DepSVG

Introduction

This page describes a dependency tree and predicate-argument structure visualizer which is implemented in Perl and saves into Scalable Vector Graphics (SVG). There are different ways to visualize dependency trees (e.g. Prague school, Melchuk, Connexor, Link Grammar, etc). Our inspiration has been the trees from the papers by Duchier and Debusmann, see e.g. XDG.

The words of the sentence are represented as nodes in a graph. Each node can have several properties which correspond to the properties of the word, such as the lemma form, the position in the sentence, the part-of-speech tag, etc. The dependencies are represented as arcs which connect the nodes. Each arc has a label for the dependency type.

DepSVG can also handle words with multiple heads. In the following picture, 'Fli-1' is both a subject and a control subject.

Download, install, usage

DepSVG is hosted by GitHub and can be downloaded from here. This page also contains a usage example.

The simplest way of obtaining the code is by cloning the repository:

git clone https://github.com/Kaljurand/DepSVG.git

Examples

DepSVG applied to the CoNLL data

data/: CoNLL06 training data for the free languages (about 100 first lines of each corpus)
img/: sample output as both SVG and PNG images

Some notes:

The morph-column is not included in the SVG (for layout purposes)
The png-formatted images were generated with Batik Rasterizer, by:
```
java -mx1024m -jar batik-rasterizer.jar -m image/png -d . *.svg
```
Note that Batik Rasterizer can also produce PDF-files (if you need a scalable image which is not SVG).

DepSVG in action

Tools/resources that use DepSVG: