Personalized interactive semantic parsing using a graph-to-sequence model
A semantic parsing method using a graph-to-sequence model, system, and computer program product include generating a syntactic graph for a sentence, generating node embeddings for each node based on other nodes the each node is connected to in the syntactic graph, generating a graph embedding over the node embeddings, performing attention-based recurrent neural network (RNN) decoding of the graph embedding and the node embeddings, and providing a logical translation of the sentence based on the decoding.
1. A computer-implemented semantic parsing method using a graph-to-sequence model, the method comprising:
generating a syntactic graph for a sentence by generating a node for each word and connecting the node of each word in the chain form, the chain form connection enabling linking of the nodes in two directions;
generating node embeddings for each node based on other nodes the each node is connected to in the syntactic graph;
generating a graph embedding over the node embeddings;
performing an attention-based recurrent neural network (RNN) decoding of the graph embedding and the node embeddings; and
providing a logical translation of the sentence based on the decoding,
wherein the syntactic graph comprises a dependency graph and a constituent parse, and
wherein the syntactic graph includes a model to learn a mapping between text and logic form in natural language processing.
2. The computer-implemented method of claim 1 , wherein the generating the graph embedding generates the graph embedding by performing max-pooling over the node embeddings,
wherein the model is jointly trained to maximize a conditional log-probability of a correct description given a source graph, and
wherein the syntactic graph does not rely on any rules and instead learns a mapping of the graph from natural language and logical forms.
3. The computer-implemented method of claim 1 , wherein the generating the graph embedding generates the graph embedding by creating an additional node and performing node embedding for the additional node as though the additional node is connected to all other nodes of the syntactic graph.
4. The computer-implemented method of claim 1 , further comprising querying a user to determine an error for the logical translation of the sentence.
5. The computer-implemented method of claim 4 , wherein, if the user returns the error for the logical translation of the sentence, repeating the generating the node embeddings, the generating the graph embedding, and the performing the attention-based recurrent neural network (RNN) decoding to produce a second logical translation.
6. The computer-implemented method of claim 1 , wherein the performing the attention-based recurrent neural network (RNN) decoding calculates a context vector to capture an attention of the generating the node embeddings and the generating the graph embedding.
7. The computer-implemented method of claim 1 , wherein the syntactic graph represents sentence level features, dependency features, and constituency features.
8. The computer-implemented method of claim 1 , embodied in a cloud-computing environment.
9. A computer program product for semantic parsing using a graph-to-sequence model, the computer program product comprising a non-transitory computer-readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform:
generating a syntactic graph for a sentence by generating a node for each word and connecting the node of each word in the chain form, the chain form connection enabling linking of the nodes in two directions;
generating node embeddings for each node based on other nodes the each node is connected to in the syntactic graph;
generating a graph embedding over the node embeddings;
performing an attention-based recurrent neural network (RNN) decoding of the graph embedding and the node embeddings; and
providing a logical translation of the sentence based on the decoding,
wherein the syntactic graph comprises a dependency graph and a constituent parse, and
wherein the syntactic graph includes a model to learn a mapping between text and logic form in natural language processing.
10. The computer program product of claim 9 , wherein the generating the graph embedding generates the graph embedding by performing max-pooling over the node embeddings.
11. The computer program product of claim 9 , wherein the generating the graph embedding generates the graph embedding by creating an additional node and performing node embedding for the additional node as though the additional node is connected to all other nodes of the syntactic graph.
12. The computer program product of claim 9 , further comprising querying a user to determine an error for the logical translation of the sentence.
13. The computer program product of claim 12 , wherein, if the user returns the error for the logical translation of the sentence, repeating the generating the node embeddings, the generating the graph embedding, and the performing the attention-based recurrent neural network (RNN) decoding to produce a second logical translation.
14. The computer program product of claim 9 , wherein the performing the attention-based recurrent neural network (RNN) decoding calculates a context vector to capture an attention of the generating the node embeddings and the generating the graph embedding.
15. The computer program product of claim 9 , wherein the syntactic graph represents sentence level features, dependency features, and constituency features.
16. A semantic parsing system using a graph-to-sequence model, the system comprising:
a processor; and
a memory, the memory storing instructions to cause the processor to perform:
generating a syntactic graph for a sentence by generating a node for each word and connecting the node of each word in the chain form, the chain form connection enabling linking of the nodes in two directions;
generating node embeddings for each node based on other nodes the each node is connected to in the syntactic graph;
generating a graph embedding over the node embeddings;
performing attention-based recurrent neural network (RNN) decoding of the graph embedding and the node embeddings; and
providing a logical translation of the sentence based on the decoding,
wherein the syntactic graph comprises a dependency graph and a constituent parse, and
wherein the syntactic graph includes a model to learn a mapping between text and logic form in natural language processing.
17. The system of claim 16 , wherein the generating the graph embedding generates the graph embedding by performing max-pooling over the node embeddings.
18. The system of claim 16 , wherein the generating the graph embedding generates the graph embedding by creating an additional node and performing node embedding for the additional node as though the additional node is connected to all other nodes of the syntactic graph.
19. The system of claim 16 , wherein the syntactic graph represents sentence level features, dependency features, and constituency features.
20. The system of claim 14 , embodied in a cloud-computing environment.