Can LLMs Identify Korean Adverbs? (Part 2)
Where we left off in round one, I was struggling with both ChatGPT 4 and Google Gemini 2.0 Flash to get them to just process the list correctly. I gave up on ChatGPT 4 because I didn’t need that kind of stress in my life. I already had decent results from Claude 3.7 Sonnet. I say “decent” because I noticed that there were words in the non-adverb table that should have been in the adverb table. I decided to have Claude reprocess each list. In the first round, Claude said that 94 words were non-adverbs. I gave it the non-adverb list again and now 68 of those words are considered adverbs and 25 of them are non-adverbs. Claude labeled 250 words as adverbs. I gave it the adverb list again and only 205 were considered adverbs. After processing about 103 entries, the prompt stopped due to message constraints. I had to prompt it with “continue” to get it to finish. I still noticed words that should be considered adverbs in the non-adverb table. “Why are 유효하게, 저조하게, 부자연스럽게, and 천진난만하게, not cons...