Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!EU.net!uknet!festival!hwcee!ceecjs
From: ceecjs@cee.hw.ac.uk (Chris J Schnurr)
Subject: Re: Text generation writing tool
Message-ID: <Cw48Eu.8tL@cee.hw.ac.uk>
Sender: news@cee.hw.ac.uk (News Administrator)
Organization: Dept of Computing & Electrical Engineering, Heriot-Watt University, Scotland
References: <Cw2ABJ.IBC@cee.hw.ac.uk> <354d6a$1d6@hobbes.cc.uga.edu>
Date: Wed, 14 Sep 1994 10:42:30 GMT
Lines: 45

In response , I wrote :
: a syntactical parser, to come up with predicted words.

They wrote : ( Excuse the badly formatted response but I haven't a clue how
this newsreader works) :

>My 5-year-old daughter would use such a system avidly, if it worked.
>
>But without knowing the subject matter and the ideas to be expressed,
>just how many words could it predict?

The system predicts a list of five words. This doesn't seem like many, but wait!It updates this list of 5 words Every keystroke. The list is updated in view
of what has been typed (the sentence to this point) and what has been type
(the word up to this point)
for example, if a child wanted to write "water", they would probably start off
with a phonetic attempt and say "wuh- Auh - Tuh". So they first type (hopefully)
a "W". the prediction system comes up with 5 word choices all begining with 
the letter "w". If the child has previously typed "water", then it will 
almost certainly be in the prediction list. At this stage, if water is in the 
list, the child selects it using a function key. If it isn't there, (s)he 
continues spelling the word : and ( again, hopefully, types an "a"). This
time, the list changes again, and give a choice of another 5 predicted words
all begining with the letters "wa". 

The system does not really need to know about the subject matter or ideas
behind the system, as it works at completing one word at a time. All it needs
is a reasonably large dictionary, and it adapts it's prediction as you type.

It currently works very well (if a wee bit slowly for fast typists) under
MsDOS on any PC.

Again, I should state that it is only completing one word at a time, rather than
predicting whole sentences. It is meant as a writing aid for users, not as a
writing tool for machines (!)


Chris

amended sig:
        The farmer sold the cow because she == gave no milk.
        The farmer sold the cow because she == needed money.  
        Stick that in your semantic net and parse it !
        ============== ceecjs@uk.ac.hw.cee =================
        The bane of HCI is the Human...

