Natural language generating apparatus, vehicle having the same and natural language generating method

1. A natural language generating apparatus comprising:

a receiver configured to receive semantic expression information including a speech act or intention of a natural language sentence to generate the natural language sentence; and

a controller configured to:

extract speech acts or intentions from a speech signal of a user,

generate a dependency tree representing a syntax structure of at least one word determined based on the semantic expression information received for a response corresponding to the speech acts or intentions, based on a neural network model,

generate the natural language sentence based on the generated dependency tree,

visit a first node among nodes of the dependency tree based on a predetermined traversal algorithm,

determine a first node value of the first node and a parent node value of a parent node of the first node as input values of the neural network model in response to visiting of the first node,

based on the determined input values, train the neural network model to output a second node value of a second node to be visited following the first node, and

determine an output value of the trained neural network model using the first node value of the first node and the parent node value of the parent node of the first node, as the second node value of the second node,

wherein, when both of the first node value of the first node and the parent node value of the parent node of the first node are input as first and second input values, respectively, the controller determines the output value as the second node value of the second node to be visited following the first node.

2. The natural language generating apparatus of claim 1 , wherein the controller is configured to visit each of the nodes based on a breadth-first search (BFS) algorithm.

3. The natural language generating apparatus of claim 1 , wherein the first node and the second node are sibling nodes.

4. The natural language generating apparatus of claim 1 , wherein the controller is configured to generate the dependency tree on which a root word determined based on the semantic expression information is a node value of a root node, based on the trained neural network model.

5. The natural language generating apparatus of claim 4 , wherein the controller is configured to initiate generating the dependency tree which includes the node value of the root node having the root word by inputting the root word to the trained neural network model.

6. The natural language generating apparatus of claim 4 , wherein the controller is configured to determine a child node of the root node visited based on a breadth-first search (BFS) algorithm and a node value corresponding to the child node, based on the trained neural network model, and to generate the dependency tree based on the determined child node and the node value corresponding to the child node.

7. The natural language generating apparatus of claim 6 , wherein the controller is configured to determine an output value of the trained neural network model using a node value of a third node and a node value of a parent node of the third node, as a node value of a sibling node of the third node.

8. The natural language generating apparatus of claim 1 , wherein the controller is configured to generate the natural language sentence based on node values of the nodes.

9. The natural language generating apparatus of claim 8 , wherein the controller is configured to visit the nodes based on a depth-first search (DFS) algorithm and to generate the natural language sentence by arranging words corresponding to the nodes based on an order of visit.

10. The natural language generating apparatus of claim 8 , further comprising an output device, wherein the controller is configured to control the output device to output the generated natural language sentence as a voice signal.

11. A method of generating natural language, comprising:

receiving semantic expression information including a speech act or intention of a natural language sentence to generate the natural language sentence;

extracting speech acts or intentions from a speech signal of a user;

generating a dependency tree representing a syntax structure of at least one word determined based on the semantic expression information received for a response corresponding to the speech acts or intentions, based on a neural network model;

generating the natural language sentence based on the generated dependency tree;

visiting a first node among nodes of the dependency tree based on a predetermined traversal algorithm;

determining a first node value of the first node and a parent node value of a parent node of the first node as input values of the neural network model in response to visiting of the first node;

based on the determined input values, training the neural network model to output a second node value of a second node to be visited following the first node, and

determining an output value of the trained neural network model using the first node value of the first node and the parent node value of the parent node of the first node, as the second node value of the second node,

wherein the determining the output value of the trained neural network model comprises

when both of the first node value of the first node and the parent node value of the parent node of the first node are input as first and second input values, respectively, determining the output value as the second node value of the second node to be visited following the first node.

12. The method of claim 11 , wherein the generating the dependency tree based on the neural network model comprises

generating the dependency tree on which a root word determined based on the semantic expression information is a node value of a root node, based on the trained neural network model.

13. The method of claim 12 , wherein the generating the dependency tree based on the neural network model further comprises:

determining a child node of the root node visited based on a breadth-first search (BFS) algorithm and a node value corresponding to the child node, based on the trained neural network model; and

generating the dependency tree based on the determined child node and the node value corresponding to the child node.

14. The method of claim 11 , wherein the generating the natural language sentence based on the generated dependency tree comprises

generating the natural language sentence based on node values of all nodes constituting the generated dependency tree.

15. A vehicle comprising:

a receiver configured to receive semantic expression information including a speech act or intention of a natural language sentence to generate the natural language sentence; and

a controller configured to:

extract speech acts or intentions from a speech signal of a user,

generate a dependency tree representing a syntax structure of at least one word determined based on the semantic expression information received for a response corresponding to the speech acts or intentions, based on a neural network model,

generate the natural language sentence based on the generated dependency tree,

visit a first node among nodes of the dependency tree based on a predetermined traversal algorithm,

determine a first node value of the first node and a parent node value of a parent node of the first node as input values of the neural network model in response to visiting of the first node,

based on the determined input values, train the neural network model to output a second node value of a second node to be visited following the first node, and

determine an output value of the trained neural network model using the first node value of the first node and the parent node value of the parent node of the first node, as the second node value of the second node,

wherein, when both of the first node value of the first node and the parent node value of the parent node of the first node are input as first and second input values, respectively, the controller determines the output value as the second node value of the second node to be visited following the first node.

16. A non-transitory recording medium storing a computer program, which, when executed by a processor, performs a method, the method comprising:

in combination with a computing device, receiving semantic expression information including a speech act or intention of a natural language sentence to generate the natural language sentence;

extracting speech acts or intentions from a speech signal of a user;

generating a dependency tree representing a syntax structure of at least one word determined based on the received semantic expression information received for a response corresponding to the speech acts or intentions, based on a neural network model;

generating the natural language sentence based on the generated dependency tree;

visiting a first node among nodes of the dependency tree based on a predetermined traversal algorithm;

determining a first node value of the first node and a parent node value of a parent node of the first node as input values of the neural network model in response to visiting of the first node;

based on the determined input values, training the neural network model to output a second node value of a second node to be visited following the first node, and

determining an output value of the trained neural network model using the first node value of the first node and the parent node value of the parent node of the first node, as the second node value of the second node,

wherein the determining the output value of the trained neural network model comprises

when both of the first node value of the first node and the parent node value of the parent node of the first node are input as first and second input values, respectively, determining the output value as the second node value of the second node to be visited following the first node.