Parsing the NEGRA corpus

1
Parsing the NEGRA corpus

• Greg Donaker
• June 14, 2006

2
NEGRA Corpus

• German language tagged corpus
• 20,602 sentences (355,096 tokens)
• Significantly smaller than Penn Treebank
• Can be used similarly to Penn Treebank
• Similar annotations, much flatter trees Dubey
  Keller 2003

3
Baseline error analysis

• Ran through Stanford Parser using NEGRA specific
  parameters
• 91.75 tagging accuracy
• PCFG f-score 66.42
• Most frequently underproposed rule
• NP -gt ART NN (98 times)
• Most frequently underproposed category
• NN (498 times three times the next category)
• These errors seem abnormally high based on the
  structure of German language.

4
Approach

• Bug modeled tag distribution of unknown words as
  baseline distribution
• Reworked unknown word model to specifics of
  German language
• Model based on first letter, capitalization of
  first letter, ending substring of words

5
Results

• Best performing (on both test and validation
  sets) model matched intuition
• Capitalization of first letter, last two
  characters of word
• Improves Tagging accuracy from 91.75 to 94.49
• Improves PCFG F-score from 66.42 to 69.87
• Reduces underproposed NP-gtART NN from 98 to 48
• Reduces underproposed NN from 498 to 73