Hacker News Reader

Comment by Asraelite | Hacker News Reader

It's a very odd set of languages that they have. French should have orders of magnitude more resources available for developing a translation engine than e.g. Bulgarian would, so why do the more obscure language first?

Maybe they just happened to be able to get volunteers/partners who know those languages.

View on HN

One of the main developers at Edinburgh is Bulgarian. :)

Possibly it has to do with who provided funding, or selection by a certain type of language families.

Can't be language families—they've got Spanish, Italian, and Portuguese, so French would be an obvious choice to include.

The inclusion of Estonian is particularly odd. It's a very small language (1.1m native) and is the only non-Indo-European language in production or development.

One of the partner universities in the original project was the University of Tartu in Estonia

Reminds me of the language learning app Lingvist, which for a while had only German, French, and Estonian for a similar reason: the company is Estonian.

Because French is a polysynthetic language. Agglutinative and polysynthetic languages have difficulty with contemporary language models.

Er, I don't think that's it. French is pretty far from being polysynthetic and is usually classified as analytic. Why do you say it's polysynthetic? Because of the few personal pronoun clitics it has?

I'd imagine the 14 grammatical cases of Estonian would be harder to handle than French grammar, and yet they have Estonian.

Reply on news.ycombinator.com