Definition
Sarcasm detection is the task of automatically identifying when an utterance’s intended meaning is the opposite of, or otherwise diverges from, its literal content. When applied to speech rather than text, it typically relies on zero-shot classification from acoustic and linguistic cues, and researchers use it to probe whether models trained in one language or context can generalize to detect sarcasm in another, such as transferring between Mandarin and English datasets.