Table 3: Category-wise evaluation scores. Likert scores are averaged over both annotators, correlation measured between individual Likert scores and percent agreement.

🔍🧠 Their experiments show that #LLMs can produce reasonable poem descriptions, but struggle with more abstract interpretion, highlighting where #NLG currently meets its #limits in #LiteraryInterpretation.

#LiteraryComputing #Evaluation #GenerativeModels