From don  Mon May 26 06:35:05 2003
Date: Mon, 26 May 2003 06:35:05 -1000
Subject: cs441 Tuesday
From: Don Colton <don@colton.byuh.edu>

Aloha,

This is a reminder that on Tuesday (tomorrow, May 27), the CS students
are expected to have their programs converted into w2p and p2p forms,
ready to run in class.  The term is zooming past.  We need to come to
closure on this part of the course.

The linguistics students have been a little less lucky to date, with
not as clear an assignment.  We have had the ongoing task of
developing rules that account for the difference between human-labeled
speech and dictionary speech.

I want to move forward in a new direction now.  Let's grab the bull by
the horns, finally, and get on with our big task: converting a stream
of phonemes into words.

I want the Linguistics students to invent a way to convert a stream of
phonemes into a stream of "reliable events" for a word.  By reliable,
I mean that there are not 20 ways to say "seven."  I mean there is
probably one, maybe two.  Feb-u-ary, Feb-ru-ary.  Not Feb-u-ary,
Feb-ya-wary, Fib-i-wary, etc.  What are the features that are most
reliably realized in human speech?  I need rules to convert a phoneme
stream into this new form.  This will be the "w2p" and "p2p" activity
for you.  What does a human listen for?  What does a human hear in the
speech it takes in?

The phonemes-to-words process will be like this (unless someone thinks
of a better idea): Starting at the front of the utterance, we look at
the first phoneme and do a reverse lookup to see if it is in our
dictionary.  If so, we report it.  We repeat with the first two
phonemes, then three, four, five, etc.  (We need some way to know when
to stop.)  We report all the words that "could" start at the beginning
of the utterance.  Then we move forward to the second phoneme and
repeat the whole process.  We will need some way to evaluate the
reliability of the match.  I want us to make substantial progress on
this task over the next week.  We will talk about it in class.

Finally, I am planning a little quiz for tomorrow.  I will give you a
stream of phonemes and ask you to translate it into words.  It
shouldn't be too hard, but I guess we will find out when I do it.

See you tomorrow.

Bro Colton