First steps in checking and comparing Princeton WordNet and Estonian Wordnet

Proceedings of the EACL 2012 Joint Workshop of LINGVIS & UNCLH, pages 25–29, Avignon, France, April 23 - 24 2012. c⃝2012 Association for Computational Linguistics First steps in checking and comparing Princeton WordNet and Esto- nian Wordnet Ahti Lohk Tallinn University of Tech- nology Raja 15-117 Tallinn, ESTONIA [ahti.lohk@ttu.ee] Kadri Vare University of Tartu Liivi 2-308 Tartu, ESTONIA [kadri.vare@ut.ee] Leo Võhandu Tallinn University of Tech- nology Raja 15-117 Tallinn, ESTONIA [leov@staff.ttu.ee] Abstract Each expanding and developing system requires some feedback to evaluate the normal trends of the system and also the unsystematic steps. In this paper two lexical- semantic databases – Princeton WordNet (PrWN) and Estonian Wordnet (EstWN)- are being examined from the visualization point of view. The visualization method is described and the aim is to find and to point to possible problems of synsets and their semantic relations. 1 Introduction Wordnets for different languages have been created for a quite a long time 1 ; also these wordnets have been developed further and updated with new information. Typically there is a special software for editing wordnets, for example VisDic2, WordnetLoom (Piasecki et al 2010), Polaris (Louw, 1998). These editing tools often present only one kind of view of the data which might not be enough for feedback or for detecting problematic synsets/semantic relations. The visualization method described here can be used separately from the editing tool; therefore it provides an additional view to data present in wordnet. For initial data PrWN version 3.03 and EstWN version 63 4 have been taken. PRWN contains of 117 374 synsets and EstWn of 51 688 synsets. The creation of EstWN started in 1998 within the EuroWordNet project5. At present the 1http://www.globalwordnet.org/ 2http://deb.fi.muni.cz/clients-debvisdic.php 3http://wordnet.princeton.edu/ 4http://www.cl.ut.ee/ressursid/teksaurus/ 5http://www.illc.uva.nl/EuroWordNet/ main goal is to increase EstWN with new con- cepts and enrich EstWN with different kinds of semantic relations. But at the same time it is necessary to check and correct the concepts al- ready present (Kerner, 2010). The main idea and basic design of all wordnets in the project came from Princeton WordNet (more in Miller et al 1990). Each wordnet is structured along the same lines: syno- nyms (sharing the same meaning) are grouped into synonym sets (synsets). Synsets are connect- ed to each other by semantic relations, like hyperonymy (is-a) and meronymy (is-part-of). As objects of analysis only noun synsets and hyperonymy-hyponymy relations are considered (of course, it is possible to extend the analysis over different word classes and different seman- tic relations). So, due to these constraints we have taken 82 115 synsets from PRWN (149 309 different words in synsets) and 41 938 synsets from EstWN (64 747 different words in synsets). 2 Method We will explain our method's main idea with a small artificial example. Let us have a small sep- arated subset presented as a matrix: Figure 1. Relation-matrix and bipartite graph In the rows of that table we have synsets and in columns hyperonyms. On the right side of 25 that figure we have presented the same data …

Content truncated to 500 words