Linguistic classification: dealing jointly with irrelevance and inconsistency

Proceedings of Recent Advances in Natural Language Processing, pages 345–352, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_040 345 Linguistic Classification: Dealing Jointly with Irrelevance and Inconsistency Laura Franzoi Faculty of Mathematics and Computer Science University of Bucharest laura.franzoi@ gmail.com Andrea Sgarro DMG University of Trieste sgarro@units.it Anca Dinu Faculty of Foreign Languages and Literatures University of Bucharest ancaddinu@ gmail.com Liviu P. Dinu Faculty of Mathematics and Computer Science University of Bucharest liviu.p.dinu@ gmail.com Abstract In this paper we present new methods for language classification which put to good use both syntax and fuzzy tools, and are capable of dealing with irrelevant linguis- tic features (i.e. features which should not contribute to the classification) and even inconsistent features (which do not make sense for specific languages). We intro- duce a metric distance, based on the gener- alized Steinhaus transform, which allows one to deal jointly with irrelevance and in- consistency. To evaluate our methods, we test them on a syntactic data set, due to the linguist G. Longobardi and his school. We obtain phylogenetic trees which some- times outperform the ones obtained by Atkinson and Gray (Gray and Atkinson, 2003; Bouckaert et al., 2012). 1 Introduction According to Ethnologue (Eth, 2018), there are around 7000 living natural languages in the world, and one of the most interesting topics (not only in the academic field, but also in the general public) is their classification. While the comparative method was the main method of classifying natural languages until the 90s, the last decades brought an increasing number of computational approaches for estimating the historical evolution of languages and their relationships. Most of the computational historical linguistics approaches rely on the use of lexical items. In contrast, very few of them take into account syntactic aspects. Moreover, fuzzy tools and information theory were employed quite sparsely in language classification tasks (Ciobanu et al., 2018), in spite the inherent fuzzy nature of the natural language data. This paper is based on previous work on fuzzy string distances and linguistic classification started in (Franzoi and Sgarro, 2017a,b; Franzoi, 2017), and inspired by the path-breaking ideas put forward back in 1967 (Muljaˇci´c, 1967) by the Croat linguist ˇZ., Muljaˇci´c. The technical tool which will be used in this paper is the general Steinhaus transform, or biotope transform, ap- plied to crisp strings which are however affected by irrelevance and inconsistency, as happens with data due to the linguist G. Longobardi and his school. Fuzziness in linguistics has been seldomly treated (Franzoi and Sgarro, 2017a,b; Dinu et al., 2018), as compared to crisp approaches. In his 1967 paper Muljaˇci´c, even if only rather implicitly, had introduced what appears to us as a natural fuzzy generalization of crisp Hamming distances between binary strings of fixed length n, and this only two years after Zadeh’s seminal work (Zadeh, 1965): the aim was showing that Dalmatic, now an extinct language, is a bridge between the Western group of Romance languages and the Eastern group, mainly Romanian. The situation is the following: Romance …

Content truncated to 500 words