incom.py - A Toolbox for Calculating Linguistic Distances and Asymmetries between Related Languages

Proceedings of Recent Advances in Natural Language Processing, pages 810–818, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_094 810 incom.py – A Toolbox for Calculating Linguistic Distances and Asymmetries between Related Languages Marius Mosbach†, Irina Stenger, Tania Avgustinova, Dietrich Klakow† Collaborative Research Center (SFB) 1102: Information Density and Linguistic Encoding † Spoken Language Systems, Saarland Informatics Campus Saarland University, Germany ira.stenger@mx.uni-saarland.de avgustinova@coli.uni-saarland.de {mmosbach, dietrich.klakow}@lsv.uni-saarland.de Abstract Languages may be differently distant from each other and their mutual intelligibil- ity may be asymmetric. In this pa- per we introduce incom.py, a toolbox for calculating linguistic distances and asymmetries between related languages. incom.py allows linguist experts to quickly and easily perform statistical anal- yses and compare those with experimen- tal results. We demonstrate the efficacy of incom.py in an incomprehension ex- periment on two Slavic languages: Bul- garian and Russian. Using incom.py we were able to validate three methods to measure linguistic distances and asymme- tries: Levenshtein distance, word adapta- tion surprisal, and conditional entropy as predictors of success in a reading inter- comprehension experiment. 1 Introduction 1.1 Related Work Linguistic phenomena may be language specific or shared between two or more languages. With regard to cross-lingual intelligibility, various con- stellations are possible. For example, speakers of language A may understand language B bet- ter than language C, i.e. [A(B) > A(C)] while speakers of language B may understand language C better than language A, i.e. [B(C) > B(A)]. For instance, Ringbom (2007) distinguishes be- tween objective (established as symmetrical) and perceived (not necessarily symmetrical) cross- linguistic similarities. Asymmetric intelligibility can be of linguistic nature. This may happen if language A has more complicated rules and/or ir- regular developments than language B, which re- sults in structural asymmetry (Berruto, 2004). Recently, in the INCOMSLAV framework (Fis- cher et al., 2015; J´agrov´a et al., 2016; Stenger et al., 2017a,b), measuring methods were devel- oped that are of direct relevance for modelling cross-lingual asymmetric intelligibility. While it has been common to use (modifications of) the Levenshtein distance (Levenshtein, 1966) to pre- dict phonetic and orthographic similarity (Beijer- ing et al., 2008; Gooskens, 2007; Vanhove, 2016), this string-edit distance is completely symmet- ric. To account for asymmetric cross-lingual in- telligibility Stenger et al. (2017b) employ addi- tional measures of conditional entropy and sur- prisal (Shannon, 1948). Conditional character adaptation entropy and word adaptation surprisal, as proposed by Stenger et al. (2017b), quantify the difficulties humans encounter when mapping one orthographic system to another and reveal asym- metries depending on stimulus-decoder configura- tions in language pairs. Similarly, the research of J´agrov´a et al. (2016) shows that Czech and Polish, both West Slavic, us- ing the Latin script, are orthographically more dis- tant from each other than Bulgarian and Russian, South and East Slavic respectively using the Cyril- lic script. Both language pairs have similar lexical distances, however, the asymmetric conditional entropy based measures suggest that Czech read- ers should have more difficulties reading Polish text than vice versa. The asymmetry between Bul- garian and Russian is very small with …

Content truncated to 500 words