Generic Language Understanding Model Configuration

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Natural language understanding (NLU) systems require significant machine learning expertise, substantial labeled data, and are not easily accessible for configuring new models, making it difficult to leverage existing data and models for new applications.

Innovation Solution

The system identifies and collects previously configured schemas for various applications, generates a generic schema, maps the collected schemas, and uses the mapped schema to train models, providing an interface for browsing and selecting models, which can be bundled for implementation on computing devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If statistical data models are used for NLU systems, then model quality can be improved, but the complexity of configuration and expertise required increases significantly

Engineering Contradiction:
Improvemodel qualityVSAvoidconfiguration complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent creates a universal schema that can be used across multiple NLU applications and domains. Instead of requiring separate statistical data models for each application, the system uses a single configurable schema that can be adapted to different domains through configuration files, reducing the need for machine learning expertise while maintaining model quality.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system copies and reuses existing schema definitions across multiple applications. Rather than creating new statistical models from scratch for each application, the patent allows copying of proven schemas and adapting them through configuration, significantly reducing the expertise and time required while maintaining reliability.

Inventive Principle:
Principle #26Copying

2Measurement precision

If application-specific labeled data is collected for training models, then model accuracy improves, but the time and resources required for data collection and model configuration increase

Engineering Contradiction:
Improvemodel accuracyVSAvoidmodel configuration time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-defining schema structures, data formats, and model configurations that can be reused across applications. Configuration files are prepared in advance with template schemas that can be quickly adapted to new applications, eliminating the need for time-consuming data collection and model training from scratch.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system allows model adaptation through parameter changes in configuration files rather than requiring complete retraining. By changing parameters such as domain-specific terms, slot definitions, and intent patterns in the schema configuration, the system can adapt to new applications quickly while maintaining the underlying trained model structure.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If multiple separate models are trained for different applications, then each model can be optimized for its specific domain, but the bandwidth and storage resources required increase

Engineering Contradiction:
Improvedomain-specific performanceVSAvoiddata and model size
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent merges multiple application-specific models into a single unified schema-based system. Instead of maintaining separate trained models for each application, the system combines them into one configurable schema that can serve multiple domains, significantly reducing the quantity of data and model files required while maintaining domain-specific performance through configuration.

Inventive Principle:
Principle #5Merging (Combining)

4Adaptability or versatility

If new NLU models are configured from scratch, then models can be customized for specific needs, but the expertise and resources required make it difficult to leverage existing models

Engineering Contradiction:
Improvemodel customizationVSAvoidmodel configuration ease
Core Design Contradiction:
Adaptability or versatilityVSEase of manufacture

Solution Approach 1:

The patent segments the model configuration into distinct layers: a universal schema structure, domain-specific adaptations, and application-specific customizations. This segmentation allows users to leverage existing schemas while easily customizing specific parts for their needs, making model configuration accessible without requiring deep machine learning expertise.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The schema acts as an intermediary between existing trained models and new application requirements. Instead of requiring direct customization of complex statistical models, the schema provides a simplified configuration interface that mediates between the underlying model and application-specific needs, enabling easy customization while leveraging existing model capabilities.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3405946B1Configurable generic language understanding models
Publication Date: 2024.07.03 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3405946B1 patent drawingFigure 1
  • EP3405946B1 patent drawingFigure 2
  • EP3405946B1 patent drawingFigure 3

AI summary

Examples of the present disclosure describe systems and methods of configuring generic language understanding models. In aspects, one or more previously configured schemas for various applications may be identified and collected. A generic schema may be generated using the collected schemas. The collected schemas may be programmatically mapped to the generic schema. The generic schema may be used to train on ore more models. An interface may be provided to allow browsing the models. The interface may include a configuration mechanism that provides for selecting on or more of the models. The selected models may be bundled programmatically, such that the information and instructions needed to implement the models are configured programmatically. The bundled models may then be provided to a requestor.