Not Just Iconic: Emoji Interpretation is Shaped by Use

Brianna O’Boyle, Gabriel DoyleACL 2023

TL;DR

This paper challenges the assumption that emoji meanings are solely derived from their visual appearance. Through a study on WeChat, researchers found that platform-specific use and user familiarity significantly alter emoji interpretation, sometimes even flipping their sentiment. This highlights that emoji, like words, develop nuanced meanings through social interaction, not just iconicity.

What Problem Does This Paper Address?

The prevailing assumption is that emoji are inherently transparent, meaning their visual form directly conveys their meaning, much like a picture. This "iconicity" is often contrasted with words, whose meanings are largely arbitrary and learned through convention. If emoji are truly iconic and universally understood, they could serve as a powerful cross-linguistic and cross-cultural communication tool. However, this paper questions this simplistic view, proposing that emoji interpretation is more complex and influenced by factors beyond pure visual representation. The core problem addressed is understanding how emoji meanings are formed and whether they are truly as transparent and universally interpretable as commonly believed, especially in the context of specific social media platforms.

Methods

The researchers conducted an experiment involving 57 participants, who were asked to rate the sentiment of 105 WeChat-specific emoji. Participants placed each emoji on a two-dimensional plot representing valence (positive/negative) and energy (arousal). They also rated their familiarity with each emoji and provided self-reported data on their WeChat usage, Chinese language proficiency, and cultural familiarity.

The analysis involved two main statistical approaches:

  1. Mann-Whitney U-tests (with Holm corrections): This was used to compare the interpretations of emoji between WeChat users and non-users, treating these as distinct groups.
  2. Kendall rank correlation tests (with Holm corrections): This was employed to examine the relationship between participants' familiarity with individual emoji and their sentiment ratings. This approach was used to gain a more granular understanding of how experience shapes interpretation, acknowledging that not all users are equally familiar with all emoji.

Datasets

The study utilized a custom dataset of 105 WeChat-specific emoji. These emoji were presented to participants for interpretation. The paper notes that three legacy emoji from WeChat's predecessor, QQ, were excluded from the analysis. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Iconicity is not enough: The study strongly suggests that emoji meanings are not solely determined by their visual forms. Familiarity with the platform and the specific emoji's usage plays a crucial role.
  • Platform-specific meanings emerge: Just like words in a language, emoji can develop "platform-specific meanings" through repeated use within a particular social media environment. These meanings can diverge from the initial, purely iconic interpretation.
  • Familiarity leads to more extreme ratings: Participants who were more familiar with an emoji tended to give more extreme valence and energy ratings. Conversely, less familiar emoji received more neutral ratings.
  • Misinterpretation can flip valence: In some striking cases, the interpretation of an emoji by experienced users was significantly different, and sometimes even opposite, to that of non-users. This highlights the potential for significant miscommunication.
  • Cultural knowledge is sometimes necessary, but not always sufficient: While some emoji (like the "666" emoji) require cultural context to be understood correctly, others showed interpretation differences that couldn't be explained by cultural knowledge alone, pointing to platform-specific usage as the driver.
  • Emoji interpretation can be treated linguistically: The findings suggest that emoji interpretation shares similarities with how words acquire meaning, advocating for a more linguistic approach to studying them.

Key Results

The study found that eight out of 105 WeChat emoji showed significant correlations between familiarity and valence ratings after statistical correction. For one emoji, a significant difference in energy ratings was also observed.

A key qualitative finding was the stark contrast in interpretation for certain emoji. For example, the "jīa yóu" emoji, used in Chinese for encouragement, was interpreted by non-users as a gesture of aggression (raising a fist to fight). Similarly, the "666" emoji, signifying "awesome" in Chinese internet slang, could be misinterpreted by Western users as referencing the devil.

The paper also highlights that for many emoji, users and non-users agreed closely on their interpretation, suggesting that the observed differences are not simply due to users being more confident in their ratings but rather to the development of richer, context-dependent meanings.

Limitations and Future Work

Limitations: * Participant Demographics: The study primarily involved American respondents, although a significant portion had lived in China. This limits the generalizability to the full spectrum of WeChat users and non-users globally. * Isolation of Emoji: The experiment evaluated emoji in isolation, detached from textual context. While this was intentional to isolate the effect of familiarity, it doesn't fully represent how emoji are typically used in richer conversational settings. * Controlling for Factors: While WeChat use and emoji familiarity were found to have the strongest impacts, the study acknowledges the need for further work to fully control for the influence of Chinese language proficiency and cultural familiarity.

Future Work: * Conveying Nuanced Meanings: The authors suggest that current emoji labels are insufficient for conveying nuanced meanings. Alternative methods are needed. * Visualization of Baseline Sentiments: Presenting users with a visualization of an emoji's baseline sentiment, perhaps derived from sentiment analysis or rating experiments, could help. * "Glossary" of Emoji Use: Creating a "glossary" that showcases sample messages illustrating the range of emoji meanings, similar to how dictionaries work, is proposed. This could be curated or automatically generated. * Real-world Conversational Context: Future research should examine how the observed user/non-user sentiment differences persist when emoji are used in conjunction with text, especially in contexts involving sarcasm, hyperbole, or irony.

Summary updated

BibTeX
@inproceedings{oboyle2023not,
  author = {Brianna O’Boyle and Gabriel Doyle},
  title = {{Not Just Iconic: Emoji Interpretation is Shaped by Use}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2023},
}
Figure