1 Introduction
Construction Grammar assumes that constructions, i.e. ‘learned pairings of form with semantic or discourse function’ (Goldberg Reference Goldberg2006: 5), are the basic units of language-related knowledge. The term form here implicitly refers to a verbal form since Goldberg continues by explaining that the definition of construction includes ‘morphemes or words, idioms, partially lexically filled and fully general phrasal patterns’ (Goldberg Reference Goldberg2006: 5). However, face-to-face communication is inherently multimodal and some research in the past decade has questioned this bias toward monomodal verbal constructions and speaks up for a multimodal orientation in Construction Grammar (see e.g. Steen & Turner Reference Steen, Turner, Borkent, Dancygier and Hinnell2013; Zima & Bergs Reference Zima and Bergs2017).
The fact that spoken interactions are inherently multimodal is well received nowadays. Early studies on gestures and their relation to language showed that speech and gesture are closely time-aligned and that both contribute meaning to the utterance, i.e. are not redundant (Kendon Reference Kendon2004; McNeill Reference McNeill2005). Others soon followed this lead and, today, there is a considerable body of research on how gestures contribute to utterance meanings (comprehensive reviews can be found in Vigliocco, Perniss & Vinson Reference Vigliocco, Perniss and Vinson2014; Feyaerts, Brône & Oben Reference Feyaerts, Brône, Oben and Dancygier2017; Perniss Reference Perniss2018). Gestures, it seems, are well-established semiotic resources that engage with semantic and pragmatic meanings. Recent advances in linguistically informed, multimodal studies suggest that the same can be claimed for semiotic resources other than (manual) gestures as well (see e.g. Feyaerts et al. Reference Feyaerts, Rominger, Lackner, Brône, Jehoul, Oben and Papousek2022 on facial expressions as response turns).
In contrast to this, the relation between grammar and semiotic resources other than language is a controversial issue. In particular, among Construction Grammarians, the notion of multimodal construction, i.e. a form–meaning pairing whose formal features comprise more non-verbal aspects, is disputed even though there are quite a few studies suggesting a close relation between syntactic, prosodic and kinesic properties for some constructions: Elvira-García (Reference Elvira-García, Beijering, Kaltenböck and Sansiñena2019) shows how the intonation contour disambiguates elliptical and independent Spanish si + indicative clauses; Zima (Reference Zima2017) shows frequent co-occurrences of distinct gestures with semantic aspects of the [all the way from X PREP Y] construction; and Hinnell (Reference Hinnell2018) shows a distinctive and iconic relation between the use of manual gestures and aspect-marking constructions, to name but a few. Independently, Ward (Reference Ward2019) introduces the notion of prosodic construction, i.e. ‘a temporal configuration of prosodic features’ with a meaning that is ‘not necessarily closely aligned with words’ (Ward Reference Ward2019: 24). In doing so, he extends the notion of construction to prosodic forms like the consider this construction, which is characterized by a prosodically highlighted beginning (high pitch, loud, slow), then followed by narrow pitch range and, finally, ends in high pitch, which is typically used to provide further information to the hearer for them to consider in an argument (Ward Reference Ward2019: 5–24).
A particular type of multimodal construction that seems to play a vital role for the present purposes is stance-related multimodal constructions. Stance is ‘a public act by a social actor, achieved dialogically through overt communicative means, of simultaneously evaluating objects, positioning subjects (self and others), and aligning with other subjects, with respect to any salient dimension of the sociocultural field’ (Du Bois Reference Du Bois and Englebretson2007: 163) and, thus, stance-related constructions are communicative means as mentioned in the definition. A stance-related construction that has received a lot of attention is the shrug. Shrugs show formal variation (shoulder shrug, mouth shrug, head tilts, raised eyebrows, etc.) and can be used to indicate a lack of knowledge, obviousness, or disengagement (Streeck Reference Streeck2009; Debras & Cienki Reference Debras and Cienki2012; Debras Reference Debras2017, Reference Debras2021; Jehoul, Brône & Feyaerts Reference Jehoul, Brône and Feyaerts2017). Other stance-related constructions include the negative assessment construction (Bressem & Müller Reference Bressem and Müller2017), which is instantiated by a throwaway gesture, and discourse management gestures that indicate strong disagreement with the interlocutor, such as pushbacks or pointing gestures that invade the shared discourse space (Wehling Reference Wehling2017).
While the studies reviewed above show a frequent co-occurrence of verbal and non-verbal resources, none of these co-occurrences seems to be obligatory in the strict sense. This has led some Construction Grammarians to argue that a statistically sufficient frequency of co-occurrence does not mean that multimodal constructions are a cognitive reality and that possible candidates for multimodal constructions need to survive a deletion test to pass for genuine multimodal constructions (see Ningelgen & Auer Reference Ningelgen and Auer2017; Ziem Reference Ziem2017). Similarly, Hoffmann (Reference Hoffmann2017) argues that, in most cases, constructions are unimodal but can be combined on-line while speaking, resulting in multimodal instantiations (see also Goodwin Reference Goodwin2017 for his notion of contextual configurations). This view resonates partially with Uhrig's (Reference Uhrig2018) notion of crossmodal collostructions, i.e. strong associations between semiotically different constructions. Essentially, crossmodal collostructions require the independent existence of non-verbal form–function pairings, which can be combined with verbal form–function pairings. Examples of such independent, non-verbal form–function pairings include the shrug and the throwaway gesture, but also prosodic constructions as reviewed above.
Independent of the question that multimodal instantiations pose, some constructional approaches favor a prototype approach to constructions in general (see e.g. Gries Reference Gries2003; Imo Reference Imo2007; Schoonjans Reference Schoonjans2018). In this spirit, Cienki (Reference Cienki2017) develops a prototype approach to multimodal constructions. He proposes that utterance constructions lie at the heart of spoken language analysis. These utterance constructions have a deep structure, which is stored in the constructicon and which contains information on ‘tools that can be drawn upon to express the construction’ (Cienki Reference Cienki2017: 3), i.e. the surface structure. These pieces of information can be verbal or non-verbal and can be more strongly or weakly associated with the utterance construction. The surface structure of the utterance construction is then a selection of relevant verbal and/or non-verbal behaviors and therefore, more often than not, stands in a metonymical relation to its deep structure. These relevant non-verbal behaviors associated with the utterance construction may also be conceptualized as crossmodal collostructions provided that there is an independently existing non-verbal construction. However, if the construction in question is associated with a non-verbal feature that does not have any independent meaning, this feature must be an integral part of the utterance construction even though it might not surface in every instance of it. In other words, if such a feature can be found, it supports the notion of multimodal constructions. Given the still thin empirical grounds, most researchers in Multimodal Construction Grammar agree on the fact that more empirical work needs to be done to come to any verifiable conclusions on the status of multimodal constructions (Hoffmann Reference Hoffmann2017; Schoonjans Reference Schoonjans2017).
The objective of the present article is twofold. The first is to provide empirical data by exploring English as if clauses and the non-verbal features they are frequently accompanied by. The second is to show that all of these features can be explained by resorting to the notion of crossmodal collostructions. Yet it will be argued that the assumptions of Utterance Construction Grammar are useful assets in explaining the different predictive power of these features. The view taken in this article is that both Utterance Construction Grammar and the notion of crossmodal collostructions are not mutually exclusive but complement each other in significant ways. Section 2 will show that as if constructions provide valuable insights for the discussion on multimodal constructions, since non-verbal features seem to be necessary for their disambiguation in at least some cases. In section 3, previous research on multimodal markers relevant for the discussion of as if clauses will be reviewed. Section 4 provides the details on the quantitative multimodal corpus study that was conducted to gather empirical evidence for non-verbal features frequently accompanying as if clauses. Section 5 presents the results. Section 6 discusses the relation between the constructions’ communicative function and the multimodal features they are accompanied by. And, finally, section 7 draws some conclusions for (Multimodal) Construction Grammar.
2 As if clauses
The use of English as if is a case in point to illustrate various degrees of what has been called insubordination, i.e. ‘the conventionalized main clause use of what, on prima facie grounds, appear to be formally subordinate clauses’ (Evans Reference Evans and Nikolaeva2007: 367). Examples (1) to (5) below illustrate the attested uses of English as if clauses, retrieved from the NewsScape Library of International Television News (Steen & Turner Reference Steen, Turner, Borkent, Dancygier and Hinnell2013).Footnote 1 Details on this archive and on the collection procedure will be provided in section 4.1. The video files from which the examples have been taken are available on the OSF platform (https://osf.io/usgw4/files/). Examples (1) and (2) illustrate genuine uses of as if clauses as subordinate clauses, while examples (3) to (5) illustrate uses with syntactic independence:
(1) It allowed him to move around as if this was Clarence Darrow in the courtroom (NewsScape 2019-01-25_0300_US_MSNBC_The_Last_Word_With_Lawrence_ ODonnell, 0:02:09-0:02:19; click to view or scan QR code)
(2) Justice Ginsburg passed away less than 48 hours ago, but it seems as if this is moving very fast and we could have a nominee very soon (NewsScape 2020-09-20_1500_US_KNBC_Meet_the_ Press, 0:03:39-0:03:54, click to view or scan QR code)
(3) I should have known you would use the video of Cuomo coming up from his basement. As if that wasn't a propaganda video (NewsScape 2020-10-06_2100_US_FOX-News_The_Five, 0:17:07-0:17:17; click to view or scan QR code)
(4) As if this year hasn't been enough (NewsScape 2020-07-31_1800_ US_KCBS_CBS _2_News_at_11AM, 0:17:49-0:17:53; click to view or scan QR code)
(5) He thought delaying me would make Republicans like me better. Yeah, right. As if (NewsScape 2010-03-09_0200_US_MSNBC_The_ Rachel_Maddow_Show, 0:58:47-0:58:58; click to view or scan QR code)
Example (1) and (2) illustrate the use of as if as a subordinating conjunction. In example (1) as if introduces an adjunct adverbial clause. It is attached to a (syntactically independent) matrix clause and functions as a manner adjunct. In example (2), the as if clause functions as a complement to the verb seems and, semantically, it introduces a possibility of ‘medium strength epistemic modality’ (Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1152). In addition to these two uses, Huddleston & Pullum et al. (Reference Huddleston and Pullum2002) also list two further functions, manner complement and adjunct of comparison (not illustrated here). Crucially, in all cases, the as if clause alone in this context would be ungrammatical without the matrix clause. Moreover, it introduces a proposition the speaker finds likely, but does not fully commit to. Given these syntactic and semantic commonalities, these uses will be referred to cumulatively as subordinate uses in the remainder of this article. Even though finer-grained analyses are possible here, treating these cases as one construction was considered feasible for the present purposes.
Example (3) is a prime example of insubordination. Here, the as if clause is syntactically independent. The speaker does not mean to say that you would use that video as if that wasn't a propaganda video, but she issues an afterthought and thereby expresses a negative attitude toward Cuomo's video. In other words, there is no matrix clause the as if clause could be attached to, and, nonetheless, it is grammatical. Like (2), it introduces a possibility, but this possibility is presented as pretty unlikely (from the speaker's perspective), if not even counterfactual. Pragmatically, the rejection of the state-of-affairs presented in the content clause usually receives an ironic interpretation (Brinton Reference Brinton2014: 96). In pragmatic terms, irony is an attributive language use, i.e. the speaker puts forward a proposition that alludes to an utterance or belief of some other person or some other version of themselves. This kind of attributive language use differs from other kinds in that it simultaneously expresses a dissociative attitude toward the proposition presented (see Wilson & Sperber Reference Wilson and Sperber2012). Example (3) fits this definition of irony: the speaker attributes the thought that that wasn't a propaganda video to some other people (referred to as you in the previous clause and, presumably, the staff of politician Andrew Cuomo, who produced the video) and, at the same time, expresses a negative stance toward this thought, because the video represents a negatively connoted propaganda video in her view. Given the semantic resemblance between (2) and (3), Brinton (Reference Brinton2014) and López-Couso & Méndez-Naya (Reference López-Couso, Méndez-Naya and Vázques2012) argue that the latter historically derives from the former.
Example (4) illustrates a special case of insubordinate as if. Like example (3), it is insubordinate, but it is lexically more constrained with the negated copula and the adjective enough being obligatory elements. In the NewsScape Library, between January 2018 and December 2020,Footnote 2 a total of 255 syntactically independent uses of as if clauses could be found, of which 74 (appr. 29%) showed this formulaic use. An independent treatment of such cases therefore seems feasible. What is more, meaning-wise, this use also behaves slightly differently from (3). While in (3) the content clause that wasn't a propaganda video is presented as unlikely, in example (4) this year hasn't been enough is presented as true (from the speaker's perspective), because further unfortunate events occurred, which are inferable from the context. Furthermore, this use of as if clauses often links two pieces of information, often bad news, while one is presented as ‘the tip of the iceberg’. In example (4), it is the corona virus pandemic (this year) that is the backgrounded bad news while the hurricane, which is reported on, is presented as foregrounded bad news. In doing so, the speaker alludes to wishful thinking that everything is going to be fine while simultaneously expressing a negative stance toward the events reported, i.e. their frustration that the wishful thinking turned out to be wrong. From such a vantage point, these uses of as if clauses can be regarded as ironic, too, while the target of the ironic criticism is a different one: in (3), particular people are addressed (Cuomo's staff), while in (4) general expectations or beliefs are addressed. There are, however, instances of this formula, e.g. as if this isn't exciting enough, which express a playful, maybe slightly mocking rather than a negative attitude. In such cases, the ironic effect is less evident. In the following, this use will be labeled ‘formulaic as if’.
And, finally, example (5) illustrates the use of as if as a bare complementizer. Here, as if is used independently in all respects – neither is it licensed by any element nor does it license any further elements. In contrast to the other two cases of insubordination, this use has found its way into the Oxford English Dictionary (2020), which states that it is ‘Typically used as a sardonic response to a stated or reported suggestion’ (s.v. as, adv. and conj.). This adequately describes example (5). As if, in combination with the ironic rejection yeah, right, assesses the thought reported previously in a negative way.
The expository paragraphs above are summarized in table 1.
Overview on as if constructions from the NewsScape Library (January 2018 – December 2020) with token frequency
The selection of examples (1) to (5) suggests that syntactic, semantic and pragmatic evidence can sufficiently disambiguate the different uses of as if. However, there are quite a few ambiguous cases (eleven in total), like the following:
(6) I just saw a video from Florida's beaches and they are absolutely packed with people sitting side by side and playing in the ocean as if this is not going on (NewsScape 2020-03-16_2100_US_ MSNBC_MTP_Daily, 0:09:40-0:09:56; click to view or scan QR code)
Technically, example (6) could be an instance of a subordinate as if clause that introduces a counterfactual possibility, with people sitting side by side and playing in the ocean being the matrix clause(s). However, it could also be syntactically independent, semantically issuing an afterthought. Contrary to what could be expected, Lehmann & Bergs (Reference Lehmann and Bergs2021) show that subordinate and insubordinate as if clauses don't show any difference as regards the tense of the verb. Essentially, the question of syntactic (in)dependence boils down to the question of which kind of attitude is conveyed in cases like (6). The obvious counterfactuality presented in the as if clause could either be treated as a neutral observation or a criticism of the way these people act, lending support to the analysis of (6) as subordinate or insubordinate, respectively. Since the attitude of the speaker is not indicated on the lexical level here, the disambiguation of (6) cannot be made on language-internal grounds, but relies on other modes than the verbal one.
In spoken language, this matter is complex and deserves systematic empirical attention – an objective that this article set itself to achieve. More specifically, this article explores the non-verbal resources speakers use in spoken interactions to mark as if constructions. The second objective, as already mentioned, is to square these empirical findings with the notion of multimodal constructions. There are quite a few possible outcomes:
1. Ironic as if clauses, irrespective of constructional type, could be associated with a particular set of non-verbal features when compared with non-ironic as if clauses. In such a case, there could be an independent set of non-verbal constructions that signals irony irrespective of verbal form.
2. Insubordinate and formulaic uses of as if clauses could be both associated with the same set of features when compared with subordinate uses. Such an outcome would lend support to the idea that there is an independent, non-verbal construction functioning as a marker for syntactic independence.
3. Insubordinate and formulaic uses of as if clauses could be associated with different sets of non-verbal features when compared with subordinate uses. Such an outcome supports an individual treatment of the two uses as different constructions. Moreover, depending on the exact nature of these non-verbal features, further conclusions might be drawn:
a. If these non-verbal features have been described elsewhere, fulfilling the same or a similar function, it could be assumed that these non-verbal features present an independent non-verbal construction that is associated with the verbal construction (i.e. forms a crossmodal collostruction).
b. If there are non-verbal features that have not been attested elsewhere (and are unlikely to function in a similar way independent of the particular verbal construction), these provide modest evidence for a genuine multimodal construction.
3 Multimodal aspects of subordination and irony
As was shown above, as if constructions vary largely on two grounds, i.e. as regards their constructional complexity and the stance the speaker takes toward the proposition expressed in the as if clause. Therefore, a review of the literature on multimodal markers of (in)subordination and irony will be provided, which served to delimit the variables for the empirical study outlined in section 4.
3.1 Marking of dependent and independent clauses
Early findings on the prosody–syntax interface observed that dependent structures are often realized with a rising pitch contour, which signals that more is going to follow (Bolinger Reference Bolinger1984; Wells Reference Wells2006). More recent, empirical studies confirm this observation (Lelandais & Ferré Reference Lelandais and Ferré2016, Reference Lelandais and Ferré2017, Reference Lelandais and Ferré2019; Elvira-García, Roseano & Fernández-Planas Reference Elvira-García, Roseano and Fernández-Planas2017; Elvira-García Reference Elvira-García, Beijering, Kaltenböck and Sansiñena2019; Maschler Reference Maschler, Doehler, Lindström and Keevallik2020). Moreover, these studies show that dependent structures can also be intonationally integrated into their host structure (Lelandais & Ferré Reference Lelandais and Ferré2016), are usually slower than more independent structures and tend to be accompanied by silent pauses (Lelandais & Ferré Reference Lelandais and Ferré2016, Reference Lelandais and Ferré2019; Köhn, Baumann & Dörfler Reference Köhn, Baumann and Dörfler2018). Other findings are less consistent. While Köhn, Baumann & Dörfler (Reference Köhn, Baumann and Dörfler2018) report a lower mean pitch for German subordinate clauses, Lelandais & Ferré (Reference Lelandais and Ferré2016) find a lowered mean pitch for English appositive clauses. Furthermore, Köhn, Baumann & Dörfler (Reference Köhn, Baumann and Dörfler2018) report a lowered intensityFootnote 4 for subordinate clauses, while Elvira-García (Reference Elvira-García, Beijering, Kaltenböck and Sansiñena2019) finds no effect of intensity on the discrimination of elliptical and independent Spanish sí-clauses.
Empirical studies concerned with kinesic information accompanying syntactic (in)dependence are rare. Notable exceptions are Lelandais & Ferré (Reference Lelandais and Ferré2017, Reference Lelandais and Ferré2019). They report that syntactically independent structures are kinesically set off from their surrounding co-text, i.e. they are often produced with non-overlapping, distinct manual gestures, gaze changes and eyebrow rises. Dependent structures, on the other hand, are often produced with overlapping hand gestures, thus creating a kinesic link to their hosts (see also Maschler Reference Maschler, Doehler, Lindström and Keevallik2020 for similar observations).
3.2 Marking of irony
Research on the so-called ‘ironic tone of voice’ has not been conclusive to date. One reason for the controversial findings is that the prosodic marking of irony is language-specific (on the difference between the prosodic profiles for English and Cantonese irony see Cheang & Pell Reference Cheang and Pell2009, Reference Cheang and Pell2011). Another reason is that scripted and unscripted irony seems to trigger different prosodic profiles (Rockwell Reference Rockwell2000). Research on unscripted English suggests that it is marked by a slower tempo (Rockwell Reference Rockwell2007; Bryant Reference Bryant2010), a higher mean pitch (Bryant & Fox Tree Reference Bryant and Fox Tree2005; Rockwell Reference Rockwell2007), greater pitch variability (Rockwell Reference Rockwell2007) and greater intensity variability (Bryant & Fox Tree Reference Bryant and Fox Tree2005).
Kinesic cues to an ironic meaning are similarly controversial. Colston (Reference Colston, Athanasiadou and Colston2020) maintains that gaze aversion is a feature of irony, while Caucci & Kreuz (Reference Caucci and Kreuz2012) report looks to the recipient to accompany an ironic remark. Other kinesic features that are sometimes reported to accompany irony include raised eyebrows and frowns (Tabacaru & Lemmens Reference Tabacaru and Lemmens2014; Tabacaru Reference Tabacaru2019, Reference Tabacaru, Athanasiadou and Colston2020), rapid blinking (Kreuz Reference Kreuz2020), tightened lips, smiles and laughter (Caucci & Kreuz Reference Caucci and Kreuz2012) as well as head nods (Caucci & Kreuz Reference Caucci and Kreuz2012; Tabacaru Reference Tabacaru2019) and head tilts (Tabacaru Reference Tabacaru2019). In contrast to these findings, Attardo et al. (Reference Attardo, Eisterhold, Hay and Poggi2003) find the so-called ‘blank face’ to be prominent in their data.
4 Study details: method and annotations
Given the fact that insubordinate as if clauses are syntactically independent structures conveying an ironic meaning, the review above suggests that they are most likely realized in a separate tone-unit and accompanied by eyebrow rises. Apart from these, no further commonalities are noted in the literature between multimodal markers of independent syntactic structures and irony. Interestingly, there are also some cues in conflict here, i.e. speech tempo and overall mean pitch: while ironic utterances are usually slower and lower in pitch, independent syntactic structures tend to be faster and higher in pitch. In the following, the details of a corpus study that investigates if and how subordinate and insubordinate as if clauses are multimodally marked in naturally occurring interactions are laid out.
The study is a corpus-based analysis of multimodal markers of irony. The multimodal archive used here is the UCLA NewsScape Library of International Television News (Steen & Turner Reference Steen, Turner, Borkent, Dancygier and Hinnell2013). This archive contains a collection of digitized television news programs. The collection extends from 2004 and runs until the present day. In March 2021, it already counted 409,532 hours of programming of American English television containing 2.94 billion words and is updated on a daily basis (Uhrig Reference Uhrig2021). The video files provided by the NewsScape Library include useful information on prosody, facial expressions, head movements and manual gestures. The archive was accessed through the facilities of the Distributed Little Red Hen Lab (which is co-directed by Francis Steen and Mark Turner), using the Edge search engine.
Given this huge archive, the search was limited to video files from January 2018 to December 2020 and to the string as if this/that. Using as if as the only search terms resulted in a considerable imbalance toward subordinate constructions. Lehmann & Bergs (Reference Lehmann and Bergs2021) suggest that subordinate as if clauses are associated with proximal demonstrative pronouns, while insubordinate as if clauses are associated with distal demonstrative pronouns. Therefore, including these kinds of pronouns in the search string was considered a useful limitation of the data. The results obtained in this way were further limited. Videos in which there was a considerable amount of overlapping speech or noise or in which the speaker's face was not visible were excluded from further analyses. Ambiguous cases that could not be assigned to a construction on syntactic grounds (like example (6) above) were not included in the analyses either. This procedure resulted in a total of 668 hits.
These hits were annotated for interaction type, construction, speaker identity, interpretation, and the syntactic form of the prosodic chunk in which the as if was embedded. Interaction type was categorized as either ‘scripted’ (TV series, movies, commercials), ‘monologue’ (stand-up routines, news reports), ‘video call’ or ‘face-to-face interaction’ since previous studies reported different markers of irony for scripted and unscripted types of interactions. As for construction, values were ‘subordinate’, ‘insubordinate’ and ‘formulaic’. The bare as if construction had to be excluded from the analysis due to the methodological considerations described above, which included using the search string as if this/that. Speaker identity was annotated manually due to the fact that the name of the speaker is not always provided in the metadata files of the NewsScape archive and needed to be extracted from the text included in the video files. If no information on the identity of the speaker was provided in the video (e.g. in the case of street interviews), the speakers were labeled as anonymous and numbered consecutively. Interpretation was categorized as ‘ironic’ or ‘non-ironic’ based on the definition by Wilson & Sperber (Reference Wilson and Sperber2012). Examples ambiguous between an ironic and non-ironic interpretation were excluded. Prosodic chunks are difficult to identify because they are fuzzy entities (Barth-Weingarten Reference Barth-Weingarten2016). Thus, boundaries between prosodic chunks were determined using a variety of features, including pauses and inbreaths (Szczepek Reed Reference Szczepek Reed2011), falling pitch, lengthening and voice creaks (Barth-Weingarten Reference Barth-Weingarten2016). The chunks containing as if were then annotated for their syntactic form, the values being ‘sentence’, ‘clause’, ‘verb phrase’, ‘as if’ and ‘other’.
In order to identify prosodic features, the video files were converted to wav format and analyzed with Praat (Boersma & Weenink Reference Boersma and Weenink2019). This software was used to measure pauses before and after the prosodic chunk as well as internal pauses. Furthermore, duration per syllable, mean pitch, standard deviation of mean pitch as well as pitch range (maximum minus minimum pitch) of the prosodic chunk containing as if was measured. Although intensity (i.e. the acoustic correlate of loudness) would have been interesting to investigate, its measurement outside the laboratory is highly unreliable and, therefore, is not considered any further in this study. Moreover, a pilot analysis of the first 100 hits revealed that all of them were produced with non-rising intonation, which is why intonation contour is not further considered in this study. One reason for this observation may be an outcome of the search string used: the demonstrative pronouns this and that are, mostly, used anaphorically, i.e. subordinate as if clauses follow the matrix clause and, thus, non-rising intonation is more likely.
In addition to prosodic features, the data were also annotated for kinesic features by the author of this article. To do so, the videos were paused at the onset of the as if clause and then viewed frame by frame. The features under consideration were gaze direction, head movements, blinking rate as well as movements in the eye, eyebrow and mouth region based on a subset of action units described in Ekman & Friesen (Reference Ekman and Friesen2003). Gaze direction was annotated broadly as either ‘directed at the camera’, ‘to recipient’, ‘to the audience’, ‘to an object’, or ‘elsewhere’. Measurements of gaze direction are imprecise without eye-tracking techniques, but here the perspective of the (uninitiated) viewer was taken. Head movements were categorized as ‘nod’, ‘shake’, ‘tilt’, ‘turn’, ‘none’ or ‘other’. The blinking rate was determined by counting the number of blinks during the utterance of the prosodic chunk divided by its total duration. Movements in the eye region were categorized as ‘blinking’, ‘closed’, ‘upper lid raised’, ‘lower lid raised’, ‘cheeks raised’, ‘other’ and ‘none’. Movements in the eyebrow region were categorized as ‘raised’, ‘frowning’, a ‘combination’ thereof, ‘other’ and ‘none’. Finally, movements in the mouth region were categorized as ‘smile or laughter’, ‘other’ and ‘none’. Another variable was ‘blank face’. Since there is no agreed-upon definition of what counts as a blank face, this variable received a positive value when all facial action units received a ‘none’-value. In all other cases, it received a negative value.
An overview on all annotated values can be found in table 2. The data are made available at https://osf.io/usgw4/files/
An overview on the variables and values used in the study
Even though the review presented in section 3 suggests that manual gestures might be relevant for discriminating between syntactically dependent and independent as if constructions, these were neglected in the present study for two main reasons. First of all, analyzing manual gestures requires the hands of the speaker to be visible to the researcher. This was the case for less than half of the data. What is more, however, a preliminary view of the data did not suggest that the speakers observed in the NewsScape Library gesture a lot when uttering as if clauses. The reasons for this can only be speculated about. One reason might be that most of the speakers are experienced TV personalities (often news anchors) who are probably aware of the fact that their hands might not be visible and who might receive some formal training in non-verbal communication. Irrespective of these reasons, manual gestures were excluded from further analyses in order to get a sufficiently large dataset for more promising features.
5 Results
In the following section, the results of the corpus study are presented. First, some observations are made on the constructions themselves. This is followed by a presentation of the results of linear mixed-effects models run using the lme4-package (Bates et al. Reference Bates, Mächler, Bolker and Walker2015) for irony and the mclogit-package (Elff Reference Elff2022) for as if constructions in the statistics program R (R Core Team 2019).
5.1 As if constructions and irony
The corpus data gained from the UCLA NewsScape Library of International Television News confirms that subordinate as if constructions typically convey a non-ironic meaning, while insubordinate constructions convey an ironic meaning. The formulaic as if construction is also more often than not used ironically, but less frequently than the insubordinate construction. This is illustrated in figure 1.
Mosaic plot illustrating the frequencies of as if constructions in spoken English and their interpretation
5.2 Modeling ironic as if clauses
Since independent as if clauses, formulaic or insubordinate, often convey an ironic meaning, one question is whether there is a prosodic or kinesic profile that is associated with irony irrespective of the construction used. Thus, a model using the lme4 package (Bates et al. Reference Bates, Mächler, Bolker and Walker2015) in R was fitted that identified multimodal markers of irony conveyed by as if clauses, ignoring construction as a potential factor. The R script is made available on the OSF platform (https://osf.io/usgw4/files/). The results are summarized in table 3.
Summary of the final model for irony
Table 3 shows that the interpretation of an as if clause as ironic and its multimodal marking show significant variance in intercepts across speakers and interaction types. In addition, ironic as if clauses are significantly more often prosodically chunked as clauses and are significantly faster than non-ironic as if clauses. None of the other features listed in the model (i.e. blinking rate, pausing, mean pitch, gaze behavior and movements in the eye region) reached a significant level, even though they improved the model fit. The features not listed in table 3 did not improve the model fit, i.e. they seem to have no influence on the interpretation of an as if clause as (non)ironic. The odds ratios and their confidence intervals are illustrated in figure 2.
Odds ratios and confidence intervals for the model terms in descending order
Figure 2 confirms that only prosodic chunking and speaking rate (tempo) can reliably predict ironic as if clauses since their confidence intervals do not cross the vertical zero-effect line. All of the other terms that have entered the model cannot confidently be used to predict an ironic interpretation, even though their odds ratios suggest some tendencies. These are the following: ironic as if clauses tend to be followed by a pause, but are not preceded by one, nor are there any internal pauses. Also, ironic as if clauses tend to be lower in pitch and tend to be accompanied by movements in the eye region, though not by frequent blinks. Finally, the speaker of an ironic as if clause tends to look more at the audience (if present) than when being non-ironic.
5.3 As if clauses
To model as if constructions a polytomous model was fitted using the mclogit package (Elff Reference Elff2022) with subordinate clauses as reference level. Table 4 summarizes the final model.
Summary of the final model for as if constructions
Table 4 shows that the relation between as if clauses and their multimodal markers is significantly influenced by the speaker and the interaction type. It also shows that prosodic chunking and mean pitch act as significant predictors of the different as if constructions. The relation between construction and prosodic chunking is further illustrated in figure 3.
Mosaic plot showing the relation between as if constructions and their prosodic chunking
Figure 3 illustrates that while subordinate constructions can be chunked prosodically in various ways, syntactically independent as if clauses (both insubordinate and formulaic) show a significant tendency to be chunked as clauses.
Despite this prosodic commonality, formulaic and insubordinate as if constructions differ in their mean pitches. That is, formulaic as if constructions are higher in pitch than subordinate constructions, while insubordinate constructions are lower in mean pitch than subordinate constructions. Other features that improved the model fit were the speaking rate (tempo) and frowning, but these did not reach a significant level. None of the other features improved the model fit. The estimates and their confidence intervals for the model are illustrated in figure 4.
Estimates and confidence intervals for the final model for as if constructions
Figure 4 illustrates the estimates and the confidence intervals for the model terms. It shows that prosodic chunking of as if clauses as clauses seems to be a reliable predictor for the formulaic and insubordinate constructions since their confidence intervals do not cross (or even come near) the vertical zero-effect line. Both kinds of constructions tend to be chunked as clauses. As regards the mean pitches, figure 4 illustrates that formulaic as if tends to be uttered with a higher pitch, while insubordinate as if is most often uttered with lower pitch. As mentioned above, none of the other terms reached a significant level, but some tendencies can be observed. Formulaic as if, for example, tends to be fast in tempo and is less likely to be accompanied by frowning, while insubordinate as if clauses also tend to be rather fast in tempo, but show a higher tendency to be accompanied by frowns. Table 5 provides some details on the absolute and relative frequencies of the factor variables (frowns, chunking) as well as mean and standard deviations of the numeric variables (mean pitch, tempo).
Frequencies of the factor variables as well as means and standard deviations of the numeric variables in the final model for as if clauses
The results presented in this section will be illustrated and discussed in the following section.
6 Discussion
The most striking finding is that both irony and syntactically independent as if clauses are significantly associated with prosodic chunks that correspond to syntactic clauses, while non-ironic, subordinate as if clauses can be uttered in various ways. This variety of chunking of subordinate clauses is illustrated by the following examples:Footnote 5
(7) uh- Justice Ginsburg is- is- is- uh passed away less than forty eight hours ago | but- it seems as if (.) uh THIS is moving very fast | and we could have a nominee VERY soon | what can you tell us (NewsScape 2020-09-20_1500_US_KNBC_ Meet_the_Press, 0:03:39-0:03:54, click to view or scan QR code)Footnote 6
(8) The president is campaigning | as if this pandemic is over | holding multiple rallies per day (NewsScape 2020-10-29_0900_ US_CNN_Early_Start_With_ Christine_Romans_and_ Laura_Jarrett, 0:04:53-0:4:59; click to view or scan QR code)
(9) But NOW it looks as if that tornado threat | is still going to be uh impactful across the deep south (NewsScape 2020-10- 10_0900_US_CNN_CNN_Newsroom _Live, 0:12:52-0:13:02; click to view or scan QR code)
Example (7) illustrates a subordinate as if clause that is uttered together with the matrix clause as one prosodic chunk. With only two syllables, the matrix clause in this example is rather short and this might be the reason why further syllables are attached to the prosodic chunk. Prosodic integration of dependent structures into host structures has also been observed in Lelandais & Ferré (Reference Lelandais and Ferré2016). In example (8), though, the matrix clause (the president is campaigning) consists of eight syllables and is thus considerably longer than the matrix clause in (7). This might be one reason why the speaker of (8) opted to chunk the as if clause in a separate prosodic unit. Another possible reason why the speaker has chunked the utterance like this becomes obvious when the video is consulted: when she utters the as if clause, the speaker is (slightly) shaking her head. In doing so, she is not presenting the news in a neutral way, but indicates her stance toward the proposition expressed in the as if clause. Thus, chunking helps the speaker to provide the hearers with clues about the scope of the stance. Finally, example (9) illustrates a kind of chunking of subordinate as if clauses that is also quite common (N = 94). It has been categorized as ‘other’ in the present study, but is probably better described as a topic-comment structure with the topic being chunked as one prosodic unit and the comment in the other, following unit (see also Wells Reference Wells2006: 72–3). In this example, the topic is the tornado threat, which is established in the first part of the utterance and is being commented on in the second prosodic chunk as still being impactful. From a syntactic point of view, in examples like this, the first prosodic chunk consists of the matrix clause, the as if, and the subject of the as if clause (usually in the form of a noun phrase), while the other prosodic chunk consists of the remaining elements of the as if clause. In contrast to example (7), the speaker of (9) can prosodically highlight both the subject and aspects of the remaining clause, while in (7) only one of the two (here: the subject) can be emphasized.
Syntactically independent (i.e. insubordinate and formulaic) as if clauses usually convey one proposition, which is also displayed as such prosodically, as the following examples illustrate:
(10) He just yells back | why you're the one always yelling the questions | as if that's something new (NewsScape 2020-11-21_ 0400_US_FOX-News_Fox_News_at_ Night_With_Shannon_ Bream, 0:37:19-0:37:25; click to view or scan QR code)
(11) And as if that wasn't enough | according to the Washington Post | the Ukrainians send a delegation to the White House in July (NewsScape 2019-10-31_0635_US_KABC_Jimmy_Kimmel_Live, 0:02:02- 0:02:12; click to view or scan QR code)
In example (10), the speaker quotes some other person who presumably said why you're the one always yelling the questions and mocks this person by claiming that this is no new information. This criticism is presented in one prosodic chunk to sufficiently distinguish the quote from the speaker's own stance toward the quote. Likewise, in example (11), the formulaic as if construction is used to link two pieces of bad news and this link is prosodically set off from its surrounding material to emphasize that the following piece of information is just ‘the tip of the iceberg’ in a series of bad news.
These findings are in line with previous research on free constituents (Ford, Fox & Thompson Reference Ford, Fox, Thompson, Ford, Fox and Thompson2002). Free constituents are syntactically and prosodically independent, but semantically related extensions of a previous utterance and are used to provide a stance toward it. Even though the free constituents described in Ford et al. (Reference Ford, Fox, Thompson, Ford, Fox and Thompson2002) are formally noun phrases, the observations made for them can be extended to syntactically independent as if clauses as well. Given these parallels, it could be argued that stance-related constructions tend to be chunked as one prosodic unit. In other words, there could be an abstract construction with [prosodic chunk] on the formal side and [information package] on the meaning side and stance-related information being the particular kind of information conveyed here. Providing direct evidence for such an assumption lies outside the scope of the present article, but the examples above provide some indirect evidence supporting it. Essentially, then, one might assume a crossmodal collostruction between the prosodic chunk construction and syntactically independent as if constructions.
A feature that is significant for an ironic interpretation, but not for as if constructions is tempo. The direction of this finding is surprising because previous research suggested that syntactically dependent structures tend to be slower, but this could not be confirmed. Rather, ironic as if clauses tend to be faster than non-ironic ones. The statistical model reported above suggests that the speaker and the interaction type have an influence on the fixed effects of the model, increasing uncertainties. The interactional data used in Lelandais & Ferré (Reference Lelandais and Ferré2016) are based on a limited number of participants and have been recorded in one setting. In contrast, the present study is based on 482 different combinations of speakers and interaction types. This use of different interaction types and speakers may explain why the findings could not be replicated. It is likewise surprising that ironic as if clauses, independent of the construction, are faster than non-ironic as if clauses, since previous research suggests that irony is slower than non-irony (see section 3.2 above). An alternative interpretation is provided in Ward (Reference Ward2019). Ward reports on a prosodic construction he calls indifference construction (Reference Ward2019: 183–5), which is characterized by a fast tempo (among other things) and usually conveys the speaker's indifference toward their interlocutor's point of view. If this is the case, a prosodic construction is superimposed on a verbal construct, independent of the grammatical construction used, i.e. this is neither evidence for crossmodal collostructions nor multimodal constructions.
In any case, it is surprising that no other feature except prosodic chunking and tempo reached a significant level. Only two further features that entered the model correspond to the ones described in the literature, namely gaze aversion (here: looks to the audience rather than the addressee) and a (slightly lowered) pitch level, albeit non-significantly. None of the other features in the model have been reported to mark irony before or, if they have, not in the predicted direction. However, given the fact that previous research on multimodal markers of irony was also inconclusive or controversial, it might be possible that verbal irony is a heterogeneous phenomenon. More specifically, it seems that the ironic function is an umbrella term (see also Gibbs Reference Gibbs2000; Simpson Reference Simpson and Dynel2011) and that the function of supposedly ironic utterances needs more fine-grained analyses including precise descriptions of the stance conveyed. Essentially, there is no evidence of a set of multimodal features that are linked to an ironic interpretation (i.e. non-verbal ‘irony’ construction(s)) and that are, in turn, associated with as if constructions.
Apart from chunking, there are two further features that distinguish subordinate from insubordinate and formulaic as if constructions, respectively. These are mean pitch and frowning. According to the model, insubordinate constructions are comparatively low in pitch and tend to be accompanied by frowns (albeit frowning was non-significant). The two features are illustrated by the following example:
(12) is now tweeting polls about |.hh America is losing faith in our democracy and our elections | as if this is winning for him | maybe it is (NewsScape 2020-11-19_0300_US_ CNN_CNN_Tonight_ with_Don_Lemon, 0:15:00-0:15:13; click to view or scan QR code)
In this example, news anchor Don Lemon first quotes one of the former US president Donald Trump's tweets after he has lost the election to president-elect Joe Biden in November 2020. After quoting the tweet, Lemon comments on it by rejecting the idea that this might be winning for Donald Trump. However, in the next utterance, he then changes his mind and finds this idea more likely. Both the quote and the as if clause are accompanied by frowning, indicating that the speaker takes a negative stance toward these propositions, while, in the following utterance, his facial expressions become neutral. This use of frowning has already been observed elsewhere. Kaukomaa, Peräkylä & Ruusuvuori (Reference Kaukomaa, Peräkylä and Ruusuvuori2014), for example, show that turn-initial frowning foreshadows trouble talk including negative evaluations and disaffiliation. Given that such use of frowning has been observed elsewhere and given that, in the example, the speaker started frowning before uttering the as if clause, this is good evidence of a crossmodal collostruction.
The mean pitch in this example is also noticeably low. Figure 5 illustrates the pitch movements of the first and the second part of this example.
Pitch movements in example (12)
Figure 5 shows that the first part of the example, the quote, is rather high in pitch, with a mean pitch of 178 Hz. The second part, i.e. the as if clause, on the other hand, is low in comparison, with a mean pitch of 109 Hz. Since syntactically independent structures have been shown to be indicated by lower pitch (Lelandais & Ferré Reference Lelandais and Ferré2016), this might serve as one explanation for the lowered mean pitch of the insubordinate as if construction here. However, the construction is preceded by a pause of more than a second, and, thus, this explanation is unlikely. Traditionally, a lowered mean pitch has also been associated with the speaker's dominance and superiority due to the fact that tall people have longer larynxes and, therefore, lower voices (Gussenhoven Reference Gussenhoven2004). This biologically motivated association still has an influence on how high and low voices are being perceived (see empirical findings in Puts et al. Reference Puts, Hodges, Cárdenas and Gaulin2007). Therefore, assuming a prosodic construction with low pitch on the form side and dominance on the function side seems likely. In example (12), the speaker probably indicates his confidence when rejecting the idea that subverting the outcome of the elections is a winning strategy. To do so, he uses an insubordinate as if construction, which is matched with the ‘low pitch construction’. Assuming a crossmodal collostruction in this case seems feasible.
Formulaic as if constructions, on the other hand, are accompanied by higher than normal pitch, which is illustrated in example (13).
(13) as if this wasn't enough news for today | the show is kind of: topsy-turvy | but let's talk about facebook (NewsScape 2018-04-09_2200_US_FOX-News_Special _ Report_With_Bret_Baier, 0:51:31-0:51:37; click to view or scan QR code)
The pitch movements of this example are illustrated in figure 6.
Pitch movements in example (13)
Figure 6 shows that formulaic as if starts with a high onset (with a maximum of 241 Hz) and then gradually declines with a mean pitch of 154 Hz. The following two utterances are lower in mean pitch (with mean pitches of 131 Hz and 135 Hz, respectively). This finding is in contrast to previous findings on syntactically independent structures (see section 3.1 above). It might be argued that formulaic as if clauses in general, and example (13) in particular, are exceptional, because they occur at the beginning of the turn and, therefore, setting them off prosodically with a lowered pitch is unnecessary. Indeed, 27 percent of formulaic as if clauses (N=20) occur at turn beginnings. However, raises in pitch can also be observed for turn-internal formulaic as if constructions (see e.g. example (11) above). Rather than signaling turn beginnings, the speaker appeals to the audience when they use a higher voice. To be more precise, the prosodic aspects of formulaic as if fit what Ward (Reference Ward2019: 182) describes as the empathy bid construction, i.e. a configuration of prosodic features speakers use when telling a story that, from their perspective, deserves an empathetic uptake. The empathy bid construction is, among other things, characterized by raised pitch, articulated speech and increased loudness. While the latter two features cannot be measured reliably in non-laboratory settings, an informal perception of these confirms their presence in example (13). The function of the empathy bid construction is to seek empathy from the interlocutor. In example (13), being the news, there are no interlocutors, but it might be argued that the news anchor appeals to his audience's empathy. This finding suggests that speakers of formulaic as if constructions tend to bond with their interlocutors by seeking empathy rather than tending to distance themselves from the proposition expressed. Essentially, there is some evidence that formulaic as if constructions are associated with the prosodic empathy bid construction.
7 Summary and conclusion: multimodal as if constructions
In the closing paragraph of section 2, several possible outcomes of this study and their implications for a Multimodal Constructional Grammar were explored. The first possibility was that there might be a set of non-verbal forms indicating irony that match up with as if clauses and support their ironic interpretation, irrespective of the construction used. This possibility could not be confirmed. Even though this might have been expected, an ironic interpretation of the utterance does not have any explanatory power for the non-verbal features. There are only two features distinguishing ironic from non-ironic as if clauses in significant ways, i.e. prosodic chunking as a clause and a fast tempo, and these features alone cannot be considered sufficient for predicting ironic utterances. What is more, only one of the two features, prosodic chunking, was also significantly associated with as if constructions and this alone provides insufficient evidence for a non-verbal irony-construction matching up with as if constructions.
The second possible outcome stated that both kinds of syntactically independent as if constructions might be accompanied by the same non-verbal features due to the fact that they are both syntactically independent. However, the findings above have shown that there are subtle differences in the non-verbal markers that accompany syntactically independent as if clauses, even though both (insubordinate and formulaic as if clauses) are similar in verbal form and function: both constructions are syntactically independent and convey a distancing attitude toward some utterance or event mentioned in the previous context. Despite these formal and functional similarities, the two constructions differ significantly in their mean pitches. As a consequence, it seems that both formulaic and insubordinate as if constructions fulfill related, albeit sufficiently different interactional functions and, crucially, seem to be accompanied by different non-verbal features.
The present study therefore supports the third possible outcome, i.e. that there are individual profiles for each as if construction. What is more, the study also supports the notion of crossmodal collostructions: all the features observed for insubordinate and formulaic as if constructions have been described to work in interaction with other constructions elsewhere, fulfilling the same (or similar) functions. However, the findings reported here also suggest that these co-occurrences cannot simply be multimodal instantiations of several overlapping unimodal constructions. Given their high (statistical) co-occurrence, it seems unlikely that language users always construct these multimodal instantiations on the fly (as suggested by Hoffmann Reference Hoffmann2017). If that were the case, language use would be quite uneconomical. It seems more likely, at least in the case of as if constructions, that language users build crossmodal collostructions, i.e. strong links between different kinds of unimodal constructions (Uhrig Reference Uhrig2018).
Apart from these substantiated conclusions, further, more tentative ones offer themselves. One of these is concerned with the different frequencies of occurrence when the features are considered. For instance, chunking the as if construction in one prosodic unit seems to be a central feature. Pitch, on the other hand, is a moderate predictor, while frowning is only a peripheral aspect of insubordinate as if constructions. The notion of crossmodal collostructions can explain these effects in terms of stronger and weaker associations between the individual constructions. And still, such a view can be complemented by Utterance Construction Grammar, which regards utterances as prototype categories with central and peripheral non-verbal associations. Given that the prototypical nature of verbal constructions has been argued for elsewhere, it seems reasonable to extend this conceptualization to the notion of multimodal constructions. The present article supports the idea that as if constructions have different multimodal profiles and that their features differ as regards cue validity. Therefore, seen as a multimodal gestalt with all features, each construction has a unique profile that enables the hearer to disambiguate the constructions in spoken English.
The present study is based on one family of constructions only and doesn't provide sufficient evidence for Utterance Construction Grammar. Still, the findings on as if constructions show that the grammar of spoken language should not be confined to the analysis of verbal elements alone, but that a multimodal perspective is worth consideration.
Open access
