109-130
Case Study: Automatic Identification of Romanian Suffixes
Authors: Verginica Barbu Mititelu
Number of views: 401
Assuming the perspective of automatically identifying derived words and their bases in the Romanian wordnet, with the aim of enriching it with derivational relations and semantic labels associated to them, in this article we present the results of a case study whose aim was to automatically identify suffixes by means of which new words are created in Romanian. In the beginning of the paper we make a brief overview of derivation in Romanian, also anticipating the challenges of our study. For the automatic identification of the suffixes in words we used generalized suffix trees for representing the lemmas in an electronic Romanian lexicon. We imposed a series of filters, identified via observation, by means of which we improved the results. For the evaluation of the identified suffixes we used gold standard lists of suffixes grouped according to the part of speech of the derived words. The evaluation of results was performed in three stages, by comparing them with the gold standard list of suffixes, with the gold standard list of suffixes and suffixoides, and with the unified gold standard list of suffixes and suffixoides of the parts of speech displaying homonymy in Romanian. We showed how the precision and accuracy of the algorithm change with the threshold imposed on the productivity of suffixes. The present study reveals the necessary knowledge for recognizing the morphological structure of words, the difficulties encountered in this process, and also the importance of this study for linguistic research, for the Artificial Intelligence domain, in tasks such as information retrieval, summarization, question answering and all the other tasks relying on natural language processing.