<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
I tend to agree, but for the sake of argument, in MeetingLine [Repeat]
is an entry-level slot, and say you get:<br>
<br>
"what was the meeting like"<br>
<br>
parse 1 / 1 (WHAT WAS THE MEETING LIKE)==|MeetingRef<br>
[MeetingReferent] ( [_meeting] ( THE MEETING ) ) <br>
<br>
Pretty reasonable. But take out the <s> construct, and you get<br>
<br>
parse 1 / 1 (WHAT WAS THE MEETING LIKE)==|GenericDialogMechanisms<br>
[Repeat] ( WHAT ) <br>
MeetingRef<br>
[MeetingReferent] ( [_meeting] ( THE MEETING ) ) <br>
<br>
Not quite right. Instead of<br>
<br>
S: What do you want to know about your last meeting?<br>
U: what was the meeting like<br>
S: Sorry, I'm not sure I understood what you said.<br>
<br>
You get:<br>
<br>
S: What do you want to know about your last meeting?<br>
U: what was the meeting like<br>
S: What do you want to know about your last meeting?<br>
<br>
Of course this only happens because that kind of query isn't in the
grammar. So the wrong interpretation here isn't that costly IMHO. In
many other cases coverage is going to resolve the right interpretation.<br>
<br>
-Thomas<br>
<br>
Alex Rudnicky wrote:
<blockquote
cite="mid:200805211534.m4LFYuGA021763@jackfruit.srv.cs.cmu.edu"
type="cite">
<pre wrap="">I'm actually puzzled that the construct is being used at all. You get
the same effect by making [Repeat] an entry net.
Coercing an input into a single word is in any case iffy given that the
speaker is not constrained in their productions. Or maybe I'm missing
something here.
Alex
</pre>
<blockquote type="cite">
<pre wrap="">-----Original Message-----
From: <a class="moz-txt-link-abbreviated" href="mailto:olympus-developers-bounces@LOGANBERRY.srv.cs.cmu.edu">olympus-developers-bounces@LOGANBERRY.srv.cs.cmu.edu</a>
[<a class="moz-txt-link-freetext" href="mailto:olympus-developers-bounces@LOGANBERRY.srv.cs.cmu.edu">mailto:olympus-developers-bounces@LOGANBERRY.srv.cs.cmu.edu</a>] On
</pre>
</blockquote>
<pre wrap=""><!---->Behalf
</pre>
<blockquote type="cite">
<pre wrap="">Of TK Harris
Sent: Wednesday, May 21, 2008 10:45 AM
To: Olympus Developers
Subject: [Olympus developers 36]: <s> in grammars
Hi,
I'd like to mandate that the <s> construct in grammars, e.g.
[Repeat]
(*sorry what's that *again)
(<s> what </s>)
(*sorry i didn't HEAR YOU)
be forbidden. It's going to be a little awkward to support by some of
the language tools, mostly because it's no longer context-free. It's
caused problems in the past where <s> gets treated like other
vocabulary. My question is, are these non-context-free grammar
</pre>
</blockquote>
<pre wrap=""><!---->elements
</pre>
<blockquote type="cite">
<pre wrap="">too useful to simply remove? If so, then we might want to keep them,
but
change the syntax so that it doesn't overload the meaning of a token,
relying on a side-effect of the recognizer.
Thanks,
-Thomas
</pre>
</blockquote>
<pre wrap=""><!---->
</pre>
</blockquote>
<br>
</body>
</html>