Method and system of text synthesis based on extracted information in the form of an RDF graph making use of templates

• US 10,210,249 B2
• Filed: 05/20/2015
• Issued: 02/19/2019
• Est. Priority Date: 03/19/2015
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY Production LLC (ABBYY Software)
• Inventors: Starostin, Anatoly, Kuklin, Dmitrii
• Primary Examiner(s): Shah, Bharatkumar S
• Application Number:
  US14/717,320
• Publication Number:
  US 20160275058A1
• Time in Patent Office: 1,371 Days
• Field of Search: 704 9
• CPC Class Codes: G06F 16/313 Selection or weighting of terms for indexing
• CPC Class Codes: G06F 16/367 Ontology
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Disclosed are system, method and computer program product for synthesis of natural-language text; receiving information objects; selecting among the received information objects information objects and an associated synthesis templates in a template library, each synthesis template including a template semantic-syntactic tree; generating for each selected information object a synthesis semantic-syntactic tree based on the template semantic-syntactic tree; and generating natural language text based on each generated synthesis semantic-syntactic tree.

Claims

1. A method comprising:

receiving, by a hardware processor, a plurality of information objects extracted from text data in one or more source text documents, the plurality of information objects having representations in a data model, wherein each information object comprises one or more properties and, for each property, a corresponding natural language string value from the text data;

selecting, by the hardware processor, at least one information object and a synthesis template associated with the selected information object, wherein the selecting of the at least one information object and the synthesis template associated with the selected information object comprises;

selecting the at least one information object from the plurality of information objects; and

selecting the synthesis template associated with the selected information object from a plurality of synthesis templates stored in a template library, the synthesis template comprising a sentence in a natural language, a plurality of substitutions comprising one or more words from the sentence and their associated properties, and a template semantic-syntactic tree, and wherein the template semantic-syntactic tree is generated based on the sentence in the natural language, wherein the selecting of the synthesis template associated with the selected information object comprises;

for each synthesis template from the plurality of synthesis templates;

comparing a property associated with a word in a plurality of substitutions of a respective synthesis template with one or more properties of the selected information object;

responsive to determining that the property of the respective synthesis template matches one of the one or more properties of the selected information object, determining whether a natural language string value corresponding to the matching property of the selected information object is nominative or not; and

responsive to determining that the natural language string value is not nominative, eliminating the respective synthesis template from the plurality of synthesis templates for selection;

generating, by the hardware processor, for each selected information object, a synthesis semantic-syntactic tree based on the template semantic-syntactic tree and the plurality of substitutions of the synthesis template associated with the selected information object; and

generating, by the hardware processor, a natural language text based on each generated synthesis semantic-syntactic tree to provide an annotation for the text data in the one or more source text documents.

2. The method of claim 1, wherein generating for the selected information object a synthesis semantic-syntactic tree further comprises:

responsive to determining that the template node is in the substitutions of the associated synthesis template and the associated property of the template node is a filled property of the selected information object, and the filled property of the selected information object has more than one natural-language string value,

for each natural-language string value, generating in the synthesis semantic-syntactic tree a node or a sub-tree based on a semantic-syntactic analysis of the natural-language string value of the filled property; and

connecting the generated nodes with a coordinating link.

3. The method of claim 1, further comprising:

forming at least one group of selected information objects associated with the same synthesis template; and

generating for the at least one group the synthesis semantic-syntactic tree based on the template semantic-syntactic tree of the associated synthesis template.

8. The method of claim 1, wherein the information object is associated with an ontological object and has a set of filled properties, a filled property having a value.

9. The method of claim 1, wherein the synthesis template is associated with an ontological object.

10. The method of claim 1, wherein the synthesis template comprises a set of required properties.

11. The method of claim 1, wherein the synthesis template comprises a set of optional properties.

12. The method of claim 1, wherein the synthesis template comprises a validation script.

13. The method of claim 1, wherein selecting the at least one information object and the synthesis template associated with the selected information object from a plurality of synthesis templates comprises:

for an information object of the plurality of information objects, determining whether there is a set of synthesis templates from the plurality of synthesis templates associated with the same ontological object as the information object;

responsive to determining that there is a set of synthesis templates associated with the same ontological object as the information object, determining whether the set of synthesis templates includes a first remaining set of synthesis templates that each have a set of required properties matching a set of filled properties of the information object;

responsive to determining that the set of synthesis templates includes the first remaining set of synthesis templates that each have the set of required properties matching the set of filled properties of the information object, determining whether the first remaining set includes a second remaining set of synthesis templates with the largest set of required properties, wherein the first remaining set comprises the second remaining set;

responsive to determining that the first remaining set includes the second remaining set of synthesis templates with the largest set of required properties, determining whether a third remaining set includes a fourth remaining set of synthesis templates with the largest intersection of the set of optional properties with the set of filled properties of the information object; and

responsive to determining that the third remaining set includes the fourth remaining set of synthesis templates with the largest intersection of the set of optional properties with the set of filled properties of the information object, adding the information object to the at least one selected information object and associating a synthesis template from the fourth remaining set of synthesis templates with the added information object.

14. The method of claim 1, wherein the selected information object has a set of filled properties, a filled property having a natural-language string value.

15. The method of claim 1, wherein the template semantic-syntactic tree comprises template nodes representing words of the template sentence in the natural language and edges representing relations between the words.

16. The method of claim 1, wherein the template sentence and the plurality of substitutions are identified by a user.

17. The method of claim 1, wherein the plurality of substitutions forming a substitution list.

18. The method of claim 1, wherein the template semantic-syntactic tree comprises for each of one or more template nodes an associated property from the plurality of substitution parameters.

19. The method of claim 1, wherein generating for the selected information object a synthesis semantic-syntactic tree comprising a root node and one or more leaf nodes comprises traversing the template semantic-syntactic tree from the root node to the leaf nodes.

20. The method of claim 1, wherein generating for the selected information object a synthesis semantic-syntactic tree comprises:

for a template node of the template semantic-syntactic tree,

generating an identical node in the synthesis semantic-syntactic tree, responsive to determining the template node is not in the plurality of substitution parameters of the synthesis template;

generating a node or a sub-tree in the synthesis semantic-syntactic tree based on a semantic-syntactic analysis of the natural-language string value of the filled property, responsive to determining the template node is in the plurality of substitution parameters of the synthesis template and the associated property of the template node is a filled property of the selected information object; and

refraining from generating a node in the synthesis semantic-syntactic tree, responsive to determining the template node is in the plurality of substitution parameters of synthesis template and the associated property of the template node is not a filled property of the information object.

4. A system comprising:

a memory; and

a hardware processor, coupled to the memory, to;

receive a plurality of information objects extracted from text data in one or more source text documents, the plurality of information objects having representations in a data model, wherein each information object comprises one or more properties and, for each property, a corresponding natural language string value from the text data;

select at least one information object and a synthesis template associated with the selected information object, wherein to select the at least one information object and the synthesis template associated with the selected information object, the hardware processor further to;

select the at least one information object from the plurality of information objects; and

select the synthesis template from a plurality of synthesis templates stored in a template library, the synthesis template comprising a sentence in a natural language, a plurality of substitutions parameters comprising one or more words from the sentence and their associated properties, and a template semantic-syntactic tree, and wherein the template semantic-syntactic tree is generated based on the sentence in the natural language wherein to select the synthesis template associated with the selected information object, the hardware processor further to;

for each synthesis template from the plurality of synthesis templates;

compare a property associated with a word in a plurality of substitutions of a respective synthesis template with one or more properties of the selected information object;

responsive to determining that the property of the respective synthesis template matches one of the one or more properties of the selected information object, determine whether a natural language string value corresponding to the matching property of the selected information object is nominative or not; and

responsive to determining that the natural language string value is not nominative, eliminate the respective synthesis template from the plurality of synthesis templates for selection;

generate for each selected information object a synthesis semantic-syntactic tree based on the template semantic-syntactic tree and the plurality of substitutions of the synthesis template associated with the selected information object; and

generate a natural language text based on each generated synthesis semantic-syntactic tree to provide an annotation for the text data in the one or more source text documents.

5. The system of claim 4, is further to:

form at least one group of selected information objects associated with the same synthesis template; and

generate for the at least one group the synthesis semantic-syntactic tree based on the template semantic-syntactic tree of the associated synthesis template.

6. A computer program product stored on a non-transitory computer-readable storage medium, the computer program product comprising computer-executable instructions for synthesis of natural-language text, comprising instructions for:

receiving, by a hardware processor, a plurality of information objects extracted from text data in one or more source text documents, the plurality of information objects having representations in a data model, wherein each information object comprises one or more properties and, for each property, a corresponding natural language string value from the text data;

selecting, by the hardware processor, at least one information object and a synthesis template associated with the selected information object, wherein the selecting of the at least one information object and the synthesis template associated with the selected information object comprises;

selecting the at least one information object from the plurality of information objects; and

selecting the synthesis template from a plurality of synthesis templates stored in a template library, the synthesis template comprises a sentence in a natural language, a plurality of substitutions comprising one or more words from the sentence and their associated properties, and a template semantic-syntactic tree, and wherein the template semantic-syntactic tree is generated based on the sentence in the natural language, wherein the selecting of the synthesis template associated with the selected information object comprises;

for each synthesis template from the plurality of synthesis templates;

comparing a property associated with a word in a plurality of substitutions of a respective synthesis template with one or more properties of the selected information object;

responsive to determining that the property of the respective synthesis template matches one of the one or more properties of the selected information object, determining whether a natural language string value corresponding to the matching property of the selected information object is nominative or not; and

responsive to determining that the natural language string value is not nominative, eliminating the respective synthesis template from the plurality of synthesis templates for selection;

generating, by the hardware processor, for the selected information object, a synthesis semantic-syntactic tree based on the template semantic-syntactic tree and the plurality of substitutions of the synthesis template associated with the selected information object;

generating by the hardware processor, natural language text based on each generated synthesis semantic-syntactic tree to provide an annotation for the text data in the one or more source text documents.

7. The computer program product of claim 6, further comprising instructions for:

forming at least one group of selected information objects associated with the same synthesis template; and

generating for the at least one group a synthesis semantic-syntactic tree based on the template semantic-syntactic tree of the associated synthesis template.