Coordination as a Parsing Problem

Declarative grammar is essentially unable to specify the variety of coordinative structures in natural languages adequately. Parsers nevertheless can be upgraded to handle coordination more subtly than by brute force. This upgrading involves knowledge of the grammar to be parsed and differentiation of parsing stages. 1. COORDINATION AS A GRAMMAR PROBLEM Natural languages have structures of the form (1) [sentence X Y1 coord Y2 Z ] where coord is an element equivalent to and or or. The substring Y1 coord Y2 is a coordination; the element coord is said to be the coordinator, and Y1 and Y2 are dubbed the left and the right coordinate, respectively. The substrings X and Z may be seen as the left and the right context, respectively, of the coordination; either of them may be empty. Coordinations in languages like Dutch and English have several unique and rather nasty properties. The most prominent among them is that the coordinates Y1 and Y2 are grammatically equivalent with respect to the context X _ Z, and that there are hardly any other wellformedness conditions on coordination in structures like (1). From this some other characteristics emanate. For example, there is no canonical class of coordinates: any locally wellformed string structure can be a coordinate. Furthermore, coordinations can occur almost anywhere in a structure, if only the equivalence condition is respected. As a consequence, coordinations may themselves be a (part of a) coordinate or occur in the context of another coordination. Moreover, the connection between Y1 and Y2 is so tight that the right coordinate can be syntactically incomplete, borrowing its combinatory power from the left coordinate. Integrating these properties, the following sentence is grammatical (coordinators and boundaries of coordinations are marked): (2) [Enkele parvenu’s met en enkele yuppen zonder] [strafblad of schulden] {[leasden en verkochten] aan ervaren of [huurden en verhuurden] voor beginnende} spekulanten die {[niet hier maar wel in A’dam] wouden beleggen of [juist wel hier maar niet in A’dam]}, {[talrijke nieuwe en enkele oudere] [woonblokken en (bedrijfsof winkelpanden)]} ’Some parvenus with and some yuppies without police record or debts leased and sold to experienced or hired and let for starting scalpers that not here but in Amsterdam wanted to invest or not here but in Amsterdam numerous new and some older housing blocks and business or commercial properties’ Coordinators, as it seems, can occur almost everywhere the main restriction here is the position adjacent to another coordinator and affect everything that is open to semantic variation. So, if we take a description like (3) instead of (1), neither X nor Z conveys specific information on suffixes or prefixes, respectively, which can be involved as a coordinate in the coordination.