Abstract

Aleksi Sahala will talk about a pipeline for automatically discovering loanwords in monolingual dictionaries. For our Coptic-Greek test case, we use the Coptic Dictionary Online and LSJ. The pipeline measures words’ semantic similarity based on their English translations, and uses a dynamic programming algorithm and a custom phonemic vector space representation to align the words phoneme-by-phoneme to measure their phonetic distance. By combining the semantic and phonetic distances, the system is able to discover the most likely candidates for etymologically related vocabulary, which can then be evaluated by a human expert.

Tommi Jauhiainen will present how the Double Mutual Rank (DOMUR) network similarity measure is calculated using text-level n-gram and Naive Bayes-based language models. Heidi Jauhiainen will present the DOMUR networks we have created so far for texts in Greek, Coptic, Cuneiform-using languages (ANEE Idiolect Network Portal), and Egyptian Hieroglyps/Demotic. As of writing this abstract, the Greek network uses 54,002 transcribed texts from Papyri.info; the Coptic network portal contains 1,968 texts from Coptic Scriptorium; the ANEE portal has 105,621 texts from Oracc; and the TLA-based sentence-level network contains over 200,000 sentences.