Organizing and Improving a Database of French Word Formation Using Formal Concept Analysis

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3969–3976 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3969 Organizing and Improving a Database of French Word Formation Using Formal Concept Analysis Nyoman Juniarta1, Olivier Bonami2, Nabil Hathout3, Fiammetta Namer4, Yannick Toussaint1 1 Universit´e de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France 2 Universit´e de Paris, CNRS, Laboratoire de linguistique formelle, F-75013, Paris, France 3 CLLE, CNRS, Universit´e de Toulouse Jean Jaures 4 ATILF, Universit´e de Lorraine, CNRS {nyoman.juniarta, yannick.toussaint}@loria.fr olivier.bonami@linguist.univ-paris-diderot.fr, nabil.hathout@univ-tlse2.fr, fiammetta.namer@univ-lorraine.fr Abstract We apply Formal Concept Analysis (FCA) to organize and to improve the quality of D´emonette2, a French derivational database, through a detection of both missing and spurious derivations in the database. We represent each derivational family as a graph. Given that the subgraph relation exists among derivational families, FCA can group families and represent them in a partially ordered set (poset). This poset is also useful for improving the database. A family is regarded as a possible anomaly (meaning that it may have missing and/or spurious derivations) if its derivational graph is almost, but not completely identical to a large number of other families. Keywords: derivation, formal concept analysis, morphology, subgraph matching 1. Introduction D´emonette2 (Hathout and Namer, 2016; Namer et al., 2019; Namer and Hathout, 2020) is a derivational database that systematically describes the derivational properties of a fragment of the French lexicon. In the database, an entry corresponds to a pair of lexemes from the same derivational family. A derivational family is a set of lexemes connected by derivational relations. Fig. 1 shows examples of five derivational families. Each family is shown with its pair(s) of lexemes and the morphological pattern relat- ing each pair. The family of cramer, for example, has only one pair describing the direct derivation from the verb cramerV to the noun cramageN using the morpho- logical pattern X-Xage. The set of derivations in the family of cramer is similar to that of haubaner, allowing them to be grouped in the same paradigm. This similarity also poses some ques- tions. Is it possible that a derivational relation (cramerV — crameN) is missing in the family of cramer? Or con- versely, is the derivation haubanerV — haubanN erro- neous? It is also possible that these derivations are cor- rect as such, the two families belong then to two differ- ent morphological paradigms. This paper deals with the following questions: 1. How to systematically represent the relation among families. 2. How to detect families having anomalies, i.e. fam- ilies having either missing or incorrect deriva- tions. The presence of anomalies in D´emonette2 is not sur- prising, given that it is built from heterogeneous re- sources. This merge generated some mistakes that should be corrected. The improvement of derivational databases of numer- ous languages has been studied: Bulgarian (Dimitrova et al., 2014), Croatian (Filko and ˇSojat, 2017; Filko et al., 2019), Czech (ˇSevˇc´ıkov´a et al., 2018), Hungarian (Tr´on et al., 2006), and Polish (Dziob …

Content truncated to 500 words