Cross-checking WordNet and SUMO Using Meronymy

Cross-checking WordNet and SUMO Using Meronymy Javier ´Alvez, Itziar Gonzalez-Dios, German Rigau LoRea Group, IXA Group, IXA Group University of the Basque Country (UPV/EHU) javier.alvez@ehu.eus, itziar.gonzalezd@ehu.eus, german.rigau@ehu.eus Abstract We report on the practical application of a black-box testing methodology for the validation of the knowledge encoded in WordNet, SUMO and their mapping by using automated theorem provers. Our proposal is based on the part-whole information provided by WordNet, out of which we automatically create a large set of tests. Our experimental results confirm that the proposed system en- ables the validation of some pieces of information and also the detection of missing information or inconsistencies among these resources. Keywords: Meronymy, Knowledge validation, Automated Theorem Proving 1. Introduction Despite being created manually, knowledge resources such as WordNet (Fellbaum, 1998) and SUMO (Niles and Pease, 2001) are not free of errors and inconsistencies. Un- fortunately, improving, revising and correcting such large knowledge bases is a never-ending task that has been mainly carried out manually. A few automatic approaches have also been applied focusing on checking certain struc- tural properties on WordNet e.g. Daud´e et al. (2003) and Richens (2008) or using automated theorem provers on SUMO e.g. Horrocks and Voronkov (2006) and ´Alvez et al. (2012). Just a few more have studied automatic ways to validate the knowledge content encoded in these resources by cross-checking them. For instance, ´Alvez et al. (2008) exploit the EuroWordNet Top Ontology (Rodr´ıguez et al., 1998) and its mapping to WordNet for detecting many on- tological conflicts and inconsistencies in the WordNet nom- inal hierarchy. In ´Alvez et al. (2017), we proposed a method for the auto- matic creation of competency questions (CQs) (Gr¨uninger and Fox, 1995), which enabled to evaluate the competency of SUMO-based ontologies. Our proposal was based on several predefined question patterns (QPs) that were in- stantiated using information from WordNet and its map- ping into SUMO (Niles and Pease, 2003). In addition, we described an application of automated theorem provers (ATPs) for the automatic evaluation of the proposed CQs. This proposal was used in ´Alvez and Rigau (2018) for a preliminary validation of WordNet, SUMO and their map- ping. The main contribution of this paper is a proposal for the au- tomatic validation of WordNet, SUMO and their mapping. For this purpose, we create a new set of CQs that is ob- tained on the basis of the part-whole data of WordNet and which is an improved version of the set proposed in ´Alvez and Rigau (2018). By means of our proposal, we demon- strate the practical capabilities of the method introduced in ´Alvez et al. (2017) for the automatic detection of semantic agreements and inconsistencies in large-scale knowledge resources. For example, the knowledge encoded in the WordNet relations part(elementary particle1 n,atom1 n), member(national1 n,country3 n) and substance(cartilage1 n,cartilaginous structure3 n) can also be inferred from SUMO. On the contrary, according to our interpretation of the meronymy relations of WordNet, the knowledge in the relations part(cell2 n,cell nucleus1 n) and substance(grape1 n,wine1 n) …

Content truncated to 500 words