On 9/11/07, <b class="gmail_sendername">John F. Sowa</b> <<a href="mailto:sowa@bestweb.net">sowa@bestweb.net</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<br>RF> I'm just saying we should at least explore the possibility<br> > formal grammars are "necessarily incomplete" descriptions of<br> > corpora, that the right way to handle language is to generalize
<br> > grammar ad-hoc from examples, as you go.<br><br>In fact, that is what many, if not most, grammar and parser<br>developers have been doing for the past 50 years. Everybody who<br>is developing broad-coverage parsers starts small and generalizes
<br>with ad-hoc examples (usually selected from one or more corpora)<br>until the coverage gets better and better.</blockquote><div><br>No, no. You are simply misunderstanding me John. When I say "generalize grammar ad-hoc from examples as you go" I don't mean "as you develop your grammar". I mean "from sentence-to-sentence."
<br><br>It is not a question that you start with a grammar where "black" is in the same class as "strong" and then gradually remove it as you add examples. "Black" will need to be in a class with "strong" for every "coffee" context, and _not_ in a class with strong for every "cloud" context.
<br><br>You need to be able to access both these generalizations. But they contradict ("black" = "strong" && "black" != "strong".)<br><br>The only solution is to keep the examples and generalize "black" = "strong" or "black" != "strong", as the context demands it.
<br><br>By all means disagree with me, but don't characterize this as "what parser developers have been doing for 50 years." It is different.<br><br>Please see the difference and argue against that.<br><br>-Rob
<br></div></div>