Information Technology Reference
In-Depth Information
The 1960s mark the appearance of the first MMD systems. ELIZA 1
[WEI 66], which we mentioned earlier, is fascinating in more than one way.
First of all, this is a written dialogue system that really works without looping
or randomly stopping. It is always possible to carry out conversations on
hundreds of speaking turns. Moreover, the chosen task is itself fascinating:
the system is supposed to play the role of a non-directing psychotherapist,
which means it simply listens to the speaker to tell it about his/her problems
(“I have a problem with my parents”) and sometimes reacts to certain
sentence (“tell me about your family”). The realism is so strong that some
users have spent hours talking with ELIZA, and J. Weizenbaum had to decide
against openly adding a dialogue-saving module, faced with accusations of
spying and violating privacy. This task has two advantages: it does not have to
carry out a complex dialogue, for example with negotiation or argumentation,
while keeping a spontaneous and natural aspect, since the user can say what
he/she wants when he/she wants to; and on the other hand, it is easy to
program, since the system does not need to understand absolutely everything:
utterances such as “what makes you say that?” or “I see, please go on” are
vastly sufficient. Indeed, and this is fascinating for NLP, AI or MMD
researchers, J. Weizenbaum managed to develop a system that appears to
master language and pass the Turing test, whereas it does not even approach
the most basic issues of automatic understanding.
Indeed, all of ELIZA's operating relies on a few well-chosen heuristic
rules. The system knows a few words, especially those linked to family:
“parents”, “mother” and “father”. It is thus able to bounce off the utterance “I
have a problem with my parents” without any understanding involved in this
process: the system just detected “parents” and answered with a new question
on “family”, a new question that actually allows it not to have to take into
account the meaning of the user's utterance. The system also knows the
personal pronouns referring to the two interlocutors, “I”, “me”, “my”, “you”
and “your” that allows it to carry out replacements and build an utterance
taking up parts of the user's utterance, such as “what makes you believe that
1 The name came from the ELIZA Doolittle character in the movie My Fair Lady (1964, G.
Cukor), itself an adaptation of the play Pygmalion (1914, G.B. Shaw), which has also been
adapted for the movies. ELIZA Doolittle is a florist from a very poor neighborhood, and
becomes the subject of a bet when an aristocrat claims that by changing her manner of speech,
he will be able to make her pass for an aristocrat herself. The idea of duping someone through
language and dialogue is thus the origin of the system's name.
Search WWH ::




Custom Search