Emoticon Style: Interpreting Differences in Emoticons Across Cultures

Emoticons are a key aspect of text-based communication, and are the equivalent of nonverbal cues to the medium of online chat, forums, and social media like Twitter. As emoticons become more widespread in computer mediated communication, a vocabulary of different symbols with subtle emotional distinctions emerges especially across different cultures. In this paper, we investigate the semantic, cultural, and social aspects of emoticon usage on Twitter and show that emoticons are not limited to conveying a specific emotion or used as jokes, but rather are socio-cultural norms, whose meaning can vary depending on the identity of the speaker. We also demonstrate how these norms propagate through the Twitter @-reply network. We confirm our results on a large-scale dataset of over one billion Tweets from different time periods and countries. Introduction The most important thing in communication is hearing what isn’t said. –Peter Drucker Body language and facial expressions can sometimes tell more about what one is trying to express than what one actually says in face-to-face interactions. Changes in vocal intonation can serve a similar purpose in exclusively spoken communications. Such cues take up an estimated 93% of everyday communication (Mehrabian 1971) and help people better communicate complex emotions like humor, doubt, and sarcasm. In text-based communication, however, these cues are not present and their absence can result in misunderstanding and confusion. The growth in computer-mediated communications has led to the use of conventions where emotion or affect is referenced pictorially using alphanumerics, punctuations, or other characters. These symbolic representations are commonly referred to as emoticons (Walther and D’addario 2001). The origin of emoticons is of dispute—especially for the basic smiley :)—but most studies suggest that they appeared in the early 1980s and have since gained massive popularity (Derks, Bos, and Grumbkow 2007). Emoticons, like nonverbal cues, help people interpret the nuance of meaning, the attitude of a conversational partner, and the level of Copyright c © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. emotion not captured by language elements alone (Lo 2008; Gajadhar and Green 2005). With the advent of mobile communications, the use of emoticons has become an everyday practice for people throughout the world. Interestingly, the emoticons used by people vary by geography and culture. Easterners, for example employ a vertical style like ^_^, while westerners employ a horizontal style like :-). This difference may be due to cultural reasons since easterners are known to interpret facial expressions from the eyes, while westerners favor the mouth (Yuki, Maddux, and Masuda 2007; Mai et al. 2011; Jack et al. 2012). In this paper, we study emoticon usage on Twitter based on complete data of tweets from the period 2006 through 2009, the first three years of this now ubiquitous microblogging platform. We focus on the macro-level trend first, and examine what emoticons are popular and how they vary stylistically. Next, viewing emoticon usage as a social norm, we study how emoticons differ across cultural boundaries defined by geography and language. We then contrast what affect categories are associated with emoticons across countries. Moving from the macro-level to the level of user-touser interactions, we use the Twitter @-reply graph to investigate the propagation processes of particular emoticons over social links, and study how the diffusion characteristics of emoticons can help us understand which emoticons have broader appeal in a new cultural setting. We make several interesting findings: 1. Emoticons are generally used in positive and light context, and as a result tweets containing extremely angry or anxious sentiment rarely accompanied emoticons. 2. Users continuously expand the meanings of emoticons by adopting variants of the normative forms such as :) with pictorial representations of facial features such as winks ;), forehead =:), and nose :-). These variants are sometimes associated with different kinds of affect than their normative forms. 3. While geography matters in determining the emoticon style, language has a higher impact. In the Philippines and Indonesia, where English is in common usage along with local languages, users utilized horizontal style emoticons as in predominantly English speaking countries. 4. European users were multi-cultural in terms of emoticon usage with both vertical and horizontal styles being employed in tweets. 5. While popular emoticons like :) and :( are adopted either spontaneously or through sources outside of Twitter, less popular ones like :P, ^^ and T_T had higher chance of diffusion through the Twitter’s @-reply friendship relationship. This diffusion may be due to influence, and occurs almost entirely within cultural boundaries. The remainder of this paper is as follows. We start by briefly reviewing the relevant literature. We then describe the Twitter dataset and our method for extracting emoticons from tweets. The next section presents the basic analysis of emoticons in Twitter, followed by cultural boundaries and diffusion processes of emoticons. Finally we discuss implications of findings and conclude. Related Work Emoticons are a crucial part of computer-mediated communication (Walther and D’addario 2001). Previous work confirmed that users reading text messages with emoticons are significantly better at interpreting the precise meaning of the author than those reading messages without emoticons (Lo 2008; Gajadhar and Green 2005). Emoticons are known to be used more frequently in socio-emotional contexts than in task-oriented contexts (Derks, Bos, and Grumbkow 2007). Recent work similarly demonstrated that Twitter users are more likely to use emoticons when conversing with others than when posting status updates (Schnoebelen 2012). Emotional valence has been shown to match well, as positive emoticons were used more in positive contexts and negative emoticons, more in negative contexts. However, emoticon usage decreased when people felt extreme emotions of anger or guilt, showing a tendency to drop emoticons for emotionally intense situations (Kato, Kato, and Scott 2009). Several studies focused on emoticon usages across different cultures. One study comparing Japanese and American emoticons based on e-mail data found that the functions of emoticons were different (Markman and Oshima 2007). American emoticons primarily established punctuation, signature, and closing of a sentence, while Japanese emoticons often had more complex shapes, mimicking offline facial expressions. Another study based on SMS data confirmed that emoticon usages varied by the gender of users, while no relationship was found across different strengths of social ties (Tossell et al. 2012). When considering emoticons as a social norm, it is crucial to consider the effects of influence and homophily. If we assume that the usage of a particular emoticon (or a set of emoticons) falls under the culture—what is called or “beliefs, attitudes and behaviors”—then influence is the mechanism whereby one’s peers influence one’s culture, and homophily is the mechanism whereby one’s culture over time affects one’s choice of peers. Foundational studies of influence (Axelrod 1986) and homophily (Lazarsfeld and R.K.Merton 1954) analyze these mechanisms on their own, but more recently researchers have begun to look at how influence and homophily interact (Axelrod 1997), especially in the context of social media to lead to complex effects on the behavior and link patterns of individuals. In this paper, we focus on influence as the mechanism of interest when it comes to the adoption of emoticons. We use a sequential adoption model that is simpler than the dynamic matched sample approach in (Aral, Muchink, and Sundararajan 2009). Our formal approach is motivated by models studying the effects of triadic closure in Twitter (Romero et al. 2011). Methodology Twitter Data We use a corpus of the Twitter data in (Cha et al. 2010) from 2006 to 2009, which contains information about 54 million users and all of their public posts. Since we are interested in studying cultural differences, we tried to include both eastern and western countries that had significant Twitter populations. We classified users and their tweets into our list of countries based on their geo-location (Kulshrestha et al. 2012) and excluded some countries such as Brazil due to difficulty in processing their language. Table 1 shows the list of countries that we focus on in this study and their population proportion within the Twitter corpus. Country Language Culture Population US English Western 57.74% UK English Western 7.33% Canada English Western 3.91% Australia English Western 2.62% Germany German Western 2.12% Indonesia English Eastern 1.46% Japan Japanese Eastern 1.45% Netherlands Dutch Western 1.16% Philippines English Eastern 0.97% France French Western 0.83% Italy Italian Western 0.65% Spain Spanish Western 0.62% Mexico Spanish Western 0.52% Singapore English Eastern 0.48% South Korea Korean Eastern 0.30% Total data analyzed: 10 million users and 1.1 billion tweets Table 1: List of countries and their data studied Extracting Emoticons We limited our focus to only those emoticons expressing human facial cues, and compiled a list of candidate emoticons from a number of sources including the Wikipedia1. Based on the compiled list, we constructed regular expressions to search our dataset. As mentioned earlier, eastern and western countries employed different emoticon styles, as highlighted in Table 2. The horizontal style, popularly used in western countries, emphasizes the mouth for expressing emotion and commonly uses the colon sign (:) for the eyes. Different mouth shapes are used to express affect (e.g., positive, negative) and meaning (e.g., happy, sad, surprise). In contrast, the vertical style, popularly used in eastern countries, emphasizes the eyes for expressing emotion. The underscore character (_) is commonly used for the mouth, while various characters are used for the eye shapes to capture affect and meaning. The following characters were used for the mouth and eye shapes in the regular expressions: Mouth variants: ( ) { } D P p b o O 0 X # | _ Eye variants: : ; ^ T @ o O X x + = > < http://en.wikipedia.org/wiki/List_of_emoticons Style Normative form Affect Meaning Variant examples :) positive happy wink ;) Horizontal :( negative sad mouth :)) :((( (expression based on :o neutral surprise nose :-) :-( :-[ the mouth shape) :P positive tongue sticking out tear :’( :*( :D positive laugh forehead or hair >:( =:-) ^^ positive happy chin (^^) Vertical T_T negative sad mouth ^___^ T___T (expression based on @@ neutral surprise nose ^.^ ^-^ T.T the eye shape) –_– negative absent-minded sweat ^^; -_-;;; o.o positive curious, amazing eyebrow –_–^ Table 2: Two different styles of emoticons: horizontal (popular in western countries) and vertical (popular in eastern countries). In addition to these basic facial cues, we captured the variants of each normative form in the regular expression, which we discuss in more detail in a later section. Inferring Affect From Tweets In order to quantitatively measure what kinds of affect are associated with a given emoticon, we used LIWC (Linguistic Inquiry and Word Count) (Tausczik and Pennebaker 2010), which is a text analysis program that counts words in various psychological categories. LIWC supports many languages including English, French, Italian, and Spanish, all of which appear in our data. How are emoticons used in Twitter? We first present the overall emoticon usage patterns. Emoticon Usage Table 3 displays the number of tweets, mentions, and retweets from our Twitter corpus. In total 7% of all tweets contained at least one emoticon, where nearly half of them (52%) were used in mentions of others appearing with the @username mark, while only few of them (4%) were used in retweets. This means that emoticons were more popularly used in conversations than in information propagation. #Tweets #Retweets #Mentions Non-emoticon 1,624,968,457 52,501,839 507,177,878 Emoticon 130,957,062 2,369,449 67,656,408 Total 1,755,925,519 54,871,288 574,834,286 Table 3: Summary of dataset The fraction of emoticon tweets starts at nearly zero during the first few months after Twitter’s launch, then since June 2007 increases slowly to reach 4%–8% of all tweets (Figure 1). Its usage remains rather steady from 2009, possibly because emoticons are not specific to Twitter (i.e., have existed since the 1980s) and also may be because it is the very rate of emoticon usage in online conversations in general. For comparison, we also show the fraction of mention tweets over the same time period, which is a Twitterspecific convention and it shows a rapid trend of adoption form 0% to 34% over the years. While 23% of all users posted at least one tweet with emoticons, 80% of the heavy users posting more than 100 tweets had used emoticons. This implies that emoticon usage is more prevalent for heavy Time (binned by month) P ro po rti on (% ) Mar'06 Oct'06 May'07 Dec'07 Jul'08 Feb'09 Aug'09 0 10 20 30 40 Emoticon Mention Figure 1: Emoticon usage over time users. Having seen the overall trend, we next focus on the different styles, variants, and context of emoticons. Emoticon Style As we discussed earlier in Table 2, there are two kinds of emoticon styles: vertical and horizontal. This division is based on which facial part carries the meaning. Because there lacks a systematic division of the two types, we propose to investigate the different emoticon styles as summarized in Table 2. Emoticons can be in either a normative form or a variation of that form, where the normative form for the horizontal style has a colon (:) as the eyes and one mouth. All other changes to this normative form can be considered variants. For the vertical style, the normative form is defined by the shape of the eyes and by default does not contain mouth. We allow the mouth to appear in the normative form as in (T_T), in cases where the normative form without a mouth (TT) has an ambiguous meaning. Based on these definitions, we captured a total of 15,059 different kinds of emoticons from the Twitter corpus. Their popularity distribution was heavy-tailed as shown in Figure 2(a), so that a small fraction of emoticons had a disproportionately large share of all usages. Only 523 emoticons appeared more than 1,000 times and 76% of all emoticons appeared fewer than 10 times. The most popular emoticon is the horizontal smiley :), which appeared in 46 million tweets. Most emoticons in the top 10 list are horizontal styles except for ^^, indicating a natural biased towards the US and other English-speaking countries. Therefore, we try to address this limitation in the next section by delving into emoticons used by people from different countries, including countries where English is not the dominant language.