Interactive processing apparatus having natural language interfacing capability, utilizing goal frames, and judging action feasibility

• US 6,275,788 B1
• Filed: 03/07/2000
• Issued: 08/14/2001
• Est. Priority Date: 09/26/1996
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Inventors: Watanabe, Keisuke, Nagai, Akito
• Primary Examiner(s): Thomas, Joseph
• Application Number:
  US09/520,365
• Time in Patent Office: 525 Days
• Field of Search: 704/1, 704/8, 704/9, 704/10, 704/257, 704/275, 707/1, 707/2, 707/3, 707/6, 707/104, 706/45, 706/71, 706/11, 706/12
• US Class Current: 704/1
• CPC Class Codes: G06F 40/20 Natural language analysis semantic analysis of natural language G06F40/30
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A user goal extracting unit extracts a user goal from an input statement 10 entered by a user. A system goal determining unit 15 determines a system goal in accordance with the user goal. A goal frame generating unit 17 generates a goal frame based on an action sequence knowledge corresponding to the system goal. An action feasibility judging unit 19 sets arguments for, and judges the feasibility of, an action in the goal frame. If the action is judged to be feasible, the action feasibility judging unit 19 outputs an action command to an external application 20. If the action is not judged to be feasible, the action feasibility judging unit 19 outputs a new system goal to the goal frame generating unit 17. The external application 20 outputs a result of execution of the action.

Claims

1. An interactive processing apparatus, comprising:

a user goal-extracting unit extracting a user goal from an input natural language statement;

a system goal determining unit determining a system goal corresponding to the user goal extracted by the user goal extracting unit;

an action sequence-generating unit generating an action sequence for achieving the system goal determined by said system goal determining unit;

an action feasibility judging unit setting arguments for an action of the generated action sequence and judging the feasibility of the action by determining whether each argument needed to execute the action has been set; and

an output outputting an action command for execution when the action is judged to be feasible.

2. The interactive processing apparatus according to claim 1, wherein said action sequence generating unit retrieves from an action sequence knowledge a system action sequence knowledge node corresponding to the determined system goal, the action sequence knowledge including system action sequence knowledge nodes needed to achieve various system goals, said action sequence generating unit generating an action sequence generating a goal frame that pairs the determined system goal with a corresponding system action sequence knowledge node.

3. The interactive processing apparatus according to claim 1, wherein

said action feasibility judging unit generates a system goal for acquiring an argument needed to execute an action when said feasibility judging unit action determines that not all arguments needed to execute the action have been set.

4. The interactive processing apparatus according to claim 1, further comprising:

an interaction history storage unit storing a stack of extracted user goals and semantic representations of input natural language statements.

5. The interactive processing apparatus according to claim 4, wherein

said action feasibility judging unit retrieves information from the interaction history storage unit to set an argument for an action when all arguments needed to execute the action have not been set.

6. The interactive processing apparatus according to claim 4, wherein

said interaction history storage unit assigns scores to data items stored therein, and

said action feasibility judging unit sets an argument needed to execute an action in accordance with the scores assigned to data items stored in said interaction history storage unit.

7. The interactive processing apparatus according to claim 6, wherein said interaction history storage unit assigns scores to data items stored therein based on a time function.

8. The interactive processing apparatus according to claim 1, further comprising:

a next action determining unit determining, based on an action execution result, an action of the generated action sequence to be executed next.

9. An interactive processing method, comprising:

extracting a user goal from an input natural language statement;

determining a system goal corresponding to the extracted user goal;

generating an action sequence for achieving the determined system goal;

setting arguments for an action of the generated action sequence and judging the feasibility of the action by determining whether each argument needed to execute the action has been set; and

outputting an action command for execution when the action is judged to be feasible.

10. The interactive processing method according to claim 9, wherein said step of generating an action sequence retrieves from an action sequence knowledge a system action sequence knowledge node corresponding to the determined system goal, the action sequence knowledge including system action sequence knowledge nodes needed to achieve various system goals, said step of generating an action sequence generating a goal frame that pairs the determined system goal with a corresponding system action sequence knowledge node.

11. The interactive processing method according to claim 9, further comprising:

generating a system goal for acquiring an argument needed to execute an action when said step of setting arguments for, and judging the feasibility of, the action determines that not all arguments needed to execute the action have been set.

12. The interactive processing method according to claim 9, further comprising:

generating an interaction history that includes a stack of extracted user goals and semantic representations of input natural language statements.

13. The interactive processing method according to claim 12, further comprising:

retrieving information from the interaction history to set an argument for an action when all arguments needed to execute the action cannot otherwise be set.

14. The interactive processing method according to claim 12, further comprising:

assigning scores to data items stored in the interaction history; and

setting an argument needed to execute an action in accordance with the scores assigned to data items in the interaction history.

15. The interactive processing method according to claim 14, wherein scores are assigned to data items stored in the interaction history based on a time function.

16. The interactive processing method according to claim 9, further comprising:

executing the output action command to generate an action execution result; and

determining, based on the action execution result, an action of the generated action sequence to be executed next.

17. An interactive processing apparatus comprising:

a user goal extracting unit referring to a user goal knowledge that includes user goals for defining intentions of input natural language statements, in order to extract a user goal from each input statement, thereby generating a semantic representation of each input statement;

an interaction history storage unit storing semantic representations of input statements and user goals output by the user goal extracting unit, as well as semantic representations output by a next action determining unit;

a system goal determining unit determining a system goal corresponding to a given user goal from a system goal knowledge that includes system goals constituting action goals of a system;

a goal frame generating unit retrieving from an action sequence knowledge a system action sequence knowledge node corresponding to a system goal which is input either from the system goal determining unit or from an action feasibility judging unit, the action sequence knowledge including system action sequence knowledge nodes needed to achieve system goals, the goal frame generating unit further generating a goal frame pairing the system goal with the corresponding system action sequence knowledge node;

a goal frame storage unit storing goal frames output by the goal frame generating unit;

the action feasibility judging unit retrieving a semantic representation from the interaction history storage unit in order to set arguments for, and judge feasibility of, an action in a goal frame being held by the goal frame storage unit;

the action feasibility judging unit outputting an action command if the action is judged to be feasible;

the action feasibility judging unit generating a new system goal if the action is not judged to be feasible, the new system goal being output to the goal frame generating unit;

an external application driven upon receipt of an action command from the action feasibility judging unit, the external application further outputting a result of execution of the action; and

the next action determining unit determining, in response to the result of action execution output by the external application, an action to be executed next in an action sequence of a goal frame being held by the goal frame storage unit, the next action determining unit further outputting to the interaction history storage unit a semantic representation acquired from the result of the execution of the action.

18. The interactive processing apparatus according to claim 17, wherein the interaction history storage unit stores in a stacked fashion semantic representations of input statements and user goals output by the user goal extracting unit, as well as semantic representations output by the next action determining unit;

and

wherein, if there exists an action for which arguments cannot be set because either a referring, a pronoun or an abbreviated expression is included in the input natural language statement, the action feasibility judging unit consecutively retrieves semantic representations stacked in the interaction history storage unit so as to set arguments for the action.

19. The interactive processing apparatus according to claim 17, wherein the interaction history storage unit assigns a score to each of a plurality of data item stored therein, and

wherein the action feasibility judging unit sets arguments for actions in accordance with the scores assigned to data items stored in the interaction history storage unit.

20. The interactive processing apparatus according to claim 19, wherein the interaction history storage unit controls the scores assigned to data items stored therein based on time function.

21. An interactive processing method, comprising:

referring to user goal knowledge that includes user goals for defining intent of input natural language statements to extract a user goal from each input natural language statement, thereby generating a semantic representation of each input natural language statement;

storing semantic representations of natural language statements and extracted user goals to create interaction history;

determining a system goal corresponding to an extracted user goal from system goal knowledge that includes system goals representing action goals of a system;

retrieving from action sequence knowledge a system action sequence knowledge node that corresponds to a determined system goal, the action sequence knowledge including system action sequence knowledge nodes needed to achieve various system goals, and generating a goal frame that pairs the determined system goal with a corresponding system action sequence knowledge node;

storing a generated goal frame;

retrieving a semantic representation from the interaction history to set arguments for, and judge feasibility of, an action represented in the stored goal frame, outputting an action command if the action is judged to be feasible, and generating a new system goal if the action is not judged to be feasible, the new system goal being used to generate a new goal frame;

driving an external application upon receipt of an output action command and outputting a result of action execution; and

determining, in response to the output result of action execution an action to be executed next in an action sequence of the stored goal frame, and outputting to the interaction history a semantic representation acquired from the result of the execution of the action.

22. The interactive processing method according to claim 21, wherein

the interaction history includes a stack of extracted user goals and semantic representations of input natural language statements, as well as semantic representations acquired from the result of action execution; and

if there exists an action for which arguments cannot be set, information from the interaction history is retrieved to set arguments for the action.

23. The interactive processing method according to claim 21, further comprising:

assigning a score to data items stored in the interaction history,

wherein arguments for actions are set in accordance with the scores assigned to data item stored in the interaction history.

24. The interactive processing apparatus according to claim 23, wherein the scores assigned to data items stored in the interaction history are assigned based on a time function.