Skip to content

Latest commit

 

History

History
362 lines (212 loc) · 11.4 KB

RULES.md

File metadata and controls

362 lines (212 loc) · 11.4 KB
Raw

PredPatt's Rules

Predicate root identification

At this stage, PredPatt identifies predicate root tokens in the input sentence. The following kinds of tokens are identified as predicate roots:

  • Governor of subjects (nsubj, csubj, nsubjpass, csubjpass), objects (dobj, iobj), or adverbial clausal modifiers (advcl).

  • Clausal modifiers: dependents of acl, or advcl.

  • (Open) clausal complements: dependents of ccomp, or xcomp.

  • Conjuncts of other predicate root token (conj).

Argument root identification

Given a predicate root token, PredPatt identifies the following tokens as its argument root tokens.

  • Nominal subjects (nsubj, nsubjpass), or objects (dobj, iobj) of the predpatt root token.

  • Nominal modifier (nmod) of the predicate root token.

  • Nominal modifier (nmod) of an adverb (advmod) which modifies the predicate root token.

Argument resolution

At this stage, PredPatt extracts additional argument root tokens for the predicate in special cases:

  • An open clausal complement (the dependent of xcomp) is viewed as part of the predicate it complements.[1] Then all argument root tokens of the open clausal complement are extracted as the argument root of the predicate it complements.

    [1]: Open clausal complements (xcomp) is a predicative or clausal complement without its own subject, which means it can’t be an independent predicate.

  • In relative clauses, PredPatt borrows an argument for adjectival clausal modifiers (the dependent of acl) from the token they modify (the governor of acl).

  • PredPatt extracts a conjunct token of other argument root token as an argument root.

Predicate phrase extraction

At this stage, PredPatt recursively extracts tokens from the subtree of the predicate root token, then adds them to the predicate phrase, or stops extraction if one of the following situations is satisfied:

  • The current token is an argument root token of the predicate.

  • The current token is another predicate root token.

  • The current token is a dependent of an dep relation.

Argument phrases extraction

At this stage, PredPatt recursively extracts tokens from the subtree of the argument root token, then adds them to the argument phrase, or stops extraction if one of the following situations is satisfied:

  • The current token is an argument root token of the predicate.

  • The current token is another predicate root token.

  • The current token is a dependent of the dep relation.

  • The current token is a dependent of coordinating conjunction (cc).

  • The current token is a dependent of the case relation (this token will be added to the predicate phrase).

Option: simplification

  • Remove a non-core argument, e.g., a nominal modifiers

  • Remove an argument of other type from the predpatt.

  • Remove an adverbial modifiers

  • Remove auxiliary in the predicate phrase.

Special predicate extraction

Besides normal predicate extraction, PredPatt extracts modifiers, such as the dependent of amod, appos, or nmod:poss, as a predicate and treats the noun it modifies as an argument of the predicate.