The literature on face emojis raises the central question whether they should be treated as pictures or conventionalized signals. Our experiment addresses this question by investigating semantic differences in visually similar face emojis. We test a prediction following from a pictorial approach: small visual features of emojis that do not correspond to human facial features should be semantically less relevant than features that represent aspects of facial expressions. We compare emoji pairs with a visual difference that either does or does not correspond to a difference in a human facial expression according to an adaptation of the Facial Action Coding System. We created two contexts per pair, each fitted to correspond to a prominent meaning of one or the other emoji. Participants had to choose a suitable emoji for each context. The rate at which the context-matching emoji was chosen was significantly above chance for both types of emoji pairs and it did not differ significantly between them. Our results show that the small differences are meaningful in all pairs whether or not they correspond to human facial differences. This supports a lexicalist approach to emoji semantics, which treats face emojis as conventionalized signals rather than mere pictures of faces.

We introduce a generic, language-independent method to collect a large percentage of offensive and hate tweets regardless of their topics or genres. We harness the extralinguistic information embedded in the emojis to collect a large number of offensive tweets. We apply the proposed method on Arabic tweets and compare it with English tweets—analyzing key cultural differences. We observed a constant usage of these emojis to represent offensiveness throughout different timespans on Twitter. We manually annotate and publicly release the largest Arabic dataset for offensive, fine-grained hate speech, vulgar, and violence content. Furthermore, we benchmark the dataset for detecting offensiveness and hate speech using different transformer architectures and perform in-depth linguistic analysis. We evaluate our models on external datasets—a Twitter dataset collected using a completely different method, and a multi-platform dataset containing comments from Twitter, YouTube, and Facebook, for assessing generalization capability. Competitive results on these datasets suggest that the data collected using our method capture universal characteristics of offensive language. Our findings also highlight the common words used in offensive communications, common targets for hate speech, specific patterns in violence tweets, and pinpoint common classification errors that can be attributed to limitations of NLP models. We observe that even state-of-the-art transformer models may fail to take into account culture, background, and context or understand nuances present in real-world data such as sarcasm.