PSY274H5 Chapter Notes - Chapter 7: Eliza, Digital Video Recorder, Morning Sickness
Chapter 7: Can computers learn language?
- We hear the common claim that computers can learn language, ex. voice recognition
- Based on the work of Alan Turing, a British mathematician. Turing developed a kind of
competition in which an interrogator (by keyboard) tried to figure out which of two respondents
was a human and which was a computer. He predicted that a computer program would be
judged hua i suh a opetitio ithi fift eas.
- In 1960s the program Eliza was developed and the result was that conversations with ELIZA
were jokingly claimed to resemble conversations with a psychotherapist. Between then and
now, other programs have been developed. Every time you use a search engine, you engage in a
rudimentary kind of conversation with a computer
- Two attempts of computer human conversations
o A.l.i.c.e.
▪ To any statement or question you input, a.l.i.c.e. responds with a
preprogrammed statement or question from its database—and that database is
impressively large.
o Daisy
▪ Not preprogrammed at all
▪ Daisy stores your input and manipulates it. So when you first talk with her, she
seems to have no intelligence whatsoever (all she can do is repeat what you
type in), but if you spend absurdly large amounts of time with her, she gradually
comes to form relatively coherent responses
- Some sample conversations and ask whether computers could have produced them.
- Consider conversation 1:
o A. Where are you going?
o B. I am going to school.
- First is that someone produced utterance A. We know computers can produce language—
whether preprogrammed (like a.l.i.c.e.) or not (like Daisy)—so utterance A could have been said
by either a human or a computer. Response B is based on interpreting utterance A. Again, we
know computers can do this to a certain extent. That is, they can analyze sentences to some
degree, recognizing verbs, you as the subject, and where as a location question word. They can
then match that sentence with the same verb(s), an appropriate subject, and possible location
responses—often those that begin with to (such as to work, to the grocery, and in this
conversation to school). However, if you say the sentences aloud, you will find them stilted.
Instead of response B, what would sound more natural? Probably this:
o C. School.
- That is, in casual conversation, we typically answer in fragments rather than whole sentences
- Both spoken and written language forms change over time, but changes in written language
usually lag behind changes in spoken language
- Consider conversation 2, which starts with this question:
o D. What’s up?
- Now consider these possible resposes ad do’t o aout ho the uestio i D ould e
written):
o E. Not much.
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
We hear the common claim that computers can learn language, ex. voice recognition. Based on the work of alan turing, a british mathematician. Turing developed a kind of competition in which an interrogator (by keyboard) tried to figure out which of two respondents was a human and which was a computer. He predicted that a computer program would be judged (cid:862)hu(cid:373)a(cid:374)(cid:863) i(cid:374) su(cid:272)h a (cid:272)o(cid:373)petitio(cid:374) (cid:449)ithi(cid:374) fift(cid:455) (cid:455)ea(cid:396)s. In 1960s the program eliza was developed and the result was that conversations with eliza were jokingly claimed to resemble conversations with a psychotherapist. Between then and now, other programs have been developed. Every time you use a search engine, you engage in a rudimentary kind of conversation with a computer. Some sample conversations and ask whether computers could have produced them. Where are you going: b. i am going to school.
