Chinese semantic parsing method and parser fusing part-of-speech features

Abstract

The invention is suitable for the technical field of Chinese semantic parsing, and provides a Chinese semantic parsing method and parser fusing part-of-speech characteristics, and the method comprises the following steps: analyzing a document structure, and carrying out sentence segmentation on a document body through an NLP technology; performing word segmentation on the sentence segmentation result by adopting a word segmentation tool to obtain a word segmentation result, part of speech and coordinate information of the word segmentation result; inputting the word segmentation result and the part-of-speech into an encoder module for encoding to obtain a semantic feature vector; and sequentially inputting the semantic feature vectors obtained by encoding into a decoder module according to a word segmentation sequence to obtain a local semantic analysis result of the sentence, and converting the local semantic analysis result into semantic feature vectors to serve as input of next decoding until a complete semantic analysis result of the sentence is obtained. According to the method, the word meaning feature vector and the part-of-speech feature vector are fused, and feature association between nodes is enhanced, so that the accuracy of sentence semantic analysis is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of Chinese semantic analysis, and in particular to a Chinese semantic analysis method and a analyzer integrating part-of-speech features. Background Art

[0002] At present, the Chinese pre-training model adopts the character-level training method. Therefore, based on this, most Chinese semantic parsing algorithms use the character-level information of Chinese, a small part uses the word meaning information, but there are few algorithms that use the part-of-speech information, resulting in inaccurate Chinese semantic parsing. Therefore, it is necessary to provide a Chinese semantic parsing method and parser that integrates part-of-speech features to solve the above problems. Summary of the invention

[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a Chinese semantic parsing method and parser that integrates part-of-speech features to solve the problems existing in the above-mentioned background technology.

[0004] The present invention is implemented as follows: a Chinese semantic parsing method integrating part-of-speech features, the method comprising the following steps:

[0005] Analyze the document structure and use NLP technology to segment the document body into sentences;

[0006] The sentence segmentation results are segmented using a word segmentation tool to obtain the word segmentation results, part of speech, and coordinate information of the word segmentation results;

[0007] Input the word segmentation results and parts of speech into the encoder module for encoding to obtain a semantic feature vector;

[0008] The semantic feature vectors obtained by encoding are input into the decoder module in sequence according to the word segmentation order to obtain the local semantic analysis result of the sentence, and the local semantic analysis result is converted into a semantic feature vector as the input for the next decoding until the complete semantic analysis result of the sentence is obtained.

[0009] As a further solution of the present invention: the step of inputting the word segmentation result and the part of speech into the encoder module for encoding to obtain the semantic feature vector specifically includes:

[0010] Use the Chinese Ernie pre-trained model to obtain the semantic feature vector of the sentence character level, and reduce the dimension of the vector through matrix mapping;

[0011] According to the coordinate information of the word segmentation result, take the semantic feature vector w of the last character of each word i Represents the semantic features of the word, and obtains the semantic feature vector W of the sentence = [w1, w2, w3, ..., w i ];

[0012] Based on the part-of-speech vector matrix, the part-of-speech feature vector N of the sentence is obtained;

[0013] Add the semantic feature vector W and the part-of-speech feature vector N according to the corresponding positions to obtain the part-of-speech weighted semantic feature vector W'=W+N;

[0014] Randomly initialize a position vector matrix to represent the word segmentation position information, and map the coordinate information of each word segmentation result into a position feature vector p through the position vector matrix i , get the position feature vector P = [p1, p2, p3, ..., pi] of all the word segments in the sentence;

[0015] The semantic feature vector W' is concatenated with the position feature vector P, dimensionally mapped and nonlinearly activated to obtain the fused semantic feature vector V of the sentence, V = relu(linear(cat([W,P])));

[0016] Perform multi-head attention calculation on the fused semantic feature vector V to obtain the final semantic feature vector S of the sentence.

[0017] As a further solution of the present invention: the step of obtaining the part-of-speech feature vector N of the sentence based on the part-of-speech vector matrix specifically includes:

[0018] Initialize an A*B part-of-speech vector matrix to represent the coordinate information of the part-of-speech of the word segmentation, where A represents the number of parts-of-speech in the word segmentation table and B represents the dimension of the part-of-speech vector;

[0019] The part-of-speech result is converted into a part-of-speech id according to the word segmentation table, and the part-of-speech id is mapped into a part-of-speech feature vector n according to the part-of-speech vector matrix i And perform nonlinear activation to obtain the part-of-speech feature vector N = [n1, n2, n3, ..., n i ].

[0020] As a further solution of the present invention: the step of performing multi-head attention calculation on the fused semantic feature vector V to obtain the final semantic feature vector S of the sentence specifically includes: dividing the fused semantic feature vector V into three vectors q, k and v according to the number of multi-heads, the vector dimension of q, k and v is dk, performing attention calculation through vector k and vector v, and performing softmax on the attention, Multiply the updated attention vector v to obtain the attention-weighted fused semantic feature vector V, V = attention(q, k)v. Input the fused semantic feature vector V into the feedforward neural network, perform two linear mappings and nonlinear activations on the fused semantic feature vector V, and obtain the final semantic feature vector S of the sentence, S = Feedforward(V).

[0021] As a further solution of the present invention: the step of obtaining a local semantic parsing result of a sentence, converting the local semantic parsing result into a semantic feature vector as an input for the next decoding, until a complete semantic parsing result of the sentence is obtained, specifically comprises:

[0022] Initialize a local feature map G0, and update the local feature map G0 to G1;

[0023] According to the edge index and node fusion feature w after adding the self-loop i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature graph G0 to G1;

[0024] The semantic feature vector of the local feature map G1 is extracted through multi-layer graph neural network, linear mapping and non-linear activation; each feature map G is weighted through matrix cutting, linear mapping, non-linear activation and attention. i The semantic features of the node are processed to obtain the node action position probability distribution R, the semantic element label probability distribution E and the semantic template label probability distribution M. The maximum value of the above three probability distribution matrices is taken to obtain the node action position, semantic element label and semantic template label of the current node;

[0025] When calculating the feature vector of the next node, it is necessary to combine the word vector feature and position feature of the current node with the semantic feature of the updated local feature graph G1, so as to calculate the semantic fusion features of the node and edge, and update the local feature graph G1 to G2;

[0026] After several cycles of updates, the complete feature graph G of the sentence and all label parsing results are obtained.

[0027] As a further solution of the present invention: the step of initializing a local feature map G0 and updating the local feature map G0 to G1 specifically includes:

[0028] Initialize a local feature graph G0, add a self-loop to the local feature graph G0, and the semantic feature vector w of the current node i and the position feature vector pi Perform linear mapping calculations and matrix concatenation respectively to obtain the fusion feature w of the current node i _p i , w i _p i =cat(linear w (w i )+linear p (p i ));

[0029] According to the edge index and node fusion feature w after adding the self-loop i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature graph G0 to G1.

[0030] Another object of the present invention is to provide a Chinese semantic parser that integrates part-of-speech features, wherein the parser is used to execute the steps in the Chinese semantic parsing method that integrates part-of-speech features, and the parser consists of an encoder module and a decoder module.

[0031] As a further solution of the present invention: the steps performed by the encoder module include: using a word segmentation tool to segment the sentence results to obtain the word segmentation results, parts of speech and coordinate information of the word segmentation results; inputting the word segmentation results and parts of speech into the encoder module for encoding to obtain a semantic feature vector.

[0032] As a further solution of the present invention: the steps performed by the decoder module include: obtaining a local semantic parsing result of a sentence, converting the local semantic parsing result into a semantic feature vector as input for the next decoding, until a complete semantic parsing result of the sentence is obtained.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention can quickly and accurately perform structural decomposition of text at the semantic level, and restore the explicit semantic information and implicit semantic information of the text to the maximum extent. In addition, the present invention fuses the word meaning feature vector and the part-of-speech feature vector through multiple matrix changes, and enhances the feature association between nodes, effectively improving the accuracy of sentence semantic analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The flowchart of a Chinese semantic parsing method that integrates part-of-speech features.

[0036] Figure 2A schematic diagram of the structure of a Chinese semantic parser that integrates part-of-speech features. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0039] like Figure 1 As shown, an embodiment of the present invention provides a Chinese semantic parsing method integrating part-of-speech features, the method comprising the following steps:

[0040] S100, analyzing the document structure and dividing the document body into sentences using NLP technology;

[0041] S200, using a word segmentation tool to segment the sentence results, and obtaining the word segmentation results, part of speech, and coordinate information of the word segmentation results;

[0042] S300, inputting the word segmentation result and the part of speech into the encoder module for encoding to obtain a semantic feature vector;

[0043] S400, inputting the encoded semantic feature vectors into the decoder module in sequence according to the word segmentation order to obtain a local semantic analysis result of the sentence, and converting the local semantic analysis result into a semantic feature vector as an input for the next decoding, until a complete semantic analysis result of the sentence is obtained.

[0044] In the embodiment of the present invention, when the document body is segmented by NLP technology, a suitable NLP library is selected according to the language of the text to be processed, and commonly used libraries include NLTK and SpaCy, and then the Chinese language model or resource is downloaded and loaded, and then the sentence segmentation function provided by the NLP library is used to segment the preprocessed text. For Chinese text, since there are no obvious separators between Chinese sentences, it is necessary to use a statistical or machine learning-based method to segment the sentences, and the sentence segmentation result will be a list containing multiple sentences, each of which is a string.

[0045] In the embodiment of the present invention, when the word segmentation tool is used to segment the sentence results, the optional word segmentation tools include jieba, HanLP, SnowNLP, etc. Most of these tools provide functions such as word segmentation and part-of-speech tagging, and some tools also support obtaining the coordinate information of the word segmentation results (i.e., the starting and ending positions of each word in the original text). Then use pip and other tools to install the selected word segmentation tool, and configure the parameters of the word segmentation tool as needed, such as custom dictionaries, word segmentation modes, etc. When performing word segmentation and part-of-speech tagging, the sentence segmentation result is used as input, and the selected word segmentation tool is used for word segmentation; the word segmentation tool will cut the text into individual words and return the word segmentation results; while segmenting, most word segmentation tools will also perform part-of-speech tagging on the word segmentation results; part-of-speech tagging will mark a part-of-speech tag for each word, such as noun (n), verb (v), adjective (a), etc. When obtaining the coordinate information of the word segmentation result, the word segmentation tool (such as HanLP, etc.) will return the starting and ending positions (i.e., coordinate information) of each word in the original text while performing word segmentation and part-of-speech tagging.

[0046] In the embodiment of the present invention, the step of inputting the word segmentation result and the part of speech into the encoder module for encoding to obtain the semantic feature vector specifically includes: using the Chinese Ernie pre-training model to obtain the semantic feature vector at the sentence character level, and reducing the dimension of the vector through matrix mapping; according to the coordinate information of the word segmentation result, taking the semantic feature vector w of the last character of each word i Represents the semantic features of the word, and obtains the semantic feature vector W of the sentence = [w1, w2, w3, ..., w i ]; Based on the part-of-speech vector matrix, obtain the part-of-speech feature vector N of the sentence; add the semantic feature vector W and the part-of-speech feature vector N according to the corresponding position to obtain the part-of-speech weighted semantic feature vector W'=W+N; randomly initialize a position vector matrix to represent the word segmentation position information, and map the coordinate information of each word segmentation result to the position feature vector p through the position vector matrix i , get the position feature vector P = [p1, p2, p3, ..., pi] of all word segmentations in the sentence; concatenate the semantic feature vector W' with the position feature vector P, perform dimension mapping and non-linear activation to get the fused semantic feature vector V of the sentence, V = relu(linear(cat([W,P]))); perform multi-head attention calculation on the fused semantic feature vector V to get the final semantic feature vector S of the sentence.

[0047] The step of obtaining the part-of-speech feature vector N of the sentence based on the part-of-speech vector matrix specifically includes: initializing an A*B part-of-speech vector matrix to represent the coordinate information of the part-of-speech, where A represents the number of parts of speech in the word segmentation table and B represents the dimension of the part-of-speech vector; then converting the part-of-speech result into a part-of-speech id according to the word segmentation table, and mapping the part-of-speech id into a part-of-speech feature vector n according to the part-of-speech vector matrix i And perform nonlinear activation to obtain the part-of-speech feature vector N = [n1, n2, n3, ..., n i ] The step of performing multi-head attention calculation on the fused semantic feature vector V to obtain the final semantic feature vector S of the sentence specifically includes: dividing the fused semantic feature vector V into three vectors q, k and v according to the number of multi-heads, the vector dimensions of q, k and v are dk, performing attention calculation through vector k and vector v, and performing softmax on the attention, Multiply the updated attention vector v to obtain the attention-weighted fused semantic feature vector V, V = attention(q, k)v. Input the fused semantic feature vector V into the feedforward neural network, perform two linear mappings and nonlinear activations on the fused semantic feature vector V, and obtain the final semantic feature vector S of the sentence, S = Feedforward(V).

[0048] In the embodiment of the present invention, the step of obtaining the local semantic parsing result of the sentence and converting the local semantic parsing result into a semantic feature vector as the input of the next decoding until the complete semantic parsing result of the sentence is obtained specifically includes: at the beginning, a local feature map G0 needs to be initialized, and the local feature map G0 is updated to G1; according to the edge index after adding the self-loop and the node fusion feature w i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature map G0 to G1; Extract the semantic feature vector of the local feature map G1 through multi-layer graph neural network, linear mapping and non-linear activation; Matrix cutting, linear mapping, non-linear activation, attention weighting are used to perform semantic feature extraction on each feature map G iThe semantic features of the node are processed to obtain the node action position probability distribution R, the semantic element label probability distribution E and the semantic template label probability distribution M. The maximum values ​​of the above three probability distribution matrices are taken to obtain the node action position, semantic element label and semantic template label of the current node. When calculating the feature vector of the next node, it is necessary to combine the word vector feature and position feature of the current node with the semantic feature of the updated local feature graph G1, so as to calculate the semantic fusion features of the nodes and edges, and update the local feature graph G1 to G2. After several cycles of updating, the complete feature graph G of the sentence and all label parsing results are obtained.

[0049] The step of initializing a local feature map G0 and updating the local feature map G0 to G1 specifically includes: initializing a local feature map G0, adding a self-loop to the local feature map G0, updating the semantic feature vector w of the current node i and the position feature vector p i Perform linear mapping calculations and matrix concatenation respectively to obtain the fusion feature w of the current node i _p i , w i _p i =cat(linear w (w i )+linear p (p i )); According to the edge index and node fusion feature w after adding the self-loop i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature graph G0 to G1.

[0050] like Figure 2 As shown, an embodiment of the present invention further provides a Chinese semantic parser that integrates part-of-speech features, and the parser is used to execute the steps in the Chinese semantic parsing method that integrates part-of-speech features, and the parser is composed of an encoder module 100 and a decoder module 200. Among them, the steps executed by the encoder module 100 include: using a word segmentation tool to segment the sentence results, and obtain the word segmentation results, part-of-speech and coordinate information of the word segmentation results; inputting the word segmentation results and part-of-speech into the encoder module for encoding to obtain a semantic feature vector. Among them, the steps executed by the decoder module 200 include: obtaining a local semantic parsing result of the sentence, and converting the local semantic parsing result into a semantic feature vector as the input for the next decoding, until the complete semantic parsing result of the sentence is obtained.

[0051] The above only describes in detail the preferred embodiments of the present invention, which is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0052] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0053] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0054] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A Chinese semantic parsing method integrating part-of-speech features, characterized in that: The method comprises the following steps: Analyze the document structure and use NLP technology to segment the document body into sentences; The sentence segmentation results are segmented using a word segmentation tool to obtain the word segmentation results, part of speech, and coordinate information of the word segmentation results; Input the word segmentation results and parts of speech into the encoder module for encoding to obtain a semantic feature vector; The semantic feature vectors obtained by encoding are input into the decoder module in sequence according to the word segmentation order to obtain the local semantic analysis result of the sentence, and the local semantic analysis result is converted into a semantic feature vector as the input for the next decoding until the complete semantic analysis result of the sentence is obtained.

2. The Chinese semantic parsing method integrating part-of-speech features according to claim 1 is characterized in that: The step of inputting the word segmentation result and the part of speech into the encoder module for encoding to obtain the semantic feature vector specifically includes: Use the Chinese Ernie pre-trained model to obtain the semantic feature vector of the sentence character level, and reduce the dimension of the vector through matrix mapping; According to the coordinate information of the word segmentation result, take the semantic feature vector w of the last character of each word i Represents the semantic features of the word, and obtains the semantic feature vector W of the sentence = [w1, w2, w3, ..., w i ]; Based on the part-of-speech vector matrix, the part-of-speech feature vector N of the sentence is obtained; Add the semantic feature vector W and the part-of-speech feature vector N according to the corresponding positions to obtain the part-of-speech weighted semantic feature vector W'=W+N; Randomly initialize a position vector matrix to represent the word segmentation position information, and map the coordinate information of each word segmentation result into a position feature vector p through the position vector matrix i , get the position feature vector P = [p1, p2, p3, ..., pi] of all the word segments in the sentence; The semantic feature vector W' is concatenated with the position feature vector P, dimensionally mapped and nonlinearly activated to obtain the fused semantic feature vector V of the sentence, V = relu(linear(cat([W,P]))); Perform multi-head attention calculation on the fused semantic feature vector V to obtain the final semantic feature vector S of the sentence.

3. The Chinese semantic parsing method integrating part-of-speech features according to claim 2 is characterized in that: The step of obtaining the part-of-speech feature vector N of the sentence based on the part-of-speech vector matrix specifically includes: Initialize an A*B part-of-speech vector matrix to represent the coordinate information of the part-of-speech of the word segmentation, where A represents the number of parts-of-speech in the word segmentation table and B represents the dimension of the part-of-speech vector; The part-of-speech result is converted into a part-of-speech id according to the word segmentation table, and the part-of-speech id is mapped into a part-of-speech feature vector n according to the part-of-speech vector matrix i And perform nonlinear activation to obtain the part-of-speech feature vector N = [n1, n2, n3, ..., n i ].

4. The Chinese semantic parsing method integrating part-of-speech features according to claim 2 is characterized in that: The step of performing multi-head attention calculation on the fused semantic feature vector V to obtain the final semantic feature vector S of the sentence specifically includes: dividing the fused semantic feature vector V into three vectors q, k and v according to the number of multi-heads, the vector dimensions of q, k and v are dk, performing attention calculation on the vector k and the vector v, and performing softmax on the attention, Multiply the updated attention vector v to obtain the attention-weighted fused semantic feature vector V, V = attention(q, k)v. Input the fused semantic feature vector V into the feedforward neural network, perform two linear mappings and nonlinear activations on the fused semantic feature vector V, and obtain the final semantic feature vector S of the sentence, S = Feedforward(V).

5. The Chinese semantic parsing method integrating part-of-speech features according to claim 2 is characterized in that: The step of obtaining a local semantic analysis result of the sentence, and converting the local semantic analysis result into a semantic feature vector as an input for the next decoding until a complete semantic analysis result of the sentence is obtained, specifically includes: Initialize a local feature map G0, and update the local feature map G0 to G1; According to the edge index and node fusion feature w after adding the self-loop i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature graph G0 to G1; The semantic feature vector of the local feature map G1 is extracted through multi-layer graph neural network, linear mapping and non-linear activation; each feature map G is weighted through matrix cutting, linear mapping, non-linear activation and attention. i The semantic features of the node are processed to obtain the node action position probability distribution R, the semantic element label probability distribution E and the semantic template label probability distribution M. The maximum value of the above three probability distribution matrices is taken to obtain the node action position, semantic element label and semantic template label of the current node; When calculating the feature vector of the next node, it is necessary to combine the word vector feature and position feature of the current node with the semantic feature of the updated local feature graph G1, so as to calculate the semantic fusion features of the node and edge, and update the local feature graph G1 to G2; After several cycles of updates, the complete feature graph G of the sentence and all label parsing results are obtained.

6. The Chinese semantic parsing method integrating part-of-speech features according to claim 5 is characterized in that: The step of initializing a local feature map G0 and updating the local feature map G0 to G1 specifically includes: Initialize a local feature graph G0, add a self-loop to the local feature graph G0, and the semantic feature vector w of the current node i and the position feature vector p i Perform linear mapping calculations and matrix concatenation respectively to obtain the fusion feature w of the current node i _p i , w i _p i =cat(linear w (w i )+linear p (p i )); According to the edge index and node fusion feature w after adding the self-loop i _p i Calculate the degree d of the current node in the feature graph G0 i ; Use mean square normalization to aggregate node and edge information to update node features; Calculate the semantic feature vectors of nodes and edges to update the local feature graph G0 to G1.

7. A Chinese semantic parser integrating part-of-speech features, characterized in that: The parser is used to execute the steps in the Chinese semantic parsing method integrating part-of-speech features as described in claim 1, and the parser consists of an encoder module and a decoder module.

8. The Chinese semantic parser integrating part-of-speech features according to claim 7, characterized in that: The steps performed by the encoder module include: performing word segmentation on the sentence segmentation result using a word segmentation tool to obtain the word segmentation result, the part of speech and the coordinate information of the word segmentation result; inputting the word segmentation result and the part of speech into the encoder module for encoding to obtain a semantic feature vector.

9. The Chinese semantic parser integrating part-of-speech features according to claim 8, characterized in that: The steps performed by the decoder module include: obtaining a local semantic parsing result of a sentence, converting the local semantic parsing result into a semantic feature vector as an input for the next decoding, until a complete semantic parsing result of the sentence is obtained.

Images