Neural machine translation method based on source language syntax enhanced decoding
A machine translation, source language technology, applied in natural language translation, neural learning methods, neural architecture, etc., can solve the problem of not studying the influence of source language syntactic information, avoid sparse word representation, improve translation quality, and improve performance. Effect
Pending Publication Date: 2022-01-07
KUNMING UNIV OF SCI & TECH
View PDF0 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
However, although these methods improve the performance on the basis of the baseline model, they only utilize the source language syntactic information in the encoder or the target language syntactic information in the decoder, without studying the effect of the source language syntactic information on the decoding process. influences
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0053] Example 1: Such as Figure 1 - Figure 2 As shown, the method based on the source language syntax enhances the decoded neuromechanical translation method, the specific steps of the method are as follows:
[0054] Step1, using the syntax analysis tool to syntax resolution in the source language sentence in parallel corners, get the syntactic dependencies, and then the resulting syntactic rely on the relationship vector, generates the syntactic sense of sensory sensation between the words and words including the source language sentence .
[0055] Step1.1, in General NEWS Commentary V11 (NC11) Yingde, Deying and IWSLT14 Deye, as well as standard low resource WMT18, IWSLT15 Yedang translation tasks experiment. The TST2012 is used as the verification set in the IWSLT15 UK mission, and TST2013 is used as a test set, and other tasks use standard verification test sets.
[0056] Step1.2, the source language sentence is analyzed by the syntactor analysis tool, resulting in the syntac...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention relates to a neural machine translation method for source language syntax enhanced decoding, and belongs to the field of natural language processing. The method comprises the following steps: analyzing a source language sentence to obtain a syntactic relationship; obtaining features of a source language sentence and syntactic related parts thereof by using a syntactic perception self-attention mechanism; extracting information in source language sentence representation and syntactic related representation through a cross attention network to jointly guide generation of a target language; and finally, predicting the vocabulary of the current sequence position by using linear transformation and a softmax function. According to the method, the syntactic information of the monolingual corpus can be explicitly utilized while the bilingual parallel corpus resources marked manually are effectively utilized; knowledge of the single statement method is an important basis for understanding semantics and constructing languages, and the problem that a neural network machine translation model cannot fully mine effective information in bilingual parallel corpora is solved.
Description
Technical field [0001] The present invention relates to a source language syntax enhanced decoded neuromechanical translation method, which belongs to the field of natural language processing. Background technique [0002] With the development of deep learning technology, the neuromechanical translation system based on deep learning methods has achieved remarkable results, making it a new fancy style of machine translation tasks. One of the best performance optimal machine translation models Transformer-based end-to-end architecture only depends only on parallel clauses, and the default model can automatically learn the knowledge in the character. This modeling method lacks explicit guidance, which cannot effectively excavate deep language knowledge, especially in low resource environments of corpus size and quality, thereby causing a decline in translation. [0003] A priori language knowledge, especially syntax, is a pre-defined language rule. Whether it is understanding semant...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06F40/58G06F40/211G06N3/04G06N3/08
CPCG06F40/58G06F40/211G06N3/08G06N3/047G06N3/044G06N3/045
Inventor 余正涛龚龙超相艳
Owner KUNMING UNIV OF SCI & TECH