Social Acceptability is an NLP task which concerns with how certain situations and actions are perceived and accepted by people and society. For example, "Killing a bear" is generally regarded as immoral everywhere, "Killing a bear to save a child". However, in any circumstance, "Killing a human" is considered regarded as morally wrong. Imbuing NLP systems with knowledge of social norms, behaviors and values is important so that they works well for the users without offending anyone.
One of the biggest problems with Social Acceptability is that it is not universal and is highly dependent on the society in which one lives. It varies greatly depending on the population's culture and demographics. For example, while "eating with hand" is considered "disgusting" in the western world, it is perfectly normal in some Asian and African countries.
In NLPositionality, we measure the demographic alignment of existing systems and datasets which model 'Social Acceptability'. We specifically look at GPT-4, the Delphi model, and the Social Chemistry 101 dataset. Click on the above buttons to explore the alignments.
