Figure 2

The specific words from Panels G, M, S and Y of Figure  1 with the greatest mismatch. Only the center histogram from Panel Y of Figure 1 is included. We emphasize that the labMT dictionary scores generally agree well with the other dictionaries, and we are looking at the marginal words with the strongest disagreement. Within these words, we detect differences in the creation of these dictionaries that carry through to these edge cases. Panel A: The words with most different scores between the labMT and ANEW dictionaries are suggestive of the different meanings that such words entail for the different demographic surveyed to score the words. Panel B: Both dictionaries use surveys from the same demographic (Mechanical Turk), where the labMT dictionary required more individual ratings for each word (at least 50, compared to 14) and appears to have dampened the effect of multiple meaning words. Panels C-E: The words in labMT matched by MPQA with scores of −1, 0, and +1 in MPQA show that there are at least a few words with negative rating in MPQA that are not negative (including the happiest word in the labMT dictionary: ‘laughter’), not all of the MPQA words with score 0 are neutral, and that MPQA’s positive words are mostly positive according to the labMT score. Panel F: The function words in the expert-curated LIWC dictionary are not emotionally neutral.