Conversion from DS to PS

Information in PS and DS PS (e.g., PTB) DS (some target DS) POS tagyes Function tag (e.g., -SBJ) yes Empty category and co-indexation yesmaybe Syntactic tagyesno Allowing crossingnoyes

Previous work (Collins, Hajič, Ramshaw and Tillmann, 1999) (Xia and Palmer, 2001) Both are based on heuristics. Need to handle non-projectivity and ambiguity.

Our approach DS  DS + : e.g., removing non-projectivity by introducing trace and co-indexation. DS +  PS + : We prefer to keep this step simple and general. PS +  PS: e.g., choose one or more phrase structures stored in PS +.

DS  DS + Whom do you think he will invite?

DS +  PS + Mary will come tomorrow

Resulting phrase structures

PS +  PS

Ambiguity in DS Ex1: young men and women Ex2: adj N 1, N 2, N 3, and N 4 ???

Remaining questions DS  DS + : Can all kinds of non-projectivity be resolved by inserting empty categories? DS +  PS + : –Are fine-grained dependency links sufficient for disambiguating all possible readings in DS? –What is the unit of conversion? A dependent link or a larger tree fragment?

Backup slides

Common approach for PS  DS For each internal node in the PS (1) Find the head child (2) Make the non-head child depend on head-child For (1), very often people use a head percolation table and functional tags.

Our approach PS  PS + : e.g., inserting internal nodes in order to make certain DS possible from the PS. PS +  DS + : same as before DS +  DS: e.g., handling crossing with the help of co-indexation.

Previous work on DS  PS

Summary Separating the conversion into three steps: –Keeping the 2 nd step simple and general –The 1 st and 3 rd steps could depend on analyses that one chooses. Rich annotation will help conversion: –Ex: function tag, empty category and co-indexation Some issues need more study.