CCGbank and TGrep2 
	------------------


CCGbank is searchable with TGrep2, an expression matcher for trees developed by Douglas Rohde. 


Obtaining TGrep2 
---------------- 
TGrep2 is available free of charge from
	http://tedlab.mit.edu/~dr/TGrep2/ 
It is governed by the GNU General Public License, version 2.
 
Searching CCGbank with TGrep2 
-----------------------------

The file ccgbank00-24.t2c contains all sections of CCGbank in TGrep2 format. 
Assuming this file is in directory $CCG_TGREP2 on your system, 
set the environment variable TGREP2_CORPUS to $CCG_TGREP2/ccgbank00-24.t2c 
or invoke  TGrep2 with the option "-c $CCG_TGREP2/ccgbank00-24.t2c", 

If TGrep2 version 1.15 or higher is used on ccgbank00-24.t2, it will run in CCG mode, 
which differs from its standard mode in the following ways: 

 - In TGrep2's CCG mode, brackets ("[" and "]"), parentheses ("(" and ")") and slashes ("\" and "/") 
   can be part of a node label, but have to be preceded by a backslash in regular 
   expression searches

 - In TGrep2's CCG mode, curly brackets ("{" and "}") are used instead of parentheses ("(" and ")") 
   to specify dominance relations and to bracket the output trees. 
 
 - In TGrep2's CCG mode, the plus and minus signs ("+" and "-") are used instead of brackets ("[" and "]")
   to group disjunctive terms (eg. "NP [ > PP  | > S ]" in standard TGrep2 becomes "NP + > PP | > S-"
   in CCG mode. 	 



A few examples: 
------------------

% tgrep2  "/.*/</^[Bb]uy/"
finds all occurrences of words starting with 'buy' or 'Buy'
and their lexical categories. 


% tgrep2  "/.*/</^[Bb]uy/"
finds all occurrences of words starting with 'buy' or 'Buy' and their lexical categories. 

% tgrep2 -C  "((S[dcl]\NP)/NP)/(S[dcl]\NP)"
finds all occurrences of the special subject-extracting category ((S[dcl]\NP)/NP)/(S[dcl]\NP), 
and prints out their file name and sentence number. 


% tgrep2 -C  "/^\(S\[.*\]\\NP\)\/NP/</^tak/"
finds all occurrences of transitive verbs (with any morphosyntactic feature) that have the prefix "tak". 
In regular expressions over category labels, all slashes, parentheses and brackets have to be escaped 
(eg. '\[', '\\'). 



% tgrep2 -wl "(NP\NP)/NP|((S\NP)\(S\NP))/NP|PP/NP<with"
finds all occurences of 'with' as head of an NP-modifying ("(NP\NP)/NP"), 
VP-modifying ( "((S\NP)\(S\NP))/NP") or argument ("PP/NP") PP, 
and pretty-prints ('-l') the entire sentence ('-w')


% tgrep2 "{S[dcl] < {NP !<< NP[nb]/N } }"
finds all occurrences of declarative sentences that have a (subject) NP which does not contain a determiner  with category "NP[nb]/N". 

% tgrep2 -l "S[dcl]<NP"
finds any sentence that contains an NP and pretty-print the results

% tgrep2 "/^S\[.*\]\\NP/< {/.*/< {/\(S\[.*\]\\NP\)\/NP\[conj\]/ } }"
finds any VP that contains transitive right node raising (coordination of "S[.*]\NP)/NP"):



% tgrep2 "NP > /^S\[.*\]/"
finds any NP that is a subject. 


% tgrep2 "NP>/^S\[.*\]\\NP/"
finds any NP that is a direct object:

% tgrep2 "NP + > PP | > S[dcl]-"
finds any NP that is dominated either by a PP or a tensed sentence. 
(In CCG mode, + and - are used instead of [ and ] for node disjunction)



Acknowledgements
----------------
We are very grateful to Doug Rohde for adapting TGrep2 to CCGbank.