Generating passages similar in style to a set of 9 examples (Question)
\n \t
\n
<\/BR>
\n
\n \t
\n
<\/BR><\/p>\n
Hello everyone
I hope I can find some guidance here for a project in generative AI.<\/p>\n
I have a set of 9 short passages from a TOEFL-like English test. I need to generate more passages that match the style of the examples set. The passages are 50 – 100 words, and are cut at the end in the middle of a sentence, and the examinees’ task is to choose the correct answer that completes the text correctly, out of 4 options.<\/p>\n
Here’s what I considered:<\/p>\n
Some questions:
1. For the scoring method, I’m considering TFIDF of POS (part of speech) and function words. Is that a good idea? Any other suggestions? I did consider embeddings, but wouldn’t that lead to passages similar in content rather than in style? 2. How would you generate 3 wrong answers that also fit the style of the wrong answers in the examples? I thought I’d cluster the examples’ wrong answers into 3 categories using k-means, figure out what distinguishes each class from the others, and ask ChatGPT to generate one wrong answer from each category (e.g. bad grammar \/ contradictory information \/ etc.). 3. Any other approaches that you’d suggest? Could i build a generative model that takes in an article (e.g. Wikipedia article) and modifies it so the format and style matches the examples’, or is the examples set too small for that?<\/p>\n<\/p><\/div>\n
submitted by \/u\/RunOrDieTrying <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span>\n<\/div>\n