Automatic generation of a speech by processing raw claims to a set of arguments

Patent metadata

Field	Value
Publication number	US-9753916-B2
Application number	US-201514698847-A
Country	US
Kind code	B2
Filing date	Apr 29, 2015
Priority date	Jun 19, 2014
Publication date	Sep 5, 2017
Grant date	Sep 5, 2017

Abstract

A method comprising using at least one hardware processor for: identifying relations between pairs of claims of a set of claims; aggregating the claims of the set of claims into a plurality of clusters based on the identified relations; generating a plurality of arguments from the plurality of clusters, wherein each of the arguments is generated from a cluster of the plurality of clusters, and wherein each of the arguments comprises at least one claim of the set of claims, scoring each possible set of a predefined number of arguments of the plurality of arguments, based on a quality of each argument of the predefined number of arguments and on diversity between the predefined number of arguments; and generating a speech, wherein the speech comprises a top scoring possible set of the possible set of the predefined number of arguments.

First claim

What is claimed is:

1. A method comprising using at least one hardware processor for:
receiving a topic under consideration (TUC), wherein the TUC is a debatable topic in a free text format;
learning a claim function by applying a machine learning technique to a claim training dataset, wherein the claim training dataset comprises: a content item, and claims selected from the content item by a group of people;
providing the TUC as input to the claim function, wherein the claim function is configured to mine at least one content resource;
applying the claim function to the at least one content resource, to extract said set of claims, wherein each claim of said set of claims is a concise statement with respect to the TUC;
learning a classification function by applying a machine learning technique to a classification training dataset, wherein the classification training dataset comprises: an example TUC, and claims that are classified with respect to the example TUC;
providing the TUC as input to the classification function;
applying the classification function to the extracted claims, to output one or more classification tags for each of the extracted claims, wherein the classification tags comprise at least one of:
a tag indicating that a certain one of the extracted claims is a pro claim with respect to the TUC,
a tag indicating that a certain one of the extracted claims is a con claim with respect to the TUC,
a tag indicating that a certain one of the extracted claims is a factual claim, and
a tag indicating that a certain one of the extracted claims is a moral claim;

automatically identifying relations between pairs of claims of a set of the extracted claims;
automatically aggregating the claims of the set of extracted claims into a plurality of clusters based on the identified relations;
automatically generating a plurality of arguments from the plurality of clusters, wherein each of the arguments is generated from a cluster of the plurality of clusters, and wherein each of the arguments comprises at least one claim of the set of extracted claims;
automatically scoring each possible set of a predefined number of arguments of the plurality of arguments, based on a quality of each argument of the predefined number of arguments and on diversity between the predefined number of arguments; and
automatically generating a speech, wherein the speech comprises a top scoring possible set of said each possible set of the predefined number of arguments.

2. The method of claim 1 , further comprising using said at least one hardware processor for automatically receiving a set of evidence supporting and associated with the set of extracted claims.

3. The method of claim 2 , wherein each argument of the set of arguments further comprises:
evidence of the set of evidence supporting and associated with each of the at least one claim, and
one or more of the classification tags associated with each of the at least one claim.

4. The method of claim 3 , wherein said evidence comprises expert evidence and wherein the method further comprises generating a description relating to the expert and incorporating said description in said speech.

5. The method of claim 1 , wherein said automatically identifying of relations between said pairs of claims comprises:
determining the existence of one or more relations between each of said pairs of claims, and
identifying the type of the one or more relations between the claims of said each of said pairs of claims.

6. The method of claim 1 , further comprising using said at least one hardware processor for:
for each pair of claims of said pairs of claims, automatically computing a relatedness score based on the extent of a relation between said each pair of claims,
wherein said aggregating of the claims into a plurality of clusters is according to their relatedness score.

7. The method of claim 1 , wherein the generating of the plurality of arguments comprises:
assessing the quality of the claims in each of said plurality of clusters and assigning each of said claims with a claim quality score, and
selecting of said at least one claim of said each argument from claims of said cluster based on their quality score.

8. The method of claim 1 , further comprising using said at least one hardware processor for automatically generating one or more paragraphs serving as at least one of a speech opening and a speech conclusion.

9. The method of claim 1 , further comprising using said at least one hardware processor for automatically generating a counter argument, the counter argument comprising one or more claims opposing one or more opponent claims.

10. The method of claim 9 , wherein the generating of the counter argument comprises at least one of:
identifying a contrast relation between each claim of the set of extracted claims and one or more of the one or more opponent claims, and
negating one or more of the one or more opponent claims.

11. A computer program product comprising a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by at least one hardware processor to:
receive a topic under consideration (TUC), wherein the TUC is a debatable topic in a free text format;
learn a claim function by applying a machine learning technique to a claim training dataset, wherein the claim training dataset comprises: a content item, and claims selected from the content item by a group of people;
provide the TUC as input to the claim function, wherein the claim function is configured to mine at least one content resource;
apply the claim function to the at least one content resource, to extract said set of claims, wherein each claim of said set of claims is a concise statement with respect to the TUC;
learn a classification function by applying a machine learning technique to a classification training dataset, wherein the classification training dataset comprises: an example TUC, and claims that are classified with respect to the example TUC;
provide the TUC as input to the classification function;
apply the classification function to the extracted claims, to output one or more classification tags for each of the extracted claims, wherein the classification tags comprise at least one of:
a tag indicating that a certain one of the extracted claims is a pro claim with respect to the TUC,
a tag indicating that a certain one of the extracted claims is a con claim with respect to the TUC,
a tag indicating that a certain one of the extracted claims is a factual claim, and
a tag indicating that a certain one of the extracted claims is a moral claim;

automatically identify relations between pairs of claims of a set of extracted claims;
automatically aggregate the claims of the set of extracted claims into a plurality of clusters based on the identified relations;
automatically generate a plurality of arguments from the plurality of clusters, wherein each of the arguments is generated from a cluster of the plurality of clusters and wherein each of the arguments comprises at least one claim of the set of extracted claims;
automatically score each possible set of a predefined number of arguments of the plurality of arguments, based on a quality of each argument of the predefined number of arguments and on diversity between the predefined number of arguments; and
automatically generate a speech, wherein the speech comprises a top scoring possible set of the possible set of the predefined number of arguments.

12. The computer program product of claim 11 , wherein said program code is further executable by said at least one hardware processor to automatically receive a

Assignees

• IBM

Inventors

• Aharoni Ehud
• Bhattacharya Indrajit
• Bilu Yonatan
• Gutfreund Dan
• Hershcovich Daniel
• Raykar Vikas
• Rinott Ruty
• Shantanu Godbole
• Slonim Noam

Classifications

• G06F40/56 Primary

  Natural language generation · CPC title

• G10L13/027 Primary

  Concept to speech synthesisers; Generation of natural phrases from machine-based concepts (generation of parameters for speech synthesis out of text G10L13/08) · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06F17/2881 Primary

  Physics · mapped topic

• G06F17/2785

  Physics · mapped topic