Presentation and representation of parallel treebanks

Abstract

We have created a small German-Swedish-French parallel treebank. The German and Swedish sentences were Part-of-Speech tagged and parsed with phrase structure in a similar way and then aligned on the sentence and phrase level. The French sentences were annotated with an external system combining constituency and dependency analysis and then aligned to the Swedish sentences. The treebanks are represented with TIGER-XML and the alignment information is contained in a separate XML document. The parallel trees can be viewed with help of the Stockholm Alignment Viewer, based on SVG-files. The program allows for many possibilities in marking and visualizing special traits of the parallel treebank.