Grammar in time: pragmatic contingency and non-restrictive which

1.1 The non-restrictive relative which-clause

Research on the non-restrictive relative clause – e.g., [She chose the cheapest cocktail [ which was non-alcoholic ]] – has focused overwhelmingly on whether this clause is, or is not, part of the syntactic representation of the sentence that contains it. Descriptions of the latter perspective – the so-called ‘radical orphanage’ approach – can be found in Fabb (1990), Espinal (1991), Burton-Roberts (1999), Schlenker (2010), Poschmann (2018), and Jasinskaja and Poschmann (2020), while the former, contrary view, is exemplified in Potts (2005), Arnold (2007), Nouwen (2007), Arnold and Borsley (2008), and AnderBois et al. (2010). There has thus been broad consensus on the analytic focus of the non-restrictive relative, if not its treatment. Indeed, this syntactico-semantic emphasis is reflected in the recasting of the restrictive/non-restrictive (‘appositive’) distinction as one between ‘integrated’ and ‘supplementary’ relative clauses in the descriptive grammar of Huddleston and Pullum (2002), given that the former do not have to be semantically restrictive. Importantly, though, in defining non-restrictive (or, as they have it, supplementary) relatives, Huddleston and Pullum note that one criterion for their definition is grounded in prosody; specifically, these relatives are “set apart prosodically (and usually marked off punctuationally), and the information they express is presented as supplementary rather than an integral part of the larger message” (2002: 211). So, for instance:

In (a), the cocktail selected was the cheapest amongst the non-alcoholic cocktails (i.e., there may have been an alcoholic cocktail on offer that was in fact cheaper), while in (b), the cocktail selected was the cheapest amongst all available cocktails (alcoholic and non-alcoholic alike).

Huddleston and Pullum’s reference to prosody in the case of the non-restrictive/supplementary relative (b) seeks to capture a feature of these clauses that has thus far been largely neglected by work in this domain: their production on particular occasions of use. Consideration of naturally occurring language data quickly reveals that the prosodic production of the which-clause can be even more disintegrated than suggested in the citation form in (b) above, in which the comma indicates that the main clause is not brought to prosodic completion before the relative clause. In fact, as we shall see, it is over twice as likely to be produced as a completely separate intonation unit, for example, after a speaker has produced the main clause with final intonation, sometimes even followed by a silence:

While in the majority of the cases, the main clause is produced with final falling intonation (as in the invented exemplar (c)), we shall also evidence cases where there is final rising intonation on the main clause. So whether the intonation on the main clause is falling or rising, in both cases the main clause is produced with a final intonation contour, with the subsequent relative clause then produced in the wake of that ostensibly completed main clause (often after silence).

Furthermore, which-clauses are also found to be produced at even greater distances from their hosts than after a simple silence – namely, occurring after intervening speech from interlocutors, but nonetheless parasitic on their own prior turn rather than that of the interlocutor, as in invented example (d):

Tao and McCarthy’s (2001) exploration of non-restrictive relatives in spoken English was the first descriptive study to note the common production of such clauses as what they called ‘turn extensions’ – namely, produced after prosodic completion of the main clause (as in (c) above). But beyond the mere existence of examples like (a)–(d) as empirically attested possibilities for speakers, little is known about what motivates speakers to mobilize these clauses in these particular ways. In other words: why should a speaker bring an utterance to completion, and then elect to speak again, and to do so using the non-restrictive relative which? These questions demand that we interrogate the syntax of these clauses in conjunction with both their prosodic and their pragmatic particulars, considerations that bear on the methodology to be used in the present study.

1.2 The present study: methods and materials

The range of possible prosodic realizations of which-clauses in English-language data necessitates close examination of the details of their deployment on actual occasions of use. It is an empirical reality that English users can and do produce such clauses at different points in the flow of talk, and with different prosodic integration as they do so. If our objective, therefore, is to uncover systematicities in the grammatical construction and temporal emergence of these clauses, we are obliged to consider their contextualized occurrence in corpora of naturally occurring conversational language data.

In addition, and relatedly, the examination of which-clauses in their on-line production and the pragmatics of their use requires analytic attention to be paid beyond the focal utterance. Since actions are not unilateral, but implemented across sequences of utterances (Schegloff 2007), it is necessary to consider not only the relationship between the non-restrictive which-relative and its host, within the target turn itself (e.g., (a)–(d)), but also the relationship between those components and what came prior in the discourse, as well as what those components occasion from recipients in response. This necessitates an approach to linguistic data that attends to the temporal, sequentially unfolding nature of interaction in its consideration of grammar (see, e.g., Clift 2001, 2016; Couper-Kuhlen et al. 2024; Raymond 2016).

In the analysis that follows, we begin by describing non-restrictive which-clauses produced as in examples (c) and (d) as one of a class of what have been called ‘increments’ in naturally occurring language use (Schegloff 2016 [2000]; see Couper-Kuhlen and Ono 2007; Ford et al. 2002). We situate this discussion within the extensive research on cross-linguistically attested universals of turn-taking (Sacks et al. 1974; Stivers et al. 2009) and use the theory and methods of Conversation Analysis (CA) to examine the data (Clift 2016; Robinson et al. 2024).

From a corpus of 25 h of recorded, transcribed interaction, we then focus on a collection of 210 instances of what we call which-increments. We first introduce our basic phenomenon: three instances showing, in these particular cases, which-increments produced in the wake of a silence after a main clause. We then proceed to discuss the pragmatic constraints on usage that have emerged out of observational studies of naturally occurring interaction to show how they are operational on all the observed uses of which-increments. To do so, we examine which-increments in all their possible instantiations, relative to a main clause: after a gap of silence following a main clause, brought to completion on either rising or falling intonation;^[1] in a next beat following that clause; after other talk from the recipient; and, finally, as a recompletion of another’s talk. It will be shown that the first two positions show increments being used to align with the recipient in terms of understanding or a common stance, and the second two show increments being used to disalign. Thus the production of the non-restrictive which as an increment is only understandable by reference to prosodic and pragmatic factors,^[2] and so implicated in profoundly contingent, interactional projects. We conclude by considering what this study of one particular clause tells us about the relationship between grammar and its implementation in interaction.

3.1 Post-gap: pursuing alignment

The following three data extracts reproduce examples (2)–(4) in their more extended sequential contexts to show clearly how the host clause in each case (ll. 10–11 in (2), ll. 56–58 in (3), and ll. 6–7 in (4)) is built for uptake and a display of alignment, but gets none:

So in (3), alignment would be manifested by some immediate display of understanding Dan’s turn from ll. 56–58; in (2) and (4), some common display of stance in the face of the assertions “he used to hold my hand” and “he said EVERY TIME I COME HERE YOU’RE EATING” (clearly designed as a punchline), respectively. In contrast to extracts (1) and (5), in which speakers through-produce the non-restrictive relative clause in pursuit of alignment, each of these host clauses shows the speaker’s orientation to a principle of minimization. Nonetheless, as we have seen, in each of these cases, the turn in question fails to receive uptake from the recipient – the silence a failure of progressivity, and a harbinger of an emergent dispreferred, disaligning response (Pomerantz 1984; Sacks 1987 [1973]; Schegloff 2007). It is in the environment of this lack of alignment that the which-increment is produced, appended to what is thereby retroactively transformed into a host TCU.

What the which-increment accomplishes in each case is to convert the silence between what otherwise would have been two different speakers’ turns, into a silence within a single speaker’s own turn – and so transform an inter-turn gap into an intra-turn pause (Lerner 2004: 176; Schegloff 1995: 198). In (2) and (4), the increment makes explicit the speaker’s stance towards that which is being reported; in both cases, these get alignment from their recipients (albeit belatedly in (4)). In (3) the increment works towards elaborating Dan’s point that, contrary to what Roger claims, therapy has improved the situation for him at home, which Roger begins to align with at line 63.

Importantly, the extended sequential context in each case makes clear the broader activity in which the participants are involved, and therefore to what ends the which-increments are being deployed. Produced in post-gap position, which-increments seek to remedy failures of progressivity: in environments where recipients do not make their understanding or stance apparent, speakers make use of the anaphoric element of the which-clause to extend their own turn, elaborating on or assessing the content therein, and simultaneously thereby furnishing another opportunity (at the increment’s end) for securing a show of alignment. In this way, we see the affordances of grammar profoundly intersecting with the real-time, situated contingencies of its deployment.

3.2 Next-beat: adjusting speaker’s stance

While examples (2)–(4) reveal failures of progressivity, and – by means of the silences greeting the host clause – incipient disalignment, increments produced in an earlier position, relative to the host clause, are so situated in attempts to forestall just such failure. This they accomplish through the explicit adjustment of their own stance in what we, following Schegloff, are calling ‘next-beat’ position. Schegloff calls these ‘next-beat’ increments (2016 [2000]: 253) because they are “begun by speakers in the next prosodic beat following the prior possible completion of a TCU (i.e., prior to any attributable silence or uptake)”. In the following cases (7)–(9), we see examples of non-restrictive which produced, in next-beat position, as what have been called ‘rush-throughs’. These are marked by a speeding up at the end of one TCU and continuation without delay into the next (Walker 2010: 62) so that the next TCU is heard as begun sooner than might have been anticipated (see Walker 2010 for evidence to support this). This articulatory practice is a means of obscuring prior turn-finality and, accordingly, the potential relevance of speaker transition (see also Local and Walker 2004; Schegloff 1998; Walker 2004, for discussions of related phenomena). In each example, the speaker uses the increment to make explicit an aspect of their stance from the prior TCU which is produced according to the principle of minimization, but immediately deemed by the speaker to be at risk of being misinterpreted:

In all three cases here, speakers produce a rushed-through which-increment in next-beat position, following what is thereby transformed into a host clause. In each exemplar as well we see the increment deployed by speakers to adjust or otherwise make explicit some aspect of stance with an assessment – making clear in (7) and (8) their own, and in (9) someone else’s, position. In (7) and (9), the recipient produces an immediate, latched acknowledgment; in (8), Matthew’s subsequent “u:h*::*”, “I mean t’say”, and clarification further “defend” (Maynard 2013) the speaker’s stance toward this wine in continued pursuit of such acknowledgment. These next-beat, rushed-through increments are thus designed as last-moment, even hurried adjustments to the stance meant to be enacted by the host clause.^[8]

It is perhaps no coincidence that all next-beat which-increments in our collection deliver assessments. Moreover, they are all produced in environments in which the speaker’s own position is vulnerable to being misunderstood: in (7), Leslie has already delivered negative assessments of Freddie Masters (ll. 6, 10–12) so the which-increment at line 21 is directed to establishing Robbie’s contrasting positive stance on Freddie Masters volunteering to “take assembly”; in (8), Matthew’s which-increment is designed to unpick a possible assumption that the “cheapest red wine” is not going to be any good; and in (9), Nancy’s which-increment is directed to clarifying Mr. Bradley’s stance as “very much against” the “hippy movement” – her earlier use of “wasn’t advocating” potentially being vulnerable to confusion as to Mr. Bradley’s actual stance, produced in a sequence devoted to conveying his views. Unlike the post-gap increments examined in the prior section (recall excerpts (2)–(4)), in these next-beat cases there is no failure of progressivity to remedy. Nonetheless, speakers display their orientation to both minimization and progressivity by endeavoring to append an additional grammatical unit to their turn-at-talk specifically before the recipient has a chance to produce a responsive action, thereby actively working to affect the design and delivery of that responsive action. In this way, next-beat and post-gap which-increments are similar: they are both designed in pursuit of the recipient’s alignment, whether in terms of intersubjectivity and common referential ground, or in terms of stance or position. These grammatical units may form the first element of a more extended pursuit, as in (2) and (3) (which eventually secure acknowledgment) and (8); they may themselves secure acknowledgment, as in (4), (7), and (9); or they may even prompt the kind of full-bodied affiliation seen in (2), “Yes that’s right. ’N he’ll cuddle up to you …”.

Both post-gap and next-beat increments, then, are produced by speakers bringing a host clause to completion according to a principle of minimization in the first instance. Speakers then display their orientation to progressivity as they contingently reconfigure the ‘end’ of their turn in the environment of a lack of alignment (post-gap), or in anticipation thereof (next-beat). In the case of next-beat increments, presented in this section, speakers actively work to momentarily stave off recipient response in favor of appending the which-increment, which makes clear the stance of the host clause for the recipient. They are thus other-directed: produced by the speaker in pursuit of alignment, whether in the form of intersubjective understanding or a common stance from a recipient.

3.3 Post-other-talk: resisting the recipient’s contribution

While the examples above (in Sections 3.1 and 3.2) offer ample evidence of speakers pursuing alignment through the use of which-relatives, it is by no means the case that all such clauses are produced to such ends. In the cases we have seen thus far, the increment is produced before the recipient’s uptake (be it in next-beat or post-gap position vis-à-vis the host), and hence in pursuit of it. But which-increments can also be positioned after a recipient has already launched a response. Consider extracts (10) and (11):

In cases such as these, speakers position which-increments in overlap with recipient responses (Drew 2009; Jefferson 1986) – that is, some way into the recipient’s responsive next turn. This may be either, as in (10), near a possible TRP, or the more strikingly disattentive instance in (11) where the increment is placed in interjacent overlap (Jefferson 1986) – that is, nowhere near the possible completion of the prior turn. In both instances, speakers disattend and thereby sequentially delete the recipient’s incipient contribution to the talk, as they work to reclaim the floor.

As opposed to endeavoring to secure alignment from recipients, which-increments produced in post-other-talk position serve positively to disalign. They do this by grammatically proposing, through the anaphoric element in the which-increment that ties back to the host, that their own turn (i.e., the host) was in fact not yet complete, and thus was in fact not yet ready for reply; in short, they disalign by casting the recipient as having come in ‘too early’ or otherwise inappositely. In (10), Nan’s continuation of her prior turn into Lesley’s response to it makes out that Lesley interrupted her. When these which-increments are elaborations, designed to secure intersubjective understanding, their placement after appropriate uptake, as in (10), constitutes the recipient as in need of such elaboration – in short, as slow on the uptake. In (10), Nan’s increment serves to give the sense of an obligation dutifully met, whereas as Lesley’s turn makes clear, it is Lesley who is doing the favor for Nan. The favor is interactionally bulldozed by the placement of Nan’s increment.^[9] In (11), Adam’s telling regarding his mother comes to a possible completion with the upshot “so it’s (.) (what,) fifty sixty miles of driving”; Mary’s launch of an assessment (and an apparently disattentive one, at that) at line 29 is produced in overlap at the first indicator of possible completion, the pitch peak on “driving” a clear instance of a speaker disaligning in the placement of their turn. Adam’s which-increment, coming as it does some way into the launch of the turn, in turn disattends Mary’s launched assessment in favor of his own – one which continues the telling, so proposing that he had not, in fact, finished. In the event, Mary takes up a stance as recipient at the end of the which-increment, subsequently relaunching her assessment at line 37 – with which Adam in turn aligns. It is clear from this extract that Adam and Mary are more broadly engaged in pursuing affiliation and thus alignment, achieved at line 40. But this does not entail that the mechanism by which this is achieved is consistently to align, as Adam’s disaligning placement of the which-increment in line 30 shows.

The following shows a comparable but extreme case of self-directed talk, where alignment is claimed by a speaker (“I understand that completely”) while in fact disaligning, at lines 16–24, where the speakers are claiming alignment in overlap. It also shows how far apart the increment may be positioned vis-à-vis its host. The speaker, Hunt, is responding to the turn at line 15, “It’ll limit my ability to help you”. What he says in response, abstracted away from the interactional sequence and the circumstances of its production, is (at lines 17, 20, and 23): “I uh- Chuck I understand that completely. Believe me. Which is why I never tried to get in touch with you’:

In all these cases then, the which-increments do the opposite of pursuing alignment by sheer dint of their placement. In bringing a host turn to grammatical, prosodic and pragmatic completion in conformity with a principle of minimization, the speaker invites uptake; by subsequently producing a grammatical continuation after that uptake is launched, and so proposing that they have been interrupted, is to be self-directed, disattentive, disaffiliative, and therefore strongly to disalign.

With such post-other-talk position cases now in view, we begin to see that whether or not which-increments constitute moves to align or disalign may thus be seen to be a function of their positioning: in the three possible places examined so far – post-gap, next-beat, and post-other-talk – it is their combined position and composition (Schegloff 1993: 121) that ultimately determines what they serve to achieve.

3.4 Other-initiated recompletion: proposing the other’s turn as incomplete

A last set of increments identified by Schegloff (2016 [2000]) consists of those produced not by the speaker of the host TCU, but by its recipient, yet in all other respects are similar to the examples examined so far. In such cases, recipients elect to append to the prior speaker’s turn a grammatical continuation which in effect recompletes that prior speaker’s turn.^[10] Consider cases (13) and (14).

In each of these extracts, the first speaker initially produces an informing. In response to the informing, the recipient in each case opts not to supply a fitted response, such as an information receipt, to the informing, but rather proposes, by means of the which-increment, that the prior speaker’s turn is incomplete. In (13), Edgerton’s recompletion of Donald’s informing makes explicit what was implicit in both Donald’s earlier assessment, “I’ve decided … that it’d be better to dispense with a clutch pedal car” (ll. 5–7), and subsequently in line 11 with the upshot marker ‘so’ (Raymond 2004) prefacing the informing. In producing an elaboration – “which is automatic” – of something (“a Volvo 345”) which needed none, and so by undermining rather than facilitating the other’s public trajectory of action, Edgerton clearly disaligns. Similarly, we can see in (14) that a recipient resists treating a turn as informative (ll. 5–6) by withholding an information receipt – although in this case there may be an institutional imperative for doing so. In this extract, the British Foreign Secretary responds at ll. 6–7 to the mention of an event he would equally well be expected to know about, his use of the grammatical continuation serving to construct the point being made as familiar to him. In both extracts, we see the potential for conflict in disalignment, however affiliative such recompletions may be designed to be. So in (13), after he attempts to complete Edgerton’s recompletion in line 13, Donald launches a recompletion of his own. His choice of the same recompleter both provides the assessment that might have been provided at line 11 (and the strength of that assessment – “magnificent” – is surely designed to be unambiguously positive, in contrast to Edgerton’s blandly noncommittal “automatic”) and is a means to deny his recipient the last word on the matter. In (14), the which-increment is produced in an already potentially hostile environment; the Foreign Secretary’s positive gloss on developments in southern Iraq is effectively squelched by the interviewer’s challenging “be-that-as-it-may” interception (itself a continuation) “and one of the governors was promptly assassinated”. The assessment in response – “which is very very serious” – resists the slick, punchline matter-of-factness of the prior turn, and is immediately followed up by a shift away from the governor assassination point (“but …”, l. 6). In both (13) and (14), then, the which-increments are used to promote, in their syntactic congruence, what might appear to be alignment; however, in proposing the other’s turn as incomplete, it is, in fact, in being self-directed and disattentive, disalignment.^[11]

Given the disaligning character of such increments, it is perhaps unsurprising that they may be produced in contexts where turn opportunities are strictly allocated. In the following extracts from broadcast radio, one speaker is given the last word, providing no opportunity for comeback from the other. In the first of these, we see the last few seconds of a daily morning news programme before the sign-off and ‘pips’ (Greenwich time signal) for 9 a.m., signaling the end of the programme. In the second, a recorded interview is brought to a close, followed by the journalist’s voiceover:

In both of these, it is evident how the recompleter, proposing the prior turn to have been incomplete, has the last word, facilitated structurally by the denial of speaking rights to the other. Such environments, as we see, are suited to the use of ironic subversion.^[12] Just as (11) shows clearly, such disaligning moves may be launched by speakers even if, more broadly, they are engaged in affiliative courses of action.

Having established the interactional constraints operating over the production of which-increments, we have now examined in some detail the positions of these increments relative to their host TCUs. It turns out that the placement of increments relative to their hosts is consequential for whether or not speakers are moving to pursue alignment or to disalign with their co-participants; increments produced before the co-participant has responded to the host TCU are seen to be other-directed and pursue alignment, whereas those produced in the wake of a response are visibly self-directed and so moving to disalign.