case classLostTokensAnalyzer(goldParseBank: ParseBank) extends ParseAnalyzer with Product with Serializable
The LostTokensAnalyzer tallies lost tokens (i.e. tokens with a different breadcrumb path in the
gold parse) according to the breadcrumb arc label of their highest misattached ancestor in the
gold parse.
Example: In the gold parse, suppose the breadcrumb path of token "red" is
--ROOT--> ate --PREP--> with --POBJ--> chopsticks --AMOD--> red
but in the candidate parse, the breadcrumb path of token "chopsticks" is
--ROOT--> ate --DOBJ--> pasta --PREP--> with --POBJ--> meatballs --AMOD--> red
then the highest misattached ancestor of "red" in the gold parse is "with" (attached to "pasta"
instead of "ate"). The arc label of "with" is "PREP" in the gold parse. So the loss of token
"red" is attributed to a "PREP" attachment error.
The LostTokensAnalyzer tallies lost tokens (i.e. tokens with a different breadcrumb path in the gold parse) according to the breadcrumb arc label of their highest misattached ancestor in the gold parse.
Example: In the gold parse, suppose the breadcrumb path of token "red" is
--ROOT--> ate --PREP--> with --POBJ--> chopsticks --AMOD--> red
but in the candidate parse, the breadcrumb path of token "chopsticks" is
--ROOT--> ate --DOBJ--> pasta --PREP--> with --POBJ--> meatballs --AMOD--> red
then the highest misattached ancestor of "red" in the gold parse is "with" (attached to "pasta" instead of "ate"). The arc label of "with" is "PREP" in the gold parse. So the loss of token "red" is attributed to a "PREP" attachment error.
a bank containing the gold parses