CSE 472/572, Spring 2002, PROJECT 1: ELIZA

CSE 472/572, Spring 2002

Programming Project 1
NATURAL-LANGUAGE PROCESSING BY PATTERN MATCHING

Read the following:
- Winograd, Terry (1983), excerpts from Ch. 2 of Language as a Cognitive Process; Vol. 1: Syntax (Reading, MA: Addison-Wesley).
- Weizenbaum, Joseph (1966), "ELIZA - A Computer Program for the Study of Natural Language Communication between Man and Machine", Communications of the ACM 9: 36-45;
  - reprinted in 26: 23-28.
  - Be sure to read the correspondence between Weizenbaum and Arbib on p. 28 of the reprint.
  - You might also be interested in:
    - Weizenbaum, Joseph (1967), "Contextual Understanding by Computers", Communications of the ACM 10: 474-480.
    - Weizenbaum, Joseph (1976), Computer Power and Human Reason: From Judgment to Calculation (New York: W. H. Freeman).
- Shapiro, Stuart C., & Kwasny, Stanley C. (1975), "Interactive Consulting via Natural Language", Communications of the ACM 18: 459-462.
You might also want to play with an Eliza implementation. Try these:
- Get into emacs and execute M-x doctor. You can get a printout by doing M-x print-buffer. To exit, either just ignore it, or use C-x k to kill the buffer.
- Get into /projects/rapaport/Allen, which contains software for James Allen's NLU text. Read eliza-doc.text and follow the directions.
- The best implementation I know of is in Franz Lisp; hence, it no longer runs in Common Lisp. If anyone has the urge to convert it to Common Lisp, see below. It's in /projects/rapaport/Doctor.

In this project, you are to create your own convincing mini-Eliza. You may do this in one of three ways:

Either implement and fully test the pattern-matching procedure whose specification is given in the material from Winograd with the modifications of exercises 2.1 and 2.13.
or else implement and fully test the mini-Eliza whose specification is given in Shapiro's Common Lisp text, in the exercises marked '(p1)' (i.e., 12.1-12.2, 13.13-13.14, 14.7-14.8, 16.15, 17.29-17.32 (hint: read 26.9 before doing 17.31!), 18.25-18.27, 24.9-24.10, 26.8-26.14, 29.32 (unless I missed some)).
(See the errata for this book.)
or else translate /projects/rapaport/Doctor to Common Lisp and fully document your code.

You should provide test runs on a variety of interesting and appropriately chosen pairs (e.g., you could try to implement Weizenbaum's original "Doctor" script or develop a "help" system for Unix).

In addition, you must design your system so that it has the following recognition-response pair:

'My 1 is 2 -> What if your 1 were not 2'

If this is applied to the sentence 'My left arm is about to fall off', it should produce 'What if your left arm were not about to fall off'.

(a) Try your mini-ELIZA on the following inputs:

(i) My head is on my shoulders.
(ii) My problem is that you hate me.
(iii) My problem is how to pay you.
(iv) My brother said your car is bigger than mine.
(v) My job is working in the mine.

(b) In each case, discuss why the response is inappropriate, and describe the changes to the procedure that would be needed to produce responses that are syntactically and/or semantically appropriate (i.e., grammatical and/or meaningful) (don't worry about their therapeutic appropriateness!:-).

Your report should include a brief description of what you have done and a discussion of Eliza-like natural-language processing. In particular:

Following both the requirements for projects as stated in the syllabus and the requirements for this project, there are 4 major items of roughly equal importance:

the project description
annotated sample runs
documented code
the Eliza question (part 5 of the project handout)

Here's a breakdown:

project description:
- description of project
- description of Eliza-like NLP
- optional: bug report (if your program doesn't work, you can recoup lost points by explaining your problems and how you might solve them with more time)
annotated sample runs:
- sample runs
- annotations
documented code:
- code ("Eliza")
- script ("Doctor")
- documentation
part 5:
- implementation of the pattern in part 5
- 5a: responses to the 5 sentences (either computer-generated or hand-done)
- 5b: discussion of the 5 sentences
- 5c: changes to code
  - sample run
  - problems
DUE AT START OF LECTURE, MON, FEB 18

Copyright © 2002 by William J. Rapaport (rapaport@cse.buffalo.edu)
file: 572/S02/proj1.28ja02.html

Programming Project 1 NATURAL-LANGUAGE PROCESSING BY PATTERN MATCHING

Programming Project 1
NATURAL-LANGUAGE PROCESSING BY PATTERN MATCHING