2.1 Properties of sluicing, swiping and wh-the-hell items

Sluicing is deletion of a wh-interrogative clause to the exclusion of the question constituent, which is derived by movement of the wh-phrase to the Spec,CP position and TP-ellipsis (e.g. Ross Reference Ross, Binnick, Davison, Green and Morgan1969, Merchant Reference Merchant2001). Sluicing can occur in matrix and embedded contexts and with any kind of wh-phrase (see Vicente Reference Vicente, van Craenenbroeck and Temmerman2018 for details and an overview of the existing literature). Thus, PP remnants as well as DP remnants are acceptable, and the latter can also strand their selecting prepositions (when the preposition has an antecedent, Chung Reference Chung, Cover and Kim2006).

Swiping is a subtype of sluicing with a prepositional phrase as remnant, in which the preposition follows the wh-constituent (recall footnote 2 above). Swiping differs from sluicing in some crucial properties. We review these below, restricting our attention to swiping with a single remnant only (see Richards Reference Richards2001, Merchant Reference Merchant, Zwart and Abraham2002 and van Craenenbroeck Reference van Craenenbroeck2010a for multiple remnants).

Two of the uncontested properties of swiping are that swiping only occurs in sluicing (the wh-complement of a preposition can only precede it in TP-ellipsis contexts) and that the preposition in swiping bears accent. The latter is illustrated with reference to (4a) and (4b), where accent is signaled by capitals. While in sluicing with a prepositional remnant, it is the wh-expression that is accented, in swiping accent must fall on the preposition. This observation is crucial for the rest of this paper.

There also appears to be a restriction on the size of wh-constituents that can occur in swiping. With a few exceptions, swiping is claimed to occur with ‘minimal’ (i.e. head-type) wh-items in Merchant (Reference Merchant, Zwart and Abraham2002: 296–297), and van Craenenbroeck (Reference van Craenenbroeck2010a), as seen in the contrast between (4a) and (4b). Hartman & Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009) and Radford & Iwasaki (Reference Radford and Iwasaki2015) on the other hand state that swiping is well-formed with complex wh-phrases as well.

The size of the allowed wh-constituent might be due to systematic variation between American and British English. Whichever way is the correct way of stating this property, the data we are concerned with, namely swiping with wh-the-hell expressions, can only contain minimal wh-expressions (so the issue has no consequence for the discussion to follow). This is due to the independent restriction that aggressively non-D-linked wh-elements are only well-formed with minimal wh-items. In other words, the hell and its ilk is only allowed after simplex wh-operators, as established by Merchant (Reference Merchant, Zwart and Abraham2002) (with reference to Pesetsky Reference Pesetsky, Reuland and ter Meulen1987). Our British English consultants share this judgement.

For this reason, we follow Merchant (Reference Merchant, Zwart and Abraham2002) in saying that the hell is a marker that attaches to wh-heads and forms a single (complex) head with them (see also Den Dikken & Giannakidou Reference Den Dikken and Giannakidou2002 for the same claim). The wh-item and the hell form a single head, and such a single head is predicted to be well-formed in swiping.Footnote ^[3]

Another property of swiping concerns the type and the information structural status of the prepositional phrases that are acceptable in swiping. Rosen (Reference Rosen, Ford, Reighard and Singh1976) was the first to state that swiping is well-formed with prepositional phrases that are non-obligatory in their clause, i.e. they can be implicit arguments or adjuncts, as seen in (7a, b) below. However, as seen in (7c, d), PPs that cannot be omitted in the antecedent, such as predicative PPs or PPs forming part of an idiom, are ill-formed in swiping.

While no explanation has been offered to date for the ill-formedness of predicative and idiomatic PPs, quite some ink has been spilled on the information structural status of those PPs that are acceptable remnants of a swiping clause (i.e. PPs that are implicit arguments and adjuncts). Whether the preposition in swiping can or cannot have an overt antecedent in these cases, in other words, whether its content can be given or new has been subject to some contention. We use the term ‘given’ preposition here with respect to a preposition occurring in the antecedent clause as is standardly done, regardless of the actual position of the correlate PP inside or outside the VP. We consider a preposition new if it is not present in the antecedent clause. See Merchant (Reference Merchant, Zwart and Abraham2002) for an account that capitalizes on this distinction to derive the observation that adjunct PPs can have overt antecedents in swiping. According to Rosen (Reference Rosen, Ford, Reighard and Singh1976), prepositional phrases that do not have an overt antecedent – i.e. PPs that are new in the swiping clause when it comes to information status – are better than those that are given. In her judgement, (8b) is better than (8a), but both are acceptable.

Merchant (Reference Merchant, Zwart and Abraham2002) & van Craenenbroeck (Reference van Craenenbroeck2010a) also consider both versions acceptable and thus allow the preposition in swiping to be given, but Sprouse (Reference Sprouse2006) and Hartman & Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009) state that no preposition in swiping can have an antecedent, ruling out cases such as (8a) and stating that the preposition in swiping must always be new. This issue is highly relevant for us as it directly informs us regarding what information structural status – and in line with it, what kind of prosodic realisation and syntactic position – the preposition has in swiping. The contradictory findings in the literature when it comes to the givenness/newness of the prepositions may potentially be due to a systematic difference between American and British English. To steer clear of such potential intervarietal differences between distinct grammatical systems in this respect, we decided to investigate this aspect of swiping experimentally, in a systematic way, and in only one variety, namely British English. This is also the variety that the rest of paper will make claims about. We leave the question if (and if so, how) American English differs from British English for future research.

Before turning to the experiment and its findings in Section 2.2 below, we close this section by mentioning another aspect of information structure on swiping, namely the contrastivity of the preposition. Radford & Iwasaki (Reference Radford and Iwasaki2015) claim that contrastivity has no effect on the acceptability of swiping: swiping is equally allowed with a non-contrastive and a contrastive preposition, where contrast is defined with respect to a preposition in the antecedent, as seen in (9):

Hartman & Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009) claim that the preposition cannot be contrastive. Since all four of our British English consultants accept examples like (9), we follow Radford & Iwasaki (Reference Radford and Iwasaki2015) in our characterisation of swiping, and claim that the preposition in swiping can be contrastively focused in the language variety we are interested in.

2.2 The givenness/newness of swiping remnants: An experimental study of British English

We have conducted a web-based acceptability judgement experiment to address the question whether the givenness/newness of the preposition impacts the acceptability of swiping. Experimental items were constructed across two factors: ellipsis (swiping vs. non-elliptical structure with P-stranding) and the information status of the preposition (given vs. new). If givenness impacts the acceptability of swiping, we expect that acceptability ratings should drop with respect to non-given prepositions in swiping (in the ellipsis condition). As a similar effect has not been reported for non-elliptical sentences in the literature (as far as we know), we do not expect such a drop in non-elliptical sentences. For this reason, we have used non-elliptical equivalents of the swiping examples (with P-stranding) as controls in the experiment in a full factorial 2 $\times$ 2 design, where the factors were the presence of ellipsis and givenness of the preposition. We created four conditions, illustrated in Table 1. There were the following seven lexicalisations for each condition: dance with, swim with, talk to, laugh at, look at, listen to, complain to, all containing monosyllabic prepositions. In each case, the optional prepositional phrase was adjacent to the verb (which is predicted to be possible in swiping – recall Rosen’s (Reference Rosen, Ford, Reighard and Singh1976) generalisation illustrated by examples (7a, b) above).Footnote ^[4]

Table 1 The 2 $\times$ 2 design of the experimental item sets.

As the experiment was designed to test the (null) hypothesis that there is no difference between given and new prepositions in swiping – thus challenging earlier published statements – we decided to test the difference between given and new prepositions in three distinct ways. The experiment is hence composed of three tasks.

The first was a forced choice task (Task 1), in which participants were presented with two sentences and they had to pick the most acceptable sentence. Comparisons were made between Conditions 1 and 2, and similarly, between Conditions 3 and 4 (refer to Table 1). The order of the sentences varied (in some cases the first sentence was one with a new preposition, in others, the one with a given preposition was first). This task included four experimental items and six fillers. It was preceded by four practice items containing sluicing and predicate-ellipsis, whose purpose was to introduce the contexts needed for judging the experimental items. A sample of the practice set-up with a context is given below:

The participants were instructed to think of similar contexts when judging the experimental items. All items were randomised.

The second task was a five-point Likert scale acceptability judgement task (Task 2), in which the participants were asked to grade how acceptable the items were (1 = completely unnatural, coloured red; 5 = completely natural, coloured green). Grades 2, 3 and 4 were not labelled but were coloured orange, yellow and blue, respectively. In addition to the four experimental items, the task also contained four practice items (with contexts specified) and six fillers. All items were randomised.

The last task was a yes/no task (Task 3), in which participants were presented with individual sentences and asked to answer the question ‘Is this sentence good?’ by clicking either ‘yes’ or ‘no’. In case of doubt, they were instructed to click ‘yes’ if the sentence was, in their judgment, closer to being good, and ‘no’ if it is closer to being bad. In addition to the four experimental items, the task also contained four practice items (with contexts specified) and six fillers. All items were randomised.

We distributed the test items with different lexicalisations among two participant lists, in order to keep the questionnaire at a manageable length. Except for the forced choice task, none of the four conditions that are tested shared the same lexicalisation. Other than the lexicalisations, the two surveys were completely identical in every respect. In total, participants judged 42 sentences in both lists.

Materials were presented and recorded using the Qualtrics platform, and distributed via the online data collection platform Prolific. A total of 108 participants were recruited online via Prolific, 52 participants completed the first list and 56 participants completed the second list. Participants received a compensation of £1.70 (first list) and £1.36 (second list). All participants had British nationality, were native speakers of British English, raised monolingual with English as their first language. They were aged between 20 years and 50 years, and all had a university degree.

2.2.1 Results of the experiment

In the forced choice task (Task 1), we tested whether participants find swiping more acceptable with new prepositions rather than with given prepositions. Table 2(a) shows the amount of preference for each test condition.

Table 2 (a) Number (and percentage) of answers selecting a condition as better than other. (b) Number (and percentage) of answers selecting a condition as better than other.

The null hypothesis is that the participants choose between given and new prepositions in the swipe at chance level, which is 50%. To test whether participants either preferred a new preposition or a given preposition, a binomial test was run using the binom.test() function in R. In the swipe condition, given and new preposition were chosen equally often (p = .773). This shows that there is no significant difference between the acceptability of swiping with given and new prepositions.

Indication that the participants were sensitive to the task, and did not judge all examples randomly comes from two observations. On the one hand, practice items that compared fully acceptable and fully unacceptable elliptical examples were judged according to expectation, as shown by the following pair of examples:

Second, in the non-elliptical conditions where we tested the effect of the information status of the preposition on non-elliptical equivalents of swiping, participants picked the sentence with a new preposition significantly more often than at chance level (p = .000065), as the raw numbers in Table 2(b) also illustrate at a glance (see footnote 5 on this).

In the Likert scale task (Task 2), we asked the participants to judge each experimental item on a scale from 1 (completely unnatural) to 5 (completely natural). The mean scores and the standard deviations are given in Table 3.

Table 3 Mean acceptability judgement ($\pm$SD) per condition.

Using the lmer() function from the lme4 package in R (Bates et al. 2013), a linear mixed regression model was run to test for differences between means. Condition was taken as a fixed effect. A random slope was included for each participant. To obtain pairwise comparisons, a post-hoc test with Bonferroni correction was performed using the emmeans() function (Lenth Reference Lenth2018). In the swipe condition, no significant difference in acceptability was found between sentences with a new preposition and sentences with a given preposition (p = .0932). This result confirms the result of Task 1 for swiping. Similarly, no difference was found in the non-elliptical condition (p = 1.00), which is in line with the expectation that non-elliptical sentences are acceptable with both given and new prepositions.

In the yes/no task (Task 3), participants had to decide whether or not a sentence was acceptable by choosing between ‘yes’ or ‘no’. The distribution of these answers is presented in Table 4.

Table 4 Raw numbers of ‘yes’ and ‘no’ answers per condition.

With the help of the pairwiseNominalIndependence() function of the rcompanion package (Mangiafico Reference Mangiafico2018), pair-wise comparisons were established with Bonferroni correction, conducting chi-square tests. In the swipe condition, sentences with a given preposition were accepted significantly less often than sentences with a new preposition (p = .0446). Note, however, that under the swiping condition, 87.96% of the participants judged the given preposition as good. In the non-elliptical condition, no significant difference was found (p = 1.00).

The results of the three tasks indicate that there is no statistically significant difference between the acceptability of given and new prepositions in swiping according to Task 1 and Task 2, and there is a marginal difference between them according to Task 3. Since the results of Task 3 approach non-significance, we take this test to point in the same direction as the other two tests: we claim that the givenness of the preposition does not play a role in the acceptability of swiping for the British English speakers we consulted.

The same can be said about the givenness of the preposition in non-elliptical contexts, according to the evidence of Task 2 and Task 3. We leave the contradictory results of Task 1 for the non-elliptical control conditions for further research, as it is tangential to our present purposes.Footnote ^[5]

2.3 The syntactic configuration of sluicing and swiping

Having established the properties of sluicing and swiping in the previous two subsections, now we are in position to set our assumptions about the syntactic configuration of swiping and sluicing.

Concerning the syntax of swiping in British English, we assume that swiping involves two leftward movement steps: a movement of the PP out of the ellipsis site, followed by a movement, which strands the preposition. We adopt the proposal by Richards (Reference Richards2001), who dissociates these movement steps from focusing, and concerning the details, we specifically follow van Craenenbroeck’s (Reference van Craenenbroeck2010a) split CP-variant of the Richards-style analysis, illustrated in (12), in which both steps of movement target a high left peripheral position, labelled CP$_{1}$ and CP$_{2}$ below.

Importantly, given the results of our five-point Likert scale acceptability judgement task (Task 2), we take it that the movement of the PP out of the ellipsis site does not take place to a specific focus position, triggered by an (information or contrastive) focus feature. Instead (again following van Craenenbroeck Reference van Craenenbroeck2010a), we term the lower CP position as an operator position and the higher one a clause-typing position.

In other words, we reject proposals in which the derivation of swiping involves a crucial movement step that manoeuvres the preposition into a focus position, to account for the fact that the preposition is always new and thus receives information focus – and focal accent as a result. We do this as we have observed in Section 2.2 that givenness/newness of the preposition has no effect on the acceptability rates among our British English consultants. Prepositions in swiping can but need not be given and can but need not be contrastive with respect to a preceding PP. On the basis of these observations, we conclude that the syntactic distribution (and the accent) of the P in swiping is not mediated by information structure. We thus claim that the position of the preposition is not due to focus movement of the PP (Kim Reference Kim1997, Hartman & Ai Reference Hartman, Ai, Grohmann and Panagiotidis2009), or focus movement of the P alone (Radford & Iwasaki Reference Radford and Iwasaki2015).

With the above syntactic derivation, we capture not only that swiping coexists with the possibility of P-stranding across languages (Merchant Reference Merchant, Zwart and Abraham2002, van Craenenbroeck Reference van Craenenbroeck2010a, Radford & Iwasaki Reference Radford and Iwasaki2015), but also that the wh-phrase in swiping can be at a long distance from its preposition, crossing a finite clause boundary (van Craenenbroeck Reference van Craenenbroeck2010a, Radford & Iwasaki Reference Radford and Iwasaki2015), as illustrated in (13):

Concerning the question why swiping (and thus swiping with WTH as well) only exists under ellipsis, we follow van Craenenbroeck (Reference van Craenenbroeck2010a) in taking this to be an instance of ellipsis repair: PP movement to CP$_{2}$ should bar further movement of a subconstituent to a higher position in non-elliptical clauses due to non-uniform chain formation; ellipsis, on the other hand, repairs this violation by eliminating illicit traces in TP.

As for sluicing, we adopt the standard analysis whereby sluicing involves movement of the wh-phrase to the Spec,CP position and TP-ellipsis (e.g. Ross Reference Ross, Binnick, Davison, Green and Morgan1969, Merchant Reference Merchant2001), as illustrated in (14).Footnote ^[6]

2.4 Back to the puzzle: Can syntax or semantics provide an explanation?

We now turn to the following question: Can the unavailability of WTH in British English sluicing examples like (2b) above be explained with reference to syntax or semantics? As it has been noted in earlier works that WTH has syntactic and semantic licensing requirements, we review these requirements one by one to see whether they are satisfied under sluicing. The outcome of this discussion will be that they are satisfied; therefore, these aspects of syntax and semantics do not provide an explanation for our puzzle.

The first observation concerning the syntactic licensing of WTH is that these phrases cannot appear in situ in English, as seen in (15a), but are perfectly well-formed when moved to Spec,CP (Den Dikken & Giannakidou Reference Den Dikken and Giannakidou2002, Huang & Ochi Reference Huang, Ochi, Moulton and Wolf2004):

Clearly, this licensing requirement is satisfied in the environment of sluicing (Sprouse Reference Sprouse2006): as Merchant (Reference Merchant2001) has argued, sluicing remnants always correspond to wh-phrases moved to Spec,CP, as (16) illustrates:

Furthermore, as Den Dikken & Giannakidou (Reference Den Dikken and Giannakidou2002) have shown, WTH phrases are polarity items that are inadmissible in veridical contexts. This means that they must occur in non-veridical environments (such as in the scope of a negative quantifier or only, being in a conditional). When it comes to their occurrence in matrix and embedded wh-clauses, antiveridicality requires that they should be licensed as polarity items. Licensing can be done by the matrix question operator Q in its c-command domain, as in (17a), interrogative-selecting verbs such as ask or wonder, as in (17b), or negated veridical verbs like know, tell and matter (but see Hoeksema & Napoli Reference Hoeksema and Napoli2008 for further refinements concerning the set of predicates), seen in (17c, d).

Den Dikken & Giannakidou (Reference Den Dikken and Giannakidou2002) claim that in these contexts, WTH phrases are properly licensed as polarity items.

Exactly the same syntactic configurations, however, do not give rise to well-formed WTH phrases if the clause in which they occur undergoes sluicing (TP-ellipsis):

Clearly, the polarity item licensors are present in these elided clauses as well, as the syntactic configuration and the lexical content of the sentences are the same: (18a) contains a Q operator c-commanding the WTH; (18b) contains an interrogative selecting predicate (wonder), (18c, d) contain the negated veridical predicate know and matter. This shows that the unacceptability of WTH in sluicing is not due to the lack of polarity item licensing as specified above.

Why would WTHs be unavailable in sluicing, then? Even though polarity item licensing is satisfied, a potential semantic problem might arise in the way the WTHs are anchored to the previous context – this is in fact the explanation that Den Dikken & Giannakidou (Reference Den Dikken and Giannakidou2002) offer to explain the puzzle. According to them, the unacceptability of WTHs in (18a–d) is due to the specific discourse context that sluicing requires, which disallows the use of WTHs. In sluicing, there is a specific indefinite with an existential presupposition, which serves as a correlate to the wh-remnant phrase. In their analysis, aggressively non-D-linked WTHs are a type of dependent indefinites, meaning that they do not assert existence and cannot be anaphoric to a previously introduced discourse referent. Since in sluicing the wh-remnant must be anaphoric to an indefinite, WTHs fail to occur in sluicing as they cannot be linked to discourse familiar entities.

The problem with this explanation is that WTH-phrases do not rule out discourse antecedents in the exact same discourse environment when they are not followed by ellipsis. It is perfectly straightforward to construe questions with WTH phrases using episodic tense as genuine information questions that seek information about the identity of a referent (as Den Dikken & Giannakidou Reference Den Dikken and Giannakidou2002 themselves note elsewhere in their article). Indeed, our consultants considered the full clause equivalents of (18a–d) well-formed with this reading – see the data in (19).Footnote ^[7]

In these cases, too, there is a presuppositional indefinite which serves as a correlate to the WTH expression and the sentences are well-formed. This shows that the unavailability of WTHs in sluicing cannot be due to the specific discourse context sluicing requires in these examples.

It is important to stress that in the examples discussed in this section, we are providing cases of WTH that have a genuine information seeking reading. As we show above in (19), such WTH constructions are well-formed with episodic tense in them and are compatible with an existential presupposition in an antecedent clause. These WTHs therefore should be discourse-licensed in sluicing. We are not making claims about the acceptability and the well-formedness of WTH constructions with a negative rhetorical meaning, such as Who the hell would do such a thing?, which constitute another type and use of WTH. WTHs with negative rhetorical reading have distinct semantic and discourse properties: as Den Dikken & Giannakidou (Reference Den Dikken and Giannakidou2002) show, negative rhetorical readings of WTHs must be licensed by a modal and expect a negative answer, which is incompatible with an existential presupposition. The way we differ from Den Dikken & Giannakidou (Reference Den Dikken and Giannakidou2002), then, is that we make a distinction between genuine information seeking WTHs and negative rhetorical WTHs when it comes to availability in sluicing, and we claim that genuine information seeking WTHs are properly discourse-licensed in sluicing as they are compatible with an existential presupposition of a correlate.

That WTHs and presuppositionality are not incompatible is clear from the observation that strongly presuppositional it-cleft questions also allow for the use of WTH, with information-seeking content, as the following examples illustrate.

Last but not least, there are languages, such as Hungarian, in which WTH expressions are well-formed in elliptical and non-elliptical sentences alike, see the case of sluicing with a WTH expression in (21).

This shows that the specific discourse context required by sluicing can support WTH phrases, further reinforcing our claim that the unavailability of WTHs in sluicing in English cannot be due to discourse semantic factors.

3.1 Main prosodic properties of swiping with and without the hell (TH) in British English

We have established the prosodic properties of swiping on the basis of the existing literature and consultations/recording sessions with four British English speakers (three female and one male). The speakers were individually consulted in speech about the acceptability of sluicing and swiping sentences with and without WTHs as well as the non-elliptical equivalents of such sentences (84 sentences in total), and were recorded when reading out the acceptable examples, which were then inspected in PRAAT (Boersma & Weenink Reference Boersma and Weenink2011). All British English examples illustrated with a pitch track below come from this pool of sentences and are given as a representative of the prosodic profile of the structures under discussion.

As for the accentual properties of the preposition in swiping, it has been reported in the literature that the preposition in swiping (with or without TH) bears a high pitch accent (i.e. a pitch peak associated with a stressed syllable, H*). When asked about the prominence on the preposition in swiping, our speakers commented that the preposition is perceived as the most prominent unit of the elliptical clause. The nature of this accent is sentence-level nuclear pitch accent (NA), which is the accent that is perceived as the most prominent accent in an intonational phrase.

Prosodically, among the phonological phrases within an intonational phrase, an NA in British English (but also German and Dutch) is the highest pitch accent (H*) within the rightmost phonological phrase of that intonational phrase (see e.g. Truckenbrodt Reference Truckenbrodt1995, Reference Truckenbrodt1999, Reference Truckenbrodt and Brown2006; Ladd Reference Ladd1996; Féry Reference Féry2011). Although the prosodic description of an NA states that the linearly rightmost accent is the NA of that sentence, this does not necessarily mean that an NA always falls onto the rightmost syntactic constituent/word of a clause. An NA may be followed by other items that do not bear phonological phrase-level pitch accent.

As for the location of an NA, we observed that not only prepositions that are new, but also prepositions that have a correlate in the antecedent clause can be the locus of the NA of that clause. Figure 1 illustrates a case in which a new preposition (in this case about) bears the NA within the intonational phrase that corresponds to the elliptical clause (i.e. within the string but I don’t know who the hell about) – the intonational phrase boundary tones are marked with a % sign next to the tone of that boundary, e.g. H% for a rising boundary, and L% for a falling boundary.

Figure 1 A sample pitch track of a new preposition in swiping+the-hell: The preposition bears the final and the nuclear accent while the-hell bears down-stepped pre-nuclear accent (Speaker 1, male).

In Figure 1, focusing on the clause with swiping, we observe that both the hell and the wh-item bear a high pitch accent. The accent on the wh-item is realised at a relatively lower level than the preceding pitch accent – maximum pitch on the wh-item is 123 Hertz (Hz), and the maximum pitch of the preceding item (i.e. don’t) is 134 Hz. This indicates a downstep from the utterance-initial H* to the accent on the wh-item, which is marked with ! on the downstepped tone: !H*. The high pitch accent on TH is also downstepped, bearing a maximum pitch of 117 Hz, which is relatively lower than the accent on the wh-item. The last accent-bearing unit in this clause is the preposition about, the maximum pitch of which is 130 Hz. The accent on the preposition is not downstepped, and P is perceived as the most prominent unit in this clause. The accent on the preposition about is hence an NA.Footnote ^[8] In this particular example, and in the rest of the tokens with WTH, we observed that TH always precedes NA.

A sample pitch tack of a swiping construction without TH is given in Figure 2. In this case, the preposition is given in the antecedent clause. Recall from the acceptability judgement experiment that similar sentences, against the statements of the previous literature, were considered acceptable.

Figure 2 A sample pitch track of a given preposition in swiping: The preposition bears nuclear accent (Speaker 1, male).

In Figure 2, the pitch peak on the preposition that is in the clause with swiping clearly shows that the preposition bears the final and hence the nuclear accent of this clause. This novel observation suggests that the accent on the preposition in swiping structures cannot be related to the information structural status of P. In addition to the accent on about, we also observe that the wh-item bears a !H* tone.

The pitch track of a swiping construction with TH is given in Figure 3. Similar to the previous case, the preposition in Figure 3, which is information structurally given, heads an adjunct PP.

Figure 3 A sample pitch track of a given preposition in swiping+the-hell: The preposition bears nuclear accent and the-hell bears a pre-nuclear accent (Speaker 1, male).

The figure shows the pitch peak on the utterance-final preposition about in the elliptical clause; about hosts the clause’s nuclear accent, which makes it the most prominent unit of this clause. We also observe a linguistic pause in the juncture of TH and P. The duration of this pause is 100 ms. On the string of WTH, there are two accents, the latter of which is a downstepped high accent. One of these accents is on the wh-item and the other on TH. Similar to the previous case of WTHs presented in this section, in this example, too, we observe that TH precedes the item that bears NA, i.e. an utterance-final preposition.

Following the prosodic examination of the pitch tracks, we conclude that the prepositions in swiping always bear the NA in swiping, regardless of whether or not they are information structurally given. The presence of TH does not affect the location of NA. As such, swiping structures with TH still locates the NA on the preposition. Additionally, TH always precedes an NA-bearing item. Moreover, we also observed that both the wh-item, and TH bear accent in swiping, however these are never NA. Lastly, in Figure 3 (and in other similar cases we recorded), we observed that in a string of WTH+P in swiping constructions, there is often a linguistic pause at the juncture of WTH and P, which indicates the presence of a prosodic boundary at this juncture.

Building on the well-attested fact that the preposition in a swiping construction always bears prosodic prominence, Sprouse (Reference Sprouse2006) suggests that the puzzle in (2) is sourced from a prosodic conflict. In line with Sprouse (Reference Sprouse2006), we maintain that the licensing conditions that constrain the well-formedness of WTHs are prosodic in nature. However, we show in the rest of this section that Sprouse’s (Reference Sprouse2006) account is untenable. In Section 4, we offer an alternative prosodic account to resolve the puzzle in (2).

In the next subsection, we introduce Sprouse’s account. In Section 3.3, we list its shortcomings. In Section 4, we posit a novel prosodic account for why swiping, and not sluicing, is a licit (prosodic) host for WTH.

3.2 A pioneering study: Sprouse (Reference Sprouse2006)

Sprouse (Reference Sprouse2006) claims that WTH is licensed in swiping as a result of the fact that the accented preposition creates the ideal prosodic environment for WTH. Sprouse’s proposal is based on a number of observations and assumptions, all of which we will discuss below.

First, Sprouse’s account is based on the observation that the preposition bears accent in the context of swiping.Footnote ^[9] He claims that the preposition in swiping is focus marked, and has a [+Focus] feature. Relating the accent on the preposition and the alleged requirement that the preposition in swiping should be discourse new, Sprouse proposes the Accent Projection Principle (APP), which states that every focus domain (i.e. the domain of a focus marked/new item) must bear phrase accent. Since the preposition in swiping bears phrase accent, Sprouse (Reference Sprouse2006) concludes that the preposition in swiping is contained within a focus domain, and is marked with [+Focus]. As pointed out by Sprouse (Reference Sprouse2006), this observation is in line with Gussenhoven’s (Reference Gussenhoven1984) Sentence Accent Assignment Rule in (22).

The rule in (22) states that as long as they are new/focused, prepositions and adjuncts receive accent and form their own focus/prosodic domains.Footnote ^[10] Building on Merchant’s (Reference Merchant, Zwart and Abraham2002) observation that the preposition of swiping cannot have an antecedent, hence must be new, Sprouse suggests that the accent on the preposition of swiping is due to the fact that it bears [+Focus], and hence forms its own focus domain, which is in line with the predictions of SAAR about adjuncts.

Secondly, following Gussenhoven’s (Reference Gussenhoven1984: 29) observation for certain moved wh-items, Sprouse assumes that every wh-item must bear a focus domain of itself. Unlike the focus domains that are generated by other kinds of arguments, the domain of the wh-word crucially cannot host another item.

Drawing on these two assumptions, Sprouse represents the organisation of focus/prosodic domains in swiping as in (23) below, and in sluicing as in (24) (Sprouse Reference Sprouse2006: 6 examples (12c) and (12a), respectively). In these and subsequent examples, following Sprouse’s (Reference Sprouse2006) notational system, each focus/prosodic domain is set in square brackets, and each phrasal accent-bearing unit is set in bold.

Lastly, Sprouse claims that TH cannot bear prosodic prominence/accent. Given Gussenhoven’s (Reference Gussenhoven1984: 28) statement that any material that does not form its own focus domain (i.e. all items that are given or cannot project accent for some reason) must be contained within the neighboring focus domain, Sprouse predicts that the prosodic environment that is generated in sluicing is prosodically unfit to host TH. This is because TH cannot project its own focus domain, and hence it must be contained within the focus domain of the item adjacent to it.

In sluicing, as illustrated in (25) (from Sprouse Reference Sprouse2006), the domain of wh-word is the only domain to which TH is adjacent. However, the focus/prosodic domains of wh-items are suggested not to be able to host other items. Therefore, in sluicing, TH forms a separate focus/prosodic domain, to the exclusion of the wh-item. However, since every focus domain must bear accent (recall APP), and since TH cannot bear accent, TH in sluicing is prosodically not licensed. The schematic representation of a WTH in sluicing in (25) demonstrates the unacceptable prosodic configuration in which TH is contained within a focus/prosodic domain, without bearing an accent.

Unlike sluicing, the prosodic structure that is generated in swiping is endowed with a focus/prosodic domain that is able to host items such as TH. This host is the domain that is generated by the preposition of swiping, which bears a phrase accent. In (26) (from Sprouse 2006: 6 example (12d)), in addition to the focus/prosodic domain of the wh-item, there is another focus/prosodic domain in the vicinity of the TH. This is the domain that contains the preposition. Unlike the wh-items, the P’s focus/prosodic domain is capable of hosting items other than the accented element. TH is included in the domain of P, which not only satisfies Sprouse’s condition that every constituent is contained within a focus domain with an accented item, but also allows TH to remain unaccented, another property that Sprouse attributes to TH.

Although Sprouse’s prosodic licensing account predicts the crucial contrast between the acceptability of WTHs in sluicing and swiping, it is also burdened with a number of shortcomings. Below we list three of these shortcomings, which lead us to posit a novel prosodic account in Section 4 for the licensing of WTHs, not only in elliptical sentences but also in non-elliptical sentences.

3.3 Shortcomings of Sprouse (Reference Sprouse2006)

The most crucial problem with Sprouse (Reference Sprouse2006) is its claim that aggressively non-D-linked items reject prosodic prominence/accent, with the help of which the unacceptability of WTHs in sluicing is explained. This claim, however, does not find empirical support. The TH part of a WTH string may bear phrasal accent (see also Merchant Reference Merchant2001). The prosodic representation given in Figure 4 below is taken from O’Connor & Arnold (Reference Joseph D. and Arnold1973), which depicts the prosody of an utterance that starts with a WTH, in this case what on earth. In the annotation system they employ, prosodically prominent/accented syllables are marked with larger dots. As seen in the annotation, the word earth, which is the last item of the WTH string, is more prominent than the preceding wh-item (TH in this case would be annotated with a high pitch accent, H*, in ToBI labelling system of Auto-segmental Metrical theory).

Figure 4 A sample representation of the intonational contour of what on earth with the British School annotation (O’Connor & Arnold Reference Joseph D. and Arnold1973).

Similar to O’Connor & Arnold’s (2001) observation, the pitch track in Figure 6 below, from our own recordings, shows that TH is not only capable of hosting a high pitch accent, but also this accent may even be more prominent than the wh-item that precedes it. In Figure 5, the mean f0 of the vowel of the wh-word is 192 Hz, and the mean f0 of vowel of the word hell is 201 Hz. In addition to this difference in the pitch register, while the wh-item bears a low pitch accent (L*), TH bears a high pitch accent (H*).

Figure 5 A sample of what the hell: The hell is accented (Speaker 3, female).

Figure 6 A sample of what the hell in swiping: The preposition is preceded by a pause (Speaker 1, male).

A second problem with Sprouse (Reference Sprouse2006) concerns the prediction about the behaviour of given prepositions. Gussenhoven (Reference Gussenhoven1984: 29) states that moved wh-items tend to form an independent focus/prosodic domain only when the moved wh-item is not followed by given material. When the wh-item is followed by given material, any constituent that is [–Focus] is included within the focus domain of the wh-item. This by itself predicts in the theory of Sprouse (Reference Sprouse2006) that when given in the antecedent clause, WTHs should be licensed in sluicing. This is not borne out:

Besides this, given Gussenhoven’s generalisation, Sprouse claims that the preposition in swiping bears accent because it is always new, and hence it cannot be fused into the domain of the wh-item.

As we have shown in Section 2, the preposition in swiping does not always have to be new. Discourse given prepositions are also perfectly acceptable in swiping. Thus, the accent on the preposition that the previous literature (including Sprouse (Reference Sprouse2006)) has observed cannot be related to its information structural status. An additional problem is that although information structurally given material may be fused into the focus/prosodic domain of the wh-item, as Gussenhoven states, this does not take place in swiping (recall Figure 2 above). Regardless of the fact that the preposition is information structurally given, see (28a), a constituency that fits to the representation in e.g. (28b) is not attestable, an observation that comes from Merchant (Reference Merchant, Zwart and Abraham2002). This behaviour cannot be explained by Sprouse’s account, as there is nothing that would motivate the accent on an information structurally given preposition.

The last problem that the proposal in Sprouse (Reference Sprouse2006) encounters is that it predicts a non-attested prosodic grouping for WTH+preposition. In Sprouse’s representation of WTH in (26) – repeated in (29) below – the wh-word and TH are contained within separate focus/prosodic domains. This implies the presence of acoustic and tonal cues that marks such a disjoint prosodic phrasing. In fact, Sprouse’s account predicts such a prosodic detachment not only in the cases of swiping but also in all occurrences of WTHs. Sprouse’s prediction is schematized in (30), in which the X, which follows TH, is an accented item that has the feature [+Focus].

Such a split prosodic grouping of wh-item and TH is simply not attested in the samples we collected. In fact, our British English consultants considered such a split grouping in swiping such as (31), and non-swiping environments such as (32), to be markedly unnatural and forced. The acoustic analysis of the recorded samples of these sentences also showed that there is no such prosodic boundary between the wh-item and the TH string (consider Figure 6).

From a syntax–prosody mapping point of view, the avoidance of a split parse is expected. This is because in syntax, wh-item and TH form a complex head, and this syntactic head is expected to be mapped into a prosodic domain.

Furthermore, Sprouse’s claim that the TH and the preposition in swiping are contained within a single prosodic domain – as illustrated in (31) – is not attested, either. In fact, the acoustic and tonal analyses of such strings clearly show that the preposition is parsed separately from WTH. Figure 6 illustrates the prosody of WTHs in swiping, where there is no metrical and audible boundary between the W and the TH, but there is a clear boundary between WTH and P.

As seen in Figure (6), the preposition is separated from the preceding string – hence from the TH – with an audible linguistic pause (duration of the pause is 130 ms). The pause immediately follows the /l/ phoneme of the TH and ends with the onset of the preposition, which is clearly visible in the waveform. Note that such a pause was present in most of the tokens we recorded. Given Sprouse’s prosodic representation, such a pause is not expected, as TH is assumed to be fused into the prosodic domain of the preposition.

3.4 Interim summary

Focusing on the prosodic properties of WTHs and the preposition in swiping, we have established the prosodic profile of the construction and made two important observations. First, WTHs may bear accent but cannot bear nuclear accent. Second, the preposition in swiping does bear nuclear accent, crucially not only when it is new but also when it is given. This indicates that the prominence on the preposition is not for marking focus.

Having ruled out syntax and semantics as possible sources for the unacceptability of WTHs in sluicing in Section 2, we now conclude that prosodic factors may play a role in licensing WTHs, in line with Sprouse (Reference Sprouse2006). Crucially, however, the prosodic account provided in Sprouse (Reference Sprouse2006) is not fully successful in accounting for the wider range of the data, and a number of its assumptions are not empirically supported. In the next section, we therefore present a prosodic account inspired by Sprouse (Reference Sprouse2006), but differing from it in certain ways, to predict the distribution of WTHs in not only elliptical clauses but also in non-elliptical ones.

4.1 Prosodic licensing of WTH in swiping/sluicing

Following Sprouse (Reference Sprouse2006), we claim that the well-formedness of WTHs in British English is prosodically conditioned. However, dissimilar to Sprouse’s claim, our proposal is based on two non-trivial facts that we reported in previous sections: (i) WTHs can bear accent but not nuclear accent, and (ii) the prepositions in swiping are accented, but this cannot be because of their information structural status.

Our account has two main ingredients. The first concerns the well-formed prosodic distribution of WTHs. The second concerns the nature and source of the accent placement on the preposition in swiping constructions.

We propose that there is a very strictly observed prosodic licensing condition on the well-formedness of WTHs in British English, which is active, not only in elliptical sentences but also in non-elliptical sentences. The condition we propose is as follows:

The condition in (33) states that the use of WTHs is well-formed in a sentence as long as it precedes nuclear accent-bearing item of that sentence. This condition also makes a number of predictions, one of which is that a WTH in sentence-final position is not well-formed, as a sentence-final WTH cannot be followed by NA. Similarly, when a WTH bears an NA, or follows an NA, (33) predicts unacceptability. This is simply because none of these prosodic configurations are fit for the well-formedness of WTHs; see (34a) below for the schematic representation of the former and (34b) for the latter. When a WTH is followed by any item capable of bearing a nuclear accent, the required prosodic environment is obtained for it; this is schematised in (34c).

As for the assignment of nuclear accent in British English, we adopt Cinque’s (Reference Cinque1993) syntax-oriented stress deepest algorithm (SD), which states that, in information structurally uniform sentences, the syntactically deepest constituent in a clause receives nuclear accent, and is perceived as the most prominent unit at the intonational phrase-level.Footnote ^[11] By information structurally uniform sentences, we mean cases in which all items in a sentence are all-new or all-given. An information structurally imbalanced sentence is a sentence with focused/given items, in which case the sentence exhibits information structure related prosodic grouping and tone insertion in intonation languages such as English and Dutch (see Zubizarreta Reference Zubizarreta, Féry and Ishihara2016). In cases of all-new or all-given contexts, the sentence bears syntactically navigated default NA (see Büring Reference Büring2016, and references in there).

To discuss the NA placement in the pronounced parts of a clause with ellipsis, we adapt SD to clauses with ellipsis. We suggest that the assignment of nuclear accent in clauses with ellipsis in British English is identical to information structurally homogeneous clauses without ellipsis. As such, unless they bear contrastive information, all items that are realised outside of the ellipsis site are considered as information structurally uniform; hence SD, which is a syntax-oriented default nuclear accent placement algorithm, is active.

Importantly, we claim that, in elliptical clauses, SD applies only to the non-elided part of the clause. This needs to be stated explicitly because, unlike in non-structural approaches to ellipsis (Culicover & Jackendoff Reference Culicover and Jackendoff2005, Sag & Nykiel Reference Sag, Nykiel and Müller2011), SD applying to the non-elided material is not self-evident in structural approaches to ellipsis, which countenance the concept of an ellipsis domain and silent structure within it. In structural approaches, the deepest item that is considered for prominence placement is not necessarily the syntactically deepest item in the entire clause. In (35), we therefore present the nuclear accent assignment algorithm that we advance for elliptical clauses in structural approaches, for which we may speculate that NA placement takes place post-ellipsis.Footnote ^[12]

The algorithm in (35) is architecturally motivated given that, in the reverse Y model of grammar, phonological operations apply after the insertion of vocabulary items (Embick & Noyer Reference Embick, Noyer, Ramchand and Reiss2007). As such, only those syntactic terminals that are not elided are visible to phonological operations (see e.g. Nespor & Vogel Reference Nespor and Vogel1986, Kentner Reference Kentner, Trouvain and Barry2007 and Truckenbrodt Reference Truckenbrodt2013, for the conclusion that ‘phonetically empty’ material such as the ellipsis site is ignored by prosody).

Returning to the puzzle in (2), with the condition in (33) and the algorithm in (35), we can account for why WTHs are not allowed in sluicing, but are allowed in swiping.

In sluicing, the deepest constituent that is outside of the ellipsis domain is the wh-item. As a result of its structural distribution, the wh-item receives nuclear accent in sluicing constructions (36). When a TH adjoins the wh-item (as in 36b), the entire string of the WTH receives the nuclear accent. This fails to satisfy the prosodic well-formedness condition in (33).

Swiping creates the desired prosodic environment to license WTH, simply by virtue of the fact that the preposition, which linearly follows WTH, bears an NA as a result of (35). This is because in swiping the structurally deepest item that is outside of the ellipsis site is the preposition and not the wh-item or WTH – see (37).

Similarly, a WTH is prosodically well-formed in any non-elliptical context, as long as the string that follows the WTH bears nuclear accent. In an example such as (38), nuclear accent falls on the lexical verb reading. See Figure 5 above for the attested prosody of this example.

In the light of new observations presented in this paper, the data are accommodated better in our account than in Sprouse’s (Reference Sprouse2006) prosodic account. First, in our account, the prominence that is observed on the preposition of swiping is not due to its information structural status, but due to its syntactic position. By employing Cinque’s (Reference Cinque1993) stress deepest algorithm, we predict prominence on the preposition regardless of whether or not it is given in the antecedent clause. Secondly, in our account, the prosodic well-formedness of WTHs is not contingent upon its ability to bear accent. As such, the condition in (33) makes no reference to the accentability of WTHs, or, more specifically, TH. The only condition on the prosodic well-formedness of WTHs is its prosodic distribution relative to the location of nuclear accent. As long as (33) is satisfied, various prosodic realisations of WTHs, including the versions in which TH bears accent, are easily accommodated in our account. Lastly, our account does not appeal to the presence of counter-intuitive prosodic chunks, such as the detached prosodic grouping of the wh-item and the string of TH.

In this section, we set out a novel prosodic account that successfully predicts the well-formedness of WTHs in swiping, and unacceptability of it in sluicing. The prosodic well-formedness condition that we propose in (33) states that nuclear accent must follow WTH, making no specific reference to ellipsis. Hence, we expect this condition to be active for all cases of WTHs, i.e. not only in sentences with ellipsis but also in sentences without ellipsis.

So far we have focused on how certain cases of ellipses, such as sluicing, accidentally create non-ideal prosodic environments for licensing WTHs. In the next section, we present more examples from elliptical clauses in support of the relevance of our account. In addition to this, we show that (33) seeks to be satisfied in non-elliptical clauses of British English, too.

4.2 Prosodic licensing of WTHs beyond swiping/sluicing

In information structurally imbalanced cases, which require the marking of focus and givenness, the default syntax-oriented NA placement algorithm in (33) above is overridden. In such cases, the NA is aligned with the item that bears information structural focus, in languages such as British English and Dutch. In such cases, the area that follows the focused item (hence NA) is deaccented (see Féry Reference Féry2011 for an overview).

Our account makes certain predictions for such cases, namely those with focus related NA placement and givenness related deaccentuation. In particular, our account predicts that in cases when all the items that follow WTHs are discourse given, and deaccented, WTHs are not licensed, as they do not precede an NA. This prediction is borne out. In (39i) below, the items that follow WTHs in the second clauses are given in the clauses that precede them. This leads to a prosodic environment in which a WTH does not precede NA. Givenness related deaccentuation is marked with italics, focus related NA is marked with capitals. Note that the unacceptability of (39i) should be attributed to the accentuation profile of the second half of these clauses, and not some other factor, such as redundant repetition of the given information, simply because, as seen in (39ii), the same sentences without TH are acceptable regardless of the repetition.

In (39), not bearing focus or any contrastive information, the strings he saw and was stolen in the second clauses are deaccented as a result of the prosodic marking of givenness in British English. In the unacceptable cases in (39), the WTH bears an NA, which fails to satisfy (33) and the sentences are rendered as unacceptable.Footnote ^[13]

A case known as why-stripping (Ortega-Santos, Yoshida & Nakao Reference Ortega-Santos, Yoshida and Nakao2014) is another elliptical environment in which the WTH is prosodically licensed due to focus related NA placement. Consider (40):

In (40B), the WTH is followed by a discourse given item, i.e. the object of the antecedent clause, Bill. However, regardless of the fact that it is given, Bill still bears the NA due to the fact that it bears a focal contrast.Footnote ^[14] As our account predicts, the utterance in (40B) is acceptable as it satisfies the prosodic well-formedness condition in (33).

Another elliptical case in which a WTH is prosodically well-formed is the case of wh-stripping. Similar to why-stripping, in wh-stripping, the item that follows the wh-word exhibits focus related NA. In (41), the constituent that follows the wh-item, i.e. the preposition after, bears focus related NA.

Since WTH precedes the NA-bearing item, it is prosodically licensed, and the sentence is acceptable.

We also predict that amalgams such as (42a) (Lakoff Reference Lakoff, Galy, Fox and Bruck1974) are not capable of hosting a WTH (see (42b)), regardless of the fact that more phonologically contentful material follows WTH.

This is because such clausal parentheticals are parsed as independent intonational phrases in English (see Dehé Reference Dehé2014). Being the last accent-bearing unit in that intonational phrase, a WTH has to host the NA of its clause. This is a prosodic environment that we predict to be ill-formed for hosting a WTH.

Lastly, our account converges with the observation that aggressively non-D-linked phrases cannot modify in-situ wh-words (Pesetsky Reference Pesetsky, Reuland and ter Meulen1987). In English, the NA aligns with the in-situ wh-word (see Truckenbrodt Reference Truckenbrodt2013). This is exactly the prosodic configuration in which our account predicts the use of TH to be unacceptable. The example in (43) demonstrates the unacceptable case of an echo question with WTH.Footnote ^[15]

In this section, we have discussed a number of acceptable and unacceptable cases of WTH, in which the NA is located on or after WTH due to information structural reasons. We have seen that the prosodic well-formedness condition in (33) successfully predicts acceptability of a number of constructions with or without ellipsis. That WTH is prosodically ill-formed in sluicing constructions in British English is due to two factors, (i) the fact that WTHs cannot bear NAs, and (ii) that, in information structurally neutral cases, the syntactically deepest item in a clause receives the NA in English. Although the second factor is a well-established observation about English prosodic grammar, the question why a WTH rejects an NA has not been discussed yet. Although we do not have a definitive answer as to why a WTH rejects nuclear accent, it seems that this property is common to other pragmatically loaded phrases with an expressive content in other languages, too.