#somewhere through this I started thinking of NMT as the ninja mutant turtles translating things for me

[redheaded-techpriest]

hey are u really studying NLP for translation? can you explain why DeepL seems to be so much better than google translate?

Hey! I'll do my best with what I've figured out through their proprietary BS!

My short answer is pretty much just due to different learning methods and dictionaries, with Google's Neural Machine Translation system and DeepL's convolutional neural net both having different advantages.

Google pulls from every google indexed site which offers a lot of languages, but of wildly varying qualities, and then runs it through their neural machine translation-inator, trying to match sentences based on its learned linguistic and semantic rules (Example Based MT), and for those without enough training data for the NMT-inator it just runs them through a statistical maximization function.

DeepL started off as Linguee, just a translation dictionary and expanded into sentences by pulling from bilingual texts, essentially trading quantity for quality of material, and then running it through their convolutional neural net, running the samples through filters trying to break down the important features, letting it have a bit more of a natural feeling translation.

#somewhere through this I started thinking of NMT as the ninja mutant turtles translating things for me #I'll probably be referring to it as the ninja mutant turtles from here on out #also apparently Linguee used a lot of canadian government documents for French training so that's cool #anyways I promise I tried keeping this short #I have a much longer rant that got shelved because it was just... too much #I hope this makes sense #nerd shit #p@ #computer science #natural language processing #nlp #comp sci #sciblr