DeepL appears to be significantly better than Google Translate

All about language programs, courses, websites and other learning resources
Cainntear
Black Belt - 3rd Dan
Posts: 3469
Joined: Thu Jul 30, 2015 11:04 am
Location: Scotland
Languages: English(N)
Advanced: French,Spanish, Scottish Gaelic
Intermediate: Italian, Catalan, Corsican
Basic: Welsh
Dabbling: Polish, Russian etc
x 8663
Contact:

Re: DeepL appears to be significantly better than Google Translate

Postby Cainntear » Sat Jun 18, 2022 5:53 pm

Le Baron wrote:Ili ankaŭ opinias, ke ĝi estas nur 'fuŝita' hispana. Estis ankaŭ la opinio de Noam Chomsky. Google traduko provos ĉiam korekti gin usante la hispanan... (try again with this spelling..etc).

They all hate Esperanto.

My old fear about learning conlangs was the idea that I might pollute my own intuition about languages by introducing data that doesn't conform to the naturally occurring rules of language.
I'd be equally hesitant to teach a conlang to a machine learning algorithm that's attempting to learn general truths about language. I'm pretty certain both Google and DeepL are trying to build a model that uses information from previous languages to help learn the next one.
2 x

User avatar
Le Baron
Black Belt - 3rd Dan
Posts: 3511
Joined: Mon Jan 18, 2021 5:14 pm
Location: Koude kikkerland
Languages: English (N), fr, nl, de, eo, Sranantongo,
Maintaining: es, swahili.
Language Log: https://forum.language-learners.org/vie ... 15&t=18796
x 9389

Re: DeepL appears to be significantly better than Google Translate

Postby Le Baron » Sat Jun 18, 2022 6:20 pm

Cainntear wrote:My old fear about learning conlangs was the idea that I might pollute my own intuition about languages by introducing data that doesn't conform to the naturally occurring rules of language.
I'd be equally hesitant to teach a conlang to a machine learning algorithm that's attempting to learn general truths about language. I'm pretty certain both Google and DeepL are trying to build a model that uses information from previous languages to help learn the next one.


There are no naturally occurring 'rules' of language. It's randomised and the product of habits and mistakes and influence and etc... Intuition doesn't work in language knowledge, because all the so called rules are permanently contravened all the time. If you use intuition (which isn't a 'sixth sense' in the magic sense, but unconscious learning) you're more likely to come to something like Esperanto because you're looking for sensible patterns.

The old fear is a mistake.
1 x

Cainntear
Black Belt - 3rd Dan
Posts: 3469
Joined: Thu Jul 30, 2015 11:04 am
Location: Scotland
Languages: English(N)
Advanced: French,Spanish, Scottish Gaelic
Intermediate: Italian, Catalan, Corsican
Basic: Welsh
Dabbling: Polish, Russian etc
x 8663
Contact:

Re: DeepL appears to be significantly better than Google Translate

Postby Cainntear » Sat Jun 18, 2022 7:07 pm

Le Baron wrote:
Cainntear wrote:My old fear about learning conlangs was the idea that I might pollute my own intuition about languages by introducing data that doesn't conform to the naturally occurring rules of language.
I'd be equally hesitant to teach a conlang to a machine learning algorithm that's attempting to learn general truths about language. I'm pretty certain both Google and DeepL are trying to build a model that uses information from previous languages to help learn the next one.


There are no naturally occurring 'rules' of language. It's randomised and the product of habits and mistakes and influence and etc... Intuition doesn't work in language knowledge, because all the so called rules are permanently contravened all the time. If you use intuition (which isn't a 'sixth sense' in the magic sense, but unconscious learning) you're more likely to come to something like Esperanto because you're looking for sensible patterns.

The old fear is a mistake.

I might be wrong, but we can't prove or disprove that.

The debate over the fundamental nature of language has not been resolved. Are things as they are just by chance, or are there some underlying principles that naturally-occurring languages are built on top of?

Certainly, there are patterns and tendencies that Esperanto doesn't follow. Is there a single natural language in the world that doesn't have true, non-derived antonyms for "good" and "bad"? Is there a single language with pronouns built around the same vowel?
2 x

User avatar
Le Baron
Black Belt - 3rd Dan
Posts: 3511
Joined: Mon Jan 18, 2021 5:14 pm
Location: Koude kikkerland
Languages: English (N), fr, nl, de, eo, Sranantongo,
Maintaining: es, swahili.
Language Log: https://forum.language-learners.org/vie ... 15&t=18796
x 9389

Re: DeepL appears to be significantly better than Google Translate

Postby Le Baron » Sat Jun 18, 2022 7:17 pm

Cainntear wrote:I might be wrong, but we can't prove or disprove that.

The debate over the fundamental nature of language has not been resolved. Are things as they are just by chance, or are there some underlying principles that naturally-occurring languages are built on top of?

If all that is so, why did you have a fear of pollution of unknown rules?
1 x

Cainntear
Black Belt - 3rd Dan
Posts: 3469
Joined: Thu Jul 30, 2015 11:04 am
Location: Scotland
Languages: English(N)
Advanced: French,Spanish, Scottish Gaelic
Intermediate: Italian, Catalan, Corsican
Basic: Welsh
Dabbling: Polish, Russian etc
x 8663
Contact:

Re: DeepL appears to be significantly better than Google Translate

Postby Cainntear » Sat Jun 18, 2022 8:18 pm

Le Baron wrote:
Cainntear wrote:I might be wrong, but we can't prove or disprove that.

The debate over the fundamental nature of language has not been resolved. Are things as they are just by chance, or are there some underlying principles that naturally-occurring languages are built on top of?

If all that is so, why did you have a fear of pollution of unknown rules?

I'm particularly concerned about the rules of Germanic and Romance languages. I've pretty much specialised in the Romance languages, and there are things that I find increasingly predictable with each language. Esperanto goes off in a totally different direction, deriving and inflecting Latin-derived roots in ways that don't sit inside the patterns of variation across Europe's languages -- even at a superficial level, there's reason to be concerned that Esperanto pollutes the system.
Esperanto has features that are demonstrably different from natural languages -- I already pointed to the antonyms thing, and the single vowel for all pronouns (which reduces redundancy and makes pronouns harder to process at speed).

Given all that, and the lack of any compelling reason to learn a conlang, I just think on balance of probabilities, it's not worth it.

And similarly, the addition of conlangs to a machine learning system doesn't offer a lot of benefits, so you'd be gambling the chance of slightly better results against the possibility of significantly worse ones. Hardly seems worthwhile.
2 x

User avatar
Le Baron
Black Belt - 3rd Dan
Posts: 3511
Joined: Mon Jan 18, 2021 5:14 pm
Location: Koude kikkerland
Languages: English (N), fr, nl, de, eo, Sranantongo,
Maintaining: es, swahili.
Language Log: https://forum.language-learners.org/vie ... 15&t=18796
x 9389

Re: DeepL appears to be significantly better than Google Translate

Postby Le Baron » Sat Jun 18, 2022 8:30 pm

Yes, I remember your opinion of Esperanto from before. Then forgot about it.
0 x


Return to “Language Programs and Resources”

Who is online

Users browsing this forum: No registered users and 2 guests