Plausibility as Commonsense Reasoning: Humans Succeed, Large Language Models Do not

1. 1Identify syntactic ambiguities in language tasks
2. 2Construct ambiguous items to test attachment preferences
3. 3Evaluate human and large language model performance on resolving ambiguities
4. 4Analyze results to understand the differences in human and model behavior

NLP researchers and AI engineers can benefit from understanding the limitations of large language models in integrating world knowledge with syntactic structure, which can inform the development of more human-like language models

💡 Large language models do not integrate world knowledge with syntactic structure in a human-like way, leading to poor performance in resolving syntactic ambiguities

🤖 Large language models struggle with plausibility as commonsense reasoning #NLP #AI