“…We converted forms in modern Slavic languages to a narrow phonetic representation using IPA transcriptions from Wiktionary (https: //www.wiktionary.org), which were used to train a neural encoder-decoder; these models were used to obtain IPA transcriptions for forms not in Wiktionary, and a portion was checked manually. In several cases we reconciled sources used in the etymological dictionary (e.g., Pleteršnik, 1894) with contemporary standardized orthographies, and made use of phonetic descriptions for languages where the training data were problematic (Schuster-Šewc, 1968;Lencek, 1982;Scatton, 1984;Comrie and Corbett, 1993;Ternes and Vladimirova-Buhtz, 1990;Landau et al, 1995;Šuštaršič et al, 1995;Dankovičová, 1997;Jassem, 2003;Gussmann, 2007;Stadnik-Holzer, 2009;Hanulíková and Hamann, 2010;Mojsijenko et al, 2010;Yanushevskaya and Bunčić, 2015;Howson, 2017Howson, , 2018Pompino-Marschall et al, 2017). For the medieval languages Old Church Slavic and Church Slavic, orthographic forms were converted to a broad phonemic transcription based on Lunt (2001).…”