<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 TRANSITIONAL//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; CHARSET=UTF-8">
<META NAME="GENERATOR" CONTENT="GtkHTML/3.10.3">
</HEAD>
<BODY>
On Wed, 2006-11-08 at 10:36 +0900, Francis Bond wrote:
<BLOCKQUOTE TYPE=CITE>
<PRE>
<FONT COLOR="#000000">G'day,</FONT>
<FONT COLOR="#000000">shouldn't you be asleep by now?</FONT>
<FONT COLOR="#000000">> I am wondering what exactly is the difference between -results --- I know</FONT>
<FONT COLOR="#000000">> what that one is --- and -nsolutions. Can anyone enlighten me here?</FONT>
<FONT COLOR="#000000">I think:</FONT>
<FONT COLOR="#000000">-results sits in the output routine and stops it printing all the</FONT>
<FONT COLOR="#000000">results. They are still all calculated.</FONT>
<FONT COLOR="#000000">-nsolutions asks cheap to only produce the top "n" parses.</FONT>
<FONT COLOR="#000000"> - with no model, I don't knwo what it does</FONT>
<FONT COLOR="#000000"> - with just a scoring model it uses some kind of beam search (I think)</FONT>
<FONT COLOR="#000000"> -- this means that the result is often not the global optimum</FONT>
</PRE>
</BLOCKQUOTE>
Yes. That's what we observed as well.
<BLOCKQUOTE TYPE=CITE>
<PRE>
<FONT COLOR="#000000"> - with selective unpacking it should produce the global best n results</FONT>
<FONT COLOR="#000000"> -- without having to explore the whole search space</FONT>
</PRE>
</BLOCKQUOTE>
Is there any documentation, i.e. paper, on how this is done?<BR>
<BR>
Thanks a lot for the summary. <BR>
<BR>
Berthold
<BLOCKQUOTE TYPE=CITE>
<PRE>
<FONT COLOR="#000000">Stephen, Bernd or Yi, please correct me if I am wrong.</FONT>
</PRE>
</BLOCKQUOTE>
</BODY>
</HTML>