Problem being addressed
Sarcasm generation is a challenging problem since the generated utterance should have at least five characteristics (a.k.a. “sarcasm factors”): 1) be evaluative; 2) be based on a reversal of valence between the literal and intended meaning; 3) be based on a semantic incongruity with the context, which can include shared commonsense or world knowledge between the speaker and the addressee; 4) be aimed at some target, and 5) be relevant to the communicative situation in some way.
An unsupervised approach for sarcasm generation based on a non-sarcastic input sentence. The method employs a retrieve-and-edit framework to instantiate two major characteristics of sarcasm: reversal of valence and semantic incongruity with the context, which could include shared commonsense or world knowledge between the speaker and the listener.
Advantages of this solution
While prior works on sarcasm generation predominantly focus on context incongruity, combining valence reversal and semantic incongruity based on the commonsense knowledge generates sarcasm of higher quality. Human evaluation shows that the suggested system generates sarcasm better than human annotators 34% of the time, and better than a reinforced hybrid baseline 90% of the time.
Solution originally applied in these industries
Possible New Application of the Work
Studies have shown that the use of sarcasm or verbal irony can increase creativity on both the speakers and the addressees, and can serve different communicative purposes such as evoking humor and diminishing or enhancing critique.
Currently sarcasm detection and generation remains a challenging task for conversational agents, so the research can contribute to creation of more adaptive and language sensitive chat bots.
Source URL: #############