Language of output

Hi there,

I’m trying to summarize Japanese sentences by inputting them into Playground and adding “Tl;dr” at the end.

Sometimes the response summary is in Japanese and sometimes in English,
What is the reason for this?

Even if the parameters and sentences are exactly the same, the language of the response changes depending on the timing.
It is the same even if you execute the API and get the result not only in Playground.
The accuracy of summarization is good regardless of the language of the response.

Any suggestions?
I’m thankful for every hint.

nico11

2 Likes

Hey - I have this exact same issue with Japanese using the text-davinchi-2 model using the API!

Someone please help!

OK - the simple way is to just add an instruction “translate this to japanese” at the end of your english strings. I’m using bubble apps so i added a field in the db with a “translate this to japanese” text and put it at the end of my text creation sequence. Works like a charm lol

2 Likes

Thank you for your reply!

As you said, when I put it at the end of the text, the summary results were output in Japanese every time.

Thanks yasha!

1 Like

By the way, how did you know to add the command “translate this to japanese” at the end of the English strings?
If you don’t mind, please let me know!

Hey @nico11 I was in the playground and found a translation example, it gave me the idea…

btw - I’m mainly working on an english app, then I’ll do a japanese version to share with my friends over there, so… this may not be a stable solution, kind of a first pass.

I tried this with davinci and it looked fine to me.

Input:

Generate a haiku in Japanese language with the topic “autumn”. The haiku:

Output:

秋風に
落ち葉が舞う
涼しい朝

Aki kaze ni
Ochiba ga mau
Suzushii asa

Thats great syntax - thanks!

This topic is very interesting and I did a test as well for Traditional Chinese. This is my test result. This is my prompt.

Please write a short summary about the impact on human society from climate change. Translate the response in Traditional Chinese.

In my first test run, I use “traditional Chinese”, the response from text-davinci-002 is mixing traditional chinese and simplified chinese, which is not correct per my request.
In my second test run, I use “Traditional Chinese”, this time the response is all in Traditional Chinese, which is correct, but few sentenses are not making sense at all.

In short, a human proofreading is still needed to verify the truthful of response from OpenAI.

2 Likes