Multi-lingual Named Entity Recognition via Synthetic Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing named entity recognition systems face challenges in accuracy, language limitations, and the cost and time-consuming process of manually creating training data, which restricts their ability to recognize a larger number of classes and operate across multiple languages.
Innovation Solution
A multi-lingual named entity recognition system is developed using labeled training sentences from monolingual and multi-lingual corpora and parallel data, with a combination of hyperlinks-based and alignment-based approaches to automatically generate and select labeled sentences, enabling the system to train machine learning components for multiple languages and classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual creation of training data is used, then quality of training data is improved, but cost and time consumption increase
Solution Approach 1:
The patent uses machine translation systems to create synthetic training data by translating source language documents into target language. This copying approach generates large volumes of parallel training data automatically, replacing the time-consuming manual creation process while maintaining quality through the use of established translation models and human-annotated source data.
Solution Approach 2:
The system leverages existing machine translation infrastructure and open-source resources to self-generate training data. By using parallel data from machine translation systems and public domain documents, the patent enables the training process to be self-sustaining without requiring proportional manual annotation efforts, thus reducing time and cost while maintaining data quality.
2Adaptability or versatility
If existing named entity recognition systems are used, then operation in English is achieved, but operation in other languages is limited
Solution Approach 1:
The patent creates a universal named entity recognition system that can process multiple languages simultaneously. By training the system on parallel data from machine translation systems and multi-lingual corpora, the model achieves cross-lingual capability while maintaining accuracy through shared representation learning and transfer from source language training data.
Solution Approach 2:
The system adapts to different languages by modifying the training data parameters and characteristics. By adjusting the input data to include multi-lingual parallel corpora and using language-specific processing parameters within a unified architecture, the patent enables the system to maintain high accuracy across multiple languages without requiring separate specialized models for each language.
3Adaptability or versatility
If limited number of entity classes are used, then system complexity is reduced, but scalability to recognize larger numbers of classes is restricted
Solution Approach 1:
The patent segments the training process into modular components that can handle different entity classes independently. By organizing training data and model parameters in a segmented, scalable architecture, the system can accommodate a large number of entity classes without proportionally increasing overall system complexity. Each class can be trained and evaluated separately while sharing common underlying representations.
Solution Approach 2:
The system manages complexity by adding dimensions to the training data structure rather than increasing model parameter dimensions. By organizing entity classes into hierarchical categories and using multi-lingual parallel data structures, the patent enables scalable recognition of numerous classes through data organization rather than model complexity, allowing the system to handle thousands of classes with manageable computational requirements.
Data Source
AI summary
Named entity recognition is described, for example, to detect an instance of a named entity in a web page and classify the named entity as being an organization or other predefined class. In various examples, named entity recognition results are used to augment text from which the named entity was recognized; the augmentation may comprise information retrieval results about the named entity mention. In various embodiments, labeled training sentences in many different languages and for many different classes, are obtained to train machine learning components of a multi-lingual, multi-class, named entity recognition system. In examples, labeled training sentences are obtained from at least two sources, a first source using a multi-lingual or monolingual corpus of inter-linked documents and a second source using machine translation training data. In examples, labeled training sentences from the two sources are selectively sampled for training the named entity recognition system.