Paraphrase Generation for QA Training Data Enrichment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The quality of responses from Question and Answer (QA) systems is limited by the quality of the ground truth used for training, as current methods rely on sparse ground truth that lacks comprehensive training data, leading to suboptimal performance in providing accurate answers.

Innovation Solution

A paraphrase generation system is employed to convert sparse ground truth into enriched ground truth by generating additional paraphrases, which are then used to train the QA system, enhancing the training data with grammatical alternatives and semantic variations, thereby improving the QA system's ability to provide accurate answers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If sparse ground truth is used for training the QA system, then the training process is simpler and faster, but the quality and accuracy of the answers provided by the QA system deteriorates

Engineering Contradiction:

Improvetraining speedVSAvoidanswer accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary action by automatically generating paraphrases of ground truth questions before the training process. This preprocessing step creates enriched training data that includes multiple syntactic variations of the same question, allowing the QA system to learn more robust answer-generation patterns without requiring manual creation of extensive training datasets.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system uses copying by generating paraphrased versions of existing ground truth questions. Instead of creating entirely new training data, it creates linguistic copies/variations of the original questions through automated paraphrase generation, preserving the semantic meaning while varying the syntactic structure to improve training effectiveness.

Inventive Principle:

Principle #26Copying

2Measurement precision

If more comprehensive ground truth is provided to enrich training data, then the quality of system training and answer accuracy improves, but the complexity and resource requirements of the training process increases

Engineering Contradiction:

Improveanswer accuracyVSAvoidtraining system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system applies self-service by using the existing ground truth questions themselves to generate the enriched training data. The paraphrase generation system automatically creates variations without requiring external tools, manual intervention, or additional complex infrastructure. The ground truth data serves its own enrichment purpose, eliminating the need for separate data collection and annotation processes.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system changes parameters by transforming the linguistic parameters of the ground truth questions through automated paraphrase generation. It varies syntactic structures, word choices, and sentence patterns while maintaining semantic equivalence, thereby enriching the training data with diverse linguistic variations without requiring fundamentally different data collection methods.

Inventive Principle:

Principle #35Parameter changes

3Quantity of substance

If automated paraphrase generation is implemented to enrich ground truth, then the comprehensiveness of training data improves, but the processing time and computational resources required increase

Engineering Contradiction:

Improvetraining data volumeVSAvoidprocessing time

Core Design Contradiction:

Quantity of substanceVSLoss of time

Solution Approach 1:

The system replaces mechanical manual processes with automated computational processes. Instead of manually creating diverse training questions, which would be extremely time-consuming, the system uses automated paraphrase generation algorithms to efficiently produce large volumes of training data variations, substituting human labor with algorithmic processing that can operate at much higher speeds.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the state of the training data from sparse to comprehensive by applying automated paraphrase generation. This transforms a limited set of ground truth questions into an expanded dataset with multiple syntactic variations, increasing the quantity and diversity of training data without proportional increases in manual processing time.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS9984063B2System and method for automatic, unsupervised paraphrase generation using a novel framework that learns syntactic construct while retaining semantic meaning

Publication Date: 2018.05.29 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A system includes a question answering system executed by a computer, a processor, and a memory coupled to the processor. The memory is encoded with instructions that when executed cause the processor to provide training for training the question answering system. The training system is configured to receive a first phrase and a second phrase, the first and second phrases being paraphrases of each other, convert the first phrase into a first logical form and the second phrase into a second logical form, generate a phrasal edit that includes a difference between the first logical form and the second logical form, convert the phrasal edit into a disjunctive logical form in two directions, and generate a first plurality of paraphrases of the first and second phrases based on the disjunctive logical form.