21 . The method of claim 20 , wherein the NLU engine comprises at least one of a local NLU engine that resides locally at the computing device and a remote NLU engine that is located remotely relative to the computing device and is in signal communication with the computing device via a network.

22 . The method of claim 1 , wherein the evidence comprises a current state of an application executing at the computing device.

23 . The method of claim 1 , wherein the evidence comprises a current state of a user that provided the natural language input.

24 . The method of claim 1 , wherein the evidence comprises at least one of i) a location, ii) a distance to a point-of-interest, iii) a date, and iv) a time of day.

25 . The method of claim 1 , wherein the evidence comprises a usage history stored at the computing device.

26 . A device comprising:
a natural language input interface configured to receive natural language input and generate a digital signal corresponding to the natural language input; and a natural language input recognizer configured to obtain evidence from an evidence source,
generate an evidence graph based, at least in part, on the evidence obtained,
generate a set of recognition hypotheses based, at least in part, on the evidence graph, and
select one of the recognition hypotheses as a recognition result for the natural language input.

27 . The device of claim 26 , wherein:
the natural language input recognizer is configured to asynchronously obtain evidence from a plurality of evidence sources, and generate the set of recognition hypothes

Assignees

• Nuance Communications Inc

Inventors

• Labský Martin
• Kunc Ladislav
• Kleindienst Jan
• Macek Tomás
• D'Hoore Bart

Classifications

• G06F40/35

  Discourse or dialogue representation · CPC title

• G10L15/18 Primary

  using natural language modelling · CPC title

• G06F17/28 Primary

  Physics · mapped topic