System and method for coupled detection of syntax and semantics for natural language understanding and generation

Patent metadata

Field	Value
Publication number	US-10402499-B2
Application number	US-201715814323-A
Country	US
Kind code	B2
Filing date	Nov 15, 2017
Priority date	Nov 17, 2016
Publication date	Sep 3, 2019
Grant date	Sep 3, 2019

Abstract

A method includes performing, with at least one processing device, natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context and (ii) a syntax generated from common semantic relations between sequential words and clauses. The semantic word and clause representation and the syntax could be generated iteratively. The method could include generating the semantic word and clause representation, where a significance of particular items of context is modified by a presence of words or clauses from a given lexicon of amplifiers along a syntax tree. The method could also include generating the syntax, where multiple semantic relations are considered to be equivalent if the multiple semantic relations have identical representations in an auto-associative memory. The semantic word and clause representation could include content semantics constructed from closest nodes in a syntax tree to a clause.

First claim

What is claimed is:

1. A method comprising:
constructing, by at least one processing device, content semantics from closest nodes in a syntax tree to a clause, each closest node having a property that a child of the node is not as common relative to the node as the node is to its parent; and
performing, by the at least one processing device, natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context, the semantic word and clause representation comprising the content semantics, and (ii) a syntax generated from common semantic relations between sequential words and clauses.

2. The method of claim 1 , wherein the semantic word and clause representation and the syntax are generated iteratively.

3. The method of claim 1 , further comprising:
generating the semantic word and clause representation, wherein a significance of particular items of context is modified by a presence of words or clauses from a given lexicon of amplifiers along a syntax tree.

4. The method of claim 1 , further comprising:
generating the semantic word and clause representation by performing disambiguation to separate instances of particular items of context that have different semantic meanings, wherein performing the disambiguation comprises clustering instances of words and clauses based on similarity of their syntactically-labeled context.

5. The method of claim 1 , further comprising:
generating the syntax, wherein multiple semantic relations are considered to be equivalent if the multiple semantic relations have identical representations in an auto-associative memory.

6. The method of claim 1 , further comprising:
measuring a sentiment of a word or clause based on components of the semantic word and clause representation coming from emotional context according to a given lexicon of emotional words and clauses.

7. The method of claim 1 , further comprising:
performing, with the at least one processing device, natural language generation using the semantic word and clause representation and the syntax.

8. A method comprising:
performing, with at least one processing device, natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context and (ii) a syntax generated from common semantic relations between sequential words and clauses; and
determining properties of words and clauses by calculating inner products of semantic vectors and semantic co-vectors, the semantic vectors and semantic co-vectors comprising coefficients corresponding to context items, the semantic vectors comprising sparse vectors, the semantic co-vectors comprising patterned vectors with coefficient values that are repeated across context items sharing a semantic property.

9. An apparatus comprising:
at least one processing device; and
at least one memory storing instructions that, when executed by the at least one processing device, cause the at least one processing device to:
construct content semantics from closest nodes in a syntax tree to a clause, each closest node having a property that a child of the node is not as common relative to the node as the node is to its parent; and
perform natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context, the semantic word and clause representation comprising the content semantics, and (ii) a syntax generated from common semantic relations between sequential words and clauses.

10. The apparatus of claim 9 , wherein the semantic word and clause representation and the syntax are generated iteratively.

11. The apparatus of claim 9 , wherein:
the at least one memory further stores instructions that, when executed by the at least one processing device, cause the at least one processing device to generate the semantic word and clause representation; and
a significance of particular items of context is modified by a presence of words or clauses from a given lexicon of amplifiers along a syntax tree.

12. The apparatus of claim 9 , wherein:
the at least one memory further stores instructions that, when executed by the at least one processing device, cause the at least one processing device to generate the semantic word and clause representation by performing disambiguation to separate instances of particular items of context that have different semantic meanings; and
the instructions that when executed cause the at least one processing device to perform the disambiguation comprise instructions that when executed cause the at least one processing device to cluster instances of words and clauses based on similarity of their syntactically-labeled context.

13. The apparatus of claim 9 , wherein:
the at least one memory further stores instructions that, when executed by the at least one processing device, cause the at least one processing device to generate the syntax; and
multiple semantic relations are considered to be equivalent if the multiple semantic relations have identical representations in an auto-associative memory.

14. The apparatus of claim 9 , wherein the at least one memory further stores instructions that, when executed by the at least one processing device, cause the at least one processing device to measure a sentiment of a word or clause based on components of the semantic word and clause representation coming from emotional context according to a given lexicon of emotional words and clauses.

15. The apparatus of claim 9 , wherein the at least one memory further stores instructions that, when executed by the at least one processing device, cause the at least one processing device to perform natural language generation using the semantic word and clause representation and the syntax.

16. An apparatus comprising:
at least one processing device; and
at least one memory storing instructions that, when executed by the at least one processing device, cause the at least one processing device to:
perform natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context and (ii) a syntax generated from common semantic relations between sequential words and clauses; and
determine properties of words and clauses by calculating inner products of semantic vectors and semantic co-vectors, the semantic vectors and semantic co-vectors comprising coefficients corresponding to context items, the semantic vectors comprising sparse vectors, the semantic co-vectors comprising patterned vectors with coefficient values that are repeated across context items sharing a semantic property.

17. A non-transitory computer readable medium containing instructions that, when executed by at least one processing device, cause the at least one processing device to:
construct content semantics from closest nodes in a syntax tree to a clause, each closest node having a property that a child of the node is not as common relative to the node as the node is to its parent; and
perform natural language understanding using both (i) a semantic word and clause representation generated from syntactically-labeled context, the semantic word and clause representation comprising the content semantics, and (ii) a syntax generated from common semantic relations between sequential words and clauses.

18. The non-transitory computer readable medium of claim 17 , wherein the semantic word and clause representation and the syntax are generated iteratively.

19. The non-transitory computer readable

Assignees

• Goldman Sachs & Co Llc

Inventors

• Burchard Paul

Classifications

• G06F40/284 Primary

  Lexical analysis, e.g. tokenisation or collocates · CPC title

• G06F40/56 Primary

  Natural language generation · CPC title

• G06F40/211

  Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06F17/2881 Primary

  Physics · mapped topic