I’m experimenting with something new. I have about a year’s streak across several languages. Today I tried something for the first time, using ChatGPT prompts.
In 2,500 words of C2-level English, tell the story of the play The Book of Mormon. give me a TXT link to download.
Now, in 2,500 words of C1-level French, tell the story of the play The Book of Mormon. give me a link to download as a TXT file.
Now, let’s proceed to N3/N4-level Japanese. In 2,500 words, tell the story of the play The Book of Mormon. Give me a link to download as a TXT file.
Now, let’s proceed to A2-level German. In 2,500 words, tell the story of the play The Book of Mormon. Use up to one dependent phrase per sentence. Give me a link to download as a TXT file.
I imported the created files into LingQ. (The parsing of the Japanese was really bad, but useable.)
I figure I can do similar with all manner of content ideas, from history, to fiction, to garbage disposal repair (which incidentally is this Saturday morning’s activity).
Further, I wonder about recall synthesis with a prompt such as:
Now, generate a multiple-choice quiz of 20-questions on the content, using questions of each language at the respective levels I’ve mentioned. Do not repeat or restate the same question across languages. Target the questions of abstract ideas and motivations to the higher language levels. Target the questions of facts and attribution of opinions to the lower language levels. At the end, provide an answer key. Then, before sharing back with me, check your work. Ensure the questions and responses offered are indeed at the language levels specified. Adjust as necessary.
There are several techniques I’ve used to work with learning more than one language at once. This is a new technique I’m going to try for a while–sequential themed reading. Has anyone done anything similar? I’m going to experiment doing this both combined with and not with my mother tongue to see how it “feels” in learning.
When GPT was first released I was very interested in this sort of thing. Initially as a business idea and later for enjoyment. I remember it was extremely powerful but still somewhat limited and repetitive in the stories it was generating. It would be interesting to give it another go with the new models.
Asking it to create content out of real things it seemed to do a much better job. My initial LingQ days were filled with generated tales of South Park into Finnishand later something like Here are a list of words (new words in my lesson) please return each with an example sentence or something like that and I would import it as a new lesson.
Using it for a multi-choice quiz setup sounds like a great idea. I will definitely look into that.
This is an example of what i had running. I generated about 5000 stories across 100 languages.
I am interested in a command that would create useable, interesting content from a static command. To some extent this is about efficiency with the free chat gpt which gives shorter outputs.
I will definitely give it a go and see how it goes. Since I have recently become a permanent resident of the United States and calling Dallas as my permanent home. I will be adding Spanish as an added language to learn along with English and German. I am getting so much “passive” exposure to Spanish in terms of receiving documents written both in English and Spanish. I also encounter Spanish native spakers in day to day interactions. I will definitely give your suggested language learning approach a try.
Currently, as a beginner of Spanish, I use English as a crutch to learn Spanish. Hence, I use a free Im translator extension which allows me to read Lessons on Lingq with Bilingual translation.
Click on Print lesson, select the text and then with right click you can access to this extension.
@ericb100 : Sequential in contrast to parallel, where in the context of language translation “parallel texts” often means where two texts in different languages are lined up, such as in two columns on a page at the paragraph or even sentence level, to aid readers fluent in one language to work with a text often in another and typically the source language of the text.
@roosterburton : I’m working with ChatGPT 4o and find it good at story summarization or regurgitation where it “knows” the story in the corpus it was trained on. In contrast, I find it quite challenged in following good, clear prompts in individual sentence generation.
@asad100101 : The screenshot you show has sentence-level parallel translation, where a known language is used to aid in the reading of a text of a lesser known language.
I’m referring to a slightly-to-substantially different technique where the comprehensible input of a general story are consumed in multiple languages successively and even at different levels (thus even looser than dynamic translation and paraphrase) and the brain is needing to not as directly build from the scaffolding of the known language.
