Dialogue Data Generation Using Interrogative Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Non-task-oriented dialogue systems face difficulties in generating question sentences to deeply delve conversations due to the lack of a structured approach and insufficient data for learning, leading to high costs and inefficient user interaction.

Innovation Solution

A dialogue data generation device and method that receives input data sets of user utterances and generates dialogue data by identifying question sentences using interrogatives, employing a neural network to learn and generate responses, thereby creating continuous and in-depth conversations at a lower cost.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If a rule base method is used to generate question sentences in a non-task-oriented dialogue system, then the system can achieve dialogue through predefined rules, but it requires a large amount of rules to be described by hand to extensively delve into the dialogue

Engineering Contradiction:

Improveease of dialogue generationVSAvoidcomplexity of rule base

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical rule-based system with a neural network-based machine learning system. Instead of manually describing rules to generate question sentences, the system learns patterns from training data and automatically generates questions, eliminating the need for extensive hand-crafted rule bases while maintaining dialogue capability

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service by allowing the neural network to automatically learn and generate question sentences from training data without requiring manual rule creation. The model autonomously improves its question generation capability through learning from examples, reducing human intervention in rule formulation

Inventive Principle:

Principle #25Self-service

2Extent of automation

If a machine learning method is used to generate question sentences, then automated learning can be achieved, but sufficient training data does not exist and it is difficult to prepare a corpus for machine learning

Engineering Contradiction:

Improveautomation of dialogue generationVSAvoidquantity of training data

Core Design Contradiction:

Extent of automationVSQuantity of substance

Solution Approach 1:

The patent performs preliminary action by using a question sentence generation model to create synthetic training data before actual machine learning training. The generation model produces artificial dialogue examples with question sentences, which then serve as training corpus for teaching the model to generate meaningful questions, solving the bootstrapping problem of insufficient initial data

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary question sentence generation model that bridges the gap between available data and the desired question generation capability. This intermediary model generates synthetic training examples that enable subsequent learning, acting as a mediator to overcome the lack of sufficient real-world training data

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If conventional non-task-oriented dialogue systems are used, then chat dialogue can be handled, but the system cannot deeply delve the utterance of the other party and cannot smoothen the interaction

Engineering Contradiction:

Improveversatility in handling chat topicsVSAvoidquality of dialogue interaction

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent changes the parameter of question generation by introducing interrogative detection and generation capabilities. The system detects whether training data contains question sentences and uses this information to adjust its generation strategy, enabling it to produce more engaging and probing questions that deepen dialogue interaction while maintaining versatility across different chat topics

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12026460B2Dialogue data generation device, dialogue data generation method, and program

Publication Date: 2024.07.02 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

To make it possible to generate dialogue data for generating a question sentence to deeply delve conversation at a low cost. For each of a plurality of pieces of data each including a set of a first utterance sentence that is a sentence uttered by a first user, a second utterance sentence that is a sentence uttered by a second user and is a response to the first utterance sentence, and a third utterance sentence that is a sentence uttered by the first user and is a response to the second utterance sentence, a dialogue data generation unit 110 generates a set of the first utterance sentence of the data and the second utterance sentence of the data as dialogue data when the second utterance sentence of the data is a question sentence using an interrogative.