Temporal Expression Normalization Using Transformer Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing normalization methods for temporal expressions in unstructured texts are inflexible and require extensive rule creation for new languages and text genres, struggling with minor spelling errors and unknown words, limiting their adaptability and accuracy.
Innovation Solution
A deep learning-based method using a transformer model to normalize temporal expressions, allowing flexible adaptation to different languages and text genres, and handling disruptions like spelling errors through context-based learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If rule-based normalization methods are used, then the system provides deterministic processing, but the system lacks flexibility when encountering new languages, text genres, spelling errors, or unknown words
Solution Approach 1:
The patent replaces the mechanical rule-based system with a deep learning model (transformer architecture) that automatically learns temporal expression patterns from training data. This substitution eliminates the need for manual rule creation and enables automatic adaptation to new languages and text genres through contextual understanding, directly resolving the contradiction between adaptability and complexity.
Solution Approach 2:
The patent changes the fundamental parameter of the normalization system from static rule sets to dynamic neural network parameters (weights and biases) that are learned from data. This transformation allows the system to adapt to different languages and text genres by learning new patterns during training, while avoiding the complexity of manually creating and maintaining extensive rule sets for each scenario.
2Measurement precision
If deep learning-based normalization is used, then the system achieves higher flexibility and accuracy, but the training data requirements and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-training the transformer model on large corpora of temporal expressions across multiple languages and text genres before deployment. This pre-training phase allows the model to learn robust patterns and representations that can be fine-tuned with smaller, domain-specific datasets, thereby reducing the training data volume required while maintaining high normalization accuracy.
3Reliability
If static rule systems are used, then the processing speed is fast, but the system cannot handle disruptions like spelling errors and unknown words
Solution Approach 1:
The patent replaces the fragile static rule system with a robust deep learning model that can handle text disruptions through contextual understanding. The transformer architecture's self-attention mechanism allows the model to infer the intended meaning of temporal expressions even when spelling errors or unknown words are present, significantly improving reliability without requiring complex error-handling rules.
Data Source
AI summary
A method and device for processing temporal expressions from unstructured texts for filling a knowledge database. A temporal expression in a text is determined. A type of the temporal expression is determined as a function of the text. The temporal expression and the type are mapped on a prediction of a value of the temporal expression in a context-free representation of the temporal expression.