Meaning representation analyzing system and meaning representation analyzing method
A meaning representation parsing system 100 includes: an input unit 110 that accepts input data 210 in a text or a graph; and a serialized graph generating unit 120 that generates a token array (serialized graph 220 ) representing a graph structure corresponding to the input data 210 . Furthermore, the token array generated by the serialized graph generating unit 120 includes at least a first token indicating a node in the graph structure corresponding to the input data 210 and a second token indicating an edge representing a relationship between the nodes.
1 . A meaning representation parsing system that parses a meaning representation of input data, the meaning representation parsing system comprising:
a storage storing a program; and
a processor executing the program to perform:
an input processing that accepts the input data in a text or a graph; and
a serialized graph generating processing that generates a token array representing a graph structure corresponding to the input data using a neural network encoder-decoder architecture,
wherein the token array includes at least a first token indicating a node in the graph structure corresponding to the input data and a second token indicating an edge representing a relationship between the nodes, and
wherein the neural network encoder-decoder architecture is trained to parse various meaning representations uniformly without designing specific actions for each meaning representation graph.
2 . The meaning representation parsing system according to claim 1 , wherein the token array generated by the serialized graph generating processing equivalently describes the meaning representation of the input data.
3 . The meaning representation parsing system according to claim 1 , wherein the serialized graph generating processing includes an identity determining processing that, with regard to the first token included in the token array representing the graph structure, determines identity of nodes on a graph and gives information indicating identity to a plurality of corresponding first tokens.
4 . The meaning representation parsing system according to claim 1 , wherein, when the input data is a text, the input processing converts the text in the input data into a predetermined character code enabling generation of the token array in the serialized graph generating processing, and inputs the input data converted, to the serialized graph generating processing.
5 . The meaning representation parsing system according to claim 1 , wherein, when the input data is a graph, the input processing converts the input data into a token array equivalent to the graph and described in a same description format as the token array generated by the serialized graph generating processing, and inputs the input data converted, to the serialized graph generating processing.
6 . The meaning representation parsing system according to claim 1 ,
wherein the processor executing the program to perform:
a serialized graph converting processing that performs conversion processing for constructing information of the edge in the graph structure for a first token array that is generated by the serialized graph generating processing and represents the graph structure corresponding to the input data, to convert the first token array into a second token array,
wherein the serialized graph converting processing executes an action corresponding to a type of the token for each token constituting the first token array in the conversion processing.
7 . The meaning representation parsing system according to claim 6 , wherein the serialized graph converting processing enables conversion of a description format between the first token array and the second token array while maintaining equivalent description of the meaning representation of the input data in the conversion processing.
8 . The meaning representation parsing system according to claim 6 , wherein the first token array and the second token array are described in either a first description format described in accordance with a graph structure corresponding to the input data or a second description format described so as to suppress a number of tokens constituting the token array.
9 . A meaning representation parsing method by a meaning representation parsing system that parses a meaning representation of input data, the meaning representation parsing method comprising:
an input step of accepting the input data in a text or a graph; and
a serialized graph generating step of generating a token array representing a graph structure corresponding to the input data based on the input data accepted in the input step using a neural network encoder-decoder architecture,
wherein the token array includes at least a first token indicating a node in the graph structure corresponding to the input data and a second token indicating an edge representing a relationship between the nodes, and
wherein the neural network encoder-decoder architecture is trained to parse various meaning representations uniformly without designing specific actions for each meaning representation graph.