Dialog Knowledge Acquisition for Social Agents

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional approaches to content authoring for social agents, such as robots and animated characters, are burdensome and limit their ability to engage in extended verbal interactions by lacking the variety and contextuality of human expression, making interactions stale.

Innovation Solution

The development of dialog knowledge acquisition systems that utilize fully-situated and semi-situated learning modes to autonomously acquire and consolidate dialog knowledge, allowing the agent to adapt and vary its responses based on contextual cues and user interactions, without retaining personally identifiable information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional content authoring approaches are used for social agents, then the system structure remains simple, but the variety and contextuality of expression are limited, making interactions stale

Engineering Contradiction:

Improvevariety of expressionVSAvoidcontent authoring burden

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The social agent autonomously acquires dialog knowledge through fully-situated and semi-situated learning modes, eliminating the need for extensive manual content authoring. The system learns from interactions itself, generating varied expressions automatically without human intervention in each interaction scenario.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system pre-acquires dialog knowledge through offline learning processes before actual interactions occur. This preliminary knowledge acquisition enables the agent to have a rich repertoire of expressions ready for deployment during social interactions, reducing the need for real-time content creation.

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If manual content authoring is used to create varied expressions, then expression variety improves, but the time and resources required for content creation increase significantly

Engineering Contradiction:

Improvecontextuality of expressionVSAvoidcontent authoring time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system automatically learns contextually appropriate expressions through dialog knowledge acquisition from interactions and crowdsourced recommendations, eliminating the time-consuming manual content authoring process while maintaining high contextuality of expressions.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system uses feedback from user interactions and crowdsourced recommendations to continuously improve its dialog knowledge. This feedback loop enables the agent to learn which expressions are most contextually appropriate, replacing manual content creation with an automated learning process.

Inventive Principle:

Principle #23Feedback

3Productivity

If the system retains detailed user interaction data for learning, then dialog knowledge acquisition improves, but privacy concerns and data management complexity increase

Engineering Contradiction:

Improvelearning efficiencyVSAvoiddata management complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system extracts only the essential dialog knowledge and patterns from user interactions, discarding personally identifiable information. This extraction approach enables effective learning while simplifying data management and reducing privacy concerns by retaining only the necessary information.

Inventive Principle:

Principle #2Taking out (Extraction)

4Adaptability or versatility

If extensive dialog knowledge is acquired through multiple learning modes, then interaction naturalness improves, but the system complexity and computational resources required increase

Engineering Contradiction:

Improvenaturalness of interactionVSAvoidsystem architecture complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The dialog knowledge acquisition system is segmented into distinct learning modes (fully-situated learning, semi-situated learning) that can operate independently or in combination. This segmentation allows the system to manage complexity by breaking down the learning process into manageable components while achieving natural interactions.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS10162815B2Dialog knowledge acquisition system and method

Publication Date: 2018.12.25 DISNEY ENTERPRISES INC

AI summary

A dialog knowledge acquisition system includes a hardware processor, a memory, and hardware processor controlled input and output modules. The memory stores a dialog manager configured to instantiate a persistent interactive personality (PIP), and a dialog graph having linked dialog state nodes. The dialog manager receives dialog initiation data, identifies a first state node on the dialog graph corresponding to the dialog initiation data, determines a dialog interaction by the PIP based on the dialog initiation data and the first state node, and renders the dialog interaction. The dialog manager also receives feedback data corresponding to the dialog interaction, identifies a second state node based on the dialog initiation data, the dialog interaction, and the feedback data, and utilizes the dialog initiation data, the first state node, the dialog interaction, the feedback data, and the second state node to train the dialog graph.