Dialogue Answer Generation Using Extractive and Generative Modes

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing interactive machine reading comprehension systems can only operate in either an extractive or generative mode, limiting their applicability.

Innovation Solution

A dialogue processing apparatus that utilizes pre-trained model parameters to generate answers in both extractive and generative modes, incorporating a question history and answer history to enhance interactive machine reading comprehension.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If interactive machine reading comprehension is implemented in extractive mode only, then the system can generate answers based on dialogue history, but the system cannot generate natural sentences as answers

Engineering Contradiction:

Improvemode applicabilityVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements a unified interactive machine reading comprehension model that can operate in both extractive mode and generative mode. The model uses a single architecture with shared components (encoder, attention mechanism) that can switch between extraction-based answer generation and natural language generation, making the system universally applicable to both modes without requiring separate dedicated systems for each mode

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments the answer generation process into two distinct pathways: an extractive mode pathway that retrieves spans from the document, and a generative mode pathway that constructs natural sentences. This segmentation allows the system to handle both modes through structured decomposition while maintaining a unified overall framework

Inventive Principle:

Principle #1Segmentation

2Ease of operation

If questions are treated independently, then the processing is simpler, but the system cannot utilize dialogue context for better answer generation

Engineering Contradiction:

Improveprocessing simplicityVSAvoiddialogue context information

Core Design Contradiction:

Ease of operationVSLoss of information

Solution Approach 1:

The patent implements feedback mechanisms where past questions and answers are fed back into the current processing cycle through attention mechanisms. The model attends to relevant historical dialogue turns when generating answers to current questions, allowing context information to flow back into the decision-making process without overwhelming computational complexity

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary encoding of dialogue history and question sequences before answer generation. By pre-processing and encoding the contextual information from past interactions, the system prepares the context in advance, making it readily available during answer generation without adding significant complexity to the core processing pipeline

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12566927B2Dialogue processing apparatus, learning apparatus, dialogue processing method, learning method and program

Publication Date: 2026.03.03 NT T INC

AI summary

A generation unit that takes a question Qi that is a word sequence representing a current question in a dialogue, a document P used to generate an answer Ai to the question Qi, a question history {Qi-1, . . . , Qi-k} that is a set of word sequences representing k past questions, and an answer history {Ai-1, . . . , Ai-k} that is a set of word sequences representing answers to the k questions as inputs, and generates the answer Ai by machine reading comprehension in an extractive mode or a generative mode using pre-trained model parameters is provided.