Hacker News Reader

Hacker News

Comment by denysvitali | Hacker News Reader

Full thread

denysvitali

> For example, here are the results of translating the English word "hello":

> Language: fr, Translation: bonjours

> Language: fr, Translation: bonsoir

> Language: fr, Translation: salutations

> Language: it, Translation: buongiorno

> Language: it, Translation: buonanotte

> Language: fr, Translation: rebonjour

> Language: it, Translation: auguri

> Language: fr, Translation: bonjour,

> Language: it, Translation: buonasera

> Language: it, Translation: chiamatemi

Is it just me or these machine translations are worse than ... Google Translate?

View on HN

These results are less accurate than Google Translate. But they are far faster to get, and far less expensive to generate: https://cloud.google.com/translate/pricing — our goal is here is speed. We want to search through many possibilities as quickly as possible.

The word vectors have been aligned in multiple languages. Using an approximate nearest neighbor search we are able to find the nearest vector to the input in multiple languages very quickly.

To keep the example simple, we did not try to filter the data through hand-built language dictionaries. In fact, we simply drop words in other languages that also appear in the English .vec file. Words like "ciao" appear frequently enough in otherwise English sentences that the example code drops it from Italian, and so is not shown in the results:

% curl -s "https://dl.fbaipublicfiles.com/fasttext/vectors-aligned/wiki..." | grep -n ciao 50393:ciao 0.0120 ...

One improvement would be to filter out any words that do not appear in a hand-curated dictionary instead of filtering out words that already appear in English. We decided not to show how to do this because we'd already introduced a few concepts, like aligned word vectors, approximate nearest neighbour searches, and wanted to keep the example as simple as possible.

Google Translate is state of the art, so I’m not sure why that would be surprising. That said, is there something wrong with the translations offered?

> That said, is there something wrong with the translations offered?

I think in French hello = "bonjour" and hi = "salut"... not sure where "bonjours" and "salutations" came from.

The Italian "auguri" means "best wishes"; "chiamatemi" means "call me". Neither is a plausible translation of "hello". The obvious one, "ciao", is missing.

I thought Hello was invented with the telephone. Prior to that, English greetings were good morning/evening. What do Italians and French say when they pick up the phone? Allora?

"Bonjours" doesn't exist, and "salutations" is a tad quirky, but OK in informal settings, especially when addressing many people at once.

No, bonjours exists (it's simply the plural form of bonjour used as a noun) but the contexts it is used are very very infrequent so it's weird to find it in that list.

Compare "I said my hellos and goodbyes." in English. It's definitely a word, just so uncommon as to be largely irrelevant in most practical contexts.

It's very clearly semantically related though? I don't understand the complaint here.

It seems to be a very domain specific solution, they are trying to present versions of words in customer requested domain names if already taken.

Like you type in “stargazer. com”, system sees it’s already registered, and returns a “sorry sir it’s taken” page, with similar words listed as “but maybe try these words: astronomer, observatory, telescope, shooting star...”.

So it’s not serious translation, more of an inexpensive quick dictionary search. I guess it’s okay for its intended purposes.

> Google Translate is state of the art

For French/English/German, DeepL is much better IME.

It would be better to run the vectors through an attention layer if you want sentence to sentence translation.

Reply on news.ycombinator.com