orchestrating ensembles in interactionhomepages.inf.ed.ac.uk/gmills/mills2013_ensembles.pdf ·...

$: Orchestrating ensembles in interactionhomepages.inf.ed.ac.uk/gmills/Mills2013_ensembles.pdf · Orchestrating ensembles in interaction Gregory J. Mills ... \Rufus Reid told Berliner$
Orchestrating ensembles in interactionGregory J. Mills

1 Interaction in an ensemble is basic

1.1 The musical score

Many accounts of musical capacity have taken a Cartesian, individualisticperspective, dominant until recently in the domain of linguistics, that fo-cuses on the perception of musical form. Such accounts concentrate on thefinal product of musical activities, represented by a musical score, and pos-tulate abstract hierarchical structures underlying the musical surface that, itis claimed, listeners assign to a musical piece during music perception [Ler-dahl and Jackendoff, 1983; Jackendoff and Lerdahl, 2006] . The idea that a“musical grammar” guides an individual’s understanding of music and pro-vides the source of musical intentions is characteristic of the computational-theory-of-mind approach to cognition that investigates both music and lan-guage at an abstract representational level isolated from online processingand sensorimotor abilities and skills that allow interaction with the envi-ronment and other individuals (see also Gregoromichelaki, this volume).However, music, like language, is an intrinsically social phenomenon. Mu-sic, like language, is learned in interaction, and the primary site where musicis performed is in interactive group settings [Cross, 2008]. Even very younginfants engage in proto-musical interactions, and their proto-linguistic andproto-musical communicative behaviours appear indissociable [Cross, 2008;Papousek, 1996], while care-giver-infant interactions frequently also exhibitproto-musical attributes [Cross, 2008].

From this perspective, interacting musically and linguistically in ensem-bles of musicians and interlocutors is the norm. When an orchestral ensem-ble gives a performance, all the prior practice, rehearsals, discussions aboutthe interpretation of the piece by the conductor and the musicians, theirprior knowledge about the piece, and previous performances by the mem-bers of the ensemble, both in the same ensemble or in different ensembles, allculminate ideally in a flawless performance. This might give the impressionthat the final performance is what matters and that faithful reproductionof some predefined musical score, as the composer intended it, constitutes

2 Gregory J. Mills

both the purpose and the unifying factor of the musical activity [Weeks,1996]. However, this is equivalent to focusing on the final stage of a scriptedtheatrical performance rather than investigating people’s natural, creativeengagement in everyday conversation. Nonetheless, even in highly con-strained types of musical engagement (as in classical performances), groupcreativity is a necessary parameter which is manifest in the eventual per-formance. This is because both a musical score and a conductor’s commu-nicative signals instructing the musicians on how to perform widely under-determine most aspects of how the performance will unfold [Sawyer, 1996;Benson, 2003; Schutz, 1964]. Music and language are similar in this respect:both linguistic expressions and musical scores are radically underspecified,relying on participant coordination for their meaningful use during interac-tion.

Joint action, joint intentions and coordinationExisting accounts of language use in conversation differ in their emphasis onwhich mechanisms are involved in underpinning meaningful use of expres-sions and how their deployment affects the course of interaction. Formalapproaches to dialogue, which were developed under the computational the-ory of mind assumption, operate under standard Gricean assumptions, (e.g.Grosz and Sidner; 1986; Cohen and Levesque, 1990; Poesio and Traum,1997; Poesio and Rieser, 2010). Such accounts see the formulation of deter-minate intentions/plans and their full recognition as the main causal mech-anism underlying dialogue comprehension and production. Performancein joint tasks then relies on the coordination of (joint) intentions/plansthrough negotiation and grounding that establish mutual beliefs and com-mon ground. Similar prominence to explicit negotiation is also given by thegrounding model [Clark and Brennan, 1991]. Here, the role of coordina-tion devices, the basis for interlocutors’ mutual expectations of each other’sindividual actions, [Clark, 1996; Schelling, 1980] is emphasized. Altermanand Garland [2001] also argue for the value of explicit negotiation in theachievement of coordination

In contrast, Pickering and Garrod [2004] observe that explicit negotia-tion is neither a preferential nor an effective means of coordination (see alsoGregoromichelaki, this volume; Gregoromichelaki et al., 2011; Mills andGregoromichelaki, 2010; Mills, 2011). If it occurs at all, it usually happensafter participants have already developed some familiarity with the task.Moreover, even when a particular approach to the task is explicitly negoti-ated and agreed by the participants they do not seem to persevere with itfor long. Instead of intention- or plan-recognition as the bases of negotia-tions and grounding, the Interactive Alignment model developed by Pick-ering and Garrod [2004] emphasizes the importance of tacit co-ordination

Orchestratring ensembles in interaction 3

and implicit common ground achieved via the psychological mechanism ofpriming. However, because this model emphasizes the similarity betweenparticipants’ representations and contributions, it does not appear ideallysuited to account for the sequential organisation of interaction which relieson complementary, rather than identical, dialogue moves.

Music is widely assumed to not possess descriptive semantic meaning,hence intention recognition is not usually assumed to be required to dis-ambiguate propositional communicative messages. In this respect, musicalintention recognition usually refers to comprehending intentional expressiveaspects of a musician’s performance of a pre-composed score. However, insocial settings, for example, in musical group improvisation, arguably themost cross-culturally natural setting for music making [Cross, 2008], thereis usually no musical score that serves as the basic coordination device. Inorder to play in unison, musicians must constantly coordinate with eachother on when, how (procedural coordination) and what to play — in fact,this is the case even in classical ensembles that are guided by a score [Keller,2012]. As in everyday conversation (see e.g. Clark, 1996), the musiciansneed to coordinate on sequentiality, e.g., their entry and exit from varioussections in the musical piece. As in conversation, they also need to deter-mine group participation and party-formation: during performance “sub-ensembles” might form and re-form, for example a soloist accompanied bythe violin section; similarly, the violin section might subdivide into separateensembles playing complementary melodies. How is this high level of pro-cedural coordination accomplished? How is the interaction orchestrated?Currently there is considerable interest in looking beyond the musical scoreand towards the mechanisms and resources, common to music and language,that enable and sustain coordination among ensemble members.

1.2 Coordination mechanisms at a fine-grained level:theoretical and experimental investigations

Performances and conversations are always situated in a social context andstrongly embodied. When groups of people, whether musicians or conver-sationalists, share the same space, they spontaneously form ensembles witheach other, which regulates interpersonal coordination in highly orderedways [Kendon, 1990]. When arranged in a circle, the central “O” space isused as a common space into which participants look, speak and gesture.This structural integrity of the ensemble is maintained via a rich set ofembodied cues, such as mutual adaptations of body position, posture andorientation [Schegloff, 1998]. However, this integrity is also highly flexible,adaptable and interactive: the size of the space is sustained when the ensem-ble moves (e.g. to let someone pass by); members of ensembles sway in uni-

4 Gregory J. Mills

son to indicate topic change, and there are specific procedures for approach-ing, inviting, joining and leave-taking (Kendon, 1990). Moreover, membersof an ensemble must pay close attention to each other’s actions. Duringperformance, musicians rely not simply on their knowledge of the piece, butalso observe bodily cues such as body orientation, posture, beat, body sway,gestures, visual cues such as holding an instrument or turning a page, as wellas auditory cues, e.g. breath, of their co-performers. But further than ob-serving and interpreting each other’s actions, it has been argued that antici-pation and prediction are key mechanisms to coordinating in both linguisticensembles (see e.g. Pickering and Garrod, 2013; Gregoromichelaki et al.,2013) and music ensembles (see, e.g. Keller, 2012). It should be noted thatsuch mechanisms are based on both social and cognitive substrata: beingpart of an ensemble with a shared history of interaction allows members topredict each other’s actions and utterances, yielding tight coupling betweenmembers’ behaviours. Participants in ensembles can also explicitly demon-strate their orientation towards such couplings as a further means of assert-ing the integrity of the group. One means of enacting membership is via an-ticipatory contributions, for example, uttering a word before another inter-locutor gets the chance to (split utterances, see e.g. Gregoromichelaki, thisvolume) to demonstrate common knowledge of the topic [Goodwin, 1986],or handing an object to another prior to their request [Clark and Krych,2004]. Besides their cognitive basis, crucially, such anticipations are alsoused for interactional purposes, for example, to demonstrate membershipboth within ensembles, and also, performatively, to outsiders [Sacks, 1995;Schegloff, 2007].

In musical group improvisation settings, similar phenomena have beennoted demonstrating their resemblance with conversation. Musicians “tradeideas with each other” [Sawyer, 2006] and these moments of “give-and-take”,of heightened, coordinated interactivity are described by musicians as oneof the aesthetic high points of musical performance:

“Rufus Reid told Berliner how he tries to weave the prior soloist’s ideasinto his own solo, but not always in an obvious way, and not always bydirect quotation; he said it was more interesting to elaborate on the prioridea.” [Sawyer, 2006].

However, similarly to what has been claimed for linguistic interaction,here there is also a temptation to treat such behaviours as being fully deter-mined by a script, plan or intention in an individual’s mind. Many ensembleshave a conductor whose role is to serve as the central coordination device[Parton and Edwards, 2009], since in large ensembles, due to the speed ofsound, and relative amplitudes, members of an ensemble might be spatiallyseparated from other members such that they are unable to coordinate au-


ditorily with each other [Malhotra, 1981]. As a result, folk-psychologicalintuitions often display a “centralizing” tendency to explain creativity inensembles [Sawyer, 2006]: while the key characteristics of group creativityare improvisation, collaboration and emergence of novel forms, we tend toattribute the creativity of a group to an individual, e.g. the director, leader,soloist or conductor, rather than to the group dynamics.

Despite the inherent interactivity of music, the majority of experimen-tal designs eschew fully interactive scenarios, for example, they only testentrainment to predetermined rhythms or recordings. Designs that do useinteracting dyads typically require participants to follow a musical score,obviating the need for musicians to develop spontaneously full coordina-tion and turn-taking. Yet turn-taking, the scaffolding for the sequentialorganisation of interaction, is perhaps the most fundamental mechanismunderpinning other forms of coordination in joint activities. This is becausedialogue coherence, the basis for defining notions such as anticipation andprediction, can only be defined when the procedural turn-taking infrastruc-ture is in place [Scaife and Bruner, 1975; Levinson, 1983].

More importantly, unlike speech which relies on prosodic, lexical andsyntactic cues to turn-taking [De Ruiter et al., 2006], requiring anticipa-tion at many different levels, when following a musical score, there is lit-tle flexibility in the cues that musicians might be able to give each otherwithin the music to guide turn-taking behaviour. In addition, the precisemetrics afforded by musical scores come at the expense of being able toinvestigate novel musical and coordinative innovations that emerge out ofthe interaction, since what emerges must match the score or rhythm to beentrained on. However, it has been shown that within minutes of inter-acting in a new ensemble, participants rapidly coordinate on pitch [Gileset al., 1991], vocabulary [Clark, 1996], semantics [Pickering and Garrod,2004], and procedures [Mills, 2011]. These emergent structures are notreducible to forms of progressive entrainment, instead they follow deter-minate courses common in multiple forms of interaction [Schwartz, 1995;Healey, 2004].

Further, focusing on similarity of behaviour (whether linguistic or sim-ilarity to a score) can lead to an impoverished conception of interaction,especially concerning error. With a score as criterion, the more similarthe actions, the greater the success; in fact this is also the basis of theclaims made for linguistic resource matching in e.g. the Interactive Align-ment model and the grounding models, where identical joint intentions haveto be derived as a criterion of success. Yet work on interpersonal coordi-nation in ensembles shows that on encountering difficulties, interlocutorscopy each other more not less, hence matching is often a signal of obsta-

6 Gregory J. Mills

cles rather than an indication of smooth interaction [Fusaroli et al., 2012;Louwerse et al., 2012; Mills, 2007; Mills, 2013b].

A further problem with restricting the interaction to scored dyadic per-formance is that many interpersonal coordination phenomena in ensem-bles only emerge through adaptive co-regulations by multiple participants[Kuhlen and Brennan, 2013; De Jaegher et al., 2010; Riley et al., 2011;Semin and Cacioppo, 2008; Semin et al., 2012]. As a result, the representa-tions used by ensembles once they are established differ qualitatively fromthose used initially [Garrod and Doherty, 1994; Mills and Healey, 2008].This is parallelled in musical ensembles: depending on the musician’s skill,coordination cues can have a differential effect, variously helping or hinder-ing coordination [Schober, 2006; Duffy and Healey, 2012].

Sharing intentions in ensemblesTransferring the individualistic focus on the musical product in the socialdomain, as an account of ensemble coordination, Keller [2008] assumes thatthe main aim of ensemble musicians is to communicate information aboutmusical structure and expressive intentions to co-performers and audiencemembers. Accordingly, the analysis is based on solo performance assump-tions, where, it is claimed, expressive devices consist in intentionally in-troducing variations in performance parameters such as tempo, intensity,articulation, and sound quality. In the ensemble case, it is claimed, themain difference is that instead of expressivity being a matter of individ-ual variation, it becomes an achievement of “inter-individual co-variation”.Under this view, the only problem that then arises for coordination is howto match these variations across performers. Keller[2008] proposes threecore cognitive-motor skills that underpin each member’s activity within theensemble: (1) adaptive mechanisms, with the core one being adaptive tim-ing, that are sensitive and respond to variations in other members’ actions,for example, adjusting the timing of one’s movements in order to maintainsynchronization with the others; (2) prioritized integrative attending whichinvolves paying close attention to one’s own musical actions (high priority),while simultaneously attending to another’s musical actions (lower prior-ity) and also paying attention to the sound of the full ensemble (see alsoKnoblich and Jordan, 2003); (3) the use of anticipatory imagery that reliesin running mental simulations of the others’ performance.

Although this description of music coordination aims to include an ac-count of the low-level cognitive processing underpinning musical joint action,it crucially relies on prior higher-level coordination and intention recogni-tion. While, on the one hand, this low-level view aims to open up a per-spective of the mechanisms that underpin communication and joint actionin general, actually it starts by presupposing the coordination it seeks to


explain. Keller states that “once shared goals are established” then theselow-level cognitive-motor ensemble skills can be used for the participants tocoordinate on those goals. However, this begs the question of how “sharedgoals” are established in the first place. What is the nature of the “sharedgoal”? How much structure needs to be shared? Is it a vague, generalcommitment to perform a piece together? Does it include each individualnote? How much leeway is there for individual expressive intentions? Anyanswers to these questions already presuppose interactive coordination: in(musical) ensembles, shared goals are typically established via verbal inter-action, in discussion, while rehearsing. A similar paradox is found in studiesof joint action where the low-level processes assumed are similarly relianton shared, high-level joint goals and representations (in experimental set-tings these are also transmitted verbally, by the experimenter). Further, thedivergent evaluations that even highly skilled musicians give of their ownand of each other’s performance suggest that fully shared intentions are notessential for skillful coordination in an ensemble [Schober and Spiro, 2013].

On this intentionalist view, then, the shared joint intentions are orches-trating the musical performance. But since we find it difficult predictinghow we will respond to others’ affective states that are in response to ourown that in turn are responses to others’, an individual musician’s expres-sive musical intentions concerning a piece of music will necessarily have acomponent that is emergent out of the interaction, it cannot come prede-fined even in highly-scripted/orchestrated performances. This process ofemergence is non-summative: putting together the intentions/plans of eachindividual performer will not generate the eventual performance withoutgoing through the interaction process. The question then becomes: how dothese emergent intentions affect the (presumed) prior sharedness of musicalgoals? No matter how much musicians describe a piece of music to eachother and establish their ostensible shared goals, the only evidence they getwill, in fact, come from the actual collaborative performance. Thus partof the process of playing a piece of music must entail continuous assess-ments of whether or not the co-performer does in fact share an intention.Participants believing that they have shared intentions is no guarantee ofsuccessful interaction; members of an ensemble must constantly adapt tothe immediate context, relying on feedback from the other participants, ina non-summative process.

Empirical research on dialogue coordination in joint activities has al-ready highlighted the problematic notion of intentions, introducing a wholeset of epistemological, ontological and also temporal issues [Arundale, 2013;Haugh, 2008; Haugh, 2012; Suchman, 2007]. In addition, although earlycognitive and AI approaches attempted to explicitly articulate and define

8 Gregory J. Mills

coordination devices such as scripts associated with particular social con-texts, e.g. going to a restaurant [Schank and Abelson, 1977], these attemptsultimately proved intractable and unsuccessful. Against this view that em-phasises the sharing and matching of intention, in the music domain, Cross,[2008] observes that collaborative musical joint action essentially involves amultiplicity of different perspectives: the “intention” or “aboutness” of apiece of music varies between different contexts, and even within the samecontext, can vary across individuals (“floating intentionality”). It is exactlythis plasticity, vagueness and flexibility of meaning that underpins musicalintersubjectivity, as it allows individuals to maintain their own, individualinterpretations while at the same time engaging in collective activity.

Although this argument is made with respect to musical joint action,a similar argument can be made for language; The underspecification andopenness of linguistic meaning is essential to having a usable communicationsystem [Cooper and Ranta, 2008; Mills and Healey, 2008; Fay et al., 2008;Fay and Ellison, 2013]. This is an inevitable feature of communication:when encountering a novel interlocutor, novel task or novel piece of music,ensemble members will necessarily be faced with violations of expectations,otherwise communication would not be profitable for each other, let aloneenjoyable [Egermann et al., 2013]. Derrida’s notion of differance, i.e., thatlinguistic communication can only function in virtue of there being an in-trinsic difference in meaning also points in the same direction [Derrida,1977].

From an analytic perspective, it seems therefore reasonable to take asa point of departure the interactive mechanisms that interlocutors use inorder to deal with violations of expectations, i.e. miscommunication. Thesemechanisms play a central role: the structure of the interaction must pro-vide members of an ensemble with the means of identifying, signalling andresolving their opposing perspectives. The purpose of engaging in conver-sational and musical interaction is to exploit these means in order to createnovel, emergent products that were not residing in an individual’s mindprior to the interaction. From this perspective, what needs to be sharedby participants is not prior intentions, scripts, plans or representations butinstead ways of knowing how to traverse an interaction so that these con-structs can emerge in a non-summative manner.

1.3 The emergence of signals and routines for orchestrating theinteraction

In musical joint action, free dance or dialogic conversational interaction,members of ensembles must coordinate with each other moment-by-momentusing a variety of signals. Even if they had a clear plan providing top-down


guidance on the sequence of actions, whether in the form of a script foreating at a restaurant or musical score, participants must still signal toeach other how to transition through the activity, as all contingencies ofthe interaction can never be fully determined by a representation, whethera plan, script or a score [Suchman, 2007]. This requires tight, moment-by-moment feedback as regards one’s own perspective as the joint actionsunfold, using interactive routines, gestures and behaviours to direct eachother’s behaviour, attention and understanding towards the contingenciesof the interaction, misunderstandings, clarifications and uncertainty. Thesignals produced by musicians during performance are sometimes describedas “coordination smoothers”, e.g., exaggerated body movements associatedwith performance, e.g. breathing, body sway and head nods. Phillips-Silverand Keller 2012 propose that these actions serve to make an individual’sactions more overt and predictable. But from the current perspective, thisis an impoverished view of interaction and communication: in interaction,the signals we use to communicate are used, not simply as outward manifes-tations of an internalized plan; instead, signals are used to propose, accept,decline, hint, suggest, interrupt, correct, confirm, allude, and play withanother’s behaviour (see e.g. Schober, 2006).

The “meaning” of these signals is not straightforwardly referential; in-stead they have a procedural function. They are used to coordinate howthe musical piece (or activity performed by the ensemble) unfolds. Differ-ent ensembles develop different cues for signalling to each other. Duringrehearsals, as an ensemble becomes more skilled, these signals become im-bued with highly ensemble-specific procedural functions. Rather than beingpredetermined, practice, within an ensemble is constitutive of the proce-dural meaning of these signals. As a result, these signals can be highlyidiosyncratic and specific to the ensemble. For example, although there isa semi-standard repertoire, the communicative signals used by conductorsto orchestrate a performance are idiosyncratic gestures, each conductor willhave their own idiosyncratic gestures that they use to coordinate during theperformance [Gumm et al., 2011] How does one find out the intention be-hind these gestures? While practising, rehearsing and discussing the piecewith the conductor, the ensemble will learn the idiosyncratic gestures usedby the conductor to coordinate the piece. However, they don’t learn byexplicit instruction, it occurs over many occasions in rehearsal while theirconscious focus is on music-making and collaboration.

Ensembles must go through a process of emergent self-organisation asa system relying on low-level, local processes that (subconsciously) directattention to the relevant problematic aspects of the performance: even inperformances involving a conductor, 15% self-report never looking at the

10 Gregory J. Mills

conductor [Malhotra, 1981]. Instead musicians pay attention to others nearthem and coordination is established and sustained via gestures, facial ex-pressions and bodily movements that, in turn, propagate the local coordi-nation achieved. These signals are not simple outward manifestations ofan inner musical pattern, they are communicative and have a coordinativefunction, e.g.:

“The first violinist’s raised eyebrow may indicate to a second violin thathe or she is playing slightly flat and must raise her pitch. The nodding ofthe first bassoonist’s head or raising of the right eyebrow can cue in thesecond bassoon” [1981, p. 106].

Note that Malhotra says “may” — the raised eyebrow might mean some-thing else at a different stage in the unfolding sequence, and this proceduralmeaning will depend on the rehearsals and the problems encountered duringrehearsal. Perhaps more importantly, this shows that the same superficialcue (raising an eyebrow) can have two different meanings by different peo-ple. It is a general observation in group coordination that multiple roundsof interactions (i.e. rehearsals) ensures that participants can have differentmeanings for the same cues at different stages of the unfolding interaction[Mills 2011; Mills 2013]. Even in highly scripted interactions, each cue ac-quires its “meaning” by reference to its position in the task, a sequentialprocedural coherence that needs to emerge via participant interaction in atrial-and-error process that allows them to coordinate their perspectives onthe task.

Coordination in dialogue is underpinnd by complementarityOne of the most fundamental insights of CA is that coordination in dialogueis underpinned by sequential structures which consist of pairs of comple-mentary contributions [Schegloff, 2007]. Canonical adjacency pairs include:question/answer; request/compliance; compliment/self-denigration. Adja-cency pairs consist of a first pair part and a second pair part that operatenormatively: production of the first pair part creates an expectation thatthe second half is accountably “due” [Heritage, 1984], leading any responseto be interpreted as pertaining to the second half (see Drew et al this vol-ume). This locally managed system of local sequential coherence betweenturns results in global coherence through the hierarchical interleaving ofembedded sequences that resolve local problems through routines of, e.g.clarification, elaboration and reformulation [Levinson, 1983]. Importantly,both the first pair part and the second pair part of complementary struc-tures in joint activities can be either an utterance or some other action[Clark and Krych, 2004]. Further, individual pair parts are often typicallyassociated with highly differentiated, complementary roles, whether briefand alternating (speaker vs. hearer) or more extended (customer vs. ticket


agent).

Complementarity of contributions introduces a new perspective on coor-dination by focusing on mismatches and how they are resolved locally bythe actions taken by participants. In psycholinguistic analyses of dialogue,e.g. Pickering and Garrod, 2004 offer a potential account of how such com-plementarity can be achieved by employing simulation of each participantby the other so that direct, non-inferential access to their respective goals isderived. Yet this perspective still relies on similarity between interlocutorsas simulations can only be effective if the same goals are already associatedwith the observed actions in a participant’s own action repertoire. Simi-larly, while it has been shown that coordination of timing and sequencingin an ensemble is underpinned by routinised adjacency pairs, conversationanalytic and cognitive studies of ensembles presuppose that adjacency pairsare already shared (and known to be shared) by interlocutors.

However, when engaging in a novel ensemble, it cannot be assumed thatinterlocutors already know the first pair-part, the second pair-part, or eventhe minimal joint action (i.e. adjacency pair) that comprises both pair-parts. Adjacency pairs can even acquire multiple conventionalized secondpair-parts [Schegloff, 2007, p. 227]. CA and cognitive approaches haveeschewed systematic analysis of how adjacency pairs are established in thefirst place. Further, once an adjacency pair is established, neither approachaddresses how the progressivity and sequential implicativeness [Schegloffand Sacks, 1973] of the same adjacency pair might change on successiveuse, in particular how an adjacency pair might acquire another second pairpart. Schegloff [2007, p. 223] remarks that activities can have very differentassociated routines which must be deployed at specific, conventionalized“junctures” within the activity. How, when establishing coordination in anensemble, do interlocutors also establish what these “junctures” are? Howdo interlocutors coordinate with each other if they disagree about whichjunctures are in fact relevant for coordinating the activity? On repeating anactivity, how do interlocutors establish that a particular sequential locationin the interaction is in fact the same juncture as that encountered on aprevious occasion?

Adjacency pairs are also normative conventions [Schegloff and Sacks,1973]. If at a particular juncture an interlocutor is unable to provide a con-ventionalized, second-pair part, interlocutors typically provide an accountfor their inability to complement the other’s action, e.g. giving reasonsfor declining an invitation or demonstrating an inability to shake hands byshowing how both hands are carrying heavy items. Normativity exacer-bates the procedural coordination problem. When establishing adjacencypairs, at what point do simple regularities of structural associations be-

12 Gregory J. Mills

tween two pair-parts acquire normative status, resulting in “accountableabsence” [Schegloff, 2007] if one of the pair-parts cannot be provided by aninterlocutor?

Solving these procedural coordination problems is in the first instancean interactive problem faced by members of an ensemble and it is indeedsolved by them during interaction. From an analytic perspective, modellingthese problems and their solutions requires giving an account of how twodifferent contributions come to be associated as part of a whole adjacencypair. Further, since one of the hallmarks of coordinated interaction is itsprogressivity (i.e. forward momentum) [Schegloff, 2007], the developmentof procedural coordination will necessarily involve the differentiation of in-terlocutors’ turns as coordination increases. Although there has been awealth of research on how interlocutors rapidly adapt to each other whilereferring to static referents, there has been almost no research on (a) howinterlocutors refer to each others’ interactive behaviour concerning how toperform joint actions together with each other; (b) how interlocutors ensurethat two complementary individual, constituent actions of a joint actionare performed correctly, whether simultaneous, or sequentially, (c) how the(inevitable) procedural coordination problems are identified, signalled andresolved (see e.g. Mills 2011, for an initial attempt to investigate proceduralcoordination separately from semantic coordination).

In addition to emergence during a single interaction, coordination alsodevelops at larger timescales - seamless performance often requires manyrehearsals. It is currently unclear how the coordination problems encoun-tered on each rehearsal, and the solutions they engender, contribute towardcoordination on the subsequent rehearsal. Moreover, while the finished per-formance might not involve overt verbal communication, rehearsals typicallyrely on linguistic coordination to identify and modify elements of the per-formance. Thus, even though the finished “product” might be coordinatednonlinguistically, this is often only made possible via expectations that wereestablished verbally, using vocabularies and routines that were interactivelyestablished during rehearsals. To this end, this paper demonstrates howprocedural coordination (of routines) develop within an ensemble, during asimple joint activity.

The data discussed in this paper come from a series of collaborative,computer-mediated maze game experiments. The experimental setup is amodified version of the original maze game experiments conducted by [Gar-rod and Anderson, 1987] and is the same basic setup as that described inHealey and Mills [2006],Mills and Healey [2006], Mills [2007]. This taskpresents pairs of participants with the procedural coordination problem ofguiding each other through the mazes, requiring them to establish routines


Figure 1. Example maze game pair

for instructing, requesting and navigating the switches and gates that blocktheir traversal through the maze. Dyads must solve 12 randomly gener-ated mazes, permitting systematic investigation of how the communicationsystem develops on each successive occasion of the joint activity.

2 Empirical data

2.1 The maze task

ProcedurePairs of participants were recruited from student pools, and participatedfor university credit. All were native English speakers. Participants weregiven 90 minutes to complete 12 mazes. Dyads consisted of a mixture offamiliar and unfamiliar pairs. Participants sit in different rooms, in front ofa computer screen which displays two applications (1) The maze applicationand (2) A text-based experimental chat tool for communicating. The mazeapplication displays a simple maze consisting of a configuration of nodesthat are connected by paths to form grid-like mazes (see Figure 1). Themazes are selected so that participants encounter an equal number of grid-like and irregular shaped mazes. The black circle shows the player’s currentposition and the cross represents the goal point that the player has to reach.Participants can move their location markers from one node to anothervia the paths. Each move is recorded and time-stamped by the server.The game requires both participants to move their location markers from astarting location to a goal. Although the basic maze topology is the samefor both participants, each participant has a different starting location, adifferent goal location, different switches and different gates, none of which

14 Gregory J. Mills

are visible to the other participant. They are also not able to see each other’slocation markers. Movement through the maze is impeded by gates thatblock some of the paths between nodes. These gates can be opened by theuse of switches (grey coloured nodes). The locations of switches and gatesare different on each maze and are not visible to the other participant.Whenever a participant moves to a node that is marked as a switch onthe other’s screen, all of the other participant’s gates open. All the gatessubsequently close when they move off the switch.

These constraints force participants to collaborate: In order for partici-pant A to open their gates, A has to guide B onto a node that correspondsto a switch that is only visible on A’s screen (and vice versa). Solving themazes, therefore (i.e. when both participants are on their respective goals)requires participants to develop coherent ways of referring to maze locations,while also requiring participants to develop procedures for requesting, de-scribing and traversing switches, gates and goals.

All participants’ communication occurs via a text-based experimentalchat tool [Mills and Healey, 2006; Mills and Healey, 2013 in prep]. Thischat tool interface is similar to proprietary chat tools (MSN, Skype, Face-book chat), allowing private turn formulation and revision prior to sending,resulting in participants’ contributions being displayed a turn at a time.

DataThe transcripts are selected from a series of maze game experiments de-scribed in [Mills and Healey, 2006; Mills, 2013a]. This paper adopts a qual-itative approach in order to describe the procedural coordination problemsand solutions encountered by participants while playing the task. The ex-cerpts in the appendix all come from separate trials. Each excerpt is a fulltranscript of all the turns from a single maze game trial. The excerpts arepresented as typed by the participants.

GeneralizabilityThis task is ideal for addressing how procedural coordination develops. Thecomputer-mediated version of the maze game task has been shown to exhibitsimilar local and global patterns of coordination to those observed in theoriginal, spoken task [Mills and Healey, 2006; Mills 2007]. The experimen-tal setup is analogous to the oft-invoked joint action problem of carrying anobject together. Consider two people, A and B, attempting to maneuver alarge piece of furniture up a narrow staircase. The piece of furniture, say acupboard, is sufficiently large such that A and B cannot see what the othersees. Due to the solidity of the object, any individual movement of theobject by A or B will have an immediate consequence for the other; theirindividual actions are (inescapably) yoked together as joint actions. How-


ever, since neither can see what the other sees, this creates a coordinationproblem, which is in part due to both A and B having privileged access tothe consequences of each others actions.

Analogously to how A only knows that A’s action has led to B’s cornerof the cupboard clearing an obstacle, in the maze game, a participant onlyknows if they are on a switch if the other person tells them that the gatesare open (and vice versa). This presents participants with the proceduralcoordination problem of instructing the other person what to do (i.e. goingto a switch location), and also informing the other what has just happened(i.e. whether or not their gates opened), forcing participants to coordinateon both.

2.2 The emergence of routines within ensembles

Initially: OK, my grey switchpoint which you have to get to now isnext to the sticking out cluster at the top

5 mins: that’s me done, can you go two down from the large blockof squares on the right

10 mins: Tell me yours and when I say, go onto the middle row, 3rdsquare from the left edge

15 mins: Now to open it, wait 3 seconds and then go to the 3rd row2nd sq from left

20 mins: wait then go 5th column top square

30 mins: went back to 5th column top

35 mins: now on 5 col, row 7 from left

40 mins: then 5c r 7. yours?

45 mins: 5, 7

Table 1:Global development of coordination in the maze game

One of the most consistent findings in studies of coordination in the mazegame is that over the course of the interaction, the turns that participants

16 Gregory J. Mills

use to coordinate procedurally become progressively concise. On each oc-casion, participants require fewer and fewer exchanges to signal to eachother how to progress through each maze. Returning to Table 1, note howinitially interlocutors use lengthy, elaborate instructions, e.g. at (0 mins):“...which you have to get to now” or at (10 mins): “Tell me yours and whenI say, go onto...”, which then becomes contracted after 40 mins to the bare“yours?”. Similarly, the instruction at (15 mins) “to open it wait 3 seconds”is shortened at (20 mins) to “wait”. Towards the end of the experiment,participants use highly elliptical language to instruct each other; the partic-ipants’ procedural language has effectively become “transparently absent”,and disappeared from the conversational surface.

This section analyzes how interlocutors establish and refine interactiveroutines for coordinating procedurally in the activity. It is helpful to ana-lyze the interaction in 3 stages. The analysis demonstrates how participantsare oriented from the outset toward establishing complementary structuresat specific sequential and temporal “junctures” within an activity. As co-ordination develops and the activity becomes sufficiently well-defined, thispaper demonstrates that these complementary structures become conven-tionalized as normative procedural conventions.

Initially: Low levels of procedural coordination: weak couplingof pairs of actionsInitially, participants are faced with the basic coordination problem of cou-pling [Riley et al., 2011; Fusaroli et al., 2013] their individual actions andutterances to allow them to carry out the joint actions of guiding each otherthrough the mazes. One of the most basic strategies is for participants toovertly assume complementary roles, e.g. in Dyad 1, Trial 0, Line 2: “menavigating and you driving” or Line 41: “I’ll direct”.

1 Participant1: Babs2 Participant1: me navigating and you driving3 Participant1: oh no4 Participant2: oh no@5 Participant1: When do we start6 Participant2: NOW7 Participant1: lemme direct you to my switch?8 Participant2: okey pokey9 Participant1: look at the third row up and second to the end block10 Participant1: i am, from the top, 2 across and down one11 Participant2: so the middle block?12 Participant2: of the third row13 Participant2: ???


14 Participant1: nah15 Participant1: start counting from the bottom16 Participant2: think rows and block numbers from left to right17 Participant1: 6 along three up18 Participant2: rows from the top!19 Participant1: from bottm20 Participant2: why make it more complecated!21 Participant1: just do it22 Participant2: row:? block?23 Participant1: ok atrat again24 Participant2: i dnt no what ur saying dunb bum25 Participant1: start again26 Participant2: ok27 Participant2: row and block?28 Participant1: if you look at the screen29 Participant2: im looking...30 Participant1: then look at the bottom left block31 Participant1: then count in a normal graph/grid fashion32 Participant1: regardless of the maze33 Participant2: OMG!34 Participant2: why so comlicated!35 Participant1: count 6 along reading frm left to right36 Participant1: then three up37 Participant2: ROW!38 Participant2: THEN BLOCK NUMBER!39 Participant1: cant you understand my directions?40 Participant2: NO41 Participant2: ill direct42 Participant1: ok its row three from the bottm43 Participant2: my first switch...44 Participant1: ok45 Participant1: ok46 Participant2: row 447 Participant2: 3rd block across48 Participant1: from bottom?49 Participant2: NO DUMB BUM i already told you, from the TOP50 Participant1: ok51 Participant2: like normal people52 Participant2: are you there?53 Participant1: row 4 has three blocks54 Participant2: yes! well done

18 Gregory J. Mills

55 Participant1: the third one across?56 Participant1: i have a gate before it57 Participant2: have you done it?58 Participant1: knob59 Participant2: ok60 Participant2: the other switch is...61 Participant2: 7th row,62 Participant2: 3rd block across63 Participant2: YAY!64 Participant1: im there65 Participant2: im at my goal!66 Participant2: where’s ur switch now,67 Participant2: USING ENGLISH68 Participant1: now you got to go to my switch u selfish cow69 Participant2: shut up bitch70 Participant1: from the top71 Participant2: if we were doing it your way, we wouldnt even be here!72 Participant1: 5 rows down73 Participant1: there are 6 on that row74 Participant1: its the 5th75 Participant1: reading from L to R76 Participant2: is it open?77 Participant1: no78 Participant2: wheres the other switch?79 Participant1: only one switch80 Participant2: just type row and block81 Participant2: just two numbers cant be that hard82 Participant1: from the top its the 5th row yeah83 Participant2: row 584 Participant2: .....85 Participant1: then it the 5th out of six86 Participant1: on that row87 Participant2: I HAVE gone over that switch ALREADY!!!!!!!!!!!!!!!!!!!!!!88 Participant2: where is your other damn switch!!!!!89 Participant1: only one switch u prat90 Participant1: listen and stop trying to b awkward91 Participant2: row 5, block 5?92 Participant1: its block 5 out of 6 yeah93 Participant1: ?94 Participant2: im doing that!95 Participant2: its not working


96 Participant1: it went away and tehn came back97 Participant2: there now?98 Participant1: you walked past it99 Participant2: there?100 Participant1: yay101 Participant2: I WAS RIGHT THERE102 Participant1: what did you do diff103 Participant2: nothing

Dyad 1. Trial 1.

At a finer grain, interlocutors’ turns are often oriented, right from theoutset, toward setting up local complementary structures between turnsproduced by both participants: “u go first, where do u need me to go?”(Dyad 2, Trial 1, Line 3).

1 Participant50 dude2 Participant40 Hello there!3 Participant50 u go first, where do u need me to go?4 Participant40 I have a gate to the last column5 Participant40 theres only one box there6 Participant50 open?7 Participant40 no8 Participant50 left, right or down?9 Participant40 if the left column is onr10 Participant40 *one11 Participant50 open?12 Participant40 then i want you to go to 713 Participant50 7 what?14 Participant50 this is gonna take a while15 Participant40 lol dude listen16 Participant40 right take note17 Participant50 ok18 Participant40 top left is 1,119 Participant50 ok20 Participant50 so like chess21 Participant40 bottom right is 7,722 Participant40 yes23 Participant50 ok24 Participant50 so where do u need me?25 Participant40 i want you to go to 7,326 Participant40 7 is x axis

20 Gregory J. Mills

27 Participant40 3 is y-axis28 Participant50 open??29 Participant40 got it?30 Participant50 yup31 Participant40 i am there32 Participant50 ok next?33 Participant40 i am in my goal34 Participant40 woohoo35 Participant40 where do i go?36 Participant50 ok i need u to go to..1,737 Participant50 goal!38 Participant40 open sesame?

Dyad 2. Trial 1.

At low levels of coordination, the interaction is replete with turns thatset up local complementary structures between turns produced by bothparticipants, e.g. “because then I let you first then I’ll follow” (Dyad 6,Trial 1, Line 8).

1 Participant26: oi2 Participant25: my gate is closed now, is urs open3 Participant25: i dont get it4 Participant26: is ur gate b4 that?5 Participant25: urs open now6 Participant25: ?7 Participant26: look in the sequence is you gate b4 e6?8 Participant26: coz den i let u fru 1st den ill follow9 Participant25: my gate is my gate is b/w c 3 and c410 Participant25: its closed now11 Participant25: is urs open12 Participant25: is urs open damit13 Participant26: yeh

Dyad 6. Trial 1.

A basic format for setting up complementary structures is “You do Xand I do Y” or “I’ll do X and then you do Y”, or “You do X so I can doY”. These differ from straightforward requests, as the requests include boththe first and second pair parts of an adjacency pair, as opposed to askingthe other simply to “do X”.

Note, also that this process is highly opportunistic. Interlocutors do


not necessarily have a clear notion of what they are doing, much less aglobal plan of how to proceed. Instead, interlocutors attempt to set up a(weak) local coupling of their respective contributions: See, for example,Dyad1, Trial 1, Line 103 who manage to solve the maze without being fullycoordinated, answering “what did you do differently?” with “nothing”. Asimilar pattern occurs in Dyad 5, Trial 1:

1 Participant5 so go back to the switch2 Participant6 do weat u did b4 man3 Participant5 gey to the switch4 Participant6 do wat u did b45 Participant5 is it open6 Participant6 no still closed7 Participant6 do wat u did b48 Participant5 now?9 Participant6 no10 Participant5 tell me when

Dyad 5. Trial 1.

Here two participants are trying to resolve the procedural problem inthe task. Note in Lines 2, 4 and 7, Participant6 requesting “do what youdid before”. Here, Participant6 doesn’t even know which particular movein the game to request. More importantly, neither does the other partici-pant, Participant5, who asks in Line 8: “now?”. Intriguingly, this transcriptshows that both participants are orienting towards a nascent complemen-tary structure, consisting of two parts, neither of which is known to eitherparticipant.

Medium levels of procedural coordination: constructing a jointactivityAfter achieving basic coupling of their actions and turns, participants be-come progressively coordinated in the task, developing a stronger sense ofthe activity, i.e. the “synergy” [Fusaroli et al., 2013] between them. Partic-ipants negotiate the sequential implicativeness [Schegloff and Sacks, 1973]

of their contributions; that is, they negotiate what they do and don’t needto confirm, inform and request of each other, and also establish the “rel-evant junctures” [Schegloff, 2007, p. 223] in the unfolding activity whereparticular contributions need to be made.

There appear to be a whole range of communicative devices used by par-ticipants to cement a routine at specific temporal and sequential juncturesin an activity. These devices work by reference to a particular sequential

22 Gregory J. Mills

location within the activity as a whole. Put differently, the dyads treat theprocedural coordination problems they encounter, not as one-off coordina-tion problems, but as a recurrent problem which occurs on each occasionof the activity. Consider Dyad 4, Trial 8, Line 6, which is from a mediumcoordinated pair.

1 Participant1 5,42 Participant1 6,63 Participant2 3,1; 3,6; 5,24 Participant2 Give me all urs5 Participant1 That’s it6 Participant2 If I do nothing it means I cannot get there

Dyad 4. Trial 8.

Here ParticipantB tells participantA “If I do nothing it means I cannotget there”. Note that ParticipantB does not simply say “I cannot get there”.This transcript is especially interesting: First, it demonstrates the explicitintroduction of a complementary convention in which silence and inactionbecome conventionalized as a “move in the activity”, within an adjacencypair whose first pair part is the ABSENCE of action. Second, convention-alization is achieved by giving a pre-emptive form of “accountable absence”[Schegloff, 2007], suggesting that there is a normative dimension to this con-vention. Third, this turn is introduced by reference to the next iteration ofthe activity, since by this stage, this dyad only gives each other the loca-tions of their switches once per trial. Finally, this example underscores thefundamental importance of looking at multiple occasions (i.e. rehearsals) ofthe same activity, as one could only know whether silence had an interactivefunction in a game by looking at multiple prior occasions of the activity,where the convention was introduced. This also suggests that the conver-sation analytic “next turn proof procedure” [Heritage, 1984] would need tobe extended to a “prior occasion of the activity proof procedure”.

Other examples of establishing junctures by reference to the activity asa whole include constructions such as “Next time do X”, instead of simplysaying “do X”. Note that this presupposes that participants are alreadysufficiently coordinated in the activity to know what the “next time” willbe. Similarly participants use “now you need to do X” (e.g. Dyad 3, Trial3, Line 19).

1 Participant9: ok wait2 Participant9: say all your switches3 Participant10: i have a one switch


4 Participant10: it is the third row from the top5 Participant9: cant reach it6 Participant10: ok where is urs7 Participant9: mine : 1 - second colmn - bottom8 Participant10: now ?9 Participant9: on the right10 Participant9: no11 Participant10: second column ?12 Participant9: second colomn from right13 Participant9: very bottom14 Participant10: cant reach it15 Participant9: ok. the very top16 Participant9: on the lesft col17 Participant9: good18 Participant9: at the goal19 Participant10: now u need to go to the third row20 Participant10: u had it21 Participant9: what colmn?22 Participant10: last column from the right23 Participant10: third row24 Participant9: now?25 Participant10: no26 Participant10: thir row from the top27 Participant9: 1st colm on the right?!28 Participant10: last column29 Participant9: now30 Participant10: yes

Dyad 3. Trial 3.

Other more explicit forms of anchoring an utterance in a particular se-quential location in the activity include saying “at this stage we do X” or“here we do X” instead of simply making the request. Uses of these de-vices tend to occur toward the middle of the task, once a certain level ofcoordination has been achieved.

Importantly, once these coordination problems have been articulated,addressed, resolved and performed successfully, they no longer need to bementioned explicitly. And since the solutions are cemented in place, theycan be elided, allowing more tacit and contracted procedural expressions,yielding an activity with greater progressivity.

24 Gregory J. Mills

High levels of coordinationOnce the participants have established successful coordination, as is appar-ent from the transcripts already seen, coordination continues to develop.Note how in Dyad 1, Trial 11, participants’ turns are shorter and also takefewer exchanges to solve than the same Dyad in Trial 5 and Trial 1.

1 Participant2: switches...2 Participant1: 1,1 and 7,23 Participant1: that it4 Participant2: this ones more complicated5 Participant2: where are ur gates6 Participant1: there at the barrier of 4,17 Participant1: and 5,18 Participant2: is that it?9 Participant2: and ur goal10 Participant1: and 6,4 and 7,211 Participant1: my goal is 2,112 Participant2: can you get to 3,1 and stay there please13 Participant2: and now to ur goal

Dyad 1. Trial 5.

1 Participant1: 2,2 4,1 6,22 Participant2: go to 2,1 please3 Participant1: ther4 Participant2: and now to ur goal5 Participant1: ther6 Participant2: shit7 Participant2: back to 2,18 Participant1: ther9 Participant2: now to the goal

Dyad 1. Trial 11.

Also, while in the earlier dialogues (e.g. Dyad 1, Trial 5), participantsprovide each other with more overt forms of language to signal differentlevels of (mis)understanding, via clarification requests or acknowledgments,in Dyad 1, Trial 11 participants are sufficiently coordinated that they cansimply perform the “next relevant contribution” (Clark, 1996). Recall thatknowing the “next relevant contribution” implies coordination on comple-mentary structures.


Importantly, although each maze can be solved in a few steps, each pairestablishes (often subtly) different complementary sequential structures forsignaling in the task. See for example Dyad 4, Trial 10 which shows anentire transcript for solving a maze.

1 Participant1 3,7 5,42 Participant2 5,1 7,63 Participant2 open4 Participant1 trapped5 Participant2 open6 Participant2 home

Dyad 4. Trial 10.

The members of this ensemble are able to signal to each other how toprogress through the maze by using three words “open”, “trapped” and“home”. Importantly, here the word “open” does not simply mean thatthe gate is open. Here, “open” means something akin to:

“My gate has opened but I can’t get to my goal, if you are able to, pleasego to your goal and then tell me when you’re on it, otherwise tell me and Iwill then go to a switch that you just mentioned and wait there either untilone of my gates opens or until you tell me to go to the goal.”

Further, since this is the same Dyad described in the preceding sectionwhich gave a pre-emptive form of “accountable absence” (Dyad 4, Trial 8,Line 6), it would also have an additional meaning of “..but if I do nothingit means I cannot get there”.

Importantly all of this is achieved with a single word, “open” withoutan overt request, that simply articulates a particular state in the activity.Note that “trapped” and “home” have similarly rich sequentially implicativefunctions. Even utterances as innocuous as “me now” in Dyad 4, Trial 12are highly dyad-specific:

1 ParticipantA 3,1..5,62 ParticipantB 4,2 5,43 ParticipantB open4 ParticipantB home5 ParticipantA Me now6 ParticipantA Home

Dyad 4. Trial 12.

Does “me now” mean “now it’s my turn to speak”, “it’s my turn to

26 Gregory J. Mills

guide you through a gate”, “my turn to go onto a switch” or “my turn towait”?. Establishing this would require looking at the prior, less ellipticalinstances to see which coordination problems were encountered and subse-quently elided.

In the most coordinated pairs, each contribution effectively becomes a“move in an activity” with a multiplicity of retrospective and prospectivesequentially implicative functions, which become progressively enriched asthe task progresses, resulting in an extremely rich procedural semantics forcoordinating the activity. As illustrated in the example above, each con-tribution can perform multiple simultaneous speech acts, i.e. participantsare simultaneously posing and answering multiple “Questions Under Discus-sion” [Ginzburg, 2012], often with a single word (see also Gregoromichelakiet al., 2011, 2013). Note that although many questions are being asked(and answered) with each turn, the participants do not overtly ask ques-tions. The turns do not have the canonical sentence structure of a question,and there is no question mark at the end of each turn. In much the sameway that different dyads might mean different things by “row” or “square”,the same words used by different dyads to coordinate procedurally can alsoacquire different procedural functions that depend on the specific conversa-tion history and coordination problems encountered during the interaction

A further hallmark of this coordinated stage is that in addition to par-ticipants’ contributions becoming increasingly sequentially implicative (i.e.developing richer and more precise prospective and retrospective functions),allowing them to make highly elliptical requests and commands for travers-ing the maze, the task actions themselves (i.e. the opening and closing ofgates by standing on switches) also acquire sequentially implicative func-tions. In the initial stages of the maze game, opening of the gates is achievedtypically after performing a request, e.g. preceded with a turn similar to“can you go to my switch on the top row, 5th square?”. Within speech acttheory [Austin, 1975; Searle, 1969], this would be conceived of as a perlocu-tionary effect, or as the second pair part of a “projective pair” [Clark andKrych, 2004]. However, towards the end of the task, participants use theopening and closing of gates to signal to each other whether or not they areable to progress unproblematically through the maze (as suggested by thestatement “If I do nothing it means I cannot get there” in Dyad 4, Trial 8),and can be used as the first pair part of a complementary structure, thatis, the opening of the other person’s gates becomes an illocutionary act,instead of a perlocutionary effect.


2.3 Developing procedures during the interaction

How is it that interlocutors are able to achieve this level of coordination?Comparing the dialogue from the early trials with the late trials is all themore striking, given that all the mazes are randomly generated and there-fore, on average, equally difficult. It appears that once a basic level of co-ordination has been achieved which allows participants to solve the mazessuccessfully, each coordination problem, once articulated and resolved, nolonger needs to be overtly mentioned, consequently enriching the sequentialimplicativeness [Schegloff and Sacks, 1973] of each turn. The proceduralfunction disappears from the conversational surface, becoming “transpar-ently absent”.

The question then emerges as to the nature of these solutions to proce-dural coordination problems. In what follows we argue that the patterns ofinteraction observed in the most coordinated pairs provide evidence that inaddition to establishing conventions for individuating locations in the maze,participants are also establishing procedural conventions for managing theprocedural coordination problems in the task. In the most coordinateddyads, this culminates in conventions that acquire normative status. Anal-ogously to Brennan and Clark’s 1996 conceptual pacts, interlocutors arealso establishing “procedural pacts” with each other.

Dyad 4, Trial 8 already showed participants explicitly establishing a con-ventional meaning for silence at a key juncture in the dialogue. Furtherevidence for procedural pacts can be seen in pairs that explicitly establishconventions at high levels of coordination. Consider Dyad 8, Trial 6.

1 Participant16 1.2 and 3.7 and 7.62 Participant17 5,1 and 1,43 Participant17 can u reach mine4 Participant17 ??5 Participant16 ATG-at your goal6 Participant16 ATG?7 Participant17 huh8 Participant16 AYG-at your goal9 Participant17 no im not i need u to open my gate10 Participant17 lol11 Participant17 ok u ATG12 Participant16 Lol

Dyad 8. Trial 6.

Note, in Line 5, Participant16 explicitly introduces “ATG” to mean “at

28 Gregory J. Mills

[your] goal”, followed by a respecification of “ATG” as a question, effectivelyasking “are you at your goal?” which is subsequently recast as “AYG”. Fourtrials later (Dyad 8, Trial 10), the dyad has developed a much richer systemof moves:

1 Participant16 lol how many more?2 Participant17 4,1 3,53 Participant17 i dunno lol4 Participant17 im havin fun5 Participant16 2.5 and 3.2 and 6.76 Participant17 :)7 Participant17 there ya go8 Participant16 wheres yours?9 Participant16 through my gate10 Participant16 AMG lol11 Participant17 4,1 and 3,512 Participant16 AYS13 Participant17 nope u sure14 Participant17 GC15 Participant17 GO16 Participant16 AYS17 Participant17 AMG

Dyad 8. Trial 10.

Here, the ensemble uses “AMG” to mean “At my goal” and “AYS” inLine 12 to ask the question “Are you at your switch?”, as well as “GC” inLine 14 to mean “gates are clear”. Of central interest here is that simplechecks “AYS”, as well as simple referential statements “AMG” about beingat a particular location both become integrated into the same system of“moves in an activity”.

It is perhaps surprising that in a task that can be solved in a couple ofsteps (listing switches, A going on another’s switch, B going on A’s switch,going to the goal) that participants develop (subtly) different ways, notsimply of solving the high-level task structure, but also of signaling to eachother how to coordinate on how the activity unfolds. For example saying“AYS”, “GC”, and the less contracted, but still highly elliptical “open?”,“goal?”, “exit?”, “done”, “blocked”. All of these have different sequentialimports that will depend on the different interactional histories of eachensemble.

A simple analogue of this is the use of “check” in chess. The utter-ance “check” is itself a move in the game, as much as moving the pieces of


chess. Similarly in the maze game uttering “AYS” and opening or closingthe other’s gates both constitute moves in the activity. Note that despiteinvolving the contraction of the name of a referent, this form of contrac-tion seen in “AYS” is separate from the contraction observed in the spatialdescriptions of the semantic model. Moreover, as they are introduced explic-itly as a convention, and then subsequently adopted by the others, and arethen used successfully, this suggests that these are stronger, more normativeconventions.

2.4 Accounting for procedural coordination

The more coordinated the dialogues, the less structure there is on the con-versational surface; the procedural expressions are highly elliptical, and theinteractive exchanges are shorter. Further complicating matters is that pro-cedural coordination is underpinned by complementary structures; a clearexample of this is Dyad 4, Trial 10 which has reached ceiling levels of coordi-nation. In this interaction, which solves the maze in an interactive sequenceconsisting simply of “open”, “trapped”, “home”, on each subsequent turn,participants are doing something different and complementary, as they carryout the “next relevant contribution” (Clark, 1996). So, although by the endof the task the participants have converged on a very small set of routines,their successful coordinated use involves no immediate, local, turn-by-turnrepetition of structure that is apparent on the conversational surface. Fromthis perspective then, once high levels of coordination have been achieved,any increase in local, turn-by-turn repetition of structure could be indicativeof a decrease in forward momentum of the interaction, and consequently beindicative of a decrease in coordination.

This raises the bar for accounting how coordination is established, as anyaccount must show how this systematic divergence develops locally betweenindividual turns, over the interaction, while the more global structures be-come elided. However, here too it is unclear how to capture this process.The patterns observed in the maze game suggest that this process is seam-less, and it is unclear how to capture the progressive enrichment of theroutines on each new occasion of use. The simple surface form obscuresmany layers from previous iterations, and consequently the same word, e.g.“open” has a very different meaning at the end of the task than at the start.Thus long-term repetition between the start and end of the interaction isdifferent from short term repetition of structure at low vs. high levels ofcoordination. It is also unclear how to capture how the moves become “partof the activity” proper.

Finally, as shown in Dyad 9, Trial 10, in highly coordinated dialogue,the participants are able to interleave their communication about the joint

30 Gregory J. Mills

activity with other, everyday conversations.

1 Participant92 we did it again2 Participant80 ooops we did it again3 Participant92 that means less nandos from todays earnings4 Participant80 a3, d55 Participant92 only 14 quid maan6 Participant80 damn i want more nandos !!!!!!7 Participant92 we shud come back in diguise8 Participant80 remember that chick from nando’s9 Participant80 that really cute one,10 Participant92 b7 and g511 Participant80 a3 or d5 mate12 Participant92 haha13 Participant92 yea14 Participant92 well15 Participant92 go fer her mate16 Participant92 she needs u17 Participant92 :D18 Participant80 I cant get to either19 Participant80 the X .. or the chick :P20 Participant80 g5 ?21 Participant92 chick maan22 Participant92 X aint gonna23 Participant92 get u nething24 Participant92 dude25 Participant80 are u on X yet ?26 Participant92 if we make it one hour and one min27 Participant92 we get paid for the extra one min effort lol28 Participant92 so go for it29 Participant80 we’re already been here for over an hour30 Participant92 hahaha31 Participant80 are you on X ?

Dyad 9. Trial 10.

Frequently the skilled, highly coordinated pairs, who are responsive toeach other’s informational needs in the task often do so in a playful manner,and once they’ve established routines, play with the expectations for effect(e.g. using “open sesame” to ask their interlocutor to open their gates). Ofcourse, being able to do this using different terminology is evidence of higher(not lower) levels of skill. Further, as in Dyad 9, Trial 10, participants often


play with the boundary between the task and the conversation (e.g. Line19).

Without knowing how difficult it is to establish coordination in the mazetask, this excerpt seems unremarkable. Notice here how the activity ofsolving mazes has receded into the background, how very few contributionsare actually concerned with coordinating in the maze game, and how thesecontributions are seamlessly interwoven with an unconstrained, everyday,playful conversation. The transitions between conversing and solving themaze are unmarked. There are no overt signals for switching between theconversation and the activity of solving the maze. Compare the ease withwhich they are solving the maze with Dyad 1, Trial 1. There is barelyany mention of what each turn is doing, the procedural function of eachturn has disappeared from the interactional surface. The participants aresimultaneously talking about the maze, and about buying and food andflirting with a worker, playing, in counterpoint, with the meanings of bothactivities.

3 Conclusion

Although this constrained task presents participants with a very simple co-ordination problem (as evidenced by the brevity of dialogues in the laterstages), the interactional patterns observed are highly complex. Ensemblesrapidly establish idiosyncratic signals used for (inter alia) initiating andsuspending actions, requesting, confirming and clarifying. The difficultiesfaced by participants initially, at low levels of coordination, suggest thatthese signals are not taken from a set of (meta)communicative devices andused unproblematically. Instead, the intentions associated with these signalsare underdetermined, ambiguous and must be adapted, interactively by theensemble to the local context. Although initially, participants encounter dif-ficulty, they rapidly conventionalize highly efficient and tacit routines witheach other, attaining “masterful” levels of coordination [Schober, 2006] and“flow” [Csikszentmihalyi, 1988; Csikszentmihalyi, 1992]. Crucially, the co-ordination problems and solutions they engender are constitutive of theintentions beind these signals, imbuing them with a rich procedural seman-tics, (see also Gregoromichelaki et al., 2011).

Under this light, consider the cues used to coordinate in musical ensem-bles. The findings discussed here strongly suggest that during rehearsalsoccurring over many weeks, the cues used to coordinate will necessarilyacquire ensemble-specific, complex interactive functions that cannot be re-duced to “coordinative smoothing” [Vesper et al., 2010]. Further, the ra-pidity with which participants develop routines suggests that even duringa single performance, ensembles might establish ad-hoc, ensemble-specific

32 Gregory J. Mills

routines. If we are to understand the cues that are used in (musical) jointaction, we must give an account of how these cues acquire communicativemeaning, which necessarily involves investigating multiple prior occasions(ie during rehearsals).

Further, although the experiment described here only focuses on or-chestrating the simplest kind of ensemble interactions, namely interactionswithin the same dyad, most ensemble interactions occur in groups largerthan two. This introduces further multiparty coordination problems, in-cluding the spontaneous forming and reforming of “sub” ensembles and theeffects this has on deriving and displaying joint intentions and plans. Initialwork [Mills, 2012] shows that the solutions to these coordination problemsare also ensemble-specific, idiosyncratic, and arise out of the interaction viathe work that participants perform during a trial-and-error process.

Finally, the picture of joint action that emerges is that we are always cre-ating novel, ad-hoc solutions to novel coordination problems; these are notfixed routines (as the conversation analytic literature sometimes seems toimply). Instead, we are constantly improvising, whether musically [Benson,2003], conversationally, or in our actions. Ensembles progressively refineand adjust their procedures and meaning-making over multiple occasions.Even in the special case of classical, orchestral music there is no final perfor-mance. We never cease rehearsing for the next rehearsal. The open-endednature of interaction precludes this.

BIBLIOGRAPHY[Alterman and Garland, 2001] Richard Alterman and Andrew Garland. Convention in

joint activity. Cognitive Science, 25(4):611–657, 2001.[Arundale, 2013] Robert B Arundale. Conceptualizing interactionin interpersonal prag-

matics: Implications for understanding and research. Journal of Pragmatics, 2013.[Austin, 1975] John Langshaw Austin. How to do things with words, volume 1955.

Oxford university press, 1975.[Benson, 2003] Bruce Ellis Benson. The improvisation of musical dialogue: a phe-

nomenology of music. Cambridge University Press, 2003.[Brennan and Clark, 1996] S E Brennan and H H Clark. Conceptual pacts and lexical

choice in conversation. Journal of Experimental Psychology: Learning, Memory, andCognition, 22:1482–1493, 1996.

[Clark and Brennan, 1991] Herbert H Clark and Susan E Brennan. Grounding in com-munication. Perspectives on socially shared cognition, 13(1991):127–149, 1991.

[Clark and Krych, 2004] H. H. Clark and M. A. Krych. Speaking while monitoringaddressees for understanding. Journal of Memory and Language, 50:62–81, 2004.

[Clark, 1996] H H Clark. Using Language. Cambridge University Press, 1996.[Cohen and Levesque, 1990] Philip R Cohen and Hector J Levesque. Intention is choice

with commitment. Artificial intelligence, 42(2):213–261, 1990.[Cooper and Ranta, 2008] Robin Cooper and Aarne Ranta. Natural languages as col-

lections of resources. Language in Flux: Relating Dialogue Coordination to LanguageVariation, Change and Evolution. College Publications, London, 2008.

[Cross, 2008] Ian Cross. Musicality and the human capacity for culture. Musicae Sci-entiae, 12(1 suppl):147–167, 2008.


[Csikszentmihalyi, 1988] I Csikszentmihalyi. The flow experience and its significance forhuman psychology. 1988.

[Csikszentmihalyi, 1992] I Csikszentmihalyi. Optimal experience: Psychological studiesof flow in consciousness. Cambridge University Press, 1992.

[De Jaegher et al., 2010] Hanne De Jaegher, Ezequiel Di Paolo, and Shaun Gallagher.Can social interaction constitute social cognition? Trends in cognitive sciences,14(10):441–447, 2010.

[De Ruiter et al., 2006] Jan Peter De Ruiter, Holger Mitterer, and Nick J Enfield. Pro-jecting the end of a speaker’s turn: A cognitive cornerstone of conversation. Language,pages 515–535, 2006.

[Derrida, 1977] Jacques Derrida. Limited inc. Northwestern Univ Press, 1977.[Duffy and Healey, 2012] Sam Duffy and Patrick GT Healey. Spatial co-ordination in

music tuition. In Proceedings of the 34th Annual Meeting of the Cognitive ScienceSociety, Sapporo, Japan, 2012.

[Egermann et al., 2013] Hauke Egermann, Marcus T Pearce, Geraint A Wiggins, andStephen McAdams. Probabilistic models of expectation violation predict psychophys-iological emotional responses to live concert music. Cognitive, Affective, & BehavioralNeuroscience, pages 1–21, 2013.

[Fay and Ellison, 2013] Nicolas Fay and T Mark Ellison. The cultural evolution of hu-man communication systems in different sized populations: Usability trumps learn-ability. PloS one, 8(8):e71781, 2013.

[Fay et al., 2008] Nicolas Fay, Simon Garrod, and Leo Roberts. The fitness and func-tionality of culturally evolved communication systems. Philosophical Transactions ofthe Royal Society B: Biological Sciences, 363(1509):3553–3561, 2008.

[Fusaroli et al., 2012] Riccardo Fusaroli, Bahador Bahrami, Karsten Olsen, AndreasRoepstorff, Geraint Rees, Chris Frith, and Kristian Tylen. Coming to terms quan-tifying the benefits of linguistic coordination. Psychological science, 23(8):931–939,2012.

[Fusaroli et al., 2013] Riccardo Fusaroli, Joanna Raczaszek-Leonardi, and KristianTylen. Dialog as interpersonal synergy. New Ideas in Psychology, 2013.

[Garrod and Anderson, 1987] S. Garrod and A. Anderson. Saying what you mean indialogue: A study in conceptual and semantic co-ordination. Cognition, 27:181–218,1987.

[Garrod and Doherty, 1994] Simon Garrod and G. Doherty. Conversation, coordinationand convention: an empirical investigation of how groups establish linguistic conven-tions. Cognition, 53:181–215, 1994.

[Giles et al., 1991] Howard Giles, Nikolas Coupland, and IUSTINE COUPLAND. 1.accommodation theory: Communication, context, and. Contexts of accommodation:Developments in applied sociolinguistics, page 1, 1991.

[Ginzburg, 2012] Jonathan Ginzburg. The interactive stance. Oxford University Press,2012.

[Goodwin, 1986] C. Goodwin. Between and within: Alternative sequential treatmentsof continuers and assessments. Human Studies, 9:205–217, 1986.

[Gregoromichelaki et al., 2011] Eleni Gregoromichelaki, Ruth Kempson, Matt Purver,Gregory James Mills, Ronnie Cann, Wilfried Meyer-Viol, and Patrick GT Healey. In-crementality and intention-recognition in utterance processing. Dialogue & Discourse,2(1):199–233, 2011.

[Gregoromichelaki et al., 2013] E. Gregoromichelaki, R. Kempson, C. Howes, and A. Es-hghi. Alignment in Communication: Towards a New Theory of Communication.,chapter On making syntax dynamic: The challenge of compound utterances and thearchitecture of the grammar. John Benjamins, 2013.

[Grosz and Sidner, 1986] Barbara J Grosz and Candace L Sidner. Attention, intentions,and the structure of discourse. Computational linguistics, 12(3):175–204, 1986.

34 Gregory J. Mills

[Gumm et al., 2011] Alan J Gumm, Sharyn L Battersby, Kathryn L Simon, and An-drew E Shankles. The identification of conductor-distinguished functions of conduct-ing. Research and Issues in Music Education, 9(1), 2011.

[Haugh, 2008] Michael Haugh. Intention in pragmatics. Intercultural Pragmatics,5(2):99–110, 2008.

[Haugh, 2012] Michael Haugh. Speaker meaning and accountability in interaction. Jour-nal of Pragmatics, 2012.

[Healey and Mills, 2006] P. G. T. Healey and G. J. Mills. Participation, precedence andco-ordination. In Proceedings of the 28th Annual Conference of the Cognitive ScienceSociety, Vancouver, Canada, 2006.

[Healey, 2004] Patrick GT Healey. Dialogue in the degenerate case? Behavioral andBrain Sciences, 27(02):201–201, 2004.

[Heritage, 1984] John Heritage. Garfinkel and ethnomethodology. Polity Press Cam-bridge, 1984.

[Jackendoff and Lerdahl, 2006] Ray Jackendoff and Fred Lerdahl. The capacity for mu-sic: What is it, and whats special about it? Cognition, 100(1):33–72, 2006.

[Keller, 2008] Peter E Keller. Joint action in music performance. Emerging Communi-cation, 10:205, 2008.

[Keller, 2012] Peter E Keller. Rhythm and time in music epitomize the temporal dy-namics of human communicative behavior: The broad implications of londons trinity.2012.

[Kendon, 1990] Adam Kendon. Conducting interaction: Patterns of behavior in focusedencounters, volume 7. CUP Archive, 1990.

[Knoblich and Jordan, 2003] Gunther Knoblich and Jerome Scott Jordan. Action coor-dination in groups and individuals: learning anticipatory control. Journal of Experi-mental Psychology: Learning, Memory, and Cognition, 29(5):1006, 2003.

[Kuhlen and Brennan, 2013] Anna K Kuhlen and Susan E Brennan. Language in dia-logue: when confederates might be hazardous to your data. Psychonomic bulletin &review, 20(1):54–72, 2013.

[Lerdahl and Jackendoff, 1983] Fred A Lerdahl and Ray S Jackendoff. A generativetheory of tonal music. The MIT Press, 1983.

[Levinson, 1983] S C Levinson. Pragmatics. Cambridge University Press, Cambridge,1983.

[Louwerse et al., 2012] Max M Louwerse, Rick Dale, Ellen G Bard, and Patrick Jeu-niaux. Behavior matching in multimodal communication is synchronized. Cognitivescience, 36(8):1404–1426, 2012.

[Malhotra, 1981] Valerie Ann Malhotra. The social accomplishment of music in a sym-phony orchestra: A phenomenological analysis. Qualitative Sociology, 4(2):102–125,1981.

[Mills and Gregoromichelaki, 2010] Gregory J Mills and Eleni Gregoromichelaki. Estab-lishing coherence in dialogue: sequentiality, intentions and negotiation. Proceedingsof the 14th SemDial, Pozdial, 2010.

[Mills and Healey, 2006] Gregory J. Mills and P. G. T. Healey. Clarifying spatial de-scriptions: Local and global effects on semantic co-ordination. In Proceedings of the10th Workshop on the Semantics and Pragmatics of Dialogue, 2006.

[Mills and Healey, 2008] Gregory J Mills and Patrick GT Healey. Semantic negotiationin dialogue: the mechanisms of alignment. In Proceedings of the 9th SIGdial Workshopon Discourse and Dialogue, pages 46–53. Association for Computational Linguistics,2008.

[Mills and Healey, 2013 in prep] G. J. Mills and P. G. T. Healey. A dialogue experimen-tation toolkit. in prep, 2013, in prep.

[Mills, 2007] Gregory J. Mills. Semantic co-ordination in dialogue: the role of directinteraction. PhD Thesis, Queen Mary University of London, 2007.


[Mills, 2011] Gregory J. Mills. The emergence of procedural conventions in dialogue.In Proceedings of the 33rd annual conference of the Cognitive Science Society, pages471–476, 2011.

[Mills, 2012] G. J. Mills. Making and breaking procedural conventions in dialogue:partner-specific effects. In Annual meeting of the cognitive science society, CgSci,Japan, 2012.

[Mills, 2013a] G J. Mills. Dialogue in joint activity: coordinating on referring intentionsand plans. In Joint Action Meeting (JAM). Berlin, July 26-29, 2013.

[Mills, 2013b] Gregory J. Mills. Dialogue in joint activity: complementarity, convergenceand conventionalization. New Ideas in Psychology, in press, 2013.

[Papousek, 1996] Hanus Papousek. Musicality in infancy research: biological and cul-tural origins of early musicality. Musical beginnings, pages 37–55, 1996.

[Parton and Edwards, 2009] Katharine Parton and Guy Edwards. Features of conductorgesture: Towards a framework for analysis within interaction. In The Second Inter-national Conference on Music Communication Science, 3-4 December 2009, 2009.

[Phillips-Silver and Keller, 2012] Jessica Phillips-Silver and Peter E Keller. Searchingfor roots of entrainment and joint action in early musical interactions. Frontiers inhuman neuroscience, 6, 2012.

[Pickering and Garrod, 2004] M J Pickering and S Garrod. Towards a mechanistic psy-chology of dialogue. Behavioural and Brain Sciences, 27(2):169–190, 2004.

[Pickering and Garrod, 2013] Martin J Pickering and Simon Garrod. How tightly areproduction and comprehension interwoven? Frontiers in psychology, 4, 2013.

[Poesio and Rieser, 2010] Massimo Poesio and Hannes Rieser. Completions, coordina-tion, and alignment in dialogue. Dialogue and Discourse, 1(1):1–89, 2010.

[Poesio and Traum, 1997] Massimo Poesio and David R Traum. Conversational actionsand discourse situations. Computational intelligence, 13(3):309–347, 1997.

[Riley et al., 2011] Michael A Riley, Michael J Richardson, Kevin Shockley, andVeronica C Ramenzoni. Interpersonal synergies. Frontiers in psychology, 2, 2011.

[Sacks, 1995] H. Sacks. Lectures on Conversation. Blackwell Publishing, 1995.[Sawyer, 1996] R Keith Sawyer. The semiotics of improvisation: The pragmatics of

musical and verbal performance. Semiotica, 108(3-4):269–306, 1996.[Sawyer, 2006] R Keith Sawyer. Group creativity: Musical performance and collabora-

tion. Psychology of Music, 34(2):148–165, 2006.[Scaife and Bruner, 1975] Michael Scaife and Jerome S Bruner. The capacity for joint

visual attention in the infant. Nature, 1975.[Schank and Abelson, 1977] Roger C Schank and Robert P Abelson. Scripts, plans,

goals, and understanding: An inquiry into human knowledge structures. PsychologyPress, 1977.

[Schegloff and Sacks, 1973] Emanuel A Schegloff and Harvey Sacks. Opening up clos-ings. Semiotica, 8(4):289–327, 1973.

[Schegloff, 1998] Emanuel A Schegloff. Body torque. Social Research, pages 535–596,1998.

[Schegloff, 2007] Emanuel A. Schegloff. Sequence Organization in Interaction: Volume1: A Primer in Conversation Analysis. Cambridge University Press,, January 2007.

[Schelling, 1980] Thomas C Schelling. The strategy of Conflict. Harvard university press,1980.

[Schober and Spiro, 2013] M. F. Schober and N. Spiro. How much do jazz players shareunderstanding of their performance? a case study. In International Symposium onPerformance Science, 2013.

[Schober, 2006] Michael F Schober. Virtual environments for creative work in collabo-rative music-making. Virtual Reality, 10(2):85–94, 2006.

[Schutz, 1964] Alfred Schutz. Collected Papers...: Studies in Social Theory. Edited andIntroduced by Arvid Brodersen. M. Nijhoff, 1964.

[Schwartz, 1995] Daniel L Schwartz. The emergence of abstract representations in dyadproblem solving. The Journal of the Learning Sciences, 4(3):321–354, 1995.

36 Gregory J. Mills

[Searle, 1969] John R Searle. Speech acts: An essay in the philosophy of language,volume 626. Cambridge university press, 1969.

[Semin and Cacioppo, 2008] Gun R Semin and John T Cacioppo. Grounding socialcognition: Synchronization, coordination, and co-regulation. 2008.

[Semin et al., 2012] Gun R Semin, Margarida V Garrido, and TA Palma. Socially situ-ated cognition: Recasting social cognition as an emergent phenomenon. Sage handbookof social cognition, pages 138–165, 2012.

[Suchman, 2007] Lucy Suchman. Human-machine reconfigurations: Plans and situatedactions. Cambridge University Press, 2007.

[Vesper et al., 2010] Cordula Vesper, Stephen Butterfill, Gunther Knoblich, and NatalieSebanz. A minimal architecture for joint action. Neural Networks, 23(8):998–1003,2010.

[Weeks, 1996] Peter Weeks. A rehearsal of a beethoven passage: An analysis of correctiontalk. Research on Language and Social Interaction, 29(3):247–290, 1996.

Gregory J MillsUniversity of Edinburgh, [email protected]

orchestrating ensembles in interactionhomepages.inf.ed.ac.uk/gmills/mills2013_ensembles.pdf ·...

Documents