ELIZA  -  A Study of Natural Language

Eliza is a program designed to make natural language communication
between Man and machine possible.
Whether or not it successfully achieves this will be answered by the user.
Eliza was originally written by Joseph Weizenbaum and described in his
paper appearing in "The Communications of the Association of Computing
Machinery" in 1966.

The appearance of Weizenbaum's paper prompted much discussion in computing
circles. Many believed Weizenbaum had crossed a barrier into the understanding
of human speech, others recognized eliza to be nothing more than a clever trick.
Amusing anecdotes surfaced about people who had used eliza. Some people
unknowingly used the program, believing their computer terminal to be connected
to that of another employee - with humourous results!
Weizenbaum added an option to the program which channeled people's
responses to a large file, which Weizenbaum was able to later read and muse
over. Weizenbaum's secretary, who claimed that she entered personal and
confidential information to the program, demanded that he remove this option.

It is often claimed that to explain is to explain away - and nowhere is this
maxim truer than in a program such as eliza. The user, initially bemused
be eliza's apparent ability to "understand" speech, quickly moves it to
the "I-could-have-done-that" basket. To help you with this transition,
here is a brief description of eliza.

Eliza first prints a greeting message, and while this is being read by the
user, reads from a large data file termed the 'script'. The script
consists of two types of dictionaries - templates and keywords.
Each template consists of a standardized sentence, followed by a number of
optional responses.

For example :

---snap---
he is %
*how long has he been %?
*is your mother also %?
*do you want to be % also?
---snap---

where the % symbol may match any number of words. For example
"he is very dead" may draw the response "is your mother also very dead?".

If eliza is unable to match the user's input sentence with any of the standard
sentences (templates), it attempts to match each word in the input sentence
with a list of keywords. Each keyword has associated with it a number
indicating the importance attached to finding that keyword in the input
sentence. A typical keyword sequence may be :

---snap---
perhaps 4
*you are being indecisive
*you are still being indecisive. I warned you
*you seem to lack self confidence
---snap---

If eliza fails to match both templates and keywords, it relies on what the
user has previously typed - keeping a "history" of keywords and using
these to prompt the conversation if it gets "bogged down".

All very simple really!

The implementation of eliza on display for Uniday in the Department of
Computer Science is running on a Digital Corporation VAX-11/750 under the
Unix version 4.2BSD operating system. It is written in the 'C' programming
language and reads from a thirty six thousand character script, comprising
eighty templates and two hundred and twenty keywords. By modifying the
standard script and increasing the priority of keywords such as 'uniday',
'computer', 'america' and 'cup', eliza can be made to give the
impression of "understanding" current events.

Any questions regarding eliza or the Department of Computer Science in
general will be gratefully received.

Chris McDonald