‘What and then a little robot brings it to you?’ The reactive what-x construction in spoken dialogue

In the process of compiling a new corpus of contemporary spoken British English, the London–Lund Corpus 2, we hit upon a construction used in the conversations recorded that had not previously been dealt with in the literature, namely the reactive what-x construction. Prompted by this discovery, we carried out a detailed analysis of its properties and constraints within the broad framework of Cognitive Linguistics, namely Construction Grammar, and found that the reactive what-x construction features the interrogative what directly followed by a phrasal or clausal complement x. Moreover, what forms one tone unit with the complement and never carries a nuclear pitch accent. The core meaning is to signal an immediate reaction to something said by another speaker in the preceding turn, and the dialogic functions include questions proper as well as expressions of disagreement. The two contributions of this study are: (i) to provide a definition of the reactive what-x construction and (ii) to propose a crucial theoretical extension of Construction Grammar involving a broadening of the concept of construction to cover not only the lexical–semantic pairing but also prosodic properties and the role of the construction in the interactive dialogic space in speech.


Introduction
Everyday conversation is the most basic use of language in all societies and cultures in the world (Chafe 1994;Clark 1996;Tomasello 1999Tomasello , 2003Du Bois & Giora 2014). It is dynamically co-constructed and demands a high degree of joint attention and coordination by the interlocutors. Being engaged in conversation entails a common conceptual ground and shared intentionality among the participants. It is therefore not surprising that spoken dialogue features expressions that support these socio-cognitive and dialogic characteristics. This study is concerned with one such construction, namely the REACTIVE WHAT-X CONSTRUCTION, and the data are from the London-Lund Corpus 2 (LLC-2), a new corpus of spoken British English with recordings taken between 2014 and 2019. An example of the construction from the corpus is given in bold in (1). 2 (1) A: we've got like fifty students they have to put up their work so we might have to put some screens in <pause/> uhm B: what for film <pause/> or screens to [separate the spaces] A: [screens to] separate the spaces The construction is produced by speaker B and contains the interrogative pronoun what and the phrasal complement for film or screens to separate the spaces. In contrast to standard uses of the interrogative pronoun what where the pronoun is an argument of the verb and typically followed by a finite verb form, what in (1) connects directly with the subsequent prepositional phrase. The construction expresses a prompt reaction to what A says in the preceding turn in order to counteract the hovering knowledge gap between the interlocutors.
To the best of our knowledge, the reactive what-x construction has not received any attention in the literature so far. The only exception is a mention in passing in Stenström (1984), who gives an example of the reactive what-x constructionalthough she does not refer to it as suchand treats it as a request for clarification (see section 2.1 below). She does not, however, define the construction on the basis of its distinct form-meaning and interactive functional characteristics in spoken dialogue. This is exactly what this study sets out to do. Couched in the broad framework of Cognitive Linguistics (Langacker 1999(Langacker , 2009Croft & Cruse 2004) with a focus on Constructions (Goldberg 1995(Goldberg , 2006, and with developments into prosody and dialogue (Du Bois 2007;Põldvere et al. 2016), we address three research questions.
1. What are the form-meaning properties of the reactive what-x construction? 2. What dialogic functions does the construction have in spoken dialogue? 3. How common is the construction in a sample from LLC-2 relative to other what-constructions?
The corpus analysis reveals that the reactive what-x construction is regularly used and that the combination of the interrogative pronoun what with a phrasal or clausal complement x has become conventionalised as a unit in English conversation. Following Linell (2009a), we argue that it is not only the internal grammatical structure, the what-x element, which links the form of the construction to its meaning, but it is the combination of the what-x element and the sequential dialogic context in which it occurs that makes it a reactive what-x construction. Furthermore, a closer inspection of the use of the construction in interaction reveals that immediate reaction is the general meaning or use potential of the construction (Paradis 2008(Paradis , 2011 and that speakers exploit it to achieve a variety of socio-communicative goals, ranging from 2 See section 3.1 for an explanation of the transcription and markup conventions used in LLC-2.

308
NELE PÕLDVERE AND CARITA PARADIS asking questions to expressing disagreement. Besides identifying and defining the reactive what-x construction, this study also makes an important theoretical contribution to Construction Grammar by taking it to the level of interaction in spoken dialogue. It furthers our understanding of the nature of constructions and shows that dialogic information is evoked by the construction through its embedding in the interactive dialogic space in speech. The article is organised as follows. Section 2 presents the background to the study, including previous work on what and spoken Construction Grammar. Section 3 introduces the sample from LLC-2 and describes the procedure of identifying the reactive what-x construction among all other what-constructions. Section 4 is concerned with determining the frequency and distribution of the reactive what-x construction in the sample and its form-meaning characteristics. Section 5 identifies the dialogic functions of the construction, and finally, section 6 summarises the study.

Background
In this section, we will first outline the different uses of what in contemporary English, which are primarily based on data from spoken sources (section 2.1). Then, we take a look at the limited number of studies that have been conducted on spoken dialogue from a constructionist perspective (section 2.2).

Different uses of what in contemporary English
What has primarily two uses in English: the pronominal use and the determiner use. Despite the grammatical differences, both the pronoun what and the determiner what are used in direct questions as placeholders for what is unknown, as illustrated in (2) and (3) respectively. 3 (2) What's your address? (Quirk et al. 1985: 371) (3) What nationality is she? (Quirk et al. 1985: 371) Closely related to the interrogative determiner is the exclamative determiner, which like what-questions appears in initial position with respect to the rest of the utterance. The function of the exclamative determiner what is, however, not to seek information but to act as an emphatic and expressive element, as shown in (4). What also functions as a subordinate element. In (5), the pronoun what introduces a nominal relative clause.
(4) What a fine watch he received for his birthday! (Quirk et al. 1985: 803) (5) I eat what I like. (Quirk et al. 1985(Quirk et al. : 1056 Due to the bias of formal written language in textbooks and grammar books about the English language (Carter & McCarthy 2017), it is not surprising that descriptions of patterns of what in informal speech or writing are typically not included. However, thanks to the availability of spoken language corpora since the mid-twentieth century, grammar books such as Quirk et al. (1972Quirk et al. ( , 1985 and Leech & Svartvik (2002) also include examples from authentic spoken material, which are accompanied by information about tone unit, pitch location and direction, etc. This development is crucial because prosody is part and parcel of linguistic units in speech and conveys important discourse-functional information about meaning in language (Bolinger 1989;Cruttenden 1997). Such grammar books reveal further uses of what that rarely feature in the written mode. For example, Leech & Svartvik (2002) discuss so-called response questions such as short questions, (6), echo questions, (7), and general requests for repetition, (8). In all cases, what typically carries a nuclear pitch accent, either falling in (6) or rising in (7) and (8) Quirk et al. (1985) give examples of what that, like (6)-(8) above, are accented but that function as expressions of surprise and incredulity rather than as questions. One such example is given in (9). In historical linguistics, these uses of what are generally referred to as pragmatic markers (e.g. Brinton 1996Brinton , 2008Lutzky 2012). Example (10) presents another use of what, namely the parenthetical what, which always forms part of the preceding tone unit and is followed by a tone unit boundary (Dehé & Kavalova 2006). Prosody explains the fixation of the parenthetical what to clause-medial position typically in front of a cardinal number, and the function of the parenthetical is to frame the cardinal number as being 'too little/few' or 'too much/ many' (Dehé & Kavalova 2006: 303). Furthermore, what can be used in split interrogatives to make a guess (Michaelis & Feng 2015). The split interrogative in (11) receives two nuclear pitch accents, the first one on what and the second one on monk.
(9) A: I paid £1000 for that picture. B: Wh/at? You must be mad. (Quirk et al. 1985 although not labelled as such, is given in Stenström (1984). It is taken from the London-Lund Corpus (henceforth LLC-1) and presented here in (12) (Stenström 1984: 59) In the example, speaker A asks B a question about B's choice of pictures, a question that B is unable to respond to. Instead, B produces his own question (in bold), constituted by the unaccented interrogative pronoun what and the noun phrase that nude. After A's confirmation ( yeah), B answers the first question. According to Stenström (1984), B's question is a request for clarification that the speaker needs to make to be able to respond to A's original question. Stenström's (1984) analysis, however, does not include any explanation about the idiosyncratic formal properties of the construction, and the one example provided by her does not account for the range of interactive functions that the construction has in spoken dialogue.

Constructionist approaches to spoken dialogue
Spoken language in all its complexity and flexibility has been the preferred object of study in Conversation Analysis (CA) and interactional linguistic approaches to language (e.g. Sacks et al. 1974;Ochs et al. 1996;Couper-Kuhlen & Selting 2001;Heritage & Clayman 2010;Thompson et al. 2015). These frameworks have embraced the dynamic and situationally grounded nature of spoken interaction and sought to explain through detailed sequential analyses of the phenomena in question the recurring practices that members of a speech community carry out when they are engaged in conversation. By concentrating on what is observable in a language rather than what cognitive processes are linked to the storage and retrieval of the units that make up language, conversation analysts have, in the words of Fischer (2015: 580), resisted 'the use of cognitive concepts and explanations'. In contrast, cognitive principles and mechanisms form the basis of grammatical theories of language within the broad framework of Cognitive Linguistics including Frame Semantics and Construction Grammar (CxG; e.g. Langacker 1987;Fillmore et al. 1988;Goldberg 1995Goldberg , 2006 Fischer 2015). These studies have concluded that the two approaches can inform each other in important ways, but they have also pointed out that this is only possible if a deeper understanding of the relation between cognition and interaction is reached within the boundaries of the respective frameworks themselves. From the point of CxG, this understanding concerns the nature of the units of a language, or the information expressed by constructions. In the early days, the study of constructions was mainly concerned with elements of language that were not words per se but idiosyncractic or maximally schematic multi-word units (see Lakoff 1987 on there;Fillmore et al. 1988 on let alone; Goldberg 1995 on argument structure; Jackendoff 1997 on time-away). They attracted researchers' interest because they did not obey traditional phrase and clause structure rules. Later on, this conservative view was relaxed and a broader view of language description and explanation was embraced (e.g. Croft 2001;Goldberg 2006;Fillmore & Baker 2010;Fillmore et al. 2012) in which language is regarded as a collection of form-meaning pairings. These form-meaning pairings, or constructions, are conventionalised elements of some size in language that are organised in a network based on family resemblance (Goldberg 2006). 5 In fact, the uses of what described in section 2.1 above are best described as a network of what-constructions where what has become conventionally associated with a particular syntagmatic context and a corresponding constructional meaning (for a discussion of wh-question constructions, see Goldberg 2006).
Although CxG takes as its starting point a relatively narrow conception of language dealing with clause structure, the broader context is not an alien concept in the framework. In fact, CxG developed out of the work by Charles J. Fillmore and colleagues on Frame Semantics (and later on FrameNet), where emphasis is put on the coexistence of the 'cognitive frame' and the 'interactional frame' and the way we 'conceptualize what is going on between the speaker and the hearer' (Fillmore 1982(Fillmore / 2006. This interactional information is represented in constructional knowledge as has been done in the representation from Croft & Cruse (2004) in figure 1. As can be seen in figure 1, the conventional meaning of a construction is all-inclusive and features properties related to semantics, pragmatics and discourse-function. For example, the What's X doing Y? construction (Kay & Fillmore 1999), famously featured in the joke Waiter, what's this fly doing in my soup?, has a special constructional meaning of surprise and displeasure that is not simply a case of conversational implicature derived from the literal interpretation (as in the waiter's reply: Madam, I believe that's the backstroke). Instead, the pragmatic interpretation of the construction as expressing disapproval is directly encoded in its meaning (cf. Michaelis & Feng 2015;Cappelle 2017).
There is a fast-growing body of research within CxG that is truly committed to extending the conception of construction into the realm of dialogicity and interaction. Other studies are concerned with linguistic categories and specific expressions in them that are either under-researched or pose a challenge to purely grammatical models of language. For example, Imo (2005) describes the prosodic, syntactic and functional properties of the phrase I mean in spoken American and British English in order to explain the relationship between I mean and other, more schematic constructions such as discourse markers and complement-taking verb-constructions. Similarly, Fried & Östman (2005) carry out a close analysis of a number of pragmatic particles in the Swedish dialect Solv and contemporary spoken Czech. They show that each particle displays properties that constitute a network of expressions, which is characterised by overlapping formal and/or functional attributes and values on the one hand and by idiosyncrasies that distinguish them from each other on the other hand. In the network, each usage pattern is a construction that incorporates not only syntactic and semantic information but also prosodic, pragmatic and culture-specific aspects of communication. Fried & Östman (2005Fried & Östman ( : 1776 conclude that 'CxG is well equipped to address the complexities of spoken language, if one allows the notion of construction to be extended in a dialogical direction.' The contributions in Deppermann & Günthner (2015) suggest that there is another key feature of spoken interaction that is central to the way grammatical constructions are conceptualised in language: the temporal unfolding of the interactional sequence. Linell (2009a: 97) takes a rather strong position and criticises CxG for suffering from 'an interactional deficit', ignoring the information that constructions encode with respect to sequential dependencies in discourse. He studies the Swedish reactive construction x-och-x (e.g. flytta och flytta 'move and move'), which he considers to belong to the category of responsive constructions alongside, for example, concessive repairs (e.g. I can switch off well not really switch off but you know relax; Couper-Kuhlen & Thompson 2005: 262-3). The x-och-x construction arises when a speaker repeats a key word or expression, i.e. the x element, from the previous utterance to problematise the situated use of that word or expression. The construction projects a segment where the speaker aims to make clear that the x element contains a Figure 1. The form-meaning correspondence of a construction (Croft & Cruse 2004: 258) number of different readings and that the reading used by the interlocutor may not have been the most appropriate one. According to Linell (2009a), the conditions that the construction sets up on the previous sequence, and the conditions that it sets up on the subsequent sequence, necessarily form part of the speakers' knowledge of its use in language. Linell (2009a) also makes an important distinction between the meaning potential (or 'functional potential') of a construction and its actualisations in natural language use (cf. Paradis 2011). The x-och-x construction, for example, 'has somewhat different semantic-pragmatic effects depending on the meaning potential of the lexical item (x) with which it interacts' (Linell 2009a: 105).
Elsewhere, Linell (2009b: 300-2) discusses the responsive and projective properties associated with what-constructions. He notes that, while what-pseudo-clefts (e.g. what I want to do is travel) have responsive properties, which explains why the construction only becomes relevant if something in the prior context requires further elaboration (see Günthner 2006 for German), interrogative what-constructions are primarily projective and make relevant certain kinds of responses, depending on the design of the question. As will be shown below, the reactive what-x construction shares important features with these constructions as it is closely tied to the sequential and dialogic organisation of speech.

Data and method
This section presents the methodology of the study. Section 3.1 introduces the corpus, LLC-2, and section 3.2 the sample from which all the utterances containing what were retrieved. The classification criteria of the what-constructions are established in section 3.3.

London-Lund Corpus 2
The data for the present study come from a new corpus of spoken British English, LLC-2, collected between 2014 and 2019. 6 The corpus contains around 500,000 words of both dialogue and monologue. The speakers are educated adult speakers of British English. An important feature of LLC-2 is that the transcriptions in the corpus are timestamped and directly link the speaker turns in the transcriptions to the corresponding places in the sound files, allowing for prosodic analysis of the data. The transcription and markup conventions used in the corpus are as follows. Speaker IDs in front of every turn are marked with letters in alphabetical order. Square brackets represent overlaps, and in the case of multiple overlaps, they are numbered. Pauses, both long and short, are represented by <pause/> and unclear transcription by <unclear/>. Truncated words are embedded within the <trunc> and </trunc> tags and anonymised names within the <anon> and </anon> tags. The tag <vocal/> describes different kinds of non-verbal vocalisations and the tag <event/> describes events. Filled pauses are spelt out as uhm.

The sample
The sample contains 50 texts and each text is approximately 5,000 words in size. The total sample size is 251,550 words. The texts are spontaneous and private face-to-face conversations from LLC-2. In the corpus, the face-to-face conversations are divided into conversations among equals and disparates. The interlocutors are equal if they are friends, peers in the workplace or related by descent or marriage (e.g. parent-child, husband-wife); they are disparates if they have hierarchically unequal positions in the workplace or the educational institution (e.g. employer-employee, supervisor-student). It should be noted that the interlocutor status is not the focus of this study and no conscious effort has been made to draw a balanced sample of the two conversational settings, primarily due to the limited number of completed transcriptions in the corpus. Nevertheless, such a distinction allows us to explore the possible effect of power and social status on the use of the reactive what-x construction, and a few remarks will be made when the results are discussed.
The sample was automatically searched to find all instances of the word what and a total of 1,661 of them were found. A considerable number of the examples, 95, were excluded for one or more of the following reasons.
• The search word could not be deciphered due to unclear transcription and poor quality of the recording.  In the examples, all instances of what, if accented, have been marked for prosody following the British tradition of contour analysis where / indicates a rising pitch contour from a low accented syllable, \ a falling pitch contour from a high accented syllable and # a tone unit boundary (Cruttenden 1997).
In what follows, we will summarise the main properties associated with the form and meaning of each construction. Interrogatives, which include both pronouns and determiners, are placeholders for what is unknown. Also, what is an obligatory element, i.e. it cannot be removed (e.g. *are you up to these days), and followed by a finite verb form. Exclamatives (determiners) are expressive and emphatic, and subordinates include, for example, nominal relative pronouns, which introduce relative clauses. So-called response questions feature three types of questions where what is typically accented: (i) short questions are elliptic and signal a request for more information; (ii) echo questions repeat part of the interlocutor's message and replace the rest with what; and (iii) general requests for repetition signal that the reiteration of the whole message is anticipated. Pragmatic markers and parentheticals are only loosely connected to the rest of the utterance or stand alone altogether. While the pragmatic marker what expresses surprise and incredulity and is always accented, the parenthetical what is clause-medial and never accented. Instead, it belongs to the same tone unit as the preceding sequence and is followed by a tone unit boundary. Finally, split interrogatives have a bipartite syntactic and prosodic structure and are used to make a guess.
Out  the use of the construction is different from all the other uses described in the literature so far. Thus, the reactive what-x construction has its own defining properties, which need to be described with respect to its frequency, form-meaning and functional properties in authentic communication. This will be done in the subsequent sections.
4 The reactive what-x construction in LLC-2 The analysis of the form and meaning of the reactive what-x construction is presented in this section. In section 4.1, we determine the frequency and distribution of the construction in the data, and in section 4.2, we describe the construction in more detail, particularly properties related to its internal structure, sequential organisation and meaning.

Frequency
The reactive what-x construction occurs 45 times in the sample. Even though the number itself is relatively small, the construction makes up almost 3 per cent (45/1,566) of all the what-constructions in the data. By way of comparison, exclamative determiners feature only 15 times and split interrogatives only 8 times. 7 These findings relating to the number of occurrences show that the reactive what-x construction is in regular use in contemporary British English speech. At the same time, it deserves to be pointed out that the construction only appears in less than half of the texts in the sample, namely in 17.5 8 out of 50, which may be due to individuals favouring certain ways of expressing themselves (Barlow 2013) or the fact that constructional use in conversation is contagious across speakers in dialogue (Garrod & Pickering 2004). A closer look at the types of conversations in which the reactive what-x construction occurs reveals that the construction is mainly used in conversations among equals; only five examples were found from conversations among disparates. This suggests that the construction is regularly used among British English speakers who know each other well.

Form and meaning
This section is concerned with defining the properties of the reactive what-x construction. We distinguish between three types of properties: internal, sequential and semantic. where what precedes and specifies the whole declarative clause she would eat it. By way of comparison, when the pronoun is moved to a clause-medial slot, as has been done in the made-up example in (14), it functions as a parenthetical and only focuses on a single element in the clause, in this case the main verb eat. Also, when what is moved to a clause-final slot, as in (15), we get the impression of an incomplete and prematurely terminated utterance rather than a fully fledged turn. The same is true when what follows a phrasal complement. Second, what in the reactive what-x construction is always followed by a phrasal or clausal complement. The clausal complements found in the sample are either declarative or interrogative, and the phrasal complements are noun phrases, prepositional phrases, an adverb phrase and a non-finite verb phrase but never a finite verb phrase. In other words, the chunk that what combines with diverts from canonical syntagmatic combinations in English. The construction, then, has a special structural status; what is only loosely connected to the complement and can therefore be omitted. Examples (16) and (17) illustrate this point. The complement that what specifies in (16) is the interrogative clause did you have a look in John Lewis or, which already displays a structure typical of yes-no questions (AUX-S-V-X). 9 Thus, as evidenced in (17), the complement does not necessarily require what as part of its syntactic structure and can stand on its own if needed. It should be noted, however, that the utterances in (16) and (17) are by no means identical and that, although the removal of what may not have major structural consequences for the utterance, the pronoun is vital for conveying the constructional meaning associated only with the combination of what and the phrasal or clausal complement (see below). 9 Note that the utterance-initial is it is a false start and not part of the reactive what-x construction.

318
NELE PÕLDVERE AND CARITA PARADIS (16) A: I think they've got lots of toys and stuff so B: is it what did you have a look in John Lewis or A: mm I did have a look in John Lewis (17) did you have a look in John Lewis or Third, the reactive what-x construction always forms one tone unit. This means that, despite being outside the phrase or clause it modifies, what never forms its own tone unit nor does it carry a nuclear pitch accent. Instead, what is the pre-head of the entire tone unit which also includes the complement, and there is no pause between the elements. By way of comparison, the utterance in (18) is a reactive what-x construction, while the utterance in (19)  Syntactically, the utterances are similar to one another; what occurs in an initial position and specifies a noun phrase, mince pies in (18) and a bird in (19). An instrumental prosodic analysis, however, reveals that the intonational contours of what are different in the two utterances. Figure 2 represents the pitch contour of the utterance in (18), drawn in the phonetics software Praat (Boersma 2001). As can be seen in figure 2, the utterance is an example of the reactive what-x construction; what is realised as an unaccented pre-head of the tone unit, and the nuclear pitch accent of the tone unit falls on pies and takes the shape of a rise-fall. Figure 3 shows that the utterance in (19) is different from the reactive what-x construction. 10 In particular, what and the complement a bird do not form one but two separate tone units, both of which take the shape of a rise-fall pitch contour. The tone unit boundary between what and the complement is evidenced by the step-down in pitch on the subsequent element, namely the indefinite article a. The intonational pattern of the utterance in (19), then, suggests that it contains the pragmatic marker what since the pronoun is prosodically prominent and used to express surprise at A's statement, i.e. that he would hit a bird with a shovel. It should be noted that 13 out of all the 39 pragmatic markers in the sample display the prosodic marking as in figure 3 (see section 5.4 below for a brief discussion of the relationship between the pragmatic marker what and the reactive what-x construction).
The formal dimension of the reactive what-x construction is symbolically linked to a specific constructional meaning, and the meaning or function of the construction is to  320 NELE PÕLDVERE AND CARITA PARADIS negotiate or call into question what is being talked about. More specifically, the construction is targeted at something said in the immediate, prior discourse, which suggests that the preceding turn plays a central role in the emergence and interpretation of the reactive what-x construction. Similar to other responsive constructions such as the ones discussed in section 2.2 above, the reactive what-x construction displays explicit responsive properties as opposed to non-responsive constructions that may only do so implicitly or with respect to other aspects of the context (Linell 2009b: 296). In this way, the construction shares important dialogic features with what-pseudo-clefts (see section 2.2) but also with the pragmatic marker what, which is used to express surprise and incredulity at what has been said or done before. Consequently, the reactive what-x construction never occurs in the first position of a conversational episode but always as a response to something said by another speaker in the immediately preceding turn. Example (20) illustrates this point. 11 In the example, speaker C produces the reactive what-x construction what the National Express as a reaction to A's description of her journey home. From the description, C is unable to successfully infer the exact referent of it (and it gets me there in the same time), or has reservations about the accuracy of her inference. In order for the conversation to continue, the speaker needs to immediately close the knowledge gap between the interlocutors, and she does so by naming the most contextually relevant and viable referent in the complement position. The example in (20) also points to the fact that the response in the subsequent turn is important because it resolves the ambiguity that inhibits the mutual understanding of the topic under discussion. In other words, the construction is a request and projects a specific response, i.e. a confirmation of what is requested in the complement. Indeed, many reactive what-x constructions in the sample display the sequential properties witnessed in (20) where A confirms the proposal with the agreement marker mhm and the conversation resumes its natural course. As mentioned in section 2.2 above, this pattern is typical of interrogative what-constructions, which anticipate and expect a certain kind of response from the interlocutor. At the same time, there is some variation in the way in which the reactive what-x construction is responded to in spoken dialogue, which suggests that the presence and nature of the subsequent turn is rather more influenced by what is going on in the local interaction than strictly associated with the formal and sequential properties of the construction itself (see section 5 below for more details).
The constructional representation of the reactive what-x construction in terms of form and meaning can be visualised as in figure 4. The formal dimension is divided into two parts: internal and external structure. The former specifies that the reactive what-x construction comprises the initial interrogative pronoun what and the subsequent phrasal or clausal complement x, which together form one tone unit (indicated by #). The latter specifies that the construction presupposes the presence of a preceding turn by another speaker, while the presence and nature of the subsequent turn is more flexible. Together the internal and external properties contribute to the meaning dimension of the construction, which is to react immediately to the preceding turn in order to negotiate or call into question something said. This means that the register that the reactive what-x construction is restricted to is (informal) spoken dialogue, where it has become a conventionalised unit with distinct grammatical, sequential and semantic characteristics.
A closer look at the discourse contexts in which the reactive what-x construction occurs, however, suggests that reaction constitutes only the meaning potential of the construction and that the representation in figure 4 does not adequately or with enough detail account for the full range of its uses in spoken dialogue. These uses are discussed in the next section.

Dialogic functions of the reactive what-x construction
This section is concerned with the dialogic polyfunctionality of the reactive what-x construction in spoken dialogue. We identify three functions: requests for verification, requests for information and adversative requests. In the following sections, each function is discussed in turn followed by an account of the theoretical implications that the findings of this study have for CxG.

Requests for verification
The most frequent dialogic function of the reactive what-x construction is requests for verification. They occur 28 times out of 45 in the sample. Requests for verification are NELE PÕLDVERE AND CARITA PARADIS made when speakers need to verify a specific referent in the preceding turn, as illustrated in (21). The example comes from a work meeting between two arts journalists, A and B, who talk about B's upcoming business trip to Moscow but have just slightly digressed from the topic. The request what to Russia is produced by speaker B as a reaction to A's prior question have you ever been out there before. Uncertain about the referent of there, possibly due to the earlier digression, B makes the request to verify it.
(21) A: they've been working <pause/> and supporting <pause/> curator shows in Venice <pause/> both with uhm the you know the Venice biennale plus the architectural [biennale] they've been doing that for quite a while <pause/> and it just means that it then has a permanent home B: [mm] B: yeah that's handy <pause/> and they know what they're working towards A: yeah <pause/> have you ever been out there before <pause/> sorry <event desc="refers to food"/> <vocal desc="laugh"/> <pause/> B: what to Russia A: mm B: no <pause/> have you As shown in (21), requests for verification indicate a breach and a possible imbalance in the common ground between the interlocutors. Common ground is defined as 'the sum of [two people's] mutual, common, or joint knowledge, beliefs, and suppositions' (Clark 1996: 93). Clark (1996: 97) also notes that speakers often have 'conflicting information' about what is in their common ground. Requests for verification emerge when one of the speakers makes this asymmetry of knowledge explicitly known. Based on contextual, perceptual and/or socio-cultural cues, however, the speaker is able to make an assumption about the identity of the problematic referent, which is then inserted into the complement position of the request.
As mentioned in section 4.2 above, the constraints that the reactive what-x construction sets up on the subsequent sequence depends on dialogic function. The data show that requests for verification require swift confirmation in the subsequent turn. This means that common ground and contextual relevance are particularly important for requests for verification and for re-establishing sufficiently good mutual understanding between the interlocutors. In (21), for example, speaker A immediately verifies the request made by B with the agreement marker mm, hence enabling B to provide a response to the interlocutor's original question.
The request for verification in (21) above is only mildly attitudinal. The attitude is that of impatience and urgency since the imbalanced common ground between the interlocutors hinders successful continuation of the conversation. However, there are instances of requests for verification in the data where the speaker takes a more explicit stance, as in (22) In the example, speaker B's request for verification (what you find that appealing) seeks not only to clarify but also to convey surprise and disbelief that A finds something appealing. The attitude is enhanced by laughter and a rising pitch contour at the end of the tone unit.

Requests for information
Another dialogic function of the reactive what-x construction is requests for information.
With 15 occurrences, they are less frequent in the sample than requests for verification, and they also differ from the latter in important ways. Particularly, speakers make requests for information to elicit new information from the addressee. Such requests correspond roughly to Stenström's (1984) eliciting questions that open either initial or non-initial superordinate exchanges. Requests for information form part of non-initial exchanges in that they do not completely change the topic of the discussion as initial exchanges do but rather shift its orientation within the boundaries established in the prior context. An example of a request for information is given in (23) In (23), the speakers discuss London transportation prices. The request for information (what do you have a lecture every day) is produced by speaker A, who does not use it to react to a specific element in the prior discourse as was the case with requests for verification but to shift the focus of the conversation from the high cost of public transportation in London to B's lecture schedule at university. We argue that what in (23) serves as a cue to this topic shift and helps the speaker signal to the addressee that what follows is at the same time a contextually relevant and a novel contribution to the conversation. The removal of what from the utterance would result in the loss of this communicative function. The topic shift is further confirmed by B's subsequent explanation of why she needs to go to the university every day. Generally, requests for information are less likely to project a confirmation, because the conceptual gap between the interlocutors is greater and hence more challenging for the speaker to fill. B: 2 [well no] <pause/> no no no no no it's because I the picture was me stood in front of <pause/> these two paintings because they're life-size but she's a lot smaller than <pause/> <unclear/> came here so the picture didn't look right Example (24) is taken from a meeting about an upcoming art exhibition. The preceding turn to the request is speaker B's expression of relief that A and C added in the work of Judith who would have been very upset if this had not been the case. The request for information produced by C, what with her photograph not being credited, is at the same time an inquiry about Judith's reason for being upset and an ironic implication about her vanity. The ironic reading of the construction is reinforced by the falling intonational contour at the end of the tone unit. Thus, C adds a novel perspective to the conversation in a way that makes his negative attitude explicitly known. The controversial nature of the assumption is confirmed by the quick and strong reaction by B (well no no no no no no) and a lengthy explanation of what she meant.

Adversative requests
The last dialogic function of the reactive what-x construction is adversative requests. Adversative requests are fundamentally different from requests proper in that they are not used as questions. Instead, adversative requests convey opposition and contrast with the view expressed by the previous speaker in a seemingly contentious and even hostile manner, which means that they are used to take a stance and explicitly express the speaker's attitude. There are only two examples of adversative requests in the sample. The scarcity of such requests in the data does not allow us to provide a complete picture of their usage constraints, but the discourse environment in which the two examples occur is clear enough to reveal the most important properties of their function and how they differ from the other requests. The examples are given in (25) and (26): what there was a <trunc>de</trunc> in (25) and what you liked all of it in (26). In (25), the last word of the clause, presumably deer, is truncated. The incompleteness of the clause is not due to interruption by another speaker but the speaker's own impatience to defend her position further.
(25) A: <anon>Jess</anon> I wouldn't have taken as someone who ate road kill but she was going to pick up a deer the other day uhm B: are you sure that wasn't a story <pause/> A: what there was a <trunc>de</trunc> we saw the deer that got knocked 1  The primary function of adversative requests is to signal disagreement with the claim made by the previous speaker. In (25), speaker B questions the legitimacy of A's story, which prompts the expression of a strong counterargument by A that her friend did indeed intend to pick up road kill. The counterargument is characterised by an increased speech rate compared to her previous turn and high pitch, contributing to the adversative nature of the utterance. The interpretation is also supported by the speaker's eagerness to provide further evidence for her claim in a prolonged turn. The same interpretation is evoked in (26) where speaker B calls into question A's negative assessment of mathcore bands. The disagreement is first expressed by the opposite assessment made by A (it was terrible -> it wasn't terrible) followed by the adversative request. The construction receives a nuclear rise-fall accent on all of it and is delivered in a clearly confrontational manner. In contrast to the example in (25), the counterargument in (26) is immediately followed by turn exchange, i.e. the interlocutor's further disagreement (no I didn't). Hence, the limited number of examples does not allow us to properly account for the nature of the subsequent turn in adversative requests.

Implications for CxG
The socio-cognitive and communicative properties associated with the three dialogic functions discussed in the previous sectionsrequests for verification, requests for information and adversative requestsprovide a sufficiently comprehensive description of the use of the reactive what-x construction in spoken dialogue. On the one hand, the properties shared by the functions allow us to make generalisations about the construction as a whole, but on the other hand, all three functions display properties that uniquely distinguish them from each other. For example, inquiry is a unifying feature of the two types of requests, for verification and information. However, the requests differ from each other in terms of speaker intent and discourse orientation; while requests for verification seek confirmation on something said in the prior discourse, requests for information elicit new information and signal topic transition. Finally, both requests proper and adversative requests express speaker stance, but there is variation in the extent to which this stance is foregrounded and made salient by the speaker. It should be noted that the properties associated with each function are the result of an observation of a limited number of occurrences. However, while exceptions may exist, all of the properties discussed above have been found to be 326 NELE PÕLDVERE AND CARITA PARADIS central to the characterisation of the reactive what-x construction and its realisations in authentic communication.
We argue that the dialogic information associated with the use of the reactive what-x construction in specific discourse contexts is clearly conventionalised and in recurrent use in contemporary English. This contextual information builds on the meaning potential of the construction, i.e. to react to and negotiate something said in the prior discourse. Without accounting for this dialogic embedding, the many ways in which the reactive what-x construction is used in spoken dialogue would not be identifiable. Thus, an extended representation, which accounts for the interplay between the meaning potential and the dialogic functions of the construction, is presented in figure 5. Moreover, in the revised figure, register is given its own separate label to highlight the constraints and possibilities that the broader social context places on constructional knowledge. What the figure aims to make clear is that the knowledge associated with any construction, and particularly a construction specific to spoken registers, goes well beyond form-meaning pairings in the strict sense and into the realm of dialogicity and interaction (cf. Fried & Östman 2005;Linell 2009a;Wide 2009;Fischer 2015). It is only by accounting for all these dimensions and domains of instantiation that we reach a comprehensive understanding of language and the constructions that constitute it.
The discussion above raises many important questions about constructional change and particularly the mechanisms that led to the emergence of the reactive what-x construction and the subsequent expansion of the schema to include the three dialogic functions. These questions go beyond the scope of this study but are addressed in detail in Põldvere & Paradis (2019). For example, Põldvere & Paradis (2019) propose that the reactive what-x construction developed from interrogative what-constructions, but that its development was supported by the non-interrogative, responsive schema. Besides inheriting the responsive properties with which the construction is so strongly associated, the reactive what-x construction also retained its interrogative function. This development seems to be identical to that of response questions as described above. The entrenchment of the constructional schema in spoken dialogue has led to the increased productivity of the construction and the development of new dialogic functions. This process necessarily involved subjectification in the sense of Traugott (1989Traugott ( , 2010 and Traugott & Dasher (2005) as the construction became increasingly grounded in the speaker's perspective and attitude and moved closer in meaning to the pragmatic marker what. Indeed, Põldvere & Paradis (2019) entertain the idea that the reactive what-x construction may be one of the source constructions of the pragmatic marker.

Conclusion
The focus of this study has been the reactive what-x construction in English conversation. We have carried out an investigation of the construction in a sample from LLC-2 in order to determine its frequency and distribution in the data and to describe the form-meaning and interactive functional properties that characterise and constrain its use in spoken dialogue.
This study is the first one to identify and define the reactive what-x construction, constituted by the interrogative pronoun what and a phrasal or clausal complement x within the boundaries of one and the same tone unit. Our analysis of a sample from a new corpus of spoken British English, LLC-2, revealed that the construction makes up a fair proportion of the data relative to other what-constructions. One possible reason for the lack of research on the construction is its novelty of use. Further research in historical linguistics is needed to explore the extent to which the construction is a new phenomenon in English as well as its developmental path in the broader network of what-constructions (Põldvere & Paradis 2019).
The study also makes an important theoretical contribution to CxG through the inclusion of discursive, dialogic and prosodic information in constructional representation. We showed that the formal dimension of the reactive what-x construction incorporates information not only about its internal syntactic structure but also the sequential and prosodic contexts in which it is used in spoken dialogue. For example, the construction never occurs in the first position of a conversational episode but always as a reaction to an immediately preceding turn by another speaker. We also showed that reaction constitutes the general meaning potential of the reactive what-x construction and that the systematic exploitation of the potential in spoken dialogue leaves us with three conventionalised dialogic functions: requests for verification, requests for information and adversative requests. Of course, a larger sample size may reveal more functions of the construction, but this question is left for future research. In sum, by combining insights from both the broad cognitive constructionist model of language, CxG, and interactional perspectives, we hope to have demonstrated the advantages of accounting for the full range of formal and functional properties that characterise the use of a construction in spoken dialogue and the much more comprehensive view that one obtains by doing so. 328 NELE PÕLDVERE AND CARITA PARADIS