CCGbank -- the machine-readable predicate-argument structure files (.parg)
	-------------------------------------------------------------------------
               Julia Hockenmaier and Mark Steedman 


The predicate-argument structure files give for each sentence a list
of the word-word dependencies in the predicate-argument structure,
including locally mediated and long-range dependencies, which are
indicated as such.  For each file in the original Treebank, there is
one corresponding predicate-argument structure file. 
Each sentence is enclosed by an <s>...</s> tag. The opening <s> tag
contains an attribute that indicates the sentence ID and is followed
by the index of the last token in the sentence.  
Each dependency appears on one line.

<s id="wsj_0001.2"> 12
1      0       N/N             1       Vinken Mr.
1      2       (S[dcl]\NP)/NP  1       Vinken is
3      2       (S[dcl]\NP)/NP  2       chairman is
3      4       (NP\NP)/NP      1       chairman of
6      4       (NP\NP)/NP      2       N.V. of
6      5       N/N             1       N.V. Elsevier
11     4       (NP\NP)/NP      2       group of
11     8       NP[nb]/N        1       group the
11     9       N/N             1       group Dutch
11     10      N/N             1       group publishing
<\s>

A dependency between the i-th and j-th word (word_i and word_j) where
the j-th word has the lexical (functor) category cat_j and the i-th
word is head of the constituent which fills the k-th argument slot of
cat_j is described as follows:

i   j   cat_j    arg_k   word_i  word_j 

Words in each sentence are numbered from 0 to n. 

In the sentence ``Mr. Vinken is chairman of Elsevier'', 'Vinken' is
the second word in the sentence and head of the constituent which
fills the first (and only) argument slot of the N/N 'Mr.'. At the same
time, the N 'Vinken' is head of the constituent which fills the first
argument slot of the (S[dcl]NP)/NP 'is', which is the third word in
the sentence. Therefore:

1      0       N/N             1       Vinken Mr.
1      2       (S[dcl]\NP)/NP  1       Vinken is


In version 1.1 of CCGbank, Penn Treebank sentences that do not receive
a CCG derivation do not appear in the PARG files.