1This study explores the connection between salience and sarcasm through analyzing the usage patterns of sarcastic interrogatives in unscripted spoken-discourse and written web-discourse within three different corpora of American-English. We focus on studying the degree of salience of these sarcastic questions in relation to the different corpus types and to the different question types. Through a large and diverse database, we reach new conclusions regarding the relationship between salience and sarcasm, based on discursive, situational, and syntactic components.
2Salience and sarcasm have been extensively studied, both together and separately. Previous research on sarcasm has often relied on the notion of salience, as exemplified in the Graded Salience Hypothesis (Giora, 1997) and The Defaultness Hypothesis (Giora, Givoni & Fein, 2015). However, the precise relationship between sarcasm and salience remains largely unexplored. Research is limited on constructions that can convey both sarcastic and literal meanings, particularly in distinguishing the factors that render sarcastic meanings more salient than their non-sarcastic, literal counterparts. Moreover, while the connection between these constructions and salience has primarily focused on the role of context in highlighting sarcastic interpretations, other influencing factors have been largely overlooked. Consequently, there is a significant gap in the literature that fails to address the comprehensive relationship between sarcasm and salience, particularly concerning factors other than context. Moving forward, it is essential to investigate these dynamics more thoroughly, exploring how not only context, but also discursive environments, and linguistic structure contribute to the salience of sarcasm in unscripted language.
3The present study aims to address these previous gaps in the literature by determining how the notion of salience features in specific interrogative constructions – WH-questions followed by a self-answer, alternative questions, and disjunctive polar questions, and how it contributes to their sarcastic expression compared with the literal one. This is done through extensive corpus research, comparing the distribution of different sarcastic questions across three distinct corpora.
4The structure of the paper is as follows: the second section presents various definitions of sarcasm and compares them, in order to clarify our use of the term. The third section briefly reviews previous research about sarcasm and salience. The fourth and fifth sections present our two main research questions and methodology. Each research question will be followed by results and a discussion. We conclude with a general discussion and suggested directions for future research.
5In this section, we address sarcasm and salience, two terms that are often used in linguistics, yet are frequently defined loosely or inadequately.
6The notion of salience deserves special attention as it is crucial to this study, particularly in its connection to sarcasm. Furthermore, despite its frequent application in linguistics, it is often considered self-explanatory, therefore commonly used without a proper definition (Boswijk & Coler, 2020). Adding to this lack of precision, the definitions of salience vary significantly across different linguistic sub-fields and theoretical frameworks.
7Many of these variations converge on the fact that salience is two-faced. On the one hand, cues that are marked, unexpected or surprising are salient, because they stand out in their surroundings. On the other hand, there are cues that are used frequently, making it easier to access and process them, and therefore rendering them more salient (Boswijk & Coler, 2020). Indeed, in psycho-linguistics studies, salience is often discussed in terms of accessibility due to high frequency of use (Giora, 2003; Cieślicka, 2006). The present study utilized this latter definition of salience. We argue that the higher frequency of constructions used sarcastically in some corpora renders their sarcastic interpretation more salient, and thus makes them more predictable and accessible. Along with this key factor, salience can also be affected by other existing expectations triggered by an interaction of linguistic, situational, and societal context (Schmid & Günther, 2016).
8The second definition that requires clarification is that of sarcasm, which is often characterized as a form of verbal irony, signifying the opposite of what is said, while conveying humor, ridicule, and criticism (Camp, 2011). However, we find this definition to be inaccurate and insufficient, since many utterances are sarcastic without signifying the exact opposite meaning of their literal content, as can be seen in the following example, adapted from Michaelis & Feng (2015:3):
(1) Someone cries disproportionally about a paper cut. Someone else looks at the trivial injury, and says the following utterance: “You’ll probably live”.
9This utterance is sarcastic, but it does not follow the aforementioned definition of verbal irony, since the utterance “you’ll probably live” does not convey the exact opposite of its literal content (you will surely die). Instead, it is an understatement and a rebuke, aimed to communicate to the injured person that their injury is frivolous (Michaelis & Feng, 2015:3).
10This is just one example of why the traditional, somewhat simplistic, definition of sarcasm is insufficient. A second issue is that not all utterances have an “exact opposite.” Notably, non-assertive utterances, such as questions, which are the construction discussed in this project. Therefore, another, more accurate and fine-tuned definition, is required.
11Following Camp (2011), we define sarcasm as a scalar meaning inversion which communicates some transposition of a pretended commitment. While pretending to make an assertion, the speaker actually assumes an evaluative standard and implicates that it has been violated negatively. In a sarcastic utterance, the speaker exploits a mutually shared assumption that they could not plausibly mean what they have really said (Michaelis & Feng, 2015), and not that what they said is the exact opposite of what they meant.
12Sarcasm is an indirect rebuke which causes the interpreter to reconstruct a scale and a presupposition (Michaelis & Feng, 2015). We can illustrate this definition with the help of example (1) – “you’ll probably live”. The scale reconstructed by the interpreter is based on the ordering of injuries by severity. The reconstructed presupposition is one concerning the circumstance in which one should be worried about the death of an injured person. Crucially, for the present case, probably not when they get a paper cut. That is, the sarcastic utterance establishes the scale necessary for evaluating the situation and signals that it is a direct opposite of a situation which allows for a sincere utterance.
13Sarcasm shows disparity between the current situation and whatever circumstances warranting a sincere speech act (Michaelis & Feng, 2015). The speaker pretends to make an assertion, while building on some presupposed standard of evaluation, implicating that it has been violated negatively (Camp, 2011). Sarcasm is a sophisticated linguistic phenomenon, conveying contempt implicitly and humorously (Savini & Caragea, 2022) while showing the speaker’s dissociation from others’ opinions (Lozano & Ruiz de Mendoza, 2022). Sarcasm is important since it can help in analyzing people’s sentiments, opinions, behaviors (Savini & Caragea, 2022), and linguistic patterns.
14Some constructions are particularly sensitive to sarcasm because they contain elements that lend themselves effectively to meaning inversion (which is the basis for sarcasm). One such construction is questions, which are the focus of our current research. The conventionalized function of a question as a speech act is to ask for information (Sadock & Zwicky, 1985). Therefore, questions usually invite some sort of feedback or answer (Michaelis & Feng, 2015).
15However, sarcastic questions are not real, genuine, questions, but a counterfeit form of pretense (Michaelis & Feng, 2015), an illocutionary speech act which expresses the speaker’s attitude as opposite of sincere (Camp, 2011). Questions are an efficient tool to convey sarcasm because under the pretense of asking for information, the sarcastic speaker performs meaning inversion, thus indirectly showing criticism and ridicule (ibid). Unlike conventional, information-seeking questions, sarcastic questions do not advance the conversation, because there is not much the addressee can say in response to one. The addressee of a sarcastic utterance detects the presupposition of the speech act, recognizes the discrepancy between the situation in the presupposition and the current one, thus inferring the negative assessment. According to this view, sarcastic speech acts are conventions of usage, and there are some grammatical patterns to which sarcastic interpretations can attach directly (Michaelis & Feng, 2015).
16In the next section, we will briefly review previous research concerning the interpretation of sarcasm, especially regarding salience.
17Sarcasm has been extensively studied for many years, primarily as a form of irony or as its subset, rather than as a distinct linguistic phenomenon. According to the standard pragmatic view (Grice, 1975), listeners must first derive the complete literal meaning of an utterance before any pragmatic information is considered to infer the additional, nonliteral messages. This is also true for indirect and figurative meanings, such as sarcasm. This standard pragmatic model has been disputed, both theoretically and empirically. Instead, subsequent psycholinguistics studies stressed the effect of context on selective access (Giora, 1999), arguing that in a rich context, addressees can access nonliteral meanings directly, without having to process the literal meaning at all. This approach is known as the direct access view (Gibbs, 1994), stating that addressees often directly understand what speakers intend to communicate when using figurative or non-literal language without having to process the literal meanings of speakers’ utterances first (Gibbs, 2002). Both these approaches addressed all non-literal and figurative language in general (e.g., metaphors, irony, idioms, and proverbs), and did not focus specifically on sarcasm.
18Later approaches targeted sarcasm specifically, using the notion of salience in order to explain the comprehension of sarcastic utterances. The best known of these approaches is the Graded Salience Hypothesis (Giora, 1997, 1999), which was followed by the Defaultness Hypothesis (Giora et al., 2015). The Graded Salience Hypothesis (Giora, 1997, 2003; Givoni & Giora, 2018) distinguishes between default meanings and default interpretations. Default meanings are salient meanings listed in the mental lexicon, ranking high on prominence due to cognitive factors (such as prototypicality or stereotypicality) or degree of exposure (e.g., experiential familiarity, conventionality, or frequency of the stimulus). When encountering a stimulus, salient meanings will spring to mind automatically, regardless of degree of figurativeness or contextual information (Levant, Fein, & Giora, 2020). According to the Graded Salience Hypothesis, default interpretations are based on the coded, salient meaning of their stimulus components. Being salience-based, their derivation is unconditional (Giora et al., 2007).
19Diverging from the Graded Salience Hypothesis, the Defaultness Hypothesis (Giora et al., 2015) focuses on interpretations, thus allowing non-salient, non-literal interpretations to be considered as default, once they are activated automatically, while meeting the conditions for defaultness (Levant et al., 2020). According to the Defaultness Hypothesis, phrasal patterns strongly weaken highly positive concepts, generate sarcastic interpretations by default, i.e., automatically, without relying on supportive context or on internal incongruity.
20The earlier approaches (the standard pragmatic view of Grice; the direct access view, Gibbs, 1994), didn’t refer to sarcasm specifically, but to all non-literal language in general. Even when later approaches (the Graded Salience Hypothesis, Giora, 1997; the Defaultness Hypothesis, Giora et. al., 2015) did address sarcasm directly, they treated it as synonymous to verbal irony, which, we argue, is an insufficient definition. Furthermore, most of the previous studies focused on sarcasm in general, and not on specific syntactic constructions. Studies that did focus on constructions tend to only explore sarcastic ones, and do not analyze those that can convey both sarcastic and literal meanings, distinguishing what supports their sarcastic interpretation over the literal one.
21There is even less research on the connection between such constructions and salience, since most of the research regarding sarcasm and salience focuses on context dependency. Also, as far as we know, there is no research comparing the usage of sarcasm specifically (and not as a form or subset of irony) in different discursive functions, arenas, and situations (regardless of its connection to salience). Therefore, the encoding of sarcasm and its relation to salience remains largely unexplored.
22The present study conducts a corpus-based analysis of three interrogative constructions capable of conveying both literal and sarcastic meanings. First, we assess the frequency of sarcastic versus non-sarcastic interpretations of these interrogatives across various corpora. Then, we investigate the occurrence rates of sarcastic interpretations among different types of interrogative constructions.
23We focused on three question types: alternative questions, disjunctive polar questions, and WH-questions followed by a self-answer (to which we refer as split interrogatives).
24AQs are closed questions (Huddleston & Pullum, 2002) asked to narrow down the answer possibilities into two (or more) distinctly mentioned alternatives (Biezma & Rawlins, 2015). The alternatives themselves are presented by the disjunctive coordinator or, which is a necessary feature of AQs. The content of an AQ is a disjunction of propositions such that each of the propositions delivers the content of one of the answers (Huddleston & Pullum, 2002).
25Answering an AQ usually consists in choosing one, several, all, or none of the suggested alternatives. AQs can be either sarcastic or literal, as shown in the following examples (all taken from the corpora research of the present study):
(2) Are you far-sighted or near-sighted? (literal AQ, the COCA corpus)
(3) Are you really a teacher or do you just hold the chalk and wipe the blackboard? (sarcastic AQ, the NLDS corpus)
26(2) is a genuine question – the speaker is interested in knowing whether the addressee is far-sighted or near-sighted. In (3), however, the speaker is not trying to receive information. Typically, it is assumed as part of the common ground that a teacher does more than just the physical action of cleaning the blackboard. Therefore, the question revokes this presupposition, and creates clear meaning inversion, which leads to sarcasm. The presupposition is made salient to form a question biased towards a negative answer.
27Consider another example of a sarcastic AQ:
(4) Is this a question or a lecture? (the COCA corpus)
28In this instance, the interrogative serves a sarcastic function, carrying an implicit rebuke and challenging the presupposition that a clear distinction exists between a question and a lecture. Typically, a question invites the speaker to pause and await a response, whereas a lecture involves continuous speech without pauses for listener feedback. This question undermines such presuppositions by suggesting that what appears to be a question does not truly seek information, but rather functions as a lengthy monologue designed to express the speaker’s opinions. This confrontation with common conversational norms highlights the question’s salient, non-literal, sarcastic nature.
29DPQs are closed questions to which the answers are usually a pair of polar opposites (Huddleston & Pullum, 2002) – a positive (yes) and a negative (no). In DPQs, the presence of the or disjunctive conjunction combines two (or more) disjuncts that denote a set of alternatives. The propositional content of one answer is expressed in the question itself, and the other is accessed by polarity-reversal (ibid). Therefore, answering a DPQ usually requires choosing between the set of explicit alternatives and the other alternatives that are available from context, though the disjunction itself implies that there is a singleton alternative (Biezma & Rawlins, 2015). Usually, the answer to a DPQ will be “yes” or “no”, verbally or by other indirect means (“uh-huh”, “mm”, head nods, etc.), or by repetitions of some or all of the DPQ (Enfield et al., 2019).
30Like AQs, DPQs can also be sincere or sarcastic, as in the following examples:
(5) Have you any links or book references to this? (literal DPQ, the NLDS corpus)
(6) Is this ancient Greece or something? (sarcastic DPQ, the NLDS corpus)
31In (5), a literal DPQ, the speaker asks the addressee to choose between the set of proposed alternatives – links and books, and an implicit alternative – nothing. A positive “yes” answer means that the addressee has links or references (perhaps only one of those) while a negative answer means that they do not have any.
32In (6), a sarcastic DPQ, the addressee is not required to choose between “ancient Greece or something” and another alternative. Instead, the meaning of the question is reliant on a presupposition – that one cannot go back in time, and therefore, this could not possibly be ancient Greece. This insinuates that the answer must be a negative one, since the positive one is impossible. The presupposition is made salient to form a question biased towards a negative answer.
33Consider another example of a sarcastic DPQ:
(7) Do you think that children should pack guns at 5 years of age or maybe younger? (the NLDS corpus)
34Like question (6), this question does not offer a genuine choice between children carrying guns at the age of five or at an even younger age. It draws on the widespread presupposition that it is inappropriate for children to handle firearms. Thus, the notion of arming children as young as five, or even younger, is patently absurd. This absurdity implies that the only reasonable answer is emphatically negative. As in question (6), this overwhelming bias toward a negative response emphasizes its sarcastic intent, overshadowing any possible literal interpretation.
35Following Michaelis & Feng (2015), we refer to this dual conversational construction as SIs. It is composed of two parts:
(i) WH-interrogative (e.g., what is this, what’s next) – the body.
(ii) a self-answer, which can be realized by a participial, nominal, adjectival, or prepositional phrase, or by a clause – the proffer.
36Usually, a WH-question invites an open answer. However, with SIs, the syntactic presence of a self-answer limits the set of possibilities to satisfy the WH-element. The possible set of answers is reduced by the juxtaposed content (Huddleston & Pullum, 2002), so that it turns into a yes-no question (Michaelis & Feng, 2015), instead of an open question.
37As with our previous examples, SIs can also be literal or sarcastic, as seen in the following examples:
(8) Where is Raleigh, in North Carolina? (literal SI, The Switchboard corpus)
(9) What’s next? Marrying an animal? (sarcastic SI, the NLDS corpus)
38Example (9) is not a genuine information-seeking question. The speaker is not interested in the addressee's opinion on whether marriage between humans and animals will become legally recognized in the foreseeable future. The sarcasm in (9) relies on the prominent presupposition that humans can marry only other humans. Thus, the speaker pretends to believe in a false notion (that one can marry an animal) while knowing it to be untrue. This creates a stark contrast between pretense and reality, making it clear that the question is insincere.
39Consider another example of a sarcastic SI:
(10) What planet are you living in? Romulus? (the NLDS corpus)
40Similarly to question (9), this inquiry does not seek any real information. The speaker is not interested in determining where the addressee actually resides. Question (10) is interpreted as sarcastic due to the salient presupposition that humans can only live on earth, and there is no human life on other planets, such as Romulus. The speaker pretends to defy this presupposition, acting as if he believes the addressee could reside outside earth's boundaries, despite knowing well that this is impossible. This clash between the hypothetical scenario that might warrant a sincere utterance (that humans can live on other planets) and the reality (that humans cannot) makes the insincere and sarcastic nature of the question salient over the literal interpretation.
41This study focuses on two main research questions. The first one is whether sarcastic interrogatives are distributed differently across three distinct corpora. That is, whether sarcasm is more salient in one corpus than in others.
42The second question pertains to the interrogatives themselves – whether some interrogatives are more frequently interpreted as sarcastic, thus rendering sarcasm more salient and accessible.
43Note that one of the key factors for salience in our definition (see section 2) is accessibility due to high frequency of usage (Boswijk & Coler, 2020). For that reason, if sarcastic questions are used in a significantly higher frequency in one corpus over another, we can determine that they are more salient in that corpus. A significant difference in the usage of sarcastic questions between these corpora can reflect different usage patterns for sarcasm in divergent discursive arenas and situations. Thus, a listener embedded in the socio-linguistic environment represented by this corpus will be able to access sarcastic meaning more readily.
44The three interrogative constructions chosen for analysis were inspected in three separate corpora of American English: the COCA corpus, the NLDS corpus and the Switchboard corpus
45The COCA (Corpus of Contemporary American English) corpus is a large and representative corpus of American English, containing more than one billion tokens of text from eight genres: spoken, fiction, popular magazines, newspapers, academic texts, TV and movies subtitles, blogs, and other web pages. We analyzed the sub-corpora of spoken, non-scripted language from two years – 2011 and 2012, which altogether contained 7,430,781 tokens (including "deleted words" and some annotations, like "[applause]").
46The NLDS (Natural Language and Dialogue Systems) corpus contains several main categories (each including several sub-corpora): online debate forums, human generated summaries, films, blogs, casual conversations, natural language generation, and hotel dialogues. Since we were interested in a corpus representing non-scripted, online web-discourse, we chose to use the sub-corpus of online debate forums. Within that category we focused on The Internet Argument Corpus (IAC) version 2 (Abbott et al., 2016), which is a collection of corpora for research in political debate on internet forums. It consists of three datasets: four forums (414,000 posts), Convince Me (65,000 posts), and a sample from Create Debate (3,000 posts).
47The Switchboard corpus is an annotated spoken corpus composed of 5-minute-long phone conversations between strangers in the United States. It consists of approximately 260 hours of speech and was originally collected by Texas Instruments in 1990/1. It is a collection of about 2,400 two-sided telephone conversations among 543 speakers (302 male, 241 female) from all areas of the United States. A computer-driven robot operator system handled the calls, giving the caller appropriate recorded prompts, selecting, and dialing another person (the callee) to take part in a conversation, introducing a topic for discussion and recording the speech from the two subjects until the conversation ended. About 70 topics were provided, of which about 50 were used frequently. Selection of topics and callees was constrained so that no two speakers would converse together more than once and that no one spoke more than once on a given topic.
48Each question was annotated as sarcastic, rhetoric or literal, as well as for its type (AQ, DPQ, or SI). For a question to be marked as sarcastic, it needed to contain meaning inversion of the expected situation, indirectly showing criticism and ridicule. Below are three examples for how this annotation criterion was implemented:
(11) Is your place pretty chic or old and historic?
49This is a sincere question, in which the asker is seeking for information in an honest way. There is no meaning inversion, no criticism, and no ridicule. Therefore, this question was marked as a literal, non-sarcastic, question.
(12) For one even if they were no religions, racial/cultural conflicts would continue. For example, were the slaves slaves because they were not Christian or they were black? I'll answer that myself. Because they were black.
50This is not a sincere question. The speaker is not seeking for information, but is trying to prove their point by asking a question to which the answer is obvious, and even goes as far as to explicate it in the following sentence. Therefore, it is not a literal question. However, there is no meaning inversion, no criticism, and no ridicule, so it is not a sarcastic question either. This question was marked as a rhetoric, non-sarcastic question.
(13) Do you have any real facts or just fallacies?
51This is also not a sincere question. The speaker is not seeking an honest response. Instead, they are trying to convey a pointed critique towards the listener by preforming meaning inversion and communicating the transposition of a pretended commitment. The speaker is only pretending to seek information, when in fact he assumes an evaluative standard and implicates that it has been violated negatively. This is done while expressing ridicule and criticism. Therefore, this question was marked as a sarcastic question.
52The next annotation criterion was the question-type. For a question to be considered as an AQ or as a DPQ, the disjunctive coordinator (“or”) had to be a clausal coordinator, as seen in the following example:
(14) Did your mommy tell you this or did you figure it out on your own?
53In this question, the disjunctive “or” coordinator coordinates between two clauses. Therefore, this question was considered as relevant for further analyses.
(15) Do you think that he or she will mind?
54In this question, the disjunctive “or” coordinator operates on a phrasal level. Therefore, this question was not considered relevant for further analyses.
55 After identifying all the sarcastic disjunctive or-questions, we divided them into AQs and DPQs. This division was not always straightforward. In general, questions that required a “yes” or “no” answer were considered as DPQs. Questions that required an answer that corresponds to one of the suggested alternatives were considered as AQs. Cases in which the question type could have not been clearly determined were put in a separate category, and were not considered for the final analysis.
56 Lastly, we analyzed the SIs. For a question to be considered as a SI, it had to consist of a WH-question followed by a self-answer ending with a question mark. Two separate questions in a row (Why? What are they saying?) were not considered to be SIs. A question asked by one person, and then answered by someone else was also not considered to be a SI (since it was not a self-answer).
57Using normalized frequencies, 6,202 utterances were manually analyzed (1,132 in the NLDS corpus, 2,481 in the COCA corpus, and 2,589 in the Switchboard corpus). These annotation criteria deemed 1,188 utterances (out of the initial 6,202) relevant for analysis. All following results will relate to these final 1,188 questions selected.
58 First, we wanted to see if there are any significant differences in the number of sarcastic questions among the three corpora examined. To do so, we compared the distribution of sarcastic questions between the non-scripted, spoken TV-corpus (COCA), the spoken telephone-corpus (Switchboard), and the written web-corpus (NLDS).
59 Out of 787 questions analyzed from the COCA corpus, 37 were sarcastic (4.7%). Out of 301 questions analyzed from the NLDS corpus, 121 of them were sarcastic (40.19%). Out of 100 questions analyzed from the Switchboard corpus, 0 of them were sarcastic (0%).
60 These results are illustrated in the following figure:
Figure 1: Distribution of sarcastic questions across corpora
61Results indicate that the proportion of sarcastic questions is significantly higher in the written web-corpus than in the spoken TV-corpus (z = -14.87, p <.001). No sarcastic questions were found in the spoken telephone corpus.
62Sarcastic questions are most frequently used in the written web-corpus (NLDS), less used in the spoken TV-corpus (COCA), and were not found at all in our sample search of the spoken telephone corpus (Switchboard). Note once again that one of the key factors for determining the salience of a linguistic structure is facilitated accessibility due to high frequency of usage (Boswijk & Coler, 2020). Therefore, our results indicate that sarcastic questions are most salient in the written web-discourse.
63We postulate that these sarcasm-distribution results are not arbitrary, and can be motivated by the discursive arena and its situational salience. In the discursive-framework, salience refers to the fact that some parts of the discourse are more activated or accessible in memory than others. Situational context plays a role in determining the interpretation of discourse. The idea that there is a discourse-mediated mental representation of what is experienced in the world is broadly agreed upon amongst discourse researchers. These mental models are called situation models (Falk, 2014), and they are integrated mental representations of a described situation which are structured in events or scenes (Zwaan & Radvansky, 1998).
64In a similar way, we propose that within a discursive setting, a scene or an arena can make certain meanings salient. We postulate that written, forum-based, web-discourse constitutes a separate discursive scene, which has its own mental model, where sarcasm holds greater salience compared to other scenes and arenas, such as spontaneous spoken discourse on television. It seems that the contextual dynamics of online forums evoke sarcasm more prominently than the situational context of spoken, non-scripted, TV discourse.
65Indeed, online discourse is a unique arena of communication, particularly evident in the language utilized within social networking platforms, such as the online forums from which the NLDS corpus is compiled. By enabling users to communicate online, social networking sites allow for a new form of sociality in which language has a pivotal role in building a community (Zappavigna, 2011). The notion of an online community has been explored since the 90s. Most notable is Rheingold’s (1993) work on virtual communities, which he defined as a social gathering that emerges online when enough people carry public discussions for long enough periods, with enough human feelings, to create webs of personal relationships in the cyberspace. Although there are still debates on the exact definition of online communities, it is agreed upon that people use language to construct social bonds and to rally around, defer or reject different values based on their usage of online language (Zappavigna, 2011).
66Language plays a crucial role in forming social bonds, particularly in online contexts where users connect over shared interests and values (Zappavigna, 2011, 2014). Social media platforms enable individuals to communicate experiences and build communities based on shared values, fostering alignments around common occurrences (ibid). It seems appropriate to consider people’s political views and affiliations as a major part of someone’s social identity, as is the case in the political NLDS corpus.
67According to Zappavigna (2014), microblogging on social media platforms, such as Twitter, has an important function of establishing or solidifying bonds of companionship rather than communicating ideas. This might be true for microblogging on Twitter; however, this may not apply to the NLDS corpus, which comprises of social media dialogues in online internet forums dedicated to political debates, which are known to be sensitive and polarizing topics. Thus, participants are in an environment that is more prone to arguments and criticism, than to ambient affiliation and bondability. Consequently, this environment has emotional and situational properties that promote the usage of sarcasm, rendering it more salient. This situational salience also promotes asking non-sincere questions, designed to convey critique and ridicule rather than to search for honest answers, which makes a fertile ground for sarcasm generally, and for sarcastic questions specifically.
68In contrast, the COCA sub-corpus we analyzed is based on spoken, non-scripted, TV discourse, mostly consisting of talk-shows, morning shows and interviews; thus, this environment has emotional and situational features which promote the usage of sincerity and literal-language rather than sarcasm. Furthermore, in such an environment, most of the questions are information-seeking, asked in order to get a real, genuine answer, rather than a sarcastic one. The difference in sarcasm-distribution between the written and the spoken corpora can be explained by these differences in situational and emotional qualities of the different discursive arenas of the corpora studied.
69Situational salience can also account for the fact that no sarcastic questions were found in the Switchboard corpus. Recall that this is a spoken corpus of two-sided phone calls, in which a computer-driven operator introduced a topic for discussion among two speakers that never spoke before and will never speak again. This is a completely artificial discursive arena, in which the participants have no previously experienced expectations or anticipated norms. Therefore, they don’t have any situational predictions or interactional conventions that may lead to sarcasm.
70The second aim of the study is to examine the distribution of the different question types among the two corpora in which sarcastic questions were found. In other words, we wanted to check if any of the three interrogative constructions is significantly more sarcastic than the others. Below are the three question types, as well as their frequency in each corpus:
71(i) Alternative Questions (AQs)
(16) Did they “discover” this before or after they smoked weed? (the NLDS corpus)
72523 Aqs were found in the spoken TV COCA corpus, and 181 Aqs were found in the written web NLDS corpus.
73(ii) Disjunctive Polar Questions (DPQs)
(17) Do you think we're dummies or something? (the COCA corpus)
74200 DPQs were found in the COCA corpus, and 80 DPQs were found in the NLDS corpus.
75(iii) Split Interrogatives (SIs)
(18) What are you, like, stuck in the 1950s? (the COCA corpus)
7664 SIs were found in the COCA corpus, and 40 SIs were found in the NLDS corpus.
77Note that no data from the spoken-phone Switchboard corpus appears regarding this research question, since no sarcastic questions were found in that corpus.
78Out of a total of 704 AQs in both corpora, 87 were sarcastic. Out of a total of 280 DPQs in both corpora, 36 were sarcastic. Out of a total of 104 SIs in both corpora, 35 were sarcastic. The exact distribution can be seen in the following figure:
Figure 2: Interrogative distribution findings
79Results indicate that SIs are significantly more sarcastic than AQs (z = -5.66, p<.001) and DPQs (z = 4.67, p<.001). The difference between AQs and DPQs is not significant (z =.21, p =.83, n.s.)
80Results show that out of the three constructions examined, SIs had the highest proportion of sarcastic interpretations. Therefore, we argue that SIs’ sarcastic meaning is more salient than the sarcastic meaning of AQs and of DPQs.
81To explain these findings, we draw on Michaelis & Feng’s (2015) analysis, which describes sarcastic SIs as pragmatically specialized constructions with a distinct syntactic pattern that differs from literal SIs (Kay & Fillmore, 1999; Goldberg, 1995). These constructions are defined by two main components: a WH-interrogative (the body), and a potential answer (the proffer). The body functions similarly to ‘trouble signals’, like 'um', indicating that the speaker should already know the answer, which undermines the sincerity of the question. This insincere usage of trouble signals manipulates conversational norms, enhancing the utterance's salience as sarcastic rather than literal.
82Michaelis & Feng (2015) further argue that Sis’ grammatical pattern makes them ideal for conveying sarcasm. The structure of SIs, consisting of the body and the proffer, serves a specific function in conversation. The body presents a question that seems to seek an answer, while the proffer, acting as a self-answer, delivers the sarcastic punchline. This setup not only reveals the speaker’s pretense of ignorance, but also the evident absurdity in the supposed question, making the sarcastic meaning more pronounced and salient. The binary form of SIs, combining the question with a self-answer that typically provides a stark contrast to expected responses, reinforces their effectiveness in sarcastic communication. The proffer resolves the setup by not just answering, but also by contrasting with the actual situation, often in a manner that distinctly deviates from conversational norms, thereby emphasizing the sarcasm.
83To illustrate, consider the following example:
(19) Who is the friend, the shotgun? (NLDS corpus)
84In this sarcastic question, the speaker uses the initial part, the body, to feign a search for an answer – a value they already know. In the subsequent part, the proffer, the speaker presents a resolution. The scenario is such that the value of the variable is patently clear since typically, only humans (and occasionally animals) can be friends with other humans, making it implausible for an inanimate object, such as a shotgun, to be considered a friend. Thus, the proffer not only conveys the assumed value of the variable, but also establishes a scalar structure that contradicts the actual situation. Theoretically, while it is conceivable that someone might regard an inanimate object as a friend, such a notion is so extreme and absurd that it sharply clashes with accepted societal norms. This deviation transforms the question from a genuine, information-seeking, question into one that is salient as sarcastic, and not as literal.
85Another possible explanation for why SIs are most salient for sarcasm among the three constructions examined may relate to the concept of predictability. As noted in section 2, salience is associated with predictability, which can manifest in two ways. On one hand, predictability can refer to something highly unexpected or surprising, thereby making it more noticeable and salient. On the other hand, predictability can also mean something is frequently used, facilitating easier access and processing, thus enhancing its salience (Boswijk & Coler, 2020). While the second aspect, frequent usage, is evident in the common occurrence of sarcastic SIs, the first aspect, unexpectedness, also plays a significant role in their sarcastic interpretation.
86Traditionally, questions as speech acts are intended to gather information (Sadock & Zwicky, 1985), typically eliciting a response from someone other than the asker. However, when the asker is the one who provides the answer, as observed in the self-answer portion of SIs, it defies conversational norms and is both unexpected and surprising. This unpredictability is in opposition to the question’s typical information-seeking role, marking it as insincere and thus, more salient in its non-literal interpretation.
87This surprising self-answer element makes SIs particularly stand out, underscoring their non-literal, sarcastic nature. As Michaelis & Feng (2015) note, SIs are not intended to receive actual responses, but to continue the speaker’s commentary, enhancing their effectiveness in conveying sarcasm. This aspect, combined with their ability to overturn presuppositions, renders SIs more salient for sarcasm compared to the other interrogatives studied, which lack the self-answer feature.
88Although the study of salience and sarcasm, both together and separately, has been conducted extensively throughout the years, there are still many gaps left in our understanding of how they work together. This is the first study, as far as we are aware, that examined the distribution of specifically sarcastic (and not ironic) interpretation across different types of discursive arenas, showing distinct uses of sarcasm in different corpora. Additionally, this study demonstrated that across different interrogative constructions, there can be a difference in the proportion of sarcastic interpretations.
89Our analysis of the results argues that these differences in distribution are motivated by different factors. The distinctions between the different corpora types are attributed to the variations in the discursive arenas that the corpora represent. The differences between the interrogatives themselves are explained by form-function relations, and the structure of the SIs. Thus, both language external and internal factors can influence how salient a sarcastic meaning becomes.
90There are several further directions for future research. First, the role of sarcastic-markers. In many of the sarcastic utterances collected (regardless of the corpus), there are certain words that appear frequently, and may make the sarcastic meaning of the interrogatives more salient than its literal one. These include discourse markers (just, really), logical operators (or something, some), attenuators (about, almost), and intensifiers (really, so). Attenuators and intensifiers, appearing together or separately, were shown as prone to convey sarcasm in several different syntactic structures, such as negations (Giora et al. 2015), rhetorical questions (Giora et al., 2018) and similes (Levant et al., 2020; Veale, 2013). We suspect that this could also be the case with our three question types, to the extent that these adverbs do not shift to epistemic markers in questions (Romero & Han 2004).
91Second, we must consider the factor of context. Although we analyzed in depth the influence of the discursive environment on the salience of sarcasm, we have not yet looked at the effect of the immediate surrounding context (the utterances appearing before and after the sarcastic question itself) on the salient meaning of these sarcastic questions. Future research would require careful analysis of the immediate context to examine if there are any commonalities in context that might make some of these questions more salient in their sarcastic meaning than in their literal one.