Stanford ’ s System for Parsing the English Web

We describe the Stanford entries to the SANCL 2012 shared task on parsing noncanonical language. Stanford submitted three entries: (i) a self-trained generative constituency parser, (ii) a graph-based dependency parser, and (iii) a stacked dependency parser using the output from the constituency parser as features while parsing. The stacked parser obtained 2nd place in the dependency parsing track. Our overall approach involved exploring techniques which improved performance consistently across domains without using many external resources.