The Extended Moral Foundations Dictionary: A new resource for coding moral content

As human beings, we are often moved to action based on moral messages. Adding a single moral-emotional word to a tweet (hate, greed, fight, safe, shame, etc.) increases the number of retweets by approximately 20% and moral content captures our attention

However, as researchers, deciding whether a particular newspaper article or social media post has a moral message can be difficult. The most common existing approach is to use a dictionary of moral terms (the Moral Foundations Dictionary). These words are classified as to whether they are a virtue or vice and which of the five moral foundations they illustrate. 

According to Moral Foundations Theory, all cultures share five moral foundations that form the basis of morality: care/harm, fairness/cheating, loyalty/betrayal, authority/subversion, and sanctity/degradation. This approach has been useful, and more than a dozen articles have used the Moral Foundations Dictionary in their research. However, there are a number of flaws with the existing dictionary. The first is that it was developed by a small group of experts and may not reflect how the general population uses moral words. A second flaw is that the dictionary takes an all-or-none approach. Each word either appears in a category or does not. This categorical approach can miss words with a heavy moral implication in some contexts but not in others.

To solve these problems, Hopp and colleagues (pictured below) took a new approach to creating a dictionary of moral terms. An article describing their new dictionary, the Extended Moral Foundations Dictionary and the process used to create it was recently published in the Psychonomic Society journal Behavioral Research Methods. 

Weber et al. 2020 Figure 3
Authors of the featured article

The research group used the power of crowdsourcing to identify the moral terms in their dictionary. Over 500 participants annotated almost 1000 news articles – highlighting sections that they believed reflected a given moral foundation. 

Weber et al. 2020 Figure 4
Flowchart of the dictionary creation process

Using those ratings, the researchers calculated how often each word was highlighted for each of the five moral foundations (number of times highlighted/number of times seen). The end result is a dictionary of 3,270 words, each associated with the probability it was highlighted for each moral foundation. 

The dictionary also includes the 689 moral words that were most closely associated with each moral foundation. 

Weber et al. 2020 Figure 5
Word clouds illustrating the words most closely related to each moral foundation

Using the new dictionary, the researchers were able to replicate many expected findings. Prior research suggests that conservatives and liberals tend to place more weight on different moral foundations. Sure enough, Breitbart (a far-right news source) emphasized loyalty and authority, while The Huffington Post (a far-left news source) emphasized care and fairness. Also, the new dictionary was a better predictor of Facebook article shares than earlier versions. Moral messaging is a strong predictor of how far an article will spread on social media. 

The authors also created new software – eMFDscore – to help other researchers use the new tool. Personally, I look forward to reading the research that will be spurred by this useful innovation.  

Psychonomic Society article considered in this post: 

Hopp, F. R., Fisher, J. T., Cornell, D., Huskey, R., & Weber, R. (2020). The extended Moral Foundations Dictionary (eMFD): Development and applications of a crowdsourced approach to extracting moral intuitions from text. Behavior Research Methods. https://doi.org/10.3758/s13428-020-01433-0

The Psychonomic Society (Society) is providing information in the Featured Content section of its website as a benefit and service in furtherance of the Society’s nonprofit and tax-exempt status. The Society does not exert editorial control over such materials, and any opinions expressed in the Featured Content articles are solely those of the individual authors and do not necessarily reflect the opinions or policies of the Society. The Society does not guarantee the accuracy of the content contained in the Featured Content portion of the website and specifically disclaims any and all liability for any claims or damages that result from reliance on such content by third parties.

You may also like