Back to results
US 7,454,344 B2Granted
Language model architecture
AssigneeMicrosoft Corporation
InventorWilliam Ramsey
Life of this patent
Estimating the term…
Abstract
An architectural design is disclosed wherein a single reusable language model component is shared by multiple applications. The language model component is loaded once for a plurality of applications, thereby reducing the amount of memory consumed by the applications independently.
Claims12
A computer-implemented system for centralizing language model functionality, the system comprising:
- a first application configured to provide a first word sequence;
- a second application configured to provide a second word sequence;
- a language model component for processing the first and second word sequences, wherein the language model component includes a framework for generating a probability of encountering the first word sequence, as well as the probability of encountering the second word sequence;
- wherein the language model component is configured to provide the first word sequence to a second language model component;
- wherein the language model component is configured to provide the second word sequence to the second language model component;
- wherein the language model component is configured to receive from the second language model component a second set of probabilities representing a likelihood of encountering at least two sub-components of the second word sequence;
- wherein the language model component is further configured to support a processing of the second set of probabilities so as to enable generation of an overall probability of encountering the second word sequence;
- and
- wherein the first language model component is configured to provide the overall probability to the second application.
6 more elements
A computer-implemented method for centralizing language model functionality, the method comprising:
- receiving a first word sequence from a first application;
- receiving from a language model component a set of probabilities representing a likelihood of encountering at least two sub-components of the first word sequence;
- generating, based at least in part on the set of probabilities, an overall probability representing a likelihood of encountering the first word sequence;
- providing the overall probability to the first applications;
- receiving a second word sequence from a second application;
- receiving from the language model component a second set of probabilities representing a likelihood of encountering at least two sub-components of the second word sequence;
- generating, based at least in part on the second set of probabilities, a second overall probability representing a likelihood of encountering the second word sequence;
- providing the second overall probability to the second application;
- and
- wherein all steps of receiving, generating and providing are performed by a controlling language model component that is separate from said language model component.
7 more elements
A computer-implemented method for centralizing language model functionality, the method comprising:
- receiving from a first application a first word sequence;
- generating a first set of probabilities representing a likelihood of encountering at least two sub-components of the first word sequence;
- receiving from a second application a second word sequence;
- generating a second set of probabilities representing a likelihood of encountering at least two sub-components of the second word sequence;
- and
- transferring to the first application the first set of probabilities;
- and
- transferring to the second application the second set of probabilities.
5 more elements
Dependencies are inferred from the claim text.