Hacker News Reader

Hacker News

Comment by thunfischtoast | Hacker News Reader

Parent Full thread

thunfischtoast

It's always a good initiative to build something on your own.

Then again, it's also good to not lie to your users.

Your courses are AI-generated and not curated by experts.

I tried the French beginner course, using German as my base language. The very first items were:

1. Hallo (hello) > Bonjour (I think salut would be better)

2. Guten Morgen (good morning) > Bonjour

Then it asked me what Bonjour means, and selecting Guten Morgen is wrong, correcting me to Hallo. Then it asked me what Bonjour means again, this time Guten Morgen is correct.

So yeah, good initiative, but please just tell me what it is and don't lie.

View on HN

The issue here is a simple bug. We have two pairs for that specific lesson: - Hello <-> Bonjour - Good morning <-> Bonjour

They are identified by the question's id. And by the id we find the answer, and this is a bug where we show you one of the Bonjours, and only one of the answers is right per id. It's just a bug, that we already have mechanics for, but it's not always perfect. The course is still tested by linguists and native speakers.

Hey, bugs can happen, no problem.

But sorry, I do not believe your last sentence. German-French should be a popular pair, and you yourself claimed to speak both languages. This happened literally in the first beginner lesson. Everyone with most basic knowledge of the languages would catch that. This wouldn't even be a big deal if you'd communicate openly. This is a problem especially as you take money for the service after the free tier.

Hey there, gotcha, sorry for the bug, it's solved for the English <-> French course, I hadn't tested the German <-> French pair (although I do speak C2 German and B2 French). Thanks for the report, we have many many courses, many people that work on these courses, and recently relaunched entirely. We do believe in the mission, and we do have linguists and native speakers paid to work on these courses.

how many of the 700 courses have been fully tested end to end by an expert human other than yourself?

Hey dear neighbor :3

I'm no German speaker, but I'm French, and without invalidating your initial claim (about the AI generated stuff), in France we do translate "good morning" by "Bonjour", which literally means good (bon) day (jour).

Any other translation would be weird: if you'd translate "good morning" by "Bonne journée" -> that would be super weird, because this is something one could say in France to say "Goodbye" xD

I lived in Germany for a short time back in 2022, and notice that saying "Hallo" is used a little bit everywhere. However I can tell you that you are NOT supposed to say "salut" in France ANYWHERE except with your friends.

Like, imagine, you're in Germany you enter a bakery, you can say "Hallo" -> no problem. Same situation in France and you say "Salut" -> either people will react badly or assume that you don't know French or maybe they'll think you're impolite for no reasons

Thanks for your input neighbor! That's interesting because other sources actually start with "Salut", without going into the usage too much. Good to know.

Bonjour for good morning is correct. The problem is that the app introduced it to me like this:

"A in German means Z in French. B in German also means Z in French. Here is the word Z without any context. What's the correct word in German? You can choose between A, B, C and D. B is wrong, I wanted A".

I can imagine that they have direct mappings between the words, without checking for collisions. Even that could be fine, if the frontpage didn't claim this was "curated by experts" and it didn't happen literally in the first lesson haha.

Agreed - the landing page specifically claims “Over 700 expert-crafted courses for any language pair imaginable.”

I’m a trust-but-verify kind of person, and I can’t find a single mention of any language-learning facilities, academies, linguists, native speakers, or anything else that would corroborate this.

In other words, the system is designed so that any learning the speaker attempts somehow ends up being scored wrong, forcing the speaker to conclude they will always be identified by the system as an error-prone, non-native speaker yearning for acceptance by ears trained from childhood to hear cracks in any facade the speaker slaps together, forever and ever...

Sounds like the author unwittingly taught you the first lesson. :)

Edit: clarification

Reply on news.ycombinator.com