Roleset comment.01 Verbnet Class: NONE "intransitive":

Roles:

Arg0:commentor


Arg1:thing commented on

Examples:

(un-named example) (-)
Judge Wapner commented publicly on a pending case.
Arg0:	Judge Wapner
REL:	commented
Argm-MNR:	publically
Arg1-on:	a pending case

Roleset comment.02 Verbnet Class: NONE "basic transitive":

Roles:

Arg0:commenter


Arg1:utterance


Arg2:hearer


Arg3:attributive

Examples:

transitive (-)
A Lorillard spokeswoman commented "This is an old story."
Arg0:	A Lorillard spokeswoman
REL:	commented
Arg1:	"This is an old story."


*Trace* > "Kent cigarettes were sold"

fronted (-)
Kent cigarettes were sold, the company commented *Trace*
Arg0:	the company
REL:	commented
Arg1:	*Trace*


*T* "What matters is what advertisers will pay"

Inverted (theoretically possible, but odd) (-)
What matters is what advertisers will pay, commented *T* Newsweek's chairman
Arg1:	*T*
REL:	commented
Arg0:	Newsweek's chairman




split (theoretically possible, but odd) (-)
"What you have to understand," commented John, "is that Philly stinks."
Arg0:	John
Arg1:	"What you have to understand, is that Philly literally stinks."




with listener (-)
John commented to Mary: "you're an idiot."
Arg0:	John
REL:	commented
Arg2-to:	Mary
Arg1:	"you're an idiot."




weird, attributive usage: (theoretically possible, but odd) (-)
"Well that's odd," commented John of the disappearance of his nose.
Arg1:	"Well that's odd,"
REL:	commented
Arg0:	John
Arg3-of:	the disappearance of his nose.

In most cases Treebank suggests an S-BAR as the object of "announce"; PReditor selects the S inside the S-BAR as the Arg1. I feel it really should be the S-BAR, because of the cases like "John said THAT he loved Mary" where the S-BAR is selected as Arg1, but in actual usage the complementizer is usually dropped. But since it would slow down throughput *enormously* to retag all of these cases, we'll accept what PReditor selects in both cases (with and without complementizer).