This is intended to be a relatively simple intro to tgrep.  See README.long
for a fuller description.


WHY TGREP?

Treebank bracketing is represented by multiple levels of parentheses that
look a lot like Lisp:

  Treebank II:
   (TOP (S (NP-SBJ my best friend)
	   (VP gave
	       (NP me)
	       (NP chocolate)
	       (NP-TMP yesterday))
	   .))

  Treebank I (approximate):
   (TOP (S (NP my best friend)
	   (VP gave
	       (NP me)
	       (NP chocolate))
	   (NP yesterday))
	.)

You can't effectively search this kind of thing using plain old grep,
because grep can only look at single lines.  So we have tgrep, which can
look for geometric relationships in the "trees" represented above.


BASICS

See the file README for set-up instructions.


Here's a simple tgrep command:

	tgrep TOP

This would find all sentences in the TGREP_CORPUS defined in your .cshrc.


Here's a more complicated example:

	tgrep -f -w 'S < /^NP-TPC/' /mnt/unagi/nldb/tgrep/wsj_skel.crp

which decomposes like this:

	-f
		print source Filenames
	-w
		print Whole sentence (not just the matching part)
	'...'
		search pattern, enclosed in quotes to keep csh happy
	/...wsj_skel.crp
		tgrepable corpus to search, instead of the TGREP_CORPUS
		defined in your .cshrc

Another possibly useful switch is -a, which prints out all matches found in
a given sentence instead of just the first one.  Not needed with -w, though.


You can't really search files directly with tgrep.  Instead you must use
pre-compiled "tgrepable" corpus files.  There are several of these
available in /nldb/tgrep; look for files with a .crp or .corpus extension
and/or see the INVENTORY file.



SEARCH PATTERNS

The syntax of search expressions is detailed in README.long.  Here are a
few examples.  All of them happen to match the above example sentences.

   'NP'
		anything called just NP.
		  In the Treebank II example above, this only matches the
		"me" and "chocolate" NPs, since the -SBJ and -TMP make the
		others different in tgrep's eyes.

   'TOP'
		anything called TOP.  Since every sentence is enclosed in
		TOP brackets, this matches everything.

   'VP < NP'
		VP immediately dominates an NP
		(VP has a child called "NP")

   'VP <3 NP'
		VP has NP as its 3rd child

   'NP > VP'
		an NP that has a VP parent

   'VP << chocolate'
		VP contains the word "chocolate"
		(i.e. "VP" dominates "chocolate", directly or indirectly)

   'S \!<< balderdash'
		S does not contain the word "balderdash"

		[the backslash (\) before the exclamation point (!) above
		 keeps csh/tcsh from doing history substitution ... hmm,
		 just take my word for it and put it in.]

   'NP $.. NP'
		two NPs have a common parent
		(first NP precedes a sister also called NP)

   '/^NP/'
		// encloses a grep-style regular expression, where "^" and
		"$" mean beginning and end of _word_ instead of line.

		Regexps can be pretty hairy, but this one is worth knowing
		-- it means "anything starting with NP", such as:
		NP, NP-SBJ, NP-TMP, NP-ADV-TPC-1, ...


If you use several operators in a row, the results are a bit unintuitive at
first.  For instance,
 'S < VP < NP'
does NOT mean
 "S has a VP child that has an NP child"
but rather
 "S has a VP child _and_ an NP child".
To get the first meaning, you need to use parentheses:
 'S < (VP < NP)'

You can put all this together in fairly complex ways.  For instance:

 'S <1 /^NP/ < (VP < (NP $.. NP))'

  means:
		Get all Ss that start with an NP (not necessarily the
		subject) and that dominate a VP that in turn has two NP
		children -- in other words, sentences with what might be
		double-object VPs.
		  (In Treebank II, they probably will really be double
		objects.  In Treebank I, they could also be things like "I
		saw yesterday my best friend from college.")


INFO ON TREEBANK

For extensive information on Treebank II bracketing, see
    ftp.cis.upenn.edu:/pub/treebank/doc/manual
  or try
    /mnt/unagi/nldb/manual/current/manual.glommed
  for a VERY long mishmash of (1) latex source of the manual and (2) minutes
  of Switchboard bracketing decisions.  Searchable, though.

For information on Treebank I (old) bracketing, see
  ftp.cis.upenn.edu:/pub/treebank/doc/old-bktguide.tex.

You can also try sending me mail, but it might take me a while to respond.

----------------------------------------------------------------
robertm@unagi.cis 3/1/96