Self Syntactico-Semantic Enrichment of LMF Normalized Dictionaries

Self Syntactico-Semantic Enrichment of LMF Normalized Dictionaries Elleuch Imen Gargouri Bilel Ben Hamadou Abdelmajid MIRACL Laboratory MIRACL Laboratory MIRACL Laboratory B.P. 1088 B.P. 1088 B.P. 242 3018 Sfax, Tunisia 3018 Sfax, Tunisia 3021 Sakiet-Ezzit Sfax, Tunisia imen.elleuch @fsegs.rnu.tn bilel.gargouri @fsegs.rnu.tn abdelmajid.benhamadou @isimsf.rnu.tn Abstract The main challenge of this paper is the syntactico-semantic enrichment of LMF normalized dictionaries. To meet this challenge, we propose an approach based on the content of these dictionaries, namely the “Context” fields and the syntactic and semantic knowledge. The proposed approach is composed of three phases. The first one deals with the data set concerning the syntactic arguments of the “Context” fields. The second consists in connect semantic arguments to the syntactic ones. The last phase links syntactic and semantic arguments. In order to evaluate the proposed approach, we have applied it to an available Arabic normalized dictionary. The results are encouraging with respect to the measurement evaluation. 1 Introduction Natural Language Processing (NLP) tasks require reliable linguistic resources such as lexicons. The latter represent lexical resources that should define for each lemma a highly valuable knowledge such as morphological features, syntactic behaviors and semantic knowledge like meanings, contexts, semantic classes and thematic roles. The availability of such knowledge favors the efficiency of NLP tools. For example, (Briscoe and Carroll, 2002) estimate that about half of the errors of parsers are based on the insufficient amount of knowledge concerning the syntactic argument structure in the used lexicons; on the other hand, (Carroll and Fang, 2004) show that the use of syntactic lexicons by a syntactic parser improves its performance. Furthermore, the lexicon is the core component for machine translation and information extraction (Surdeanu et al., 2011). Unfortunately, we find that the lexicons that combine syntactic and semantic knowledge (i.e., representing semantic predicates) are shallow for some languages and unavailable for many others. Among the first lexicons dealing with the syntactico-semantic knowledge, we note the framework of (Gross, 1975), which was a revelation in this field. However, the enrichment of the proposed structure and even that of the lexicons proposed thereafter was such a hard task that it could not be accomplished due to the varied and abundant knowledge to be represented and requiring a high linguistic expertise. Thus, this enrichment task is an expensive and time- consuming process. Some other researchers like (Medelyan et al., 2013) have proposed to enrich such lexicons automatically using statistical methods. Nevertheless, the obtained content of such lexicons lacks reliability compared to the expert enrichment work. We feel that the enrichment issue of syntactico- semantic lexicons cannot be dealt with independently of their models. In this context, the International Organization of Standardization (ISO) has published the LMF-ISO 24613 (Lexical Markup Framework) standard (Francopoulo and George, 2008). LMF provides a unified model for constructing lexical resources covering all linguistic levels and dealing with the majority of languages. It offers a finely-structured model including the syntactico-semantic part. Many compliant lexicons to the LMF standard have been developed such as Wordnet-LMF (Henrich and Hinrichs, …

Content truncated to 500 words