Yes, with MT, you can teach broad concepts like seq2seq and attention, which is why I wrote this NMT tutorial 3 years ago https://github.com/tensorflow/nmt . One may also view translation as the first area that deep learning works at scale for NLP.
-
-
-
Valid points.
Kraj razgovora
Novi razgovor -
-
-
I explored MT at a smaller company, settling on a simpler method (fast-align for learned dictionaries) - current MT models are overkill for most use-cases and paying for an MT API doesn't always make sense.https://engineering.squarespace.com/blog/2018/template-search-in-six-languages-using-machine-translation …
- Još 2 druga odgovora
Novi razgovor -
-
-
"approximately nobody needs to make MT models" Hm. Translation is a big multibillion dollar industry, no? Is MT a reasonable fraction of that? Abstractly I wonder how to estimate the number of practitioners doing MT versus, say, IE or something else.
-
Sure, people work on MT...But nobody's like, *dabbling* in MT, professionally. Plenty of people need to do quick-and-dirty custom IE projects.
- Još 1 odgovor
Novi razgovor -
-
-
There is demand from people who underestimate the difficulty/investment and want to try it themselves. EX: I wrote a hands-on beginner-level ML book and said basically what you said and pretty much the only negative feedback I get on the book is about not providing a MT project.
-
Also, a surprising number of the people who ask me about MT are wanting to build MT for their otherwise unsupported native language, so maybe a big part of the demand is just the long tail of smaller languages that Google isn't covering yet.
- Još 3 druga odgovora
Novi razgovor -
-
-
The translation task presents a lot of linguistic challenges, there is public datasets available and it has a pretty clear goal that everybody can relate to. Also: If you want to get into MT you need to start somewhere.
What would be your favorite task for a seq2seq tutorial? -
If you want to get into MT you should start in a PhD program
. Agree about the public datasets, and also that it's a relatable task. On the other hand it's also extremely particular...And most people can't read the data, which is kind of bad for an example. - Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
Author of the
Founder