1 Introduction
The reactive what-x construction has been noticed very recently in spoken British English. It consists of the wh-phrase what followed by another phrase (Põldvere & Paradis Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020), as illustrated by the bolded utterances in (1)–(3). The majority of the examples cited throughout this article are from the London–Lund Corpora (LLC) of spoken British English, either the first (LLC–1) or second edition (LLC–2).Footnote 2 If no source is indicated, the examples are made up.
(1)
A: after you'd read it over to her <pause/> that's uh about half past two <pause/> she had a rest <pause/> didn’t she
B: yes we both had a rest yes
C: what she went to sl\eep # did she
B: she went to sleep yes (LLC–1)
(2)
A: but this museum is double the size of Tate Modern even with the the uh the extension
B: what the extension that’s not \open
A: yes (LLC–2)
(3)
A: and we managed to add in Judith Jacobs
B: yeah thank god you added in Judith she’d’ve been so upset <pause/>
C: what with her photograph not being cr\edited
B: well no <pause/> no no no no no it's because I the picture was me stood in front of <pa use/> these two paintings because they're life-sized (LLC–2)
According to Põldvere & Paradis (Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020), this construction has a set of key properties, which we briefly summarize below and return to in section 2. First, it never appears discourse-initially. Note that in each example above, there is a linguistic antecedent (some part of) which the reactive what-x construction is a reaction to. Second, the slot that follows what may be filled by a range of syntactic categories: S, NP, PP, VP or CP. The first three are illustrated by examples (1)–(3), respectively. Third, the reactive what-x construction is a matrix-clause phenomenon with a distinct intonation pattern: what and the phrase following it belong to a single tone unit, which we indicate by the absence of a tone unit boundary between them in the examples above and elsewhere. To these general properties we add two more. The entire construction distributes like the syntactic category following what. In (1), for instance, it distributes like a matrix clause and in (3) it corresponds to the PP complement of upset, suggesting that the grammatical function of what is that of an adjunct. This observation is both relevant for the analysis we spell out in section 5 and a departure from Põldvere & Paradis (Reference Põldvere and Paradis2020), who propose a head–complement structure for the reactive what-x construction, with what serving as the head. Finally, what itself seems to have little interrogative meaning, which instead is associated with the entire construction's function as an interrogative utterance. As we will see below, our data can be classified as either reprise (i.e. clarificational) or direct (i.e. non-reprise) utterances, which brings it in line with the classification Ginzburg & Sag (Reference Ginzburg and Sag2000) propose for interrogative clauses.
What we are concerned with here is a specific kind of phrase that follows what in well over half of our data. It has thus far escaped notice that the reactive what-x construction hosts ellipsis, that is, phrases that seem syntactically incomplete. This is illustrated in (2)–(3). The phrases that follow what are utterances smaller than sentences (i.e. an NP in (2) and a PP in (3)), but they still receive sentential interpretations on the basis of the preceding context. We demonstrate that such examples of the reactive what-x construction have all the relevant characteristics of what we more generally term fragments here (we spell out the characteristics of fragments in section 4) and argue that they are analyzable as either reprise or direct fragments. To anticipate our proposal, it is an analysis of the reactive what-x construction drawing on Ginzburg & Sag’s (Reference Ginzburg and Sag2000) constructional Head-driven Phrase Structure Grammar (HPSG) account of a comprehensive network of interrogative constructions, which includes both elliptical (fragments) and nonelliptical ones, the former analyzed as genuinely non-sentential utterances.
The rest of this article proceeds as follows. In section 2, we elaborate on the properties of the reactive what-x construction that are familiar from previous research. Section 3 describes our corpus data, along with the criteria for selecting the relevant instances of the reactive what-x construction. In section 4, we present our reasons for arguing that a large portion of our data instantiates ellipsis and that these examples are classifiable as reprise or direct fragments. We offer a theoretical analysis of them in section 5, based on Ginzburg & Sag’s (Reference Ginzburg and Sag2000) non-sententialist (also known as direct-interpretation) approach to ellipsis, that is, one where fragments are generated ‘as is’ as opposed to the sententialist approach generating them from underlying sentential sources via operations like movement and deletion (Merchant Reference Merchant2001, Reference Merchant2005). Section 6 concludes.
2 Properties of the reactive what-x construction
During the compilation of the London–Lund Corpus 2 (LLC–2) of spoken British English, Põldvere & Paradis (Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020) isolated a construction that had not previously been noticed in the literature. To reflect its idiosyncratic syntactic and dialogic properties in conversation, the construction has come to be known as the reactive what-x construction, a conventionalized unit in English. What differentiates the reactive what-x construction from other interrogative constructions in English is that what connects directly with the phrase following it. As mentioned in the Introduction, the two elements always form one and the same tone unit with the nuclear pitch accent falling somewhere on the latter element (for more detail, see Põldvere & Paradis Reference Põldvere and Paradis2020: 319). In examples (1)–(3) above, the falling nuclear pitch accent is on the last lexical words of the utterances (sl\eep, \open, cr\edited). This intonation pattern sets the reactive what-x construction apart from, specifically, the pragmatic marker what (4), and split questions (5).
(4)
A: what are you going to play us <pause/>
B: I was looking for those <pause/> Indian records we’ve got
C: wh\at # Panna Lal G/osht
A: no <pause/> no not Panna Lal Gosht (LLC–1)
(5)
A: uhm general thing I would say about getting yourself out of the defensive mode is <pause/> if you can sprinkle a few rewards around yourself <pause/> then that is a very good way of engaging
B: wh\at is that # ch/ocolate # what do you mean
A: well <pause/> do you know it kind of so <pause/> the reward system <pause/> does respond to food (LLC–2)
The pragmatic marker what produced by speaker C in (4) looks very similar to the reactive what-x construction with one non-trivial exception: what is separated from the following phrase by a tone unit boundary (#). This means that what is prosodically and interactionally prominent, expressing speaker stance, particularly, surprise, incredulity and positive or negative emotion (Lutzky Reference Lutzky, Markus, Iyeiri, Heuberger and Chamson2012), and the fragment that follows it should be analyzed in its own right. Split questions such as the one produced by speaker B in (5) have a bipartite prosodic structure constituted by two separate tone units, and are used to make a guess (Arregi Reference Arregi2010; Michaelis & Feng Reference Michaelis and Feng2015).
Furthermore, the reactive what-x construction is used as a prompt reaction to what an interlocutor says in a preceding turn to negotiate and call it into question. The construction thus presupposes the presence of a preceding turn by another speaker, which becomes most relevant in spoken dialogue. For instance, in (6) this meaning potential is exploited to give rise to a specific dialogic function, namely, to verify a referent in the preceding turn, which is the reason for speaker B feeling creeped out (see Põldvere & Paradis Reference Martin and McElree2019, Reference Merchant2020 for the classification of the construction based on dialogic function).
(6)
A: okay I don’t really know what to expect
B: I’m a bit creeped out by this <pause/>
A: what the the fact that we're playing Minecraft <pause/> times are getting desperate you know few monetisations
B: it is really (LLC–2)
Examples like (6) could also be analyzed as other-initiated self-repairs (Schegloff et al. Reference Schegloff, Jefferson and Sacks1977), a category familiar from the framework of Conversation Analysis, to solve a difficulty in conversation. On the surface, what the fact that we're playing Minecraft looks like an other-initiated self-repair used to perform a non-serious action such as a tease (Kendrick Reference Kendrick2015); however, since speaker B does not actually accept or reject what speaker A has said, its status as a genuine repair is unclear. The extent to which the reactive what-x construction is used as a repair in conversation is not part of Põldvere & Paradis’ (Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020) analysis either.
Based on the form–meaning properties of the reactive what-x construction, Põldvere & Paradis (Reference Põldvere and Paradis2020) propose a constructional representation, in line with the basic tenets of Construction Grammar (e.g. Fillmore et al. Reference Fillmore, Kay and O'Connor1988; Croft & Cruse Reference Croft and Cruse2004; Fried & Östman Reference Fried and Östman2005; Goldberg Reference Goldberg1995, Reference Goldberg2006; Linell Reference Linell, Bergs and Diewald2009; Hilpert Reference Hilpert2014). This representation is given in figure 1.Footnote 3
Figure 1. Constructional representation of the reactive what-x construction (Põldvere & Paradis Reference Põldvere and Paradis2020: 327). The symbol # indicates a tone unit boundary.
In addition to what has already been mentioned above, figure 1 reveals a distinction between internal formal structure and external formal structure. While the internal structure relates to the syntactic and prosodic properties of the construction, the external structure is concerned with the sequential placement of the construction in the surrounding discourse. As mentioned in the Introduction, the representation in figure 1 assumes that what is a head followed by a complement. The complement may be of any syntactic category and there is nothing here to separate out elliptical from nonelliptical complements, the latter ranging over polar interrogative clauses, in-situ interrogative clauses, and, rarely, declarative clauses (see section 3 for descriptive statistics). In contrast, our focus throughout is specifically on fragments as distinct from nonelliptical phrases used in the reactive what-x construction. This also means that, for the purposes of this article, we ignore constructions that can be seen as related to the reactive what-x construction in various formal ways, such as the pragmatic marker what we saw illustrated in (4). An overview of these constructions and relationships among them can be found in Põldvere & Paradis (Reference Põldvere and Paradis2019).
3 Data
Following Põldvere & Paradis (Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020), we extracted our data from the London–Lund Corpora of spoken British English, consisting of LLC–1 (Svartvik & Quirk Reference Svartvik and Quirk1980; Greenbaum & Svartvik Reference Greenbaum, Svartvik and Svartvik1990) and LLC–2 (Põldvere et al. Reference Põldvere, Johansson and Paradis2021). An important difference from the earlier work is that the data in this study are from complete corpora rather than subsets of spontaneous face-to-face conversation. This gives us access to the reactive what-x construction from a diverse range of discourse contexts. For instance, example (6) above is part of the new data. Both LLC–1 and LLC–2 contain around 500,000 words (one million words in total) of spoken dialogue and monologue recorded with educated adult speakers of British English. There are seven broad discourse contexts: face-to-face conversation, distanced conversation, broadcast media, parliamentary proceedings, spontaneous commentary, legal proceedings and prepared speech. The recordings are from a time period covering approximately sixty years, from the 1950s to the 2010s, thus giving us a good overview of the behavior of the reactive what-x construction, and its relationship to other constructions, in present-day English. The corpora were accessed via the online corpus interface Corpuscle (LLC–1; Meurer Reference Meurer and Andersen2012) and offline XML files (LLC–2), alongside corresponding audio files. We searched the corpora automatically to find all instances of what and found a total of 5,388 of them (2,723 in LLC–1 and 2,665 in LLC–2). Access to the audio files of the LLCs, and therefore to the prosodic information of what-constructions, proved useful in identifying the reactive what-x construction and excluding superficially similar constructions like the pragmatic marker what and split questions. Thus all the what-constructions that exhibited the syntactic structure of the reactive what-x construction were subject to close auditory and instrumental analysis, the latter of which was conducted in the phonetics software Praat (Boersma Reference Boersma2001). Only those constructions where what was realized as an unaccented pre-head of the tone unit, with a nuclear pitch accent somewhere in the following element, were included. In the end, we identified 61 instances of the reactive what-x construction.
Our sample contains 15 instances of nonelliptical phrases following what. Among them, there are 7 instances of in-situ interrogative clauses (see example (1)), 5 instances of polar interrogative clauses (7) and 3 instances of declarative clauses (8). As these data are not the focus of our article, we do not discuss them further with the exception of footnote 9.
(7)
A: well I think they’ve got lots of toys and stuff so
B: is it <pause/> what did you have a look in John Lewis or
A: mm (LLC–2)
(8)
A: well as long as it’s not mathcore again
B: I like mathcore
A: it was terrible <vocal desc="laughs"/>
B: it wasn’t terrible what you liked all of them
A: the first three no I didn't the first three were terrible (LLC–2)
The remaining items are 45 instances of fragments (phrases smaller than a sentence) and one instance of Verb Phrase ellipsis (VPE). The latter is shown in (9).
(9)
A: but what about Nuffield Leisure they're a very posh gym they've taken over they've taken over the one in Moorgate that's like […]
B: they won't do it they won't do it you you
A: what they won't <pause/> but they have but they have in their foyer they have (LLC–2)
Unlike fragments, VPE produces a sentential structure, but the VP is seemingly incomplete, as it appears to consist only of a stranded auxiliary. The fragments’ syntactic categories range over NPs, PPs, VPs and CPs; their counts are given in table 1.
Table 1. Counts for all syntactic categories of fragments in our data sample
We have illustrated NPs in examples (2) and (6) and PPs in (3). A CP and a VP are illustrated in (10) and (11), respectively.
(10)
A: I started falling asleep a little bit in the middle yeah the cornflake one was good
B: did you really <pause/> what cause you were knackered <pause/> or cause it was boring
A: cause I was knackered and it was boring (LLC–2)
(11)
A: I must tell Michael, I haven't told Michael, he proposes to go to Manchester on Thursday Catrina
B: what and leave you to move
A: no I'm moving on Friday (LLC–1)
We take a closer look at fragments in the next section.
4 The reactive what-x construction as fragments
In this section we turn to a property of the reactive what-x construction that has never been addressed before, namely the fact that a large portion of its uses feature ellipsis. Recall that we have extracted the total of 46 (75.4%) instances of ellipsis out of 61 instances of the reactive what-x construction (but we exclude our one instance of VPE from the counts for fragments in this section and return to it in section 5.3). Why should we define these instances as elliptical? First, their syntax is apparently incomplete in the sense that the phrases following what look non-sentential, resembling a type of ellipsis known as fragments. Second, these examples have sentential semantics despite the unusual syntax: they are fully equivalent to what their sentential counterparts would mean, but their resolution requires access to the preceding context, which we henceforth term antecedent, following the literature on ellipsis. Such apparently incomplete syntax and context-dependence are two hallmarks of ellipsis. We concede that this definition of ellipsis is firmly entrenched in work within theoretical syntax, and not at all so in work focused on spoken interaction (e.g. Goodwin Reference Goodwin1995; Jacoby & Ochs Reference Jacoby and Ochs1995; Lerner Reference Lerner2004). This is purposeful, as we offer a novel theoretical analysis of the reactive what-x construction as an instance of ellipsis according to the definition above, though we do not assume the syntax of fragments is incomplete in any way, given our non-sentential approach.
We next address four further features that align the reactive what-x construction more closely with fragments: the presence of a correlate in the prior context, uses as either merger or sprouting, an ability to drop prepositions, and uses as reprise or direct utterances (these four features are also summarized in table 2).
Table 2. Summary of the shared properties of the reactive what-x construction and fragments
The first feature is the presence of a correlate for the phrase preceded by what. A correlate is a phrase found in the antecedent that is associated with a fragment. For instance, the fragment in (12) is associated with the PP About it, serving as its correlate.
(12)
A: He's sorry about it.
B: About what?
Every fragment has a correlate to be associated with this way, which constitutes the basis for retrieving its sentential semantics: after a fragment has been paired with its correlate, it can be integrated into the proposition expressed by its antecedent (Culicover & Jackendoff Reference Culicover and Jackendoff2005; Harris & Carlson Reference Harris and Carlson2019; Nykiel et al. Reference Nykiel, Kim and Sim2023).Footnote 4 That the reactive what-x construction parallels fragments in this respect is clear from (13), where the correlate is the NP headed by the pronoun it in the antecedent.
(13)
A: you fancy it yourself do you <pause/>
B: what the men's doubles
A: yeah (LLC–1)
But correlates do not have to be overt phrases, which brings us to the second feature.
Phrases preceded by what may pick out overt or covert correlates, just like fragments. A covert correlate is illustrated in (14).
(14)
A: I've had so many emails <pause/> asking me the same question <pause/>
B: what today
A: today (LLC–2)
The NP following what is associated with an unexpressed adjunct in the antecedent (see also example (3) for a PP following what and taking as its correlate a PP complement to the adjective upset located in the antecedent). The configuration where fragments pick out covert correlates is referred to as sprouting, while the one where they pick out overt correlates is known as merger. We can manipulate example (12) to illustrate sprouting with fragments, as in (15): now the fragment picks out a covert PP complement to sorry.
(15)
A: He's sorry.
B: About what?
The distribution of sprouting versus merger in our data is 17 (37.7%) and 28 (62.3%), respectively.
Related to the sprouting-merger distinction is the third feature, the possibility of omitting prepositions from fragments. Consider an NP variant of the fragment in (12), depicted in (16).
(16)
A: He's sorry about it.
B: What?
The NP what in (16) has the interpretation that the PP about what does in (12), i.e. What is he sorry about?, even though the preposition is omitted, which is a grammatical possibility.Footnote 5 The possibility of omitting prepositions from fragments is limited to merger, a constraint that was first noted by Chung (Reference Chung, Cover and Kim2006). In other words, prepositions may be omitted from fragments only if they have been realized overtly in the antecedents. Thus the fragment in the sprouting example in (15) is required to be a PP, with no possibility of omitting the preposition. Like fragments, the reactive what-x construction permits omission of prepositions in merger, but not in sprouting environments. Example (17) illustrates this (note that the preposition omitted by speaker B and present in the antecedent is to).
(17)
A: I am going to your neck of the woods did I tell you that <pause/>
B: what Gloucestershire
A: yeah (LLC–2)
The reactive what-x construction closely tracks fragments not only with respect to when prepositions are omitted, but also how often. There are altogether eight environments where prepositions could have been omitted and actually were 87.5 percent of the time (i.e. seven out of eight times). For comparison, prepositions are omitted just under 70 percent of the time in spoken US English fragments (Nykiel & Hawkins Reference Nykiel and Hawkins2020).
Finally, the set of uses of the reactive what-x construction ranges over reprise and direct utterances. All uses of fragments fit into this binary classification described in detail in Ginzburg & Sag (Reference Ginzburg and Sag2000). We begin by briefly summarizing their criteria for classifying both fragments and other interrogatives as reprise as opposed to direct.
Reprise utterances are interrogatives that serve to request clarification of a part of a previous utterance. Typical reasons for their use are failure to hear the part that is being clarified (this type is called echo questions) and failure to understand the reference of the part that is being clarified (this type is called reference questions). Both types are illustrated in (18)–(19) for nonelliptical utterances (adapted from Ginzburg & Sag Reference Ginzburg and Sag2000: 260). Speaker B's utterance in (18) is an echo question on the assumption that B failed to hear the name Jan correctly. B's utterance in (19) is a reference question in the sense that the reference of the pronoun them is unclear to B although they heard the pronoun itself correctly.
(18)
A: Chris is annoyed with Jan.
B: Chris is annoyed with who(m)?
(19)
A: Chris is annoyed with them.
B: Chris is annoyed with who(m)?
The syntax of nonelliptical reprise utterances typically involves in-situ wh-phrases, as is the case in (18)–(19), but this is not mandatory.Footnote 6 Although it is beyond the scope of this article to go into the formal semantics of reprise utterances, we would like to mention that it departs in only one way from the semantics of other interrogatives: their semantics partially overlaps with that of the immediately preceding utterance (for more detail, the reader is referred to Ginzburg & Sag Reference Ginzburg and Sag2000; Ginzburg & Cooper Reference Ginzburg and Cooper2004; Ginzburg Reference Ginzburg2012). This brings us to the second category of interrogatives, direct utterances. We agree with Ginzburg & Sag (Reference Ginzburg and Sag2000) that these are all other interrogatives that do not express clarification requests.
When used as reprise utterances, fragments have two main readings. These are clausal confirmation and intended content readings, both falling into the category of reference questions (Ginzburg & Cooper Reference Ginzburg and Cooper2004; Ginzburg Reference Ginzburg2012; Ginzburg & Kolliakou Reference Ginzburg and Kolliakou2018). We exemplify them in (20)–(21), from Ginzburg & Kolliakou (Reference Ginzburg and Kolliakou2018: 348), along with the immediately following context to help disambiguate the readings. The fragment shown in (20) has a clausal confirmation reading, paraphrasable as Are you asking if I ever play with Mark?. This class of readings represents a request for confirmation of how the constituent repeated in the fragment fits into the proposition expressed by the preceding utterance. The fragment in (21), on the other hand, has an intended content reading, that is, it queries the reference of the NP Adam in A's preceding turn. This reading is paraphrasable as Who is Adam? or Who do you mean Adam?.
(20)
A: Do you ever play with that wee boy Mark that lives next door to me?
B: Mark?
A: Yes, does he come up?
B: He doesn’t play with me. He doesn’t know me.
(21)
A: Because it belongs to Adam.
B: Adam?
A: Adam’s my little boy.
We should note at this point that some fragments solely consist of, or host, wh-phrases, and have sentential semantics equivalent to wh-interrogatives. We saw some examples of wh-phrases used either alone as fragments or as complements within PPs that were fragments in (12) and (15)–(16) above; this class of fragments is referred to as sluicing, a term going back to Ross (Reference Ross, Binnick, Davison, Green and Morgan1969). Sluicing often functions as reprise utterances, and when doing so, it conveys intended content readings in the sense that the hearer is requesting clarification of some aspect of the content of an utterance present in the prior context. In (22), from Ginzburg & Kolliakou (Reference Ginzburg and Kolliakou2018: 345), the referent of the pronoun he is being clarified.
(22)
A: What a useless fairy he was.
B: Who?
Even though we have not observed any instances of reprise (or non-reprise) sluicing among our instances of the reactive what-x construction (these would be realized as what followed by another wh-phrase), we cannot rule out that they are possible.
We classified our data as either reprise or direct utterances, the latter understood as exhibiting no clarificational properties. We identified 28 (62.3%) reprise uses and 17 (37.7%) direct uses among the 45 fragments (the one example of VPE was also classified as reprise (clausal confirmation)). This result strongly suggests that the reactive what-x construction is biased towards reprise utterances.
It is interesting to note that the distribution of reprise versus direct uses and of merger versus sprouting is the same. In fact, all the sprouting cases of the reactive what-x construction were classified as direct utterances and the merger cases as reprise utterances. This follows from the fact that in sprouting environments the hearer has clearly not failed to understand the prior utterance. Consider the reactive what-x construction produced by A in (23). It is clear that speaker A has understood B's response and is requesting additional information, and hence their utterance is direct. This is also the case for the remaining instances of sprouting in our data.
(23)
A: do you do you drive <pause/>
B: no I'm too short-sighted to drive <pause/> […]
A: what with glasses you can wear glasses
B: oh yes I wear them but I'm ever so short-sighted even with glasses on (LLC–1)
In contrast, merger environments open themselves up to clausal confirmation readings in our data. Various constituents in antecedent clauses are queried by being paraphrased or expanded upon in reprise utterances, but never by being repeated verbatim. To see this, consider (24)–(25). The constituent in speaker B's utterance in (24) that is being paraphrased by A is the pronoun that. Similarly, the NP the extension in speaker A's utterance in (25) is being expanded upon by B.
(24)
A: they're all worried <unclear/> he's going to drop dead <pause/> having had two heart attacks <pause/> rightly […]
B: how can we tactfully acknowledge that in the session with him there
A: what that they're worried
B: that they're worried he's gonna drop dead (LLC–1)
(25)
A: but this museum is double the size of Tate Modern even with the the uh the extension
B: what the extension that’s not open
A: yes (LLC–2)
We classified the readings of all 28 reprise examples as clausal confirmation, given that all of them are paraphrasable as polar questions of the type Are you referring to X?. However, we acknowledge that reprise fragments in general are regularly ambiguous between clausal confirmation and intended content readings (see Ginzburg & Cooper Reference Ginzburg and Cooper2004). Ginzburg & Cooper (Reference Ginzburg and Cooper2004) also point out that one diagnostic for teasing these readings apart are weaker identity requirements (i.e. morphosyntactic category, but not necessarily segmental, identity between the fragment and the constituent being queried) that clausal confirmation readings have compared to intended content readings. We are seeing this pattern in our data as well: none of the 28 instances exhibit segmental identity. Whether an intended content or a clausal confirmation reading is involved in every case has no bearing on our syntactic analysis of the reactive what-x construction to be presented shortly, but it does impact the construction's semantics. Table 2 summarizes the four properties of the reactive what-x construction we have discussed, comparing them with fragments.
Having now established that there are good reasons for assimilating the reactive what-x construction to fragments, we turn to our theoretical analysis of it in the next section. Because we are concerned with the elliptical side of this construction, we do not extend our analysis to uses not involving any ellipsis here (but see footnote 9).
5 Analysis
The ingredients of our analysis come from Ginzburg & Sag (Reference Ginzburg and Sag2000), who articulate a constructional Head-driven Phrase Structure Grammar (HPSG) analysis of a wide range of interrogative constructions, including of reprise and direct fragments. The analysis of both reprise and direct fragments that Ginzburg & Sag (Reference Ginzburg and Sag2000) propose is non-sententialist in the sense that the syntax of fragments does not involve unpronounced structure at any level. It is also the kind of theoretical analysis reprise fragments typically attract (see Ginzburg & Cooper Reference Ginzburg and Cooper2004 and Ginzburg Reference Ginzburg2012 for other non-sententialist analyses, and see Griffiths et al. Reference Griffiths, Güliz and Lipták2023 for a recent attempt at a sententialist analysis of reprise fragments), a tradition we continue in the spirit of here. But it is fair to say that, besides the bias toward, and a well-developed formalism of, non-sententialist approaches to reprise versus direct fragments, there is nothing about our current data that would motivate non-sententialist approaches to ellipsis over sententialist ones more generally.
We begin by presenting Ginzburg & Sag's (Reference Ginzburg and Sag2000) non-sententialist approach to fragments preceded by a brief introduction to the main tenets of constructional HPSG (for more detail, see Sag Reference Sag1997; Ginzburg & Sag Reference Ginzburg and Sag2000; Sag Reference Sag, Boas and Sag2012).
5.1 Fragments in constructional HPSG
In constructional HPSG linguistic expressions are defined as signs. These range from words to phrases (to be precise, phrases are configurations of signs termed constructs) and are modeled as feature structures carrying phonological, morphological, syntactic, semantic and contextual information. Given that signs and families of signs often have shared features, generalizations over such shared features are captured via type hierarchies. Each family of signs or constructs in a hierarchy must satisfy the constraints imposed on it as well as any constraints imposed on the type(s) it is a subtype of. It is important to note at this point that any type may belong to, or inherit from, more than one supertype – this is known as multiple inheritance. This property of type hierarchies will be important for our analysis of the reactive what-x construction. Finally, we should point out a terminological difference between constructional HPSG and other versions of Construction Grammar, such as Goldberg's Cognitive Construction Grammar (1996, 2006): constructions refer here to constraints that license (configurations of) signs, and not to linguistic expressions understood as pairings of form and meaning, as in Põldvere & Paradis (Reference Põldvere and Paradis2019, Reference Põldvere and Paradis2020). We will, however, be using the term constructions to refer to families of constructs and constraints that license them for brevity.
Ginzburg & Sag's (Reference Ginzburg and Sag2000) approach to fragments has come to be known informally as a WYSIWYG (what you see is what get) approach. In other words, there is no sentential structure that fragments are embedded in, although they are still analyzed as sole daughters of an S node. The construction that licenses fragments (the head-fragment construction) is given in (26). The version in (26) is a slightly modified version of the construction proposed in Ginzburg & Sag (Reference Ginzburg and Sag2000: 304), which goes back to Nykiel & Kim (Reference Nykiel and Kim2022) (for another version, see Abeillé & Kim Reference Abeillé and Kim2022). Let us now look at the mechanism underlying (26). The mother is an S, allowing fragments to distribute like clauses. There is one daughter, the head daughter, which may be a phrase of any (but typically nonverbal) category, and whose semantics is supplied by the surrounding context in the following way.
(26) Head-fragment construction
First, the surrounding context represented as the feature CNXT in (26) supplies the attribute called the focus-establishing constituent (FEC).Footnote 7 The FEC serves as a fragment's correlate and constitutes the only locus of information about elements of the antecedent structure, such as syntactic category, case specifications and argument structure. Ginzburg & Sag's (Reference Ginzburg and Sag2000) head-fragment construction requires syntactic category identity between a fragment and the FEC, but later research has demonstrated that this is too strict a requirement and no longer assumes such identity (for detailed discussion, see Nykiel & Kim Reference Nykiel and Kim2022). The construction in (26) follows Nykiel & Kim (Reference Nykiel and Kim2022) in this respect, requiring only that a fragment and the FEC be saturated phrases and share their semantic index. Being co-indexed with the FEC gives the fragment access to the FEC's morphosyntactic specifications.
The FEC is a sub-utterance of the second contextual attribute, the MAX-QUD. This is the question that is currently under discussion and whose semantics constitutes the basis for interpreting a fragment. On the assumption that the context keeps track of every question that is under discussion and updates as new ones come in, the MAX-QUD is understood as the most recent question under discussion. For fragments, it typically arises from a preceding wh-interrogative, as in (27).
(27)
A: What did Harvey learn?
B: That you can score from halfway.
Speaker's A question leaves one element unresolved, the thing that Harvey learned, which serves as the FEC. The fragment in B is matched with the FEC and the proposition it expresses (i.e. Harvey learned that you can score from halfway) is built up from the MAX-QUD. The mechanism underlying the process of matching a fragment and an FEC is known as a direct-access mechanism, that is, a non-serial search for the target phrase among the representations previously stored in memory (Martin & McElree Reference Martin and McElree2011). The morphosyntactic specifications of a fragment are used as retrieval cues in this search (Harris Reference Harris2015; Nykiel et al. Reference Nykiel, Kim and Sim2023), making the morphosyntax of the FEC equally relevant for the search, exactly as predicted by (26). However, notice that the syntactic categories of the fragment and the FEC in (27) do not match, the former being clausal and the latter nominal, a pattern that is allowed by the constraint in (26).Footnote 8 Let us now illustrate the non-sentential syntax and semantics of the utterance That you can score from halfway in (28) below. If we were to reconstruct this fragment's sentential syntax, it would be an embedded declarative clause (Harvey learned that you can score from halfway). Here instead, it is a stand-alone CP, i.e. the only daughter of the S node.
(28)
Besides the information about the FEC, we have spelled out the propositional semantics of the fragment, which is built up from the MAX-QUD supplied by the antecedent in (27), and represented here as lambda abstraction (λi [learn(h,i)] can be read as the property of being a thing that Harvey learned, but the technical details need not concern us here; the reader is referred to Ginzburg & Sag Reference Ginzburg and Sag2000 and Sag Reference Sag2010 for more detail on the semantics of different types of clauses).
Recall from section 4 that fragments take overt or covert correlates, the latter environment called sprouting. The fragment whose structure we illustrated in (28) has an overt correlate, that is, the FEC is an overt phrase. But consider example (15) again, repeated here as (29). Recall that the fragment About what? has a covert correlate.
(29)
A: He's sorry.
B: About what?
A sprouting fragment like this has the structure in (30).
(30)
The difference between (28) and (30) is that the FEC is annotated as an ini (for noninstantiated indefinite null) argument in the latter. This solution goes back to Ruppenhofer & Michaelis’ (Reference Ruppenhofer, Michaelis, Bourns and Myers2014) proposal that the list of arguments of a lexical head can include noncanonical arguments that are not realized in the overt syntax. This makes them available for serving as FECs. Thus the mechanism for deriving fragments we have seen so far is able to handle both sprouting and merger cases.
Before closing this section, we turn to Ginzburg & Sag's (Reference Ginzburg and Sag2000) type hierarchy of clauses to see which clauses fragments inherit their properties from. The type hierarchy is given in (31) (it is a slightly revised version of the type hierarchy in Ginzburg & Sag Reference Ginzburg and Sag2000: 333).
(31)
A declarative fragment like That you can score from halfway in (27) is analyzed as a decl-frag-cxt (declarative fragment construction). This construction inherits from two schematic types, headedness and clausality, and further down from a decl-cxt (declarative construction) on the one hand, and a hd-cxt (headed construction) on the other. The construction that licenses fragments directly is the hd-frag-cxt (head-fragment construction) we already saw in (26). Notice that the hd-frag-cxt has another subtype, a slu-int-cxt (sluiced interrogative construction), which is also a subtype of an inter-cxt (interrogative construction). The slu-int-cxt licenses direct sluicing, some examples of which we saw in (12) and (15)–(16) in section 4.
When fragments function as reprise utterances they are still analyzed as a hd-frag-cxt, but they are also analyzed as a dir-is-int-cxt (direct in-situ interrogative construction) one level up. In other words, the hd-frag-cxt is analyzed as the head daughter of the dir-is-int-cxt (this structure is shown in (36) below). The dir-is-int-cxt is a subtype of an is-int-cxt (in-situ interrogative construction) and is otherwise used to analyze nonelliptical in-situ interrogative clauses like the bolded utterance in (32).
(32)
A: I'm too short-sighted to drive.
B: You're too short-sighted to drive even with glasses?
These are separate from reprise nonelliptical in-situ interrogative clauses like the bolded utterance in (33), which inherit from a different subtype of is-int-cxt, which Ginzburg & Sag (Reference Ginzburg and Sag2000: 289) call a rep-int-cxn (reprise interrogative construction).
(33)
A: We're leaving for Venus.
B: We're leaving where?
Applying the dir-is-int-cxt to both direct in-situ interrogatives and reprise fragments means that Ginzburg & Sag (Reference Ginzburg and Sag2000) do not propose a designated construction type for reprise fragments, although they do so for reprise nonelliptical in-situ interrogative clauses. The reason is that there is nothing about the syntax of reprise fragments that offers cues to their reprise uses; these uses are instead determined by context.
A desirable property of the dir-is-int-cxt with respect to reprise fragments is that it blocks them from being embedded by virtue of having to satisfy the constraints imposed on its supertype, the is-int-cxt, which is specified as [I(ndependent)C(lause) +]. Reprise fragments are not embeddable, and neither are any in-situ interrogative clauses. Furthermore, the dir-is-int-cxt successfully captures a polar-question-like property of reprise fragments associated with their clausal confirmation readings. This analysis extends to direct fragments that also express polar questions and fail to embed, depicted in (34) (for this point, see Ginzburg & Sag Reference Ginzburg and Sag2000: 312).
(34)
A: Somebody needs a ride home.
B: Harvey?
Because such fragments do not express propositions, they cannot be licensed the same way declarative fragments are, that is, by the decl-frag-cxt. We assimilate direct uses of the reactive what-x construction to polar-question-expressing direct fragments in the next section.
With this mechanism in place, we are ready to analyze the uses of the reactive what-x construction with ellipsis.
5.2 Reactive what-x construction with ellipsis
Our analysis relies on two key assumptions. First, the wh-phrase what has no interrogative content regardless of the presence of ellipsis. Its contribution to the meaning of the construction could be best described as procedural in the sense of Blakemore (Reference Blakemore1987), that is, as a guide to the contextual information surrounding the propositional content of the construction. Among linguistic expressions that have been suggested to carry procedural meaning are pronouns, interjections and expletives (Wilson & Sperber Reference Wilson and Sperber1993; Wharton Reference Wharton2003; Blakemore Reference Blakemore2011), as well as parenthetical what (Dehé & Kavalova Reference Dehé and Kavalova2006). It seems reasonably clear both from the current data and Põldvere & Paradis’ (Reference Põldvere and Paradis2020) description of the semantics of the reactive what-x construction in figure 1 that what is a cue to an upcoming context-dependent utterance, typically a request for clarification of some content in the preceding context. Second, with one exception, the presence of what is neither mandatory nor impacts the syntactic or semantic properties of fragments (cf. Põldvere & Paradis Reference Põldvere and Paradis2020: 318), an assumption warranted by the data in table 2, which demonstrate that fragments and the reactive what-x construction in fact have identical properties. The exception is that the reactive what-x construction may never be embedded, which closely tracks the properties of reprise fragments and direct fragments expressing polar questions rather than propositions.Footnote 9 As we will see below, this property of the construction can be easily captured the same way reprise and direct fragments are analyzed. To put it differently, there is nothing unpredictable about the structure of the reactive what-x construction to give us a reason to posit a new construction type to license it in addition to the ones we saw in the type hierarchy in (31).
This analysis is less than fully satisfactory, of course, in that it neither predicts nor explains the distinct intonation pattern of the reactive what-x construction with ellipsis. The road we are not taking here would be to propose a new construction type, quite plausibly as a subtype of a hd-adj-cxt (head-adjunct construction), which permits an adjunct to combine with another phrase to form a new phrase whose syntactic category is unaffected by that of the adjunct. In fact, Abeillé & Kim (Reference Abeillé and Kim2022) argue that a subtype of a hd-adj-cxt with a mandatory adjunct and ellipsis, which they call a hd-adj-frag-cxt (head-adjunct fragment construction), is independently motivated for those English fragments that are mandatorily followed by the adverb too (i.e. the so-called me too fragments). As we have explained above, what is optional with fragments, casting doubt on this line of analysis.Footnote 10
Let us now see what the structure of the reactive what-x construction looks like. What is an adjunct combining with a dir-is-int-cxt, whose head daughter is a decl-frag-cxt. Analyzing a decl-frag-cxt as a dir-is-int-cxt is exactly the analysis applied to reprise and direct fragments, which enables us to capture the fact that each of the uses of the reactive what-x construction expresses a polar question. We extend this analysis to both reprise and direct uses of the reactive what-x construction, leaving it to the context to distinguish between reprise and direct interpretations, as is the case with other fragments. In (36) the structure of the construction is represented schematically for speaker A's utterance in (35). Note that the adjunct what selects the head it modifies via the feature MOD (Modified), as indicated by the indices. This is licensed by a hd-adj-cxt (head-adjunct construction), which is a subtype of a hd-cxt (see (31)).
(35)
A: do you do you drive <pause/>
B: no I'm too short-sighted to drive <pause/ […]
A: what with glasses you can wear glasses
B: oh yes I wear them but I'm ever so short-sighted even with glasses on (LLC–1)
(36)
Because the mother, the dir-is-int-cxt, inherits the specification [IC +] from the is-int-cxt nothing else needs to be said about the unembeddability of the reactive what-x construction.
Recall from section 5.1 that our theoretical apparatus for licensing fragments can handle both merger and sprouting ones. This enables us to combine what with either kind of fragment, correctly predicting the data summarized in table 2. Furthermore, we can license what we referred to as preposition omission in section 4: we can do so by generating fragments the usual way via the hd-frag-cxt and blocking the possibility of omitting prepositions for sprouting fragments with the help of an additional constraint that Nykiel & Kim (Reference Nykiel and Kim2022) introduce for fragments. What can then combine with fragments realized as either PPs or NPs (the latter resulting from preposition omission) for merger cases, but not for sprouting cases. Finally, we would like to note that we predict that what can combine with fragments representing sluicing, though our current data remain inconclusive with regard to the correctness of this prediction. Reprise sluicing fragments are analyzed in a similar way to the reprise fragments we have seen, and direct sluicing fragments are licensed by a separate slu-int-cxt (see Ginzburg & Sag Reference Ginzburg and Sag2000). If the reactive what-x construction permitted sluicing fragments, it would also be analyzable as a slu-int-cxt, and in the reprise case, as a dir-is-int-cxn. Before concluding this discussion, we briefly outline an analysis of the reactive what-x construction with VPE.
5.3 Reactive what-x construction with VPE
Everything we have said so far in this section relates to fragments – the focus of this special issue – but recall that one example of the reactive what-x construction features a different type of ellipsis, VPE. Consider this example again in (37).
(37)
A: but what about Nuffield Leisure they're a very posh gym they've taken over they've taken over the one in Moorgate that's like […]
B: they won't do it they won't do it you you
A: what they won't <pause/> but they have but they have in their foyer they have (LLC–2)
Given the in-situ syntax and reprise use of this example (recall from section 4 that we classified it as expressing a clausal-confirmation reading), its structure is analyzable as consisting of the adjunct what combined with a rep-int-cxt, which licenses reprise nonelliptical interrogatives as a subtype of is-int-cxt (see section 5.1). The rest is rather straightforward. We simply need to license the ellipsis within the VP headed by the negated auxiliary, which will have the desired effect of requiring an antecedent in the surrounding context. Like fragments, VPE is generated without deletion on the non-sententialist approach to ellipsis. The apparently missing complement to the auxiliary is defined as a pro expression, absent from the overt syntax but listed on a separate level of representation as one of the arguments required by the auxiliary and dependent on an antecedent for its interpretation (see accessible overviews in Ginzburg & Miller Reference Ginzburg, Miller, van Craenenbroeck and Temmerman2018; Kim & Michaelis Reference Kim and Michaelis2020; Nykiel & Kim Reference Nykiel, Kim, Stefan, Abeillé, Borsley and Koenig2021; Kim & Runner Reference Kim and Runner2022).
6 Conclusion
Our focus in this article has been on the reactive what-x construction intersecting with ellipsis. Based on data from the London–Lund Corpora of spoken British English, we have demonstrated that a large portion of the uses of this construction involve ellipsis, aligning it specifically with fragments. This, we have argued, opens up the possibility of analyzing the reactive what-x construction the same way reprise and direct fragments are analyzed in the framework of constructional HPSG in Ginzburg & Sag (Reference Ginzburg and Sag2000). We have done so by analyzing the reactive what-x construction as consisting of an adjunct combining with a dir-is-int-cxt whose head daughter is a decl-frag-cxt. Our analysis does not require the addition of a new construction to license the reactive what-x construction when ellipsis is also involved, but such a move is motivated for cases without ellipsis.
Open access
