I grew up in Istanbul, in what is left of the Armenian community there. Western Armenian was standardised in that city in the nineteenth century, when the dialect of Constantinople became the written standard. It is still spoken there, by fewer people every year.
I went to Mkhitaryan through middle school, an Armenian school founded in Istanbul in 1825 by the Mekhitarists, the order that has run an Armenian press and a scholarly library out of San Lazzaro in Venice since 1717. So I learned to read and write the language formally, rather than only hearing it at home. After the entrance exams I went to the Deutsche Schule Istanbul, which opened in 1868 and is one of two schools in Turkey that can award a German Abitur. That is how I ended up in Germany.
I moved to Karlsruhe in 2015 to study computer science at KIT and finished the master's in 2023. I chose the subject because I liked video games, and then I learned how they are made. The trick is always better than what is holding it up.
Language is the exception. Taking it apart has not cost me anything yet, and either I have not got far enough in or there is no backend to reach. Words survive the inspection: where they came from, what they are related to, whether I can guess one in a language I do not speak by looking at it long enough. Armenian and Turkish came from home, German and English from school, and I picked up Italian for the Romance foothold, although all I can really do with it is order food, badly.
During the master's I got interested in AI, and particularly in NLP. My professor Jan Niehues showed me NLLB, Meta's translation model covering two hundred languages, Eastern Armenian among them and Western Armenian not. Extending it looked possible. I wanted to put my mother language in front of the research community and build something the community itself could use.
There turned out to be more Western Armenian on the internet than I expected. The problem was that it was scattered, unlabelled, and in no state to train anything on. So the first half of the thesis became collecting it: libraries in three countries, a lot of scanning, and a lot of correcting by hand.
I could not do that and keep a student job, so I left the job. The Calouste Gulbenkian Foundation covered my living costs with a project grant, and Mkhitaryan, my old school, together with TEAOV, the mutual aid foundation for teachers at the Armenian schools in Turkey, paid for the machine I still work on. Both of those are community money. That is how nearly everything in this language gets done.
I have kept at it since. New sources, more sentences, and increasingly the same data pointed at other problems: speech recognition, language models, the parts of a language that a translation system on its own does not reach. The fact of the matter is that very few people are working on any of this. Few enough that not knowing what the others are doing is not an option. If you work on Western Armenian, or on another language with nobody obliged to keep it working, or you are interested in this kind of thing, write to me and we can see how to help each other.
Apart from the occasional walk or bike ride, I collect things, real and virtual. Coins, stamps, Pokémon cards, achievements, in-game pets and mounts. The one I have kept up longest is the Panini World Cup album, since 2002, and only the World Cups, because the Euros are more or less the same teams every time.