Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!seismo!rochester!kodak!elmgate!mj
From: mj@elmgate.UUCP
Newsgroups: comp.ai
Subject: Re: In layman's terms.
Message-ID: <639@elmgate.UUCP>
Date: Fri, 29-May-87 17:44:43 EDT
Article-I.D.: elmgate.639
Posted: Fri May 29 17:44:43 1987
Date-Received: Sat, 30-May-87 11:47:23 EDT
References: <13263@watmath.UUCP>
Reply-To: mj@elmgate.UUCP (Mark A. Johnson)
Organization: Eastman Kodak Company, Rochester, NY
Lines: 44
Keywords: English, Backus-Naur Form, syntax.


	I recently finished an experimental graduate course in 
	Natural Language Processing at Purdue, where I am working on
	my MSEE.  It was offered for graduate credit jointly by
	the schools of Electrical Engineering and English, and co-taught
	by an EE and a linguistics professor.  Due to
	other time constraints, I did not complete the final project,
	but understood the material well enough to have done so if
	I'd not been so rushed.  The course used Terry Winograd's 
	LANGUAGE AS A COGNITIVE PROCESS: VOLUME I: SYNTAX, which 
	discusses context-free phrase-structure grammars, 
	active chart parsing, transformational grammars, and 
	describes recursive and augmented transition networks (RTN's 
	and ATN's) well enough that implementation is 
	really not too difficult.  One of the nice things about the ATN
	parser is that (for the language described by the ATN),  it can
	be written to produce a recursive data structure that describes
	all of the possible meanings of a sentence available to the ATN.
	This means that, with a carefully described grammar and for a
	particular subset of a language, syntactically unambiguous 
	sentences will produce a unique parse,  and ambiguous sentences
	("Time flies like an arrow") will produce multiple parses.
	Semantic analyzes specially designed for acting on the frames
	produced by the ATN parser can then choose an appropriate parse,
	ask for clarification, update knowledge bases, etc.  
	
	I can give a limited recommendation for Winograd's book: it is fairly
	understandable, but it was quite hard to find anyone who
	liked his algorithm notation.  Winograd introduces an algorithm
	description language called "DL", and I can't say I like it much.
	It does have the advantage of separating the topic from the
	technology, i.e., you don't have to know any particular programming
	language to understand the algorithms.  Ain't my cup 'o' tea, tho'.

	One of the posters mentioned "waiting FOREVER for Volume II"
	of Winograd's book to come out.  Don't hold your breath.  I
	think I'll get the final volume (9 is it?) of THE ART
	OF COMPUTER PROGRAMMING before I buy Winograd Vol II.  I could be
	wrong, though, and would love to see Vol II in print.

-------------------------------------------------------------------------------
Mark A Johnson - Eastman Kodak Co. - Dept 646 - KEEPS - Rochester, NY 14650
The opinions expressed above are not necessarily those of Eastman Kodak Company
and are entirely my responsibility.