Layering Semantics (Putting meaning into trees) Treebank Workshop Martha Palmer April 26, 2007
Treebank Workshop/NAACL07 Outline Semantic role labeling, verbs and nouns Treebank/PropBank discrepancies Treebank/NomBank discrepancies Dependencies vs. Phrase Structure
Treebank Workshop/NAACL07 A TreeBanked phrase a GM-Jaguar pact NP S VP would VP give the US car maker NP an eventual 30% stake NP the British company NP PP-LOC in A GM-Jaguar pact would give the U.S. car maker an eventual 30% stake in the British company. Marcus, et al, 1993
Treebank Workshop/NAACL07 The same phrase,PropBanked* a GM-Jaguar pact would give the US car maker an eventual 30% stake in the British company Arg0 Arg2 Arg1 give(GM-J pact, US car maker, 30% stake) A GM-Jaguar pact would give the U.S. car maker an eventual 30% stake in the British company.
Treebank Workshop/NAACL07 Frames File example: give Roles: Arg0: giver Arg1: thing given Arg2: entity given to Example: double object The executives gave the chefs a standing ovation. Arg0: The executives REL: gave Arg2: the chefs Arg1: a standing ovation
Treebank Workshop/NAACL07 NomBank Frame File example: gift (nominalizations, noun predicates, partitives, etc. Roles: Arg0: giver Arg1: thing given Arg2: entity given to Example: nominalization Nancy’s gift from her cousin was a complete surprise. Arg0: her cousin REL: gave Arg2: Nancy Arg1: gift
Treebank Workshop/NAACL07 Status of TB/PB merge Extracted and classified mismatch examples Reconciled different policy decisions An example of TB changes: Modified list of verbs which take small clauses and sentential complements (e.g. keep their markets active) An example of PB changes: A different approach to annotation of empty categories: John seems [*T* to know the answer]-Arg1 Some mismatches require manual adjudication (e.g. PP-attachment ambiguity): The region lacks necessary mechanisms for handling the aid and accounting items. Unresolved issues: gapping
Treebank Workshop/NAACL07 Treebank-PropBank Reconciliation Problem: One PropBank arg can involve many parse nodes Solution: Single argument – single parse node analysis Problem: Different semantic analyses Solution: Reconcile semantic assumptions
Treebank Workshop/NAACL07 NomBank Issues Treebank has no structure for noun phrases Many nominal event arguments do not correspond to tree nodes
Treebank Workshop/NAACL07 Dependency Trees vs Phrase Structure CATEGORYDTPST Finding constituent boundaries√ Noun phrase structure√ Empty categories√ Coordination – what is the head?√ Phrasal verbs√ Discontinuous constituents√
Treebank Workshop/NAACL07 Dependency Trees vs Phrase Structure Really depends on choices with respect to guidelines and the annotation tool display Penn Treebank chose not to bracket NP’s Dependency trees can be displayed visually to clarify constituent structure Empty categories can be a help or a hindrance depending on the annotator’s task Coordination heads are defined in the guidelines Penn Treebank could define phrasal verbs