US 10,592,519 B2Granted

Computational-model operation using multiple subject representations

AssigneeMicrosoft Technology Licensing, LLC

InventorsXiaodong He, Li Deng, Jianfeng Gao, Wen-tau Yih, Moontae Lee, Paul Smolensky

Life of this patent

Estimating the term…

Loading patent…

Abstract

A processing unit can determine multiple representations associated with a statement, e.g., subject or predicate representations. In some examples, the representations can lack representation of semantics of the statement. The computing device can determine a computational model of the statement based at least in part on the representations. The computing device can receive a query, e.g., via a communications interface. The computing device can determine at least one query representation, e.g., a subject, predicate, or entity representation. The computing device can then operate the model using the query representation to provide a model output. The model output can represent a relationship between the query representations and information in the model. The computing device can, e.g., transmit an indication of the model output via the communications interface. The computing device can determine mathematical relationships between subject representations and attribute representations for multiple statements, and determine the model using the relationships.

Claims20

Dependent claims

An apparatus, comprising:

at least one processor;
and
a computer-readable medium including instructions to, when executed by the at least one processor, cause the at least one processor to:
analyze a natural-language statement to determine therein a first subject, a second subject, and a predicate;
determine a first-subject tensor representation associated with the first subject of the natural-language statement, a predicate tensor representation associated with the predicate of the natural-language statement and a second-subject tensor representation associated with the second subject of the natural-language statement;
use one or more tensor arithmetic operations to determine a tensor-based computational model of the natural-language statement based at least in part on the first-subject tensor representation, the predicate tensor representation, and the second-subject tensor representation, the model expressing a relationship between the first subject, the second subject, and the predicate;
determine, based at least in part on a query, at least two query representations comprising at least two of a first query-subject tensor representation, a second query-subject tensor representation, and a query-predicate tensor representation;
and
operate the tensor-based computational model using the two query representations to provide a model output corresponding to the query, wherein the model output represents a relationship between the two query representations and at least some information in the computational model.

6 more elements

A system, comprising:

a communications interface;
at least one processor adapted to execute modules;
and
one or more computer-readable storage media having thereon a plurality of modules, the plurality of modules comprising:
a module of a representation engine that is configured to:
determine a first-subject tensor representation associated with a first subject of a natural-language statement;
determine a second-subject tensor representation associated with a second subject of the natural-language statement;
and
use one or more tensor arithmetic operations to determine a tensor-based computational model of the natural-language statement based at least in part on the first-subject tensor representation and the second-subject tensor representation,
wherein the first-subject tensor representation and the second-subject tensor representation include at least one of a vector or a matrix, and wherein the tensor-based computational model includes the at least one of a vector or a matrix;
a module of a query engine that is configured to:
determine an entity tensor representation based at least in part on a query;
and
operate the tensor-based computational model using the entity representation to provide a model output corresponding to the query;
and a module of a communications engine that is configured to:
receive the query via the communications interface;
and
transmit an indication of the model output via the communications interface.

15 more elements

A method, comprising,

using one or more processors, performing operations comprising:
determining respective subject tensor representations associated with respective natural-language statements of a plurality of natural-language statements, wherein at least one of the subject tensor representations lacks representation of semantics of the respective natural-language statement;
determining respective attribute tensor representations associated with the respective natural-language statements of the plurality of natural-language statements;
determining at least one mathematical relationship between one or more individual ones of the subject tensor representations and one or more respective ones of the attribute tensor representations;
using one or more tensor arithmetic operations to determine a tensor-based computational model based at least in part on the at least one mathematical relationship;
determining, based at least in part on a query, at least one query representation;
operating the tensor-based computational model based on the at least one query representation to provide a model output, wherein the model output represents a relationship between the at least one query representations and at least some information in the tensor-based computational model,
wherein the subject tensor representation and the attribute tensor representation include at least one of a vector or a matrix, and wherein the tensor-based computational model includes the at least one of a vector or a matrix.

5 more elements

Dependencies are inferred from the claim text.