Method for follow-up expression for intelligent assistance

Patent metadata

Field	Value
Publication number	US-10929606-B2
Application number	US-201815904196-A
Country	US
Kind code	B2
Filing date	Feb 23, 2018
Priority date	Dec 29, 2017
Publication date	Feb 23, 2021
Grant date	Feb 23, 2021

Abstract

A method for intelligent assistance includes identifying one or more insertion points within an input comprising text for providing additional information. A follow-up expression that includes at least a portion of the input and the additional information at the one or more insertion points is generated for clarifying or supplementing meaning of the input.

First claim

What is claimed is:

1. A method for intelligent assistance, comprising:
identifying one or more insertion points within a natural language input comprising text for providing additional information;
training a backwards language model to predict a previous word given following words in the natural language input;
representing a text expression from the natural language input as a graph;
creating nodes in the graph that represent the additional information;
using the trained backwards language model for: inserting the nodes within the graph, generating and scoring edge weights linking the nodes to the graph, determining node paths within the graph, and determining backwards scores for edges in a back half of the inserted nodes; and
generating a follow-up expression based on the backwards scores and the prediction from the trained backwards language model, the follow-up expression including at least a portion of the natural language input and the additional information at the one or more insertion points for clarifying or supplementing meaning of the natural language input.

2. The method of claim 1 , wherein identifying the one or more insertion points within the natural language input is based on a natural language model, and the backwards language model comprises a neural network.

3. The method of claim 2 , wherein the additional information is determined to be consistent with intent of the natural language input.

4. The method of claim 3 , further comprising:
selecting a set of words from the natural language input while maintaining sequential order of the words present in the natural language input.

5. The method of claim 4 , further comprising:
partitioning the set of words into a first subset and a second subset.

6. The method of claim 5 , further comprising:
determining a word distance between the first subset and the second subset based on the natural language model; and
determining the one or more insertion points based on the word distance.

7. The method of claim 6 , further comprising:
for each insertion point, identifying words to be inserted for forming the additional information, using both a trained forward language model and the trained backwards language model, wherein the trained forward language model predicts probability of a next word given prior words in the natural language input, the trained backwards language model predicts probability of the previous word given following words in the natural language input; and
using the trained forward language model for: inserting the nodes within the graph, generating and scoring of the edge weights, determining of the node paths within the graph, and determining forwards scores for edges in a front half of the inserted nodes, wherein the node paths are further scored with the forwards scores, and generating the follow-up expression is based on the backwards scores, the forward scores, and the prediction from both the trained backwards language model and the trained forward language model.

8. An electronic device comprising:
a memory storing instructions; and
at least one processor executing the instructions including a process configured to:
identify one or more insertion points within a natural language input comprising text for providing additional information;
train a backwards language model to predict a previous word given following words in the natural language input;
represent a text expression from the natural language input as a graph;
create nodes in the graph that represent the additional information;
use the trained backwards language model to: insert the nodes within the graph, generate and score edge weights linking the nodes to the graph, determine node paths within the graph, and determine backwards scores for edges in a back half of the inserted nodes; and
generate a follow-up expression based on the backwards scores and the prediction from the trained backwards language model, the follow-up expression including at least a portion of the natural language input and the additional information at the one or more insertion points for clarifying or supplementing meaning of the natural language input.

9. The electronic device of claim 8 , wherein a natural language model is used to identify the one or more insertion points within the natural language input, the backwards language model comprises a neural network.

10. The electronic device of claim 9 , wherein the additional information is determined to be consistent with intent of the natural language input.

11. The electronic device of claim 10 , wherein the process further comprises:
selection of a set of words from the natural language input while maintaining sequential order of the words present in the natural language input.

12. The electronic device of claim 11 , wherein the process further comprises:
partitioning the set of words into a first subset and a second subset;
determining a word distance between the first subset and the second subset based on the natural language model; and
determining the one or more insertion points based on the word distance.

13. The electronic device of claim 10 , wherein the process is further configured to:
for each insertion point, identify words to be inserted to form the additional information, using both a trained forward language model and the trained backwards language model, wherein the trained forward language model predicts probability of a next word given prior words in the natural language input, and the trained backwards language model predicts probability of the previous word given following words in the natural language input; and
use the trained forward language model to: insert the nodes within the graph, generate and score the edge weights, determine the node paths within the graph, and determine forwards scores for edges in a front half of the inserted nodes, wherein the node paths are further scored with the forwards scores, and generation of the follow-up expression is based on the backwards scores, the forward scores, and the prediction from both the trained backwards language model and the trained forward language model.

14. The electronic device of claim 8 , wherein the process is performed on at least one of a server device, a smart portable device, a smart appliance, or a combination thereof.

15. A non-transitory processor-readable medium that includes a program that when executed by a processor perform a method comprising:
identifying one or more insertion points within a natural language input comprising text for providing additional information;
training a backwards language model to predict a previous word given following words in the natural language input;
representing a text expression from the natural language input as a graph;
creating nodes in the graph that represent the additional information;
using the trained backwards language model for: inserting the nodes within the graph, generating and scoring edge weights linking the nodes to the graph, determining node paths within the graph, and determining backwards scores for edges in a back half of the inserted nodes; and
generating a follow-up expression based on the backwards scores and the prediction from the trained backwards language model, the follow-up expression including at least a portion of the natural language input and the additional information at the one or more insertion points for clarifying or supplementing meaning of the natural language input.

16. The non-transitory processor-readable medium of claim 15 , wherein identifying the one or more insertion points within the natural language input

Assignees

• Samsung Electronics Co Ltd

Inventors

• Martineau Justin C
• Ray Avik
• Jin Hongxia

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06F16/3344 Primary

  using natural language analysis · CPC title