Automated natural language inference system

• US 20040044515A1
• Filed: 08/30/2002
• Published: 03/04/2004
• Est. Priority Date: 08/30/2002
• Status: Abandoned Application

• Current Assignee: Applied Voice & Speech Technologies Incorporated (Open Text Corporation)
• Original Assignee: Applied Voice & Speech Technologies Incorporated (Open Text Corporation)
• Inventors: Dingus, Peter, Metcalf, Michael
• Application Number:
  US10/231,552
• Publication Number:
  US 20040044515A1
• Time in Patent Office: Days
• US Class Current: 704/1
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

An automated natural language inference system that interprets KEYWORDS within a natural language stream to carry out a particular action inferred from such stream and, if necessary, queries the user for more KEYWORDS or natural language CONFIRMATION via a natural language dialog until an inferred action can be fired. The inference engine of the system populates rule TEMPLATES with different permutations of the KEYWORDS, constrained by word-to-slot size and the KEYWORD-to-slot number, to find a rule match. If more than one rule match is found, the rule with the highest priority is selected and fired.

Claims

1. An automated a natural language interactive inference system comprising:

an automatic speech recognition unit for recognizing natural language and identifying keywords therein, each keyword has a word size;

a template associated with a rule for firing an action wherein the template has slots each slot has a slot size; and

,

an inference engine that populates slots of the template with the keywords having a word size equal to the slot size wherein if a populated template matches the rule, the action is fired.

2. The system according to claim 1, wherein the inference engine populates the slots using keyword permutations constrained by slot size and a number of the keywords to a number of the slots and creates an active rule list therefrom.

3. The system according to claim 2, wherein each slot has a weighting factor that is used to determine which rule in the active rule list takes preference if more than one template permutation is matched to the rule.

4. The system according to claim 1, wherein the inference engine includes an interaction procedure that queries a user for more information based on a partially-populated template.

5. The system according to claim 4, wherein the partially-populated template has a highest priority of a list of partially-populated templates.

6. The system according to claim 1, wherein the automatic speech recognition unit converts the natural language into text and identifies keywords within the text and communicates the keywords to the inference engine.

7. The system according to claim 1, wherein each slot has a set of slot variables, the slot variables include the number of the slots of the template;

a weighting factor of each slot;

dialogue context;

optional words; and

, equivalent word occurrences.

8. The system according to claim 7, wherein the dialogue context is a function of the keyword recognition and includes email applications, voicemail applications or banking applications.

9. A method of automatically inferring natural language for an interactive natural language system comprising the steps of:

placing keywords in slots of rule templates in various permutations to create populated templates;

scanning production rules to determine which populated template has a production rule match;

during the scanning step, retaining the populated templates with complete or partial rule matches in an active rule list; and

, firing a rule in the active rule list that has highest priority.

10. The method according to claim 9, wherein the step scanning step includes the steps of:

if no rules can be made active, during the retaining step, determining closest probable rules with largest percentage of populated slots over some minimum cut off; and

,

querying using a natural language dialog a leading question in an attempt to satisfy a rule.

11. The method according to claim 10, further comprising the step of:

if no rule can be made active in a number querying steps, indicating a miss-recognition.

12. The method according to claim 9, further comprising prior to the placing step, the steps of:

receiving natural language speech;

converting the speech into text; and

,

parsing the text into the keywords.

13. The method according to claim 9, wherein the placing step is constrained by a slot sizes in the template and the sizes of available keywords.

14. The method according to claim 13, wherein the placing step is constrained by a number of slots in the template and a number of the available keywords.

15. The method according to claim 9, wherein the slots have weighting factors to determine which of the rules has the highest priority.

16. The system according to claim 9, wherein the keywords are a function of industrial application which includes email applications, voicemail applications or banking applications.

17. An automatic interactive natural language system comprising:

means for placing keywords in slots of rule templates in various permutations to create populated templates;

means for scanning production rules to determine which populated template has a production rule match;

means for retaining the populated templates with complete or partial rule matches in an active rule list; and

,

means for firing a rule in the active rule list that has highest priority.

18. The system according to claim 17, further comprising:

means for determining, if no rules can be made active, closest probable rules with largest percentage of populated slots over some minimum cut off; and

,

means for querying using a natural language dialog a leading question in an attempt to satisfy a rule.

19. The system according to claim 17, further comprising:

means for receiving natural language speech, converting the speech into text; and

, parsing the text into the keywords.

20. The method according to claim 17, wherein:

the placing by the placing means is constrained by a slot sizes in the template and the sizes of available keywords;

the placing step is constrained by a number of slots in the template and a number of the available keywords; and

,

the slots have weighting factors to determine which of the rules has the highest priority.