Moral Essential Elements: MEE-A Dataset for Moral Judgement

Moral judgments, including moral polarity judgments, moral intensity judgments, and moral type judgments, are important in helping people to understand the moral characteristics of behaviors. Moral judgments do not rely on all elements in behavior, but on those strongly related to them, which are named as moral essential elements in this paper. By conducting research on the moral essential elements in machine morality, it can help machines adapt to downstream tasks such as eliminating moral bias while making existing artificial intelligence more ethical. Existing research on machine morality knowledge contains moral vocabulary, moral sentences, etc. This information has been able to help machines achieve relatively good results when making moral judgments, but there are still problems such as incomplete injection of information and multiple actions present in a sentence with different moral judgments. To address the above problems, this paper summarizes and designs a moral judgments system, proposes the concept of moral essential elements, constructs a moral essential elements dataset containing more than 10,000 behaviors, identifies behavioral words and their moral essential elements in sentences using the traditional model and the large model ChatGPT respectively, then makes moral judgments.