Filtering Left Dislocation Chains in Parsing Categorial Grammar 1 Parsing Left Dislocation

This paper reports on a way to reduce the complexity of the process of left dislocation (re)construction for categorial grammar in the case of lexically assigned gaps, as an additional restriction on the complexity arising from lexical polymorphism in general. Specifying extraction sites lexically has the advantage that the combinatory explosion can be contained in the preparsing track by a specialized constraint on the expansion of sequences of categories. This constraint is called the Left Dislocation Chain Filter and is implemented by a Finite State Transducer. It is shown that the Filter can reduce the number of full string assignments under consideration prior to parsing with an average of one half to one order of magnitude, depending on the nature of the sentence. Left dislocation is a very common, almost universal phenomenon in natural languages. It establishes the relation between an element at the left periphery of a clause and a particular, lexically open position in its right context. The leftward nature of dislocation is explained in Kayne (1994). The most prominent of these relations are invoked by so called wh-elements at the leftmost edge of questions and relative clauses. If a language has left dislocation, however, many other constituents can occur in a left dislocated position. Here are some examples from Dutch; the distinguished position in the right context is marked by t. (1) Mimi vroeg zich af wie] Jan dacht dat t zou gaan winnen Mimi wondered (herself) who Jan thought that would go win`Mimi wondered who Jan thought would win' (2) De lm die] ik heb t gezien] kan jij niet t gezien hebben The movie that I have seen can you not seen havèYou cannot have seen the movie I have seen'