Linguistic Model Training via Grammar Graph Frequency Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition technologies require large amounts of corpus data for linguistic model training, leading to high storage and computing resource consumption and prolonged training times.
Innovation Solution
A method for training linguistic models involves obtaining grammars and slot values from sample texts using semantic analysis, generating grammar graphs, calculating weights for grammars, slots, and slot values, and determining grammar frequencies to train the model efficiently without extensive corpus expansion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large amount of corpus data is used for linguistic model training, then model performance is improved, but storage resources and computing resources are significantly consumed
Solution Approach 1:
The patent extracts only the essential linguistic patterns and grammar structures from corpus data through semantic analysis, rather than using the entire corpus. By extracting grammars, slots, and slot values from sample texts and representing them in a compact grammar graph format, the method achieves model training with minimal data while maintaining performance.
Solution Approach 2:
Instead of training the model directly on large amounts of raw corpus data, the patent inverts the approach by first analyzing a small set of sample texts to extract grammatical structures, then using these extracted structures to train the linguistic model. This inversion transforms the training process from data-intensive to structure-intensive.
2Reliability
If a large amount of corpus data is used for linguistic model training, then model performance is improved, but training time is prolonged
Solution Approach 1:
The patent performs preliminary semantic analysis on sample texts to extract grammars, slots, and slot values before model training. By pre-processing a small set of texts into structured grammar graphs with calculated weights and frequencies, the method eliminates the need for time-consuming training on large corpora, significantly reducing training time while preserving performance.
3Reliability
If traditional corpus-based training is used, then model performance can be improved, but computing resources are significantly consumed
Solution Approach 1:
The patent changes the fundamental parameters of the training approach by transitioning from raw corpus data to structured grammar representations with calculated weights and frequencies. By representing linguistic knowledge in terms of grammars, slots, slot values, and their associated probabilities, the method achieves efficient training with minimal computing resources while maintaining model performance.
Data Source
AI summary
The present disclosure provides a method for training a linguistic model, related to fields of speech, natural language processing, deep learning technologies. A method includes: obtaining grammars corresponding to a plurality of sample texts and a slot value of a slot in each grammar by using semantic analysis; generating a grammar graph corresponding to each grammar based on the corresponding grammar and the slot value of the slot in the corresponding grammar; obtaining a weight of each grammar, a weight of each slot, and a weight of each slot value in each grammar graph based on the sample texts; determining at least one grammar frequency of each order based on the weight of each grammar, the weight of each slot, and the weight of each slot value in each grammar graph; and training the linguistic model based on the at least one grammar frequency of each order.