class
OracleParser[L, L2, W] extends SafeLogging
Value Members
-
final
def
!=(arg0: AnyRef): Boolean
-
final
def
!=(arg0: Any): Boolean
-
final
def
##(): Int
-
final
def
==(arg0: AnyRef): Boolean
-
final
def
==(arg0: Any): Boolean
-
final
def
asInstanceOf[T0]: T0
-
def
clone(): AnyRef
-
final
def
eq(arg0: AnyRef): Boolean
-
def
equals(arg0: Any): Boolean
-
def
finalize(): Unit
-
-
final
def
getClass(): Class[_]
-
-
def
hashCode(): Int
-
final
def
isInstanceOf[T0]: Boolean
-
def
logger: Logger
-
-
final
def
ne(arg0: AnyRef): Boolean
-
final
def
notify(): Unit
-
final
def
notifyAll(): Unit
-
def
oracleMarginalFactory(trees: IndexedSeq[TreeInstance[L2, W]]): Factory[L, W]
-
def
oracleParser(constraintGrammar: Factory[L, W], trees: IndexedSeq[TreeInstance[L2, W]])(implicit deb: Debinarizer[L]): Parser[L, W]
-
final
def
synchronized[T0](arg0: ⇒ T0): T0
-
def
toString(): String
-
final
def
wait(): Unit
-
final
def
wait(arg0: Long, arg1: Int): Unit
-
final
def
wait(arg0: Long): Unit
Inherited from AnyRef
Inherited from Any
Finds the best tree (relative to the gold tree) s.t. it's reacheable given the current anchoring. Best is measured as number of correct labeled spans, as usual. If the given treebank symbol is correct, bonus points can be awarded for getting the right refinement.
On the training set, the "best" reachable tree will not always (~5% of the time) be the correct tree, because pruning will remove the right answer. We don't want to try to train towards an unreachable tree, because the training algorithm will do bad things. Instead, we want the best possible tree that our parser could conceivably produce. That is why this class exists.
If backupGrammar is provided, it will be used to find such a tree in the case that no tree can be found with grammar (given the current constraints).
Typically, the first grammar will be a treebank grammar that has no horizontal markovization (i.e. it is not forgetfully binarized) and it also remembers the functional tags like -TMP. The backup grammar is usually the grammar with which the pruning masks were produced; because of the way we prune, that parser will always be able to find a tree (assuming that it was able to find a tree without pruning.)
TODO: should be a cascade of grammars