Syllable boundaries

[wollmers]

In the default German text the word 'Mithilfe' (Worttrennung: mit·hil·fe, IPA: [mɪtˈhɪlfə]), is transliterated to Ancient Greek 'Μιθιλφε', which IMHO should be 'Μιτἱλφε'.

[davidpomerenke]

Agree about the example.

The current implementation uses simple rules that do not take syllable boundaries into account. I suspect it would be very effortful to add consideration for syllable boundaries, and the benefit is very limited. (For German, the transliteration works relatively well and syllable boundaries are not a big concern; and for English there's a lot of other problems besides syllable boundaries.)

One approach could be converting to IPA before transliterating, see #8. At the time of writing of that issue, I had not found an IPA converter for German.

Another more general approach could be combining gpt-4o-mini or similar, which is very cheap now, with some rule-based scaffolding.

@wollmers do you know about German (or ideally, multilingual) IPA converters, or do you have ideas about alternative approaches (I see you've been working on syllable counting)?

[wollmers]

@davidpomerenke

There is a German IPA dictionary https://github.com/devio-at/german-ipa-dict. With some modification of the Python scripts, it should be possible to extract IPA for EN, which includes syllable boundaries like 'lexicon' IPA: /ˈlɛk.sɪ.kən/.

For counting syllables I use https://metacpan.org/pod/Text::Hyphen::DE, which uses the Knuth-Liang algorithm (made for TeX). Maybe you find an implementation in JavaScript or you port it.