Graphemes from Graphemes: Reverse-engineering transcriptions from the 2nd century BCE

Abstract

Historical (medieval and antique) systems of phonetic transcription use the graphemes of one natural language alphabet to encode the phonemes of another language. Since such systems are restricted to the script of the target language, they are heavily under- and at times overdetermined, especially if the languages and writing systems are typologically diverse.

Hexapla Secunda is a very small corpus of Biblical Hebrew transcribed into the Greek alphabet, attesting such a system. We describe this system by composing a formal grammar. With the help of transduction and learned character-level alignments it is possible to reconstruct the Hebrew vocalic graphemes from the Greek transcription. From there we infer reconstructions for Biblical Hebrew even outside of the attested corpus.