Parse Thicket Text Generalization for Training Data Expansion

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current classification models in linguistics are limited by the quality of their training data, which affects their performance and ability to generalize beyond the original data set.

Innovation Solution

The use of parse thickets to represent syntactic and discourse information in text, allowing for the identification of common entities and rhetorical relations, and the creation of generalized text fragments that enhance the quality and coverage of training data sets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If statistical and deep learning language systems are used to improve classification performance, then recognition accuracy is improved, but the quality and coverage of training data sets deteriorate due to inability to cover corner cases

Engineering Contradiction:

Improverecognition accuracyVSAvoidcoverage of training data

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The text is segmented into parse thickets that represent syntactic and discourse information separately. By dividing the training data processing into syntactic parsing and discourse parsing components, the system can systematically cover corner cases that holistic statistical methods miss, thereby improving both accuracy and coverage simultaneously

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Parse thickets serve as an intermediary representation between raw text and classification models. This intermediate structured representation captures syntactic and discourse features explicitly, enabling the model to generalize better to corner cases while maintaining high recognition accuracy on standard cases

Inventive Principle:

Principle #24Intermediary (Mediator)

2Adaptability or versatility

If more training data is added to improve model generalization, then adaptability is improved, but the complexity of data processing and validation increases

Engineering Contradiction:

Improvemodel generalizationVSAvoiddata processing complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary syntactic and discourse parsing on training data before feeding it to classification models. By pre-structuring the data into parse thickets with explicit syntactic and discourse relationships, the system reduces the complexity of subsequent model training and validation while improving generalization to unseen cases

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention changes the parameter representation of text from raw strings to structured parse thickets with syntactic and discourse features. This parameter transformation enables more efficient processing of expanded training datasets while systematically capturing corner case patterns that improve model adaptability

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11455494B2Automated building of expanded datasets for training of autonomous agents

Publication Date: 2022.09.27 ORACLE INT CORP

AI summary

Improved systems and methods for generating training data for classification models are disclosed. In an example, a training application accesses two fragments of text. The application represents each fragment of text as a parse thicket. The parse thickets jointly represent syntactic and discourse information. From the parse thickets, the application generalizes the text by identifying common entities or common rhetorical relations between parse thickets. The generalized text is added to a training data set, thereby increasing the coverage of the training set.