430 likes | 1.29k Views
Augmented Transition Networks:. Combining different AI Technologies for Language Understanding. Motivations. In the last lecture, you have learned to use regular transition networks to parse a grammar efficiently.
E N D
Augmented Transition Networks: Combining different AI Technologies for Language Understanding
Motivations • In the last lecture, you have learned to use regular transition networks to parse a grammar efficiently. • This technology has trouble dealing with subject-verb agreement, and other inflectional aspects of the language such as person, tense, and number. • Today you will learn about augmented transition networks to handle these difficulties by adding AI technologies previously learned in this course to the basic transition network.
Objectives • Augmented transition networks • Dog-world grammar • Dog-world knowledge base • Information extraction from the Web • Database front-end
Regular transition network • Invoke parse(verb_phrase) will tell us whether the next section of the input stream is a verb phrase or not. • But it does not tell us what kind of verb phrase it is. What person-number-tense? • We are going to re-define sentence as a frame and attach procedures to the arcs (transitions) to make augmented transition network. A transition network grammar Trace "dog bites"
Augmented transition network ATN parsers Terminals and non-terminals (grammar symbols) are represented as frames. Grammar symbol characteristics are represented as slots:
subject-verb agreement The dog likes a man Parse tree for “The dog likes a man”
.NOUN_PHRASE Attach procedures to arcs ←textbook typo p. 609 checking for agreement→ Frame data structures
Dog-world knowledge base Type hierarchy Conceptual graphs for the concept like and the concept bite
Parse tree to semantic representation 1. As usual, to begin, call sentence 2. sentence calls noun_phrase 3. noun_phrase calls noun 4. noun returns concept for dog (1) 5. Article is definite "the dog". Bind a marker #1 to dog (2) 6. sentence calls verb_phrase 7. verb_phrase calls verb which retrieves concept for like (3) 8. verb_phrase calls noun_phrase which calls noun to retrieve man (4) 9. "a man" (5). 10. verb_phrase procedure restricts entity to man and returns (6). 11. sentence procedure joins dog: #1 to experiencer (7).
SQL Database query What is John Smith's salary? select salary from employee_salary where employee ="John Smith" English: SQL:
Database query 2 List the salaries of employees who work for Ed Angel. select salary from employee_salary, manager_of_hire where manager = "Ed Angel" and manager_of_hire.employee = employee_salary.employee
Combine entity-relationship diagram with conceptual graph Database entity-relationship diagram Who hired this employee? AI knowledge base conceptual graph
Conclusion • Natural language understanding is one of the most difficult problems in AI, even though little kids seem to learn languages effortlessly. • You have seen how the following AI technologies have been used to process natural language: FOL, grammars, parsing, semantic networks, transition networks, frames, augmented transition networks, and conceptual graphs. • These AI techniques have been successful applied to database front-ends and expert system interfaces.