20080519
Added the keys for test.wsj and test.gold as *.GOLD files.

20080507
Added the two test corpora (see README for description).

20080421
Fixed several bugs in the data:
- Fixed a bug in the conversion of NomBank arguments from the constituent to 
the dependency based representation. Due to this bug, the same head was 
assigned to different NomBank arguments. For example, for the phrase "goods 
retailer" and the predicate "retailer" the correct assignment is A0 to 
"retailer" and A1 to "goods". In the incorrect output both arguments were 
assigned the same head, "retailer". 
- Improved the integration of BBN named-entity annotations in the syntactic 
dependency representation. We now annotate TITLE dependencies (e.g., "Mr."), 
post-honorifics (POSTHON) (e.g., "Corp."), and NAMEs much more consistently.
- Possessives (POS) such as "'s" are now consistently treated as SUFFIX.
- Fixed a bug which allowed C-A* arguments without the preceding A* argument.
- We now use gold lemmas in the PRED column. Previously we used lemmas 
extracted from WordNet using the most common sense and the predicted POS tag. 
Using gold lemmas increases significantly the match with the PropBank and 
NomBank frame files. Observation: to alleviate the difference between gold and 
predicted lemmas, we score only senses not lemmas when scoring predicate sense 
classification. So a system gets full credit for a predicate sense if the sense
produced is correct, even if the system generates incorrect lemmas in the PRED 
column.  

20080327
Fixed several bugs in the conversion from NomBank to our format. 
The main bug was caused by the fact that some titles, e.g., "Dr." in 
"Dr. Smith" where not recognized as such. The syntactic dependency 
from the actual name to the title was incorrectly marked as an 
apposition, and any semantic arguments were also incorrectly headed 
by the title. Fixed: the correct syntactic dependency (TITLE) is
marked between title and name; semantic arguments are headed by the 
actual name rather than the title.
Other minor conversion bugs fixed (the "luxury auto maker" bug).

20080317
Fixed head assignment for NomBank arguments assigned to coordinated structures.
Originally, NomBank assigns the conjunction as head. But, according to our
dependency representation, the head is the first element in the coordination.
Heads are now assigned to be consistent with the syntactic dependencies.

20080316
Removed predicates from the Penn Discourse TreeBank (PDTB) that were 
previously included here erroneously. Fixed several other minor NomBank bugs 
like incorrect lemmas etc.

20080312
Normalized the Penn function tags for syntactic dependencies. Previously, 
the function tags were sometimes unnormalized: for instance, the LOC-PRD 
tag sometimes (but rarely) appeared as PRD-LOC. Now the function tags are
displayed in alphabetical order.

20080311
Several bug fixes:
1. Replaced all white spaces with underscores for lemmas fetched from WN 
   (the "comic strip" bug).
2. Merging of PropBank and NomBank frames fails for approximately 1.7% of 
   the sentences due to different tokenizations in the two corpora. In this
   release all these sentences are discarded.

20080310
Released the training and development corpus.

20080228
Initial release. Includes only the trial corpus.