jeisner / treebank-scripts

Suite of scripts for preprocessing the Penn Treebank, primarily to extract lexical subcategorization frames and dependencies.
MIT License
7 stars 1 forks source link

arg-marking -- NOT a bug #2

Open jeisner opened 8 years ago

jeisner commented 8 years ago

[item from the old TO-DO file dated 2002-04-07]

Contrast these two lines in train0-15.simple.len5. They suggest that we are missing arg-marking on the RHS if the LHS is arg-marked! No: it is just that the first line was originally NP-CLR and PP-CLR, so marking was correctly suppressed according to Collins 1997. So this is just fine. (There are in fact plenty of lines where both LHS and RHS have arg marks.)

   03/wsj_0349.mrg:765: take    S~  TO @ NP PP
   13/wsj_1377.mrg:512: take    S   TO @ NP~ PP