Besides Donald Trump, its most famous user, some 330 million people use Twitter as a platform for communication, much of it political. Yet, given the 280 character limit, how much can you say in a tweet? Although much has already been written about Twitter, little attention has been given to the nature of the argument found there. To begin filling this gap, it is necessary to identify the basic units of such an argument. Identifying them as speech acts, we demonstrate here by discourse analysis how by virtue of the enthymematic quality of public argument, much argument can be communicated even by singular speech acts and even by speech acts other than assertion.
As is well-known, the micro-blogging platform known as Twitter allows the exchange of messages of up to 280 characters. Besides U.S. President, Donald Trump, its once most famous political user, some 330 million people use Twitter as a platform for communication.
Yet how much can you say in a Tweet? Although Twitter is used to communicate content other than argument, for argument, the 280-character limit would seem to be constraining. Thus, the animating question motivating this study is the possibility for and nature of political argument on Twitter and how it might vary by the news outlet to which it is addressed.
Ultimately, we seek a quantitative answer to that question employing content analysis. Given, however, how little attention has so far been paid to argument on Twitter, there are some theoretical questions that need to be answered first: Given Twitter’s character limitation, is an argument even possible or attempted? What do we even mean by argument and what constitutes its content? If, as we do here, we propose to understand argumentative content as speech acts, what do we mean by that and what kinds of speech acts could be used on Twitter to comprise an argument? Again, given Twitter’s character limitation, how briefly can an effective argument be made? Can it be done even by a single speech act, and if so speech acts of what kind?
These more theoretical questions prior to any kind of rigorous content analysis are the research questions explored in this paper. In a sense, the paper aims to establish a broader theoretical framework for any quantitative study of argument on Twitter. One possible unit of argumentative content is an entire dispute or thread of disputing tweets, but an even more basic unit of analysis within a single tweet is an individual speech act, a speech act defined as an act performed through an utterance (see Austin, 1962; Searle, 1970).
With individual speech acts in mind, the purpose of this paper is not yet to make statistical generalizations but just to uncover the range of possibilities, i.e. the kinds of speech acts to be found on Twitter and any arguments they support. For that purpose, it suffices to focus on illustrative cases. Here, we focus on tweeted responses to the initiating Tweets from Fox News and its polar opposite, MSNBC in October 2018 as they covered the arrest of Cesar Sayoc for mailing pipe bombs to Trump critics.
We think our following findings important not just in themselves but for future, more quantitative study: (i) given the great enthymematic power of even individual utterances, powerful arguments can be and are advanced even by tweets consisting of single speech acts; (ii) not needing to be assertions, such speech can be of a variety of kinds, including interrogatives and expressives; (iii) responsive tweets in our study to both FOX and MSNBC were dominated by Trump critics; and (iv) although others have observed humor to be pervasive on Twitter (see Davis et al., 2018), we find them more specifically to be powerful ways to advance the argument.
In an amazingly comprehensive literature review, assisted by computerized search, Karami et al. (2020) canvass some 18,000 manuscripts on Twitter published between 2006 and 2019. Politics shows up among some 40 other most frequent topics covered, which range from social movements and public relations to sentiment analysis. Indeed, the vastness of literature on Twitter is further indicated by yet another recent literature review that focuses on just the study of Twitter that uses sentiment analysis (Zimbra et al., 2018).
So far, most studies of Twitter have been quantitative. Much of it focuses on Twitter’s structural features. There have been a number of studies, for example, that employ social network analysis to gain a deeper understanding of the relationships between social platform users, as well as within whole online communities (Bruns and Burgess, 2011; Williamson and Ruming, 2015; Lycarião and dos Santos, 2017). Quite a few studies have likewise examined the use of hashtags, which, it is argued, make information dissemination faster and more effective than traditional media (Cheong and Cheong, 2011). Bruns and Moe (2014) distinguish three layers of information exchange within Twitter, where hashtags facilitate the most general, macrolevel. More recently there have been studies of how hashtagging styles relate to cultural values (Sheldon et al., 2019) and how hashtags function linguistically (De Cock and Pedraza, 2018).
Beyond hashtags, Gentzkow and Shapiro (2011) and Barberá (2015) study user segregation along ideological lines, while Puschmann (2015) examines the form and function of quoting in digital media. In another seminal piece on formal aspects of Twitter, Boyd et al. (2010) examined the practice of retweeting. Again using quantitative methods, Waterloo et al. (2018) have compared emotion norms on Twitter with those on other platforms. Similarly, Guntuku et al. (2019) examine what Twitter postings reveal about anxiety and depression.
There has also been some important qualitative research on Twitter content. Beyond hashtag use, Herring et al. (2004) began looking at lexical features of Twitter content. Similarly coming from linguistics, Zappavigna (2011) has studied the meta, interpersonal, and ideational features of Twitter content. Some scholars, focusing on performativity (e.g., Baym, 1995; Papacharissi, 2012), have examined speech acts without necessarily identifying them as such. Their concern, however, has been more with identity formation than political argumentation.
There has also been researching concerned with speech acts on Twitter. Hemphill and Roback (2014) in particular have examined the speech acts that constituents use to lobby Congress. Mostly, however, the study of speech acts on Twitter has so far come from outside of communication or even the central social sciences. Much of this research is trying to define speech acts in ways conveyable and recognizable by machine learning technologies (see, e.g, Vosoughi and Roy, 2016). That research began outside of Twitter study with the early effort of Cohen et al. (2004) to develop an algorithm for translating email into speech acts. Oraby et al. (2017) try the same for tweets related to customer service, which form a relatively confined linguistic module. Still using a computer-mediated methodology, Nemer (2016) studies the various speech acts employed by celebrities on Twitter, such as inquiries, requests, invitations, elaborations, and claims.
In their study of speech acts on Twitter, Zhang et al. (2012) probe deeper into speech act theory, trying to get machines to detect differences between assertive, directive, and expressive forms of utterance. These are already useful distinctions, but what makes this study even more significant is the corpus it utilized and the way it used the speech act basis to construct topic summaries. Drawing their Twitter data on six pre-selected topics in March 2011, they found, for example, that on the topic of a Japanese earthquake, statements, suggestions, and commissives dominate the content. While this breakdown is important and an accomplishment for machine learning, for the discipline of communication, there is a need to test the findings further on other topics and to go beyond them toward the study of argument on Twitter.
Speech Act Theory
Speech Act Theory is generally thought to have been originated by Searle (1970), who built on Austin (1962). Both draw from Wittgenstein’s (2009) notion of language games, which holds language to have more agency than just telling or describing. Austin coined the term performatives for utterances that did rather than described something. Examples would include promising, commanding, and authorizing. Searle expanded the idea, detailing what Habermas would go on to describe as validity claims, the presumptions governing speech acts making each what they are. Assertions, for example, presuppose that the asserter believes that what is asserted is true, making assertions that deny truth performative contradictions (Apel, 2003).
Searle’s work also popularized Austin’s (1962) distinction of three levels of speech act: locutionary; illocutionary; and perlocutionary. The locutionary level is simply the concrete performance of a speech act in some natural language. An example would be, “Please open the window.” The illocutionary level is the broad type of speech act performed, in the case of the foregoing example, a request. The perlocutionary level is the effect, in this case, compliance. In the case of assertions, the perlocutionary effect might be to inform or persuade. The perlocutionary effect of a joke might be laughter, although as can be argued (e.g., Davis et al., 2018), humor has its own ability to persuade, making it an important rhetorical device in its own right.
One of the interesting aspects of speech acts that makes their identification complicated is that their form and perlocutionary effects may be indirect (Green, 2014). In an example given by Austin (1962), when a bridge player bids three clubs, he indirectly informs his partner that he has no diamonds. Such subtle differences are difficult to catch by formal content analysis and require discourse analysis to be fully appreciated.
Arguments and enthymemes
Philosophically speaking, an argument encompasses a set of assertions, an assertion being one specific type of speech act that advances a claim to truth. In order to constitute a valid argument, the comprising assertions need to be deductively related, following the pattern of a syllogism of the form “if premises, then conclusion.” Formal arguments of this nature are prevalent in philosophy and scholarly discourse.
As Aristotle observed long ago, however, in popular discourse, arguments tend to be enthymematic rather than formal. An enthymeme is an argument in which some of the premises or even the conclusion are not explicitly stated but left implicit. Consider, for example, the simplest kind of argument that can be made, consisting of two premises and a conclusion:
If P then Q
An enthymeme would present the above argument with either premise (1) or (2), or even the conclusion (3) missing. To give a concrete example, consider the comment in the Chicago Sun-Times by the late Father Andrew Greeley about the impending attack on Iraq back in 2002.
So without proof of the seriousness or the imminence of an Iraqi attack…The United States may still stumble into a war that is evil and unjust and in which thousands and perhaps tens of thousands of people will die horrible deaths (Greeley, 2002, p. 20).
Essentially, Greeley is advocating against the attack by providing only a single premise without even the conclusion. He does not say there is no proof or that we should not go to war but presents only the implication that if there is no proof, then we will stumble into something evil and unjust. It is enough. Greeley does not even include the logically necessary premise that we should not do what is evil or unjust. The words evil and unjust are what are called thick descriptors with mini-arguments built into them, arguments saying we should not do what they characterize (Appiah, 2010).
Because the enthymematic form allows much to go unsaid, it allows arguments to be made even by singular speech acts of few characters. This makes even singular speech acts a potent way of expressing even moral arguments in limited space such as character-restricted tweets. Admittedly, it can be disputed whether enthymematic claims actually make valid arguments, but on the most natural read, they do.
Humor as a rhetorical device
Both Highfield (2016) and Davis et al. (2018) report that humor and irreverence are “core elements” of political discussion on Twitter. Our expectation, therefore, was that humor in general and sarcasm, in particular, would be prevalent rhetorically, used even to make moral points. Davis et al. go on to show that political humor on Twitter tends to serve three primary functions: Discrediting the opposition; establishing political subjectivity; and bolstering civic support.
Although no one really understands the essence of humor—if indeed there is one, as Kuipers (2011) observes, it is generally thought connected to the incongruity of some kind. Although Zhang and Liu (2014) go onto identify multiple other linguistic features associated with humor, they, like Highfield and Davis and Killen agree with Kuipers that incongruities lie at the center of much humor.
Our conjecture is that this element of incongruity is one reason why sarcasm is so prevalent a form of rhetorical humor and allows us to go beyond Davis et al. to identify some of the ways in which humor accomplishes the rhetorical task of discrediting. Sarcasm, for example, can be used as a humorous way of signaling various incongruities relating to intellectual bad faith such as hypocrisy (i.e., a lack of congruity between one’s self-claims and reality) or foolishness (what one opines vs. what should be opined). Simultaneously, humor can be employed to exclude, to create solidarity, and either to establish or level superiority. Leveling is connected to the benign violation theory of humor, which captures what is putatively humorous in pratfalls, as when someone dignified slips on a banana peel. Going as far back as the institution of court jesters (see Turner, 1969), the leveling function of humor as a political device is generally associated more with the left rather than the right (Dagnes, 2012), the latter expressing itself less often by humor than by outrage Young (2019).
Method and data
If as we have argued the basic constituents of argument are speech acts, then preliminary to any quantitative content analysis of argument on Twitter is a basic understanding of which speech acts we find there and how they might function argumentatively. Producing that preliminary understanding is the objective of this paper. Achieving it is a task of illustration rather than statistical representativeness. Thus, although we have collected a large data set of Tweets associated with the 2018 national midterm elections in the United States, for our purposes here, we focus on tweeted responses to initiating tweets from Fox News and its polar opposite, MSNBC as they covered the arrest of Cesar Sayoc for mailing pipe bombs to Trump critics.
How or why from our larger data set containing some 64,000 tweets directed at 36 different news sites, did we arrive at Fox and MSNBC and the case of Cesar Sayoc? Given the illustrative task of this paper, the choice could have been arbitrary, but in fact, it was not. Since our ultimate goal is to see how argument and argumentative form might vary across the political spectrum, it made sense to counterpoise Fox and MSNBC. As is well-known, Fox News is a conservative news site that has closely aligned itself with the Trump movement, so much so that during the Trump administration, it almost came to be considered the state channel. In contrast, even more to the left than CNN, MSNBC is kind of the anti-Fox, often criticizing Fox directly. Thus, if we wanted to see the contrast between left and right arguments, it made sense for a preliminary study to focus on these two networks. The Cesar Sayoc case also made sense as a focus as it was both particularly salient with the public and, we hypothesized, very likely to draw sharply contrasting arguments.
Although there were scattered comments before and renewed attention later, responsive discussion of the Sayoc case on Fox News was concentrated across 10 tweets Fox released on the subject between 4:50:42 p.m. on October 26 and 6:39:18 p.m. On MSNBC, the discussion was concentrated on seven tweets released by MSNBC between 3:15:15 p.m. of the same day and 3:33:39.
Speech acts can be categorized in varying ways. Although in this paper, using qualitative discourse analysis, we identify them more granularly, for reliable content analysis, broader categorization was necessary. Using such broader categorization, Table 1 identifies the distribution by type of the first speech act in the first 100 tweeted responses to Fox’s initial tweet on Sayoc that a male suspect had been arrested at 4:50:42 and, likewise, in the first speech act in all 42 tweeted responses to MSNBC’s tweet on “Major Response by FBI and other law enforcement” at 3:20:47. In terms of interrater reliability, an agreement was 88% with Cohen’s κ = 0.81. Expressives, which often appear in the form of assertions, was the kind of speech act for which agreement was most difficult. Although other kinds of speech acts also show up, as can be seen, in the initial speech acts of responsive tweets to both MSNBC and Fox, assertions dominated.
Table 2 treats individual responsive tweets as the unit of analysis. The results are at least suggestive for future research. It is remarkable, for example, how many of the tweets consist of no more than single speech acts—a few even only an image without anything said. Interrater agreement on this variable was again 88% with κ = 0.74. To the extent that such distribution proves representative, it becomes all the more important to discern what arguments might be conveyed by single speech acts. And even with so many tweets containing even just a single speech act, roughly two-thirds—64% on Fox and 72% on MSNBC—made some kind of argumentative point (again 88% interrater agreement, κ = 0.81).
A greater percentage of responses to Fox was accompanied by images. This was often because Tweeters were trying to show the van that they perceived Fox to be deliberately hiding. The relationship between text and image in tweets is an entirely overlooked dimension that should be a focused area of study in itself but is beyond the scope of this paper. We ourselves hope to address it in future research.
It is further remarkable that responses to Fox as to MSNBC were dominated by left-wing tweeters. Interrater agreement on political orientation was 84% with κ = 0.65, lower when controlling for chance because, as seen, the variable is so sharply skewed toward left-wing tweeters. Such being the case, we really were not able from these data to compare left and right styles of argumentation.
We also tried to code for humor, but, so far, our sensibilities differed too widely. As Nissenbaum and Shifman (2020) note, humor is a polysemic phenomenon particularly resistant to highly reliable identification, particularly in the form of sarcasm. We can still say, however, that humor characterized somewhere between 20% and 60% of tweets. Like Nissenbaum and Shifman, we attempted to identify the “butt” of any humor and can say that to the extent that one or the other coder could identify a butt, which was very often, the butt was almost always either Fox News (on the Fox site) or Trump, the Trump administration, or Trump followers.
Given that, like others (e.g., Baym, 1995; Davis et al., 2018; Zappavigna, 2011; Zhang et al., 2012), we quote directly from people’s tweets, we need to address the ethics of this practice. The guidelines of the Association of Internet Researchers (AOIR 2012) remain flexible to context, counseling common-sense avoidance of harm and respect for any privacy that might be expected. As Bolander and Locher (2014) observe, care is most required when dealing with vulnerable populations talking about personal matters. We take note of these considerations. In comparison with platforms like Facebook, it is generally agreed that privacy concerns are less applicable to Twitter, as messages are clearly intended to be publicly available to the entire Internet (Bruns et al., 2014). Admittedly, although we present no identifiers with the statements, enterprising individuals could still trace them back to the tweeters who submitted them. Still, as the statements presented do not come from a vulnerable population nor express any personal concerns but exclusively commentary on the newsfeed, the potential for harm is minimal.
Evidence of enthymematic argument
Within our data set, we found multiple examples of enthymematic effects. The first example follows the breaking news tweets from Fox over the police arrest of Cesar Sayoc, the New York man then suspected of sending pipe bombs to President Trump’s various critics. The earliest Fox tweet presented a video of Sayoc’s van, covered by a tarp, being towed by the police.
One of the responsive tweets begins with a speech act that could be described as an evaluative report: “Pretty good reporting today.” The tweet’s next speech act lends support for that assessment—“They give info as it is happening.” The final speech act of the tweet—“Fox didn’t show van or much else,” also a report could be a defensive response to a previous tweet.
In the tweet just previous to the one above, we see one instance of what could be an enthymematic argument. That tweet consists of a single speech act: “Fox is reporting without showing van.” Technically, this speech act is a report. But standing alone as it does, it can also function as a different kind of speech act—a complaint. As a complaint, the implicit message is that the van should have been shown uncovered, that Fox was deliberately not showing it. Per enthymematic form, the statement then functions as a mini-argument impugning Fox’s intellectual integrity: Full candor calls for the van being shown; the van is not being shown; therefore, Fox news is less than fully candid.
Why should the image of the van have been such an issue? The answer comes from two following tweets that present images of the van, accompanied by speech acts that could be classified as announcements:
“This is his van.”
“Here is the van in all its glory.”
“Picture that Fox will not show u.”
In all cases, what the accompanying image shows is a van, the side windows of which are completely covered with posters and stickers celebrating President Trump. As one other tweet puts it in an evaluative statement: “It’s a ferking shrine to Trump!” The implicit humor in that remark and its rhetorical effect we postpone discussing until a later section. We also leave aside the question of how all the referenced tweets and speech acts they include work in the context of embedded images or videos. That co-relation between image and text is one of the communicative strengths of Twitter.
For now, our focus is on the enthymematic nature of the tweeted comments bulleted above. Although the clearest case is the last bulleted announcement, all three could in context be considered as offering mini-arguments.
What is the context? We have to remember from Saussure how meaning is partially built from paradigmatic and syntagmatic relations. Whereas paradigmatic relations refer to presence vs. absence, i.e., to what is said as opposed to what could have been said but was not, syntagmatic relations refer to the meaning that arises from juxtaposition, as in the juxtaposition suggested above of text and image.
When we speak of context here, however, we speak of intertextuality (see Bakhtin, 1981). Tweets have a temporal order and hence a relation to one another. The context for one tweet, therefore, is the relevant tweets that came before. So in the case, we have been examining, previous questions raise a question concerning a paradigmatic matter—the absence of an image of the suspect’s van in Fox news reporting.
The question is whether the absence is innocent or bad faith. As none of the tweeters is privy to the minds of Fox executives, any answer can only be based on external evidence. Knowing the close relationship between Fox News and the Trump administration, it could be presumed that Fox has an interest in protecting Trump. That interest would suggest the possibility that Fox was deliberately refusing to show the van because of the implication that Trump himself had been a motivating cause of the suspect’s behavior. Such reasoning, whether right or wrong, is an implicit argument in support of which an image of the van with its “shrine” to Trump would be an evidentiary clincher. Actually, one tweet explicitly says just that: “Van had to be covered in tarps to hide the obscene numbers of pro-@realDonaldTrump bumper stickers and messaging.”
The point of this section has been simple but important: Much argumentative content can be and is conveyed on Twitter. Because of the way in which text and context work together, rather complex enthymematic argumentation can be and often is conveyed through brief, even singular speech acts.
Arguments without explicit or direct assertions
We saw in our previous section on speech act theory that locutionary acts may be indirect rather than direct, that one can, for example, request that a window be closed merely by asserting that one is cold. We should not be surprised therefore to find arguments conveyed enthymematically not merely by assertives but also by other kinds of speech acts such as expressives and interrogatives.
Even in the one case we examine, we find evidence of such a phenomenon. Some such tweets are questions:
“Why is the Van Covered?”
“Will arrest of guy, his van bedecked with right-wing stickers, end FOX News speculation of left-wing conspiracy on bomb mailings?”
“Any connection to GOP yet?”
And even after Fox evidently did start showing some images of the van, questions remained:
“Why are you blurring pictures of the van?”
“And just calling them “political bumper stickers”? It’s a ferking moving shrine to Trump.” (The full tweet from before)
“Still don’t think @realDonaldTrump’s vitriol doesn’t lead to terrorizing Americans? Cesar Sayoc proves otherwise. Words matter and the truth is important.”
A possibly implied answer to the first question on the first bulleted list above could be a reprise of the argument that Fox was being deliberately mendacious in its coverage. Contextually, however, given the placement of that particular tweet early in the queue of responses, it was more likely an actually non-rhetorical request for information. It was not Fox after all that placed the tarp on the van.
It is otherwise, however, for the other questions bulleted. They are rhetorical in nature, implying enthymematic arguments. For example, although an interrogative, the second question on the first bulleted list implicitly argues that this arrest of Sayoc and his van should put to rest Fox’s speculation about a left-wing conspiracy. And actually, the note of sarcasm in the question implies even more: a suggestion that that speculation was foolish or mendacious to begin with.
It is the same with the questions in the second bulleted list. By explicitly answering the questions asked, the replies presented in the final two bullets actually identify the implied arguments behind the questions, i.e., that Fox is not being intellectually honest.
Although we need to test it with a fully rigorous content analysis, our initial impression from our data is that Fox News attracts much more hostile replies than do other outlets. Thus, when on the same topic we turn to MSNBC, we find the same use of non-assertives to make arguments but with less suspicion directed at the news source itself. To an original MSNBC tweet reporting on the Sayoc arrest, some of the replies were the following:
“So the pipe bomb sender is a trumplican? That explains why some packages were sent to the wrong address, and why he was caught so quickly.”
“Why did they cover the Van? Do they Not want people to see all the Trump Garbage on it?”
“Why the tarp? Lol hmmm”
“What’s with the tarp? America has a right to see those (hundreds of) bumper stickers!”
What is displayed in the above list are the individual tweets in their entirety. The first thing to notice is how brief they are, well below the 280 characters the medium allows. This suggests there is little difficulty making a point in the space permitted.
The second thing to notice is that the interrogatives are in fact all making a point. Moreover, in all of the cases above, the tweets are structured as hypophoras, raising questions in order to immediately provide answers. By answering their own questions, authors make explicit the points or rhetorical thrust behind the questions. In the first tweet, the pejorative trumplican already suggests a disparaging view of the targets’ intelligence, on which the answer expands.
Similar to the responses to Fox, the remaining replies to MSNBC express suspicion, but in contrast with the former case, this suspicion is not directed at MSNBC. From the answers, the tweets themselves suggest that it is the current government of which the tweeters are suspicious.
Although as MSNBC tweets continued their news coverage, most responses were not interrogatives, we still find some such as the one below:
So the bomber is a Trumpazee with a van covered in Trump stickers from a town named, ‘Plantation’? All we need is Alanis Morissette singing Ironic in the background.
Again, we see the interrogative is making a point, explained by the accompanying answer: the irony of the suspect’s being a Trump follower, again designated by the use of a pejorative. Again, the target of the ire is not MSNBC but Trump and his movement. In the subsequent replies to MSNBC, we see that interrogatives are not the only non-assertives that can be used to make a point. Although the following tweets are all technically assertions, they are assertions that are also expressives—in this case expressions of the utterer’s attitude.
“I’m really curious about this fruitcake.”
“lol at them covering it with a tarp.”
“Gee, I never would’ve guessed this would end up with the arrest of a white male who drives a creepy kidnaper van covered in Trump stickers. Oh wait, actually that’s exactly how I pictured this to end!”
“I cannot wait to hear Trump’s comments or see his tweets on the arrest of a white male in his 50s Trump supporter in Florida in relation to the mail bombs sent to enemies of Trump.”
Although as noted all of the above commentary is expressed as assertions, what is asserted are not facts about the world but facts about the utterer’s state of mind. They thus function as expressives. Yet in conveying the author’s state of mind, some facts about the world are still implied. Thus, for example, the first tweet above suggests that Sayoc actually is a fruitcake. The second in laughing at the tarp suggests that the government was using it deliberately in embarrassment at Sayoc’s shrine to Trump. The adjectives in the third tweet suggest that the creepiness of the van and its kidnaper nature are not just in the tweeter’s mind but beyond dispute. Finally, the last tweeter’s eagerness to hear from Trump suggests that Trump at least ought to be embarrassed.
But in this case, in response to MSNBC, we find points being made not just by direct assertions, interrogatives, and expressives, but also by commands and calls. Consider the two tweets below.
“Lock him up! Lock him up! Lock him up!”
“Trump NEEDS to apologize to America for this! He caused it with his rhetoric everyday!”
The first tweet above is technically a command or plea. Repeated as it is, it is obviously meant to echo the chant that Trump’s former national security advisor, Michael Flynn, led against Hilary Clinton. The rhetorical effect of the entire locutionary act is to turn the rhetoric of the Trump movement back on itself, to suggest that it is its proponents, like Michael Flynn, then locked up himself, who should be imprisoned.
The first sentence of the second tweet is again technically an assertion but one that functions as a call. By saying that Trump needs to apologize, it is calling for him to do so. And if the premise of that call is that Trump needs to apologize, there must be something for him to apologize for. Thus, a call for an apology is simultaneously an accusation of something requiring it. That is why just saying an apology is needed is simultaneously calling for that apology. The second sentence of the tweet again makes explicit the implied grievance behind the call.
If the point of the previous section was that entire arguments can be made enthymematically by brief, even singular speech acts, the point of this section is that those speech acts need not be simple or direct assertions. They might be questions, calls, expressives or commands as well. We now turn to the rhetorical use of humor on Twitter.
Humor and morality
It is not unexpected to find humor frequently deployed on Twitter. Again, determining just how frequent requires a formal, quantitative content analysis, which in turn requires a reliable way to identify it, which, as we have seen, is not so easy. That humor is frequent, however, is indicated by how many of the tweets we reviewed previously for other reasons actually trade on humor.
Humor is ultimately in the eye of the beholder, its appreciation in part dependent on whose ox is gored. We are much more likely to be amused when the butt of the joke is not a friend – or ourselves – but an enemy. What is significant, however, is that humor can be used rhetorically to target enemies.
As explained in our previous section on humor, humor’s ability to serve as a rhetorical weapon stems from the way in which much humor seems to work, that is by playing on differences from expectation. The expectation can be of different kinds. Formal jokes work by leading us to expect one thing and delivering another. In more informal usage, Merriam-Webster defines irony as “the use of words that mean the opposite of what you really think especially in order to be funny.” Irony is often invoked in sarcasm to highlight an opponent’s departure from a norm. The norm may be a convention of rationality such as consistency, but even then the departure also carries moral freight, for to accuse another side of condemning what it allows itself is to call it hypocritical, which is a term of moral opprobrium. To suggest that a news report is less forthcoming than it should be is similarly not just to highlight an epistemic lapse but also to accuse the report of being disingenuous or less than honest, which are again moral faults. In such cases, we find humor and morality closely tied.
There is, however, an additional reason to expect to find sarcastic humor frequent on Twitter. In the liminal state of communitas (see Turner, 1969) momentarily created by mirth, humor binds those in on the joke against the target. Humor is thus a community-building device that unites users across the Twitterverse.
We return to some of the tweets we have already examined to observe now their humorous and moral features, but as a baseline, it is useful to look at a tweet that makes a moral point without humor. Consider again the following tweeted response to the arrest of Cesar Sayoc and his van:
“Trump NEEDS to apologize to America for this! He caused it with his rhetoric everyday!”
The first thing to note is that there is no apparent humor in this tweet. As we previously observed, the tweet is a straight-forward call for Trump to apologize. Apologies are a major form of moral interaction (Tavuchis, 1991). Today, calls for them are major ways in which opponents accuse each other of moral infractions. Why is that? Again, as Merriam-Webster defines it, an apology is a regretful admission of error or wrong-doing. Thus, if an apology is deemed necessary, it is because the party needing to apologize is considered to have erred or done something wrong. The magnitude of the error or wrong-doing can vary. A simple lapse like missing an appointment is usually not morally grave, amounting perhaps just to bad etiquette. At the opposite extreme, responsibility for someone’s engagement in terrorist acts would seem to be great moral culpability. As the above tweet goes onto make explicit, it is of such moral culpability that the tweet enthymematically accuses Trump just by citing his need to apologize. There is in other words a mini-moral argument being made just by that singular speech act. Let us contrast that tweet now with others in which humor is in play.
“So the bomber is a Trumpazee with a van covered in Trump stickers from a town named, ‘Plantation’? All we need is Alanis Morissette singing Ironic in the background.”
“Gee, I never would’ve guessed this would end up with the arrest of a white male who drives a creepy kidnaper van covered in Trump stickers. Oh wait, actually that’s exactly how I pictured this to end!”
“Will arrest of guy, his van bedecked with right-wing stickers, end FOX News speculation of left wing conspiracy on bomb mailings?”
“And just calling them “political bumper stickers”? It’s a ferking moving shrine to Trump.”
“Why the tarp? Lol hmmm”
“What’s with the tarp? America has a right to see those (hundreds of) bumper stickers!”
“Lock him up! Lock him up! Lock him up!”
None of these tweets will have you rolling on the floor with laughter, but they do seem at least intended to be at least amusing to a politically left audience. It is in fact striking that all of these tweets with humorous aspects do come from the left. Again, it would take a formal, quantitative analysis to confirm more definitively, but our early results here do provide initial support for how as a rhetorical device, humor is deployed more by the left than the right (again see Dagnes, 2012; Young, 2019).
But are the above tweets humorous? We can only comment on the features that may make them so. One low form of humor is simple, derogatory name-calling. Trump himself is a master of the practice, and the first tweet above returns the jab with the derogatory “Trumpazee.” Its opening “So…” sets us up for humor by echoing the famous Willy Wonka meme (see Richmond, 2019). The subsequent comment, which explicitly mentions irony, is sardonic, which Merriam-Webster tells us means disdainfully humorous or mocking, which in turn means to ridicule, which in turn means “to make fun of.”
For the second tweet above, it is helpful to begin again by consulting Merriam-Webster on the word “gee.” That dictionary tells us the word is an interjection connoting enthusiasm or surprise. Beginning with “gee” and followed by “I never would have guessed,” the second tweet sets us up for surprise, which already involves the potentially humorous element of incongruity, in this case between what we might expect and what we actually receive. But the humor of the statement lies in its sarcasm, which rests on a different, epistemic incongruity between what was presumably expected by Trump and his supporters and what they should in fact rationally have expected. Lest we miss the irony, the following statement drives it home, again in a way with the initiating word wait that is intended to be humorous.
Is the third tweet humorous? It is perhaps the most borderline on the list and worth exploring for that very reason. It is reflective of what might make humor difficult to detect reliably by a formal content analysis. On one interpretation, the tweet can be read as a straightforward and reasonable question.
What might lend the question an element of humor? To left-wing readers, Fox News is ipso facto a target of humor as is what left-wingers consider its ridiculous conspiracy theory that the left was orchestrating the attacks on its own. In other words, for a left-wing audience, there is an element of humor with its community-building function just in mentioning, let alone questioning Fox and its theory. Some of the wording too adds to the appearance of humor— bedecked, right-wing, and conspiracy are all words meant to intensify the disparagement of the target. For all that, as a case of humor, the tweet remains borderline.
Arguably, the next tweet is much more clearly humorous, trading as it does on the incongruity between the understated “political bumper stickers” and the manifest reality that the tweeter describes as a veritable “shrine.” The use of the word ferking is doubly funny. First, the invented word is an amusing way to evade the censors, and second the inclusion of the actual word in such syntax is always a humorous intensifier. Although arguably quite apt, the word shrine for the bumper stickers almost qualifies as a comic exaggeration for effect.
The next two tweets may be borderline too, but asking “Why the tarp?” In that flippant manner suggests jockularity. That suggestion is confirmed by the “lol” that follows. Although it is not followed by anything so clearly suggesting jockularity, the next text’s “What’s with…” is also of a flippant form, suggesting a snide critique, which does in fact follow.
Is the “Lock him up..” funny? Again, trading on the incongruity crucial to humor, the tweet is funny to the extent that it deploys the ability to use the opposition’s own offensive line against it. As that line was morally offensive in the first place, the tweet could also be considered to be making a moral point, but in this case, both the moral and humorous aspects are rather implicit.
Despite all that has already been written about Twitter, there has been scant attention to the nature of political argument we find there. The major purpose of this study has been to lay the theoretical groundwork for approach to such study. Thus, although this paper began with a number of theoretical questions, most basic was whether there is political argument on Twitter and how we recognize it.
Argument is a specific form of qualitative content, but again relatively little of the vast literature on Twitter actually examines qualitative content. Theoretically, we thus had to start more or less from scratch. We began with a basic understanding of formal argument as a series of logically connected assertions that result deductively in a conclusion. We took note, however, of what rhetoricians have been telling us since Aristotle that in popular discourse, the argument is much more enthymematic, with logical links—sometimes even the conclusion—left unsaid. And given what are called thick descriptors like “unjust,” which have mini-arguments built into them, sometimes even a single speech act or even a simple word can suffice as an implicit argument.
Our aim was thus to examine the kinds of speech acts to be found on Twitter and the kinds of arguments they might support. At this stage of inquiry, we focused on single speech acts. We drew on the five most basic speech act types and were able to show that they can be reliably identified in tweets and even counted.
Although at this stage, our investigation was exploratory, even our quantitative findings are very suggestive for future research. We found, for example, that character space on Twitter may be largely underused. Half of the tweets we examined contained no more than a single speech act. Thus, although arguments are being made on Twitter, they may often be of a very simple nature. Further study is needed to determine how general this pattern is.
We further found that while assertions dominate as tweet openings, other speech acts—interrogatives, expressives, directives and declaratives—are also to be found. And we found that even when tweets consisted of single speech acts, they more often than not expressed some kind of argumentative point. In the cases we examined, the point was generally a left-wing critique of the right. The domination of left-wing critics even on Fox was something of a surprise, and another question for future research is whether this pattern holds generally on Fox and other right-wing news outlets.
The major thrust of our paper was, however, qualitative in nature. The aim was to show qualitatively the different ways that different kinds of speech acts can and do make argumentative points. We were able to show that due to conversational implicature, the types of speech acts making argumentative points encompass more than just assertions. Interrogatives, expressives, and directives are all devices through which rhetorical points can be made.
There are two workhorses behind this power to say more than what is actually said. First are thick descriptors that have mini-arguments built into them. Second is the enthymematic nature of informal discourse, which lets implication carry much of the load. That power at work is what we mostly observed in this paper.
The power of implication applies not just to words, but also, although we did not explore it here, to images. Images too can function rhetorically. As when someone says—rightly or wrongly—“here is the van that Fox will not show you,” a shown van is a disclosure, a premise in an implied argument. Images of cartoons might similarly function as analogies or suggestions, as in one case in our corpus an image of Bugs Bunny sawing Florida off from the United States. And of course the power of enthymemes will have purchase even beyond Twitter. One such natural application would be to memes as studied, for example, by Shifman (2014), who, like Richmond (2019) as well, shows that whatever arguments memes might enthymematically convey, they certainly apply to politics.
Humor too we found to be a frequent rhetorical device, even to make points of an important moral nature. Although we could not yet code it formally with enough reliability to count, we can see from our qualitative analysis that the butt of much humor is not just a personal target like Trump or Fox News but also a particular offense. As in the case above about what Fox would putatively not show, a very frequent offense in our corpus was intellectual dishonesty or hypocrisy, a failure in some way to weigh the evidence fairly or give the other side its due. It is an important finding because sincerity or honesty is not among the moral foundations listed by the now very influential Moral Foundations Theory (see Haidt, 2012). If reliable identification can be attained, a quantitative study of the moral points made by tweeters would thus be one more very fruitful line of future research.
All data analyzed in this study are included in this published article.
Apel KO (2003) The response of discourse ethics to the moral challenge of the human situation as such and especially today. Peeters, Belgium
Appiah K (2010) Experiments in ethics. Harvard University Press, Cambridge
Austin JL (1962) How to do things with words. Oxford University Press, Oxford
Bakhtin M (1981) The dialogic imagination. University of Texas Press, Austin
Barberá P (2015) Birds of the same feather tweet together: Bayesian ideal point estimation using Twitter data. Pol Anal 23(1):76–91
Baym NK (1995) The performance of humor in computer-mediated communication. J Comp-Mediat Com 1(2):JCMC123
Bolander B, Locher MA(2014) Doing sociolinguistic research on computer-mediated data: A review of four methodological issues Discourse Context Media 3:14–26
Boyd D, Golder S, Lotan, G (2010) Tweet, tweet, retweet: Conversational aspects of retweeting on twitter. Paper presented at the 2010 43rd Hawaii international conference on systems science
Bruns A, Moe H (2014) Structural layers of communication on Twitter. In: Twitter and society, vol 89. Peter Lang, pp. 15–28
Bruns A, Weller K, Zimmer M, Proferes NJ (2014) A topology of Twitter research: disciplines, methods, and ethics. Aslib J Inf Manag 66 (3):250–261
Bruns, Axel & Burgess, Jean (2011) The use of Twitter hashtags in the formation of ad hoc publics. In Bruns, A & De Wilde, P (Eds.) Proceedings of the 6th European Consortium for Political Research (ECPR) General Conference 2011. The European Consortium for Political Research (ECPR), United Kingdom, pp. 1–9
Cheong F, Cheong C (2011) Social media data mining: a social network analysis of tweets during the Australian 2010–2011 floods. Paper presented at the 15th Pacific Asia Conference on Information Systems (PACIS)
De Cock B, Pedraza AP (2018) From expressing solidarity to mocking on Twitter: pragmatic functions of hashtags starting with #jesuis across languages. Lang Soc 47(2):197–217
Cohen WW, Carvalho VR, Mitchell TM (2004) Learning to classify email into “speech acts”. In: Proceedings of the 2004 conference on empirical methods in natural languange processing, pp. 309–316
Dagnes A (2012) A conservative walks into a bar: the politics of political humor. Palgrave Macmillan, New York
Davis JL, Love TP, Killen G (2018) Seriously funny: the political work of humor on social media. New Media Soc 20(10):3898–3916
Gentzkow M, Shapiro JM (2011) Ideological segregation online and offline. Q J Econ 26(4):1799–1839
Greeley A (2002) Hawks offer no proof to justify attack. Chicago Sun-Times 20
Green M (2014) Speech acts. Stanford Enc Phil (Winter 2017 Edition), Edward N. Zalta (ed.), https://plato.stanford.edu/archives/win2017/entries/speech-acts/
Guntuku SC, Preotiuc-Pietro D, Eichstaedt JC, Ungar LH (2019) What twitter profile and posted images reveal about depression and anxiety. In: Proceedings of the international AAAI conference on web and social media, PKP publishing services network. vol. 13, pp. 236–246. Retrieved from https://ojs.aaai.org/index.php/ICWSM/article/view/3225.
Haidt J (2012) The righteous mind: why good people are divided by politics and religion. Vintage, New York
Hemphill L, Roback AJ (2014) Tweet acts: how constituents lobby congress via Twitter. In: S. Fussell and W. Lutters (Program Chairs) Proceedings of the 17th ACM conference on computer supported cooperative work & social computing, Association for Computing Machinery, New York, pp. 1200–1210
Herring SC, Barab S, Kling R, Gray J (2004) An approach to researching online behavior. In: SA Barab, R. Kling, JH Gray (Eds.), Designing for Virtual Communities in the Service of Learning. Cambridge University Press, New York, p. 338
Highfield T (2016) News via Voldemort: parody accounts in topical discussions on Twitter. N Media Soc 18(9):2028–2045
Karami A, Lundy M, Webb F, Dwivedi YK (2020) Twitter and research: a systematic literature review through text mining. IEEE Access 8:67698–67717
Kuipers G (2011) Good humor, bad taste: a sociology of the joke, vol 7. Walter de Gruyter, New York
Lycarião D, dos Santos MA (2017) Bridging semantic and social network analyses: the case of the hashtag# precisamosfalarsobreaborto (we need to talk about abortion) on Twitter. Inf Commun Soc 20(3):368–385
Nemer D (2016) Celebrities acting up: a speech act analysis in tweets of famous people. Soc Netw 5(1):1
Nissenbaum A, Shifman L (2020) Laughing alone, together: local user-generated satirical responses to a global event. Inf Commun Soc 1–18, Retrieved online. https://doi.org/10.1080/1369118X.2020.1804979.
Oraby S, Gundecha P, Mahmud J, Bhuiyan M, Akkiraju R (2017) How May I Help You?: modeling Twitter customer service conversations using fine-grained dialogue acts. In: Papadopoulos GA and Kuflik T (Program Chairs) Proceedings of 22nd international conference on intelligent user interfaces. ACM, pp. 343–355
Papacharissi Z (2012) Without you, I’m nothing: performances of the self on Twitter. Int J Commun 6:18
Puschmann C (2015) The form and function of quoting in digital media. Discourse Context Media 7:28–36
Richmond JC (2019) Pepe for president: a critical examination of memes as political discourse and humor. Unpublished doctoral dissertation, Drexel University
Romero DM, Meeder B, Kleinberg J (2011) Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on Twitter. In: S. Sadagopan and Krithi Ramamritham (Program Chairs) Proceedings 20th international conference on world wide web. ACM, pp. 695–704
Searle J (1970) Speech acts: An essay in the philosophy of language. Cambridge University Press, Cambridge
Sheldon P, Herzfeldt E, Rauschnabel PA (2019) Culture and social media: the relationship between cultural values and hashtagging styles. Behav Inf Technol 1–13
Shifman L (2014) Memes: In D\digital culture. MIT Press, Cambridge
Tavuchis N (1991) Mea culpa: a sociology of apology and reconciliation. Stanford University Press, Stanford
Turner V (1969) Liminality and communitas. The ritual process: structure and anti-structure, vol 94. Routledge, New York, p. 193
Vosoughi S, Roy D (2016) Tweet acts: a speech act classifier for twitter. In: Tenth international AAAI conference on web and social media, PKY network services. vol 10 (1). Retrieved from https://ojs.aaai.org/index.php/ICWSM/article/view/14821.
Waterloo SF, Baumgartner SE, Peter J, Valkenburg PM (2018) Norms of online expressions of emotion: comparing Facebook, Twitter, Instagram, and WhatsApp. New Media Soc 20(5):1813–1831
Williamson W, Ruming K (2015) Who’s talking, who’s listening: exploring social media use by community groups using social network analysis. Paper presented at the 14th International Conference on Comps in Urban Planning and Urban Management. Massachusetts Institute of Technology, Cambridge
Wittgenstein L (2009) Philosophical investigations. Wiley-Blackwell, New York
Young D (2019) Irony and outrage: the polarized landscape of rage, fear, and laughter in the United States. Oxford University Press, New York
Zappavigna M (2011) Ambient affiliation: a linguistic perspective on Twitter. New Media Soc 13(5):788–806
Zhang R, Li W, Gao D, Ouyang Y (2012) Automatic twitter topic summarization with speech acts. IEEE Trans Audio Speech Lang Process 21(3):649–658
Zhang, Liu N (2014) Recognizing humor on Twitter. In: CIKM 14 Proceedings of the 23rd ACM International Conference on Information and Knowledge Management, Shanghai, China, November 3–7, 2014. pp. 889–898
Zimbra D, Abbasi A, Zeng D, Chen H (2018) The state-of-the-art in Twitter sentiment analysis: a review and benchmark evaluation. ACM Trans Manag Inf Syst 9(2):1–29
The authors are grateful for the comments on earlier drafts of two anonymous reviewers.
The authors declare no competing interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Elliott-Maksymowicz, K., Nikolaev, A. & Porpora, D. How much can you say in a tweet? An approach to political argumentation on Twitter. Humanit Soc Sci Commun 8, 118 (2021). https://doi.org/10.1057/s41599-021-00794-x
This article is cited by
Characterization Frames Constructing Endoxa in Activists’ Discourse About the Public Controversy Surrounding Fashion Sustainability
Twitter Activists’ Argumentation Through Subdiscussions: Theory, Method and Illustration of the Controversy Surrounding Sustainable Fashion