Parse Viewer for Czech Dependency Structures

Introduction

Part of the fun of writing Parsers is looking at how they parse. The Prague Dependency Bank associates with each sentence a dependency structure. In this dependency structure, each word has associated with it a governing word or start symbol. These dependency parses are one way to show the syntactic structure of natural language.

As part of the CLSP Workshop 98, we have written a simple graphical viewer to display dependency structures graphically and to compare them. In the interest of portability, this program was written in Java 1.1.

Fonts and Java

If your machine has difficulty with the Czech characters (latin 2, iso 8859-2) you may wish to visit Sun's page outlining the Internationalization features of Java 1.1 for information pertaining to using these fonts. Java 1.1 enabled Browsers should open a copy of the Parse Viewer on loading this page. If your Browser does not support Java 1.1, the Parse Viewer may not loa d by itself. You may still be able to use the Parse Viewer by loading this page in your AppletViewer, part of the Java Runtime Environment, and the Java Development Kit.

Using the Parse Viewer

To run the Viewer on your machine, you will need to first download some parse files to look at. Save the following to a local file. Once you have the Parse Viewer running (see above, if the applet does not automatically start), open these files f rom within the Viewer.

      bcb01aba.cst
      bcb02aba.cst

When the files load, you should be presented with a list of sentence id's, view and close buttons, and radio buttons for compare, reference, g, and MDg. Each SGML data file will have for each word either [digit] or [digit] tags which specif y its governing word. The tags are found in truth data, are machine disambiguated - determined by a parser. Usually, the correct checkbox will be checked for you. If no file is found to have tags, you will have to select which file to com pare the others to, by clicking on the appropriate Reference radio button.

To view the parses from one individual file, click on the View button associated with that file. To compare multiple files, click on the compare button at the bottom of the parse viewer. Files to be compared should be specified by selecting the co mpare checkbox for each of the files to be compared. You can compare any number of files greater than 2, though it gets hard to see after about 4. Try clicking on the words to see what other associated tags are available.

Getting the Parse Viewer

To get the Parse Viewer, either download the compressed class files or try to run the applet online. Downloading the class files is more reliable and the program will run significantly faster locally than over the net, but you will need version 1.1 of the java runtime environment in order to run them. Once the class files are downloaded and gunzipped, either add that directory to your java CLASSPATH or run them from their own directory.

The command line (for unix) is

java PV1

      Click here for the parse viewer applet.
	(Applet may not work yet with all browsers)