Skip to content

Latest commit

 

History

History
63 lines (36 loc) · 2.03 KB

README.md

File metadata and controls

63 lines (36 loc) · 2.03 KB

TuLiPA-frames

This is an extension for the existing TuLiPA (Tuebingen Linguistic Parsing Architecture), a parser for Lexicalised Tree Adjoining Grammars. It is extended with semantic frames paired with the elementary trees.

Documentation on the parser and the resources is provided in the Wiki.

Cite

If you use TuLiPA-frames, please cite this paper for attribution:

Arps, D. & Petitjean, S. (2018), A Parser for LTAG and Frame Semantics. In: N. Calzolari, K. Choukri, C. Cieri, T. Declerck, S. Goggi, K. Hasida, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, S. Piperidis & T. Tokunaga, eds, ‘Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)’, European Language Resources Association (ELRA), Paris, France.

Generalized wrapping

Predict-wrapping stays the same

Old Complete-wrapping:

Antecedents:

  • $[γ,eps_T,i,j,Γ1◦<<f1,f2,Y>>◦Γ2,ws?]$
  • $[α,(p·m)⊥,f1,f2,Γ3,yes]$

Side conditions:

  • label(α,p·m) = Y
  • label(γ,eps) = label(α,p)

Consequent:

  • $[α,(p·m)⊥,i,j,Γ1◦Γ3◦Γ2,no]$

Generalized CW:

Antecedents:

  • $[γ,q_T,i,j,Γ1◦<<f1,f2,Y>>◦Γ2,ws?]$
  • $[α,m_⊥,f1,f2,Γ3,yes]$

Side conditions:

  • $m$ is an integer (i.e. $α(m)$ is the daughter of the root of $α$)
  • label(γ,q) = label(α,m)
  • label(α,m) = Y

Consequents:

  • $[α,m_⊥,i,j,Γ1◦Γ3◦Γ2,no,[γ,q_T,i,j,Γ1◦<<f1,f2,Y>>◦Γ2,ws?]]$ Put differently, the consequent item has a direct pointer back to the antecedent target item of the wrapping so that we can jump back to parse the upper part of the wrapping target tree later

Finish Generalized CW

A new rule is necessary: After going to the root of the wrapping tree, jump back to the target tree of the wrapping and continue going upwards

Antecedents:

  • $[α, eps, TOP, i, j, Γ1, false, [γ, q_T,k,l,Γ2,ws?]]$

Consequent:

  • $[γ,q_T,i,j,Γ1,false]$