Encoder Model Semantic Matching for Natural Language Actions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automated assistants often fail to perform intended actions in response to user queries due to lack of explicit mappings, leading to resource wastage and user frustration.

Innovation Solution

Training an encoder model to generate semantic encodings of natural language inputs, allowing for accurate determination of responsive actions by comparing encodings to pre-determined mappings, even when explicit mappings are absent.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If automated assistants use explicit mapping for query-action pairs, then action accuracy is improved, but adaptability to new queries deteriorates

Engineering Contradiction:

Improveaction determination accuracyVSAvoidquery understanding flexibility

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent introduces an encoder model as an intermediary between the natural language query and the action determination system. The encoder transforms diverse query formulations into a unified semantic representation space, enabling both accurate action matching and adaptability to new queries without explicit mappings.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the parameter representation from discrete explicit mappings to continuous semantic encodings. By representing queries and actions in a shared vector space with learned embeddings, the system achieves flexible adaptation while maintaining determination accuracy through similarity computation.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If automated assistants process each query explicitly, then action determination accuracy is improved, but computational resource usage worsens

Engineering Contradiction:

Improveaction determination accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The encoder model is pre-trained on large datasets to learn semantic representations offline. During inference, the pre-trained encoder rapidly encodes queries into the action space without requiring intensive processing, thus maintaining accuracy while reducing online computational resources.

Inventive Principle:

Principle #10Preliminary action

3Adaptability or versatility

If automated assistants use semantic encoding for query processing, then adaptability to new queries is improved, but measurement precision of action determination may worsen

Engineering Contradiction:

Improvequery understanding flexibilityVSAvoidaction determination accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent replaces traditional mechanical rule-based matching with a learned semantic encoding system. The encoder model learns to map queries to actions through training data, substituting rigid explicit mappings with flexible learned representations that maintain or improve determination accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20250384350A1Training encoder model and/or using trained encoder model to determine responsive action(s) for natural language input

Publication Date: 2025.12.18 GOOGLE LLC

AI summary

Systems, methods, and computer readable media related to: training an encoder model that can be utilized to determine semantic similarity of a natural language textual string to each of one or more additional natural language textual strings (directly and/or indirectly); and/or using a trained encoder model to determine one or more responsive actions to perform in response to a natural language query. The encoder model is a machine learning model, such as a neural network model. In some implementations of training the encoder model, the encoder model is trained as part of a larger network architecture trained based on one or more tasks that are distinct from a “semantic textual similarity” task for which the encoder model can be used.