HOW LANGUAGE MODELS AND HUMANS INTERPRET MEANING IN ENGLISH
Authors/Creators
Description
The mixed-methods design combined a survey of 300 specialists in linguistics, AI research, and ethics with case studies of model responses to naturally occurring discourse. The study drew examples from social media, academic articles, and spoken dialogue corpora. Standardized prompts were used across platforms. Descriptive and inferential analyses showed that model responses were closer to human interpretations on tasks involving structured grammar and contextual inference. Greater differences emerged when tasks depended on lived experience, sensitivity to emotion, or ethical judgment. The findings support evaluating models on their ability to interpret cultural and emotional cues and making the use of AI-generated text transparent. They also raise a question for cognitive linguistics about how to account for AIgenerated language as it becomes more common in everyday communication.