Syntax: Structural Descriptions of Sentences

Penn Treebank (PTB)

• Syntactically annotated corpus (phrase structure)
• Contains 1 miilion words of Wall Street Journal sentences marked up with syntactic structure.
• Can be converted into a dependency Treebank.
• need for head percolation tables
• Completely flat structure in NP
• brown bag lunch, pink-and-yellow child seat
• Represents a particular linguistic theory
• PropBank
• PTB with some grammatical relations made explicit

Unification

• Mechanism needed to pass and check constraints.
• Constraints, syntactic and semantic:
• Subject-verb agreement
• S  NP VP
• the boy reads / the boys read / * the boys reads
• Subject/Auxiliary inversion: (Yes-no-question)
• S  AuxVerb NP VP
• Do you have flights / * does you have flights
• Selectional restrictions:
• An apple reads a book
• Need a mechanism to encode these constraints
• Refine the non-terminal set to encode these constraints.
• S  3sgAux 3sgNP VP ; 3sgAux  does | has …
• S  Non3sgAux Non3sgNP VP; Non3sgAux  do | have | can
• We need to split the NP rule into the 3sgNP and Non3sgNP.
• Size of the grammar grows;
• can we factor these constraints out of the structure of the rules?

Unification – contd.

• Attribute value matrix:

• boy :

• Number

• Person

• sg

• 3

• Cat

• N

• read :

• Number

• Person

• pl

• 3

• Cat

• V

• Subj

• agr

• NP.number = VP.subj.agr.number
• NP.person = VP.subj.agr.person

• S  NP VP

• reads:

• Number

• sg

• Cat

• V

• Subj

• agr

• VP  V

• VP.number = V.subj.agr.number
• VP.person = V.subj.agr.person

• Percolate Constraints

• Check Constraints

• The boy reads / * the boys reads / the boys read

• boys :

• Number

• Person

• pl

• 3

• Cat

• N

• Number

• sg

• Person

• 1|2