Conversational Text Error Injection for Human-like Interaction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conversational agents generate text that appears unnatural and lacks errors typical of human communication, making it recognizable as machine-generated, which can lead to a lack of engagement and authenticity in interactions with human recipients.

Innovation Solution

A system that injects context-appropriate errors into conversational text generated by conversational agents, using AI to determine the acceptability of these errors based on user profiles and ensuring the meaning and context remain consistent with the original text, thereby mimicking human-generated text.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conversational agents generate text using a pre-determined format and tone, then the text is clear and error-free, but the text appears unnatural and noticeably machine-generated

Engineering Contradiction:

Improvetext accuracyVSAvoidnaturalness of text

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The system dynamically changes text parameters by selectively introducing errors (typos, slang, colloquialisms) based on user profiles and context analysis. This transforms the static, perfect machine-generated text into dynamic, variable text that mimics human typing patterns while maintaining contextual appropriateness

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system applies different error types and densities to different parts of the text based on local context. User-specific error patterns are applied selectively rather than uniformly, creating natural variation in different sections of the conversation while preserving overall meaning

Inventive Principle:

Principle #3Local quality

2Reliability

If conversational agents use formal language and predetermined formats, then the text maintains professional standards, but the text lacks informalities and errors associated with human typing

Engineering Contradiction:

Improvetext professionalismVSAvoidhuman-like characteristics

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The system adjusts formality parameters dynamically by selecting error types that match the user's communication style. Formal users receive minimal or subtle modifications, while informal users receive more pronounced error patterns, maintaining appropriate professionalism levels for each user

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The text generation transitions from static formal language to dynamic adaptive language that adjusts formality based on user profiles. The system dynamically selects which errors to introduce and at what density, creating a living, breathing text that adapts to each user's communication preferences

Inventive Principle:

Principle #15Dynamics

3Ease of manufacture

If the system introduces text entry errors into conversational text, then the text becomes more natural and human-like, but the context and meaning may become inconsistent

Engineering Contradiction:

Improvenaturalness of textVSAvoidcontext consistency

Core Design Contradiction:

Ease of manufactureVSLoss of information

Solution Approach 1:

The system implements feedback loops where introduced errors are validated against user profiles and contextual constraints. The AI analyzes whether errors maintain acceptable meaning consistency, and adjusts error selection based on this feedback, ensuring naturalness without sacrificing core message integrity

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system uses disposable, context-specific error patterns that are selected and applied temporarily rather than permanently. Errors are chosen based on immediate contextual needs and user preferences, then discarded after application, allowing flexible adjustment without long-term consistency constraints

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

4Ease of manufacture

If the system analyzes user profiles to determine error acceptability, then the text becomes more personalized and engaging, but the processing complexity and time increase

Engineering Contradiction:

Improvetext personalizationVSAvoidsystem processing complexity

Core Design Contradiction:

Ease of manufactureVSDevice complexity

Solution Approach 1:

The system performs preliminary analysis of user profiles during setup or initial interactions, caching error preferences and patterns for future use. This pre-processing reduces real-time complexity by having user preferences ready before text generation occurs

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes processing parameters by adjusting the depth of profile analysis based on context. For routine messages, simplified profile matching is used, while for important communications, more comprehensive analysis is applied, optimizing the balance between personalization and processing load

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS10812417B2Auto-incorrect in chatbot human-machine interfaces

Publication Date: 2020.10.20 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Context appropriate errors are injected into conversational text generated by conversational agents. The conversational agent creates an imperfect conversational text containing at least one text entry error added to the original conversational text. A confidence level that at least one of context and meaning of the imperfect conversational text is consistent with the context and meaning of the original conversational text is determined, and the imperfect conversational text is communicated to a human recipient if the confidence level is above a pre-defined threshold.