HOMEWORK 3 (CS 1573): An Intelligent Dialogue Agent

Assigned: February 26, 2004

Due: March 18, 2004

Project Description

The goal of this project is to use natural language processing (rather than simple pattern matching as in Eliza) to implement a dialogue agent that can suggest a CS 1500-level course to take next semester, based on user preferences regarding class time and instructor.

Your system should take as input English sentences, and extract enough information to query the information in the following url, based on the Hours and/or Instructor fields. Feel free to create a local database with only the subset of courses and fields required for this assignment.

Here are some example dialogues:

• Agent: Hello, this is the course assistant agent. Do you prefer a morning, afternoon, or evening class?
• User: I want a class that is held in the afternoon and is taught by Professor Hwa.
• Agent: (queries database with 12:00 < HOURS < 6:00 and INSTRUCTOR=Hwa) You're in luck, Introduction to Artificial Intelligence (1571) satisfies your request. Would you like any more recommendations?

• Agent: Hello, this is the course assistant agent. Please tell me a preferred class time or instructor.
• User: I want a class that begins before 12:00.
• Agent: Please state morning, afternoon, or evening.
• User: Oh, sorry, morning.
• Agent: What instructor do you prefer?
• User: I really loved Dr. Litman's 1573 class.
• Agent: (queries database with HOURS<12:00 and INSTRUCTOR=Litman) You're out of luck, no course next fall satisfies your request. Would you like any more recommendations?

Your agent should use a finite-state automaton to implement its dialogue manager, with transitions based on semantic analysis of the user input. You should implement either the D-Recognize or ND-Recognize algorithms (from class and the handout) to process inputs using your dialogue FSA.

To implement semantics (e.g., to compute contains-instructor(answer)), your implementation should use simple pattern matching. However, you will only be required to process temporal expressions of the form "morning", "afternoon", "evening", "in the morning", "in the afternoon", "in the evening", "before hh:mm", "after hh:mm", "at hh:mm". For instructors, you only need to recognize expressions of the form "Dr. Surname", "Professor Surname". The wildcard "I don't care" is also an acceptable answer for either names or times.

You should be able to handle user inputs that provide preferences in one shot (as in the first dialogue), or incrementally (as in the second). You should also have a repair strategy for handling user inputs that your system cannot initially process (as in the second dialogue), and for handling other types of complications (e.g., specification of multiple professors).

Your grade will be based both on a set of training sentences to be provided, and an unseen test set.

Please submit (at least in hardcopy) a graphical representation of your dialogue manager. Please also submit both hardcopies and electronic versions of your source code, a README file describing how to run it, and a transcript highlighting your program's best features. Bring the hardcopies to class, and submit electronically following the class submission policies.

IMPORTANT NOTE: Beginning with this homework, points will be deducted if the submission procedure is not followed.