I learned some more vocab and did some context learning.
I also started using GPT for some really basic story translation. I’d been using children’s board books (18 month old kid kind of materials) for a while, but with GPT it can give me some tailored short stories, then some feedback on my translation.
My prompt:
“I’d like to do some german to english translation practice. Could you please write a 3 paragraph story in German. The language level is 18 months, like a board book kind of level. I’ll try to translate it.”
The feedback after I type in a translation is usually about like this:
Good work! You caught most of the meaning correctly. A few small notes:
“Der kleine Hund Max läuft in den Garten.”
You wrote: “The small dog Max ran in the garden.”
Better: “The little dog Max runs into the garden.”
Two things:
läuft is present tense (“runs”), not past.
in den Garten means “into the garden” (motion toward), not “in the garden.”
“Die Sonne scheint.”
You wrote: “The son shines.”
Just a slip: Sonne = sun, not son. So: “The sun is shining.”
Is it better than in-person work? Nope. Can I do it on the train while I commute? Yup. Is it better than many of the tools I’ve run into? Naturlich.
I know a lot of folks around these parts are pretty down on LLMs, but I think they are fairly strong for this kind of use case. Doesn’t really matter if there are occasional parts it gets outright wrong, and it’s very good at producing reasonable sounding language and proofreading your messages for mistakes.
I feel it’s a reasonable level of use. Some of the big hurdles they have are pronunciation, idioms, and complex grammar. I’m operating at a text only 2 year old level, so it should be okay. :-)
The immediate feedback on why my German isn’t totally correct is invaluable. The big ones are conjugation and the variants on implied motion (a huge part of German ad a language), which it’s helping me a ton with.
Glad to hear you are working on your german still! Have you gotten some in-person tries in yet? I know those are intimidating.
LLMs can be useful for early language training the way you use it. I’ve used it to help me figure out dialect words I just can not find online at all in german.
And AFAIK language is specifically what they are trained for no? So it’s a good usecase. But a word of caution: be careful with trusting it with new information as always. Once asked an LLM out of curiosity what would be the closest non-direct translation for a “Fence” (merchant of stolen goods) in Finnish, it gave me a racist word that probably was in its training set. But for training easier stuff in the way you described it seems good :)
I learned some more vocab and did some context learning.
I also started using GPT for some really basic story translation. I’d been using children’s board books (18 month old kid kind of materials) for a while, but with GPT it can give me some tailored short stories, then some feedback on my translation.
My prompt:
“I’d like to do some german to english translation practice. Could you please write a 3 paragraph story in German. The language level is 18 months, like a board book kind of level. I’ll try to translate it.”
The feedback after I type in a translation is usually about like this:
Good work! You caught most of the meaning correctly. A few small notes:
“Der kleine Hund Max läuft in den Garten.”
You wrote: “The small dog Max ran in the garden.” Better: “The little dog Max runs into the garden.”
Two things:
läuft is present tense (“runs”), not past. in den Garten means “into the garden” (motion toward), not “in the garden.”
“Die Sonne scheint.”
You wrote: “The son shines.” Just a slip: Sonne = sun, not son. So: “The sun is shining.”
Is it better than in-person work? Nope. Can I do it on the train while I commute? Yup. Is it better than many of the tools I’ve run into? Naturlich.
I know a lot of folks around these parts are pretty down on LLMs, but I think they are fairly strong for this kind of use case. Doesn’t really matter if there are occasional parts it gets outright wrong, and it’s very good at producing reasonable sounding language and proofreading your messages for mistakes.
I feel it’s a reasonable level of use. Some of the big hurdles they have are pronunciation, idioms, and complex grammar. I’m operating at a text only 2 year old level, so it should be okay. :-)
The immediate feedback on why my German isn’t totally correct is invaluable. The big ones are conjugation and the variants on implied motion (a huge part of German ad a language), which it’s helping me a ton with.
Mit freundlichen Grüßen!
Glad to hear you are working on your german still! Have you gotten some in-person tries in yet? I know those are intimidating.
LLMs can be useful for early language training the way you use it. I’ve used it to help me figure out dialect words I just can not find online at all in german.
And AFAIK language is specifically what they are trained for no? So it’s a good usecase. But a word of caution: be careful with trusting it with new information as always. Once asked an LLM out of curiosity what would be the closest non-direct translation for a “Fence” (merchant of stolen goods) in Finnish, it gave me a racist word that probably was in its training set. But for training easier stuff in the way you described it seems good :)
Und: Hast du schon ein Oktoberfestbier getrunken?
Danke für die Gedanken.
Nein, ich habe kein Oktoberfestbier getrinken. Ich trinke fast kein alkohol. Aber, ich bin in Berlin und Oktoberfest ist klein hier.
Achso! Ich trinke auch nicht so viel. Ich glaube es ist besser für unsere gesundheit.