This demo shows which words are semantically related. Type a word in the empty box and click 'Search' (names are also allowed). The result will be a network of semantically related words. Double click on any of the words to get their nearest neighbours.

At the bottom of the page you can zoom in and rotate.

This demo resulted from research within the framework of the NWO project: Question Answering for Dutch using Dependency Relations. Large quantities of text were searched to automatically find semantically related words using three methods. This resulted in three demo's: a syntax-based demo, an alignment-based demo, and a proximity-based demo.

The alignment-based method uses multilingual parallel texts (the same text translated in multiple languages) to find semantically related words. The idea behind this method is that words that are translated with the same words in multiple languages might be semantically related or even synonymous. Whereas we used newspaper text in the proximity-based and syntax-based method, the corpus we used for the alignment-based method is the Europarl corpus. It consists of the proceedings from the European Parliament. The difference in domain accounts for some differences in output. For example, the polysemous Dutch word appel 'apple' 'appeal' gets neighbours from the legal domain, whereas in the other two demos it gets nearest neighbours that are all types of fruit.

More background info can be found in the thesis of Lonneke van der Plas, in the on-line presentation (on ISLA TV), or in the web article of KennisLink (in Dutch).

The interface was made by Ismail Fahmi within the framework of the SWHi project (Semantic Web for History).