The emergence and decay of multilogue.
Self-regulation of a scholarly mailinglist

Abstract

This paper explores the self-organizing nature of a mailinglist activity system that has been sustained for more than ten years with a dynamically stable subscriber community. The emergence and decay of cascades of spontaneously unfolding "multilogue" (Shank, 1993) is displayed and analyzed by means of link maps of intermessage references. The paper spotlights the inverse power law patterns in the distribution of messages over participants and the "lifetime" of messages in the system, and reflects on some episodes in the local history where efforts have been made towards more planned activity.

Introduction

In this article, which is built upon substantially longitudinal material from one particular setting, the Xlists, I will explore the nature of a scholarly mailinglist as a self-regulating activity system, analyzing the dynamics of spontaneous emergence and decay of multilogue. The patterns of rapidly developing coordination between distant players, converging on particular objects of joint attention, and the following phase of divergence and decoupling are examined by means of empirical material from the cyber-archaeology (Jones, 1997) of the Xlist archives, notably from three major multilogical episodes from different eras in list history. The resulting link maps offer resources for a structural analysis of the emergence and decay of multilogues. In addition the temporal separation of the three episodes provides an opportunity to explore the changes in temporal dynamics of multilogue as the mailflow on a scholarly mailinglist increases: the way coordination and remembering plays out under different conditions of mailflow intensity. The present paper is written in dialogue with the model suggested in the companion paper, where the mailinglist is regarded as a consumer-producer ecology with the available time of list participants as the constraining, renewable resource. Finally, the implications of the self-regulating nature of mailinglist ecologies have been studied in relation to one type of more planned activity on the Xlists: four case studies of episodes of joint reading.

The Xlists: a dynamically stable setting

  

active years	10	9	8	7	6	5	4	3	2	1
contributors	5	9	12	19	18	32	46	106	217	656

As Table 1. shows, a small number of contributors have been present in the setting for most of its existence, many others have been present for four or five years, larger numbers for two or three years, and the largest number contributing actively in one single year. The generations of contributors overlap, as Fig. 4 shows. The community changes constantly but never abruptly.

Self-regulation on a scholarly mailinglist

[T]here is a heterogeneous but constrained set of topics which comes up once or 
twice a year such that in some cases (zone of proximal development, context, 
learning/development/play, Vygotsky vs Piaget, genetic methodologies, activity 
centered instruction, etc.) have reappeared over a number of years. By itself this is 
no problem. The discussions are always interesting. But I get the nagging sense that 
it should be possible for the discussions of these key conceptions in socio-cultural-
historical-activity-centered theories to CUMULATE. 
(M.C. 92-10-01)

A number of folk have commented, myself included, that in those periods when the
discussion gets hot, as it has at least once in each of the subconferences over the 
last year, one begins to get a sense that a sort of Vygotskian "chaining" or 
"thinking in complexes" takes place. At first the conceptual flow is exciting, but at 
some point, one longs to lay out the messages side by side and get close to a 
simultaneous/paradigmatic summary of the flow, e.g., to create a product that might 
approximate what LSV calls true concepts. 
(M.C. 89-04-04)

The activity on a scholarly mailinglist is self-regulating in several senses. For one thing, participation is voluntary. In contrast to CMC in settings of institutionally regulated education, there is neither anybody with the formal authority to coerce Xlist participants into reading and writing Xlist mail, nor any formal credit for participating. In this sense mailinglist activity is regulated by self as opposed to being regulated by others. Then, in a mailinglist setting participants will not only be scattered in geographical space, but also in time: across time zones around the globe and in terms of access habits. The setting of Xlist activity is telematic and heterochronous and so the activity is self-regulating in the sense that the only coordination between participants is provided by the messages they circulate within the activity system, without either central or external control. It is also self-regulating in the sense that the written conversation flows between self-initiating participants. The only feedback cues given in text-only CMC are those of the arrival of messages and those given in the message text (Herring, 1999). Thus mailinglist activity is self-regulating in at least three senses: first in the social sense of being voluntary, with a distribution of authority throughout the community, second in the technical sense that any coordination of the activity is heterochronous and telematic, and third but not least in the sense of being a process of semiotic self-regulation, proceeding across the spatiotemporal gaps in the virtual setting (and with at least a certain amount of collective memory over the latest few generations of messages). Syverson, writing about the mailinglist as an ecology of composition (1994, 1999) found several indications of self-organizing criticality in her study of a heated episode on the XLCHC, and the dynamic and decentralized features of mailinglist multilogue, make it reasonable to continue exploring whether, and to what extent, the community and its activity can also be modelled as self-organizing system of mathematical complexity (Barowy, 1999). Unravelling mailinglist self-organization will scarcely provide instruments for harnessing the conversation, but an understanding of the contingencies might provide participants with an antidote to impending frustration.

Much of the Xlist mailflow is produced by participants responding to earlier messages - in this my research object is no different from the electronic fora studied by so many other researchers (e.g. Gruber, 1998; Galegher et al. 1998; Rafaeli & Sudweeks, 1996; Palme, 1989, 1995; Herring, 1999). This prevalent chaining of responses was what allowed Palme (1989, 1995) to construct a probabilistic model for calculating the lower size limit for a successful group (when the purpose is the exchange of experience). His model illustrates that if each message evokes, on average, more than one message in reply, then a chain reaction arises that keeps the activity going. Given a certain probability for each participant (excepting the sender) to reply to a message, Palme shows that group size must be above a certain threshold if activity within the group is to be sustained. This is a plausible explanation of the empirical observations on group size. However, while a simple probabilistic model like this can describe moments of expanding mailflow (or conversation that never starts), it is too simple for explaining why the chain reaction once started does not proceed exponentially over time (see companion paper). Some kind of self-regulating mechanism is called for in order to produce the typical oscillations of mailinglist activity, and it is reasonable to frame this mechanism in terms of the time expenditure of participants.

Time in mailinglist activity

On a mailinglist the balance between too much mail and too little is precarious (Syverson, 1999). Too far between messages and subscribers forget that a list exists - or at least there is nothing to respond to, so they don't. The result is an extinction of the mailflow on the list, a shrinkage of shared time and public space to nothing. But also the opposite case, where the mailflow on a list is too intensive and time demanding, runs the risk of overloading readers so that they do not have any time left for writing. An overload of mail is commonly given as a reason for dropping out of a list (Rojo, 1995) and the Xlists have their own share of subscribers asking for help to be removed because they are "drowning in print", or get "too much mail" while lacking "enough time to read it or do anything with it right now". On the other hand, it only takes a couple of days without mail over the Xlist channel before someone posts a message asking what has happened. A dynamically stable and successfully active list seems to settle into oscillation around a recognizable average mailflow, where the occasional references by list contributors to episodes of either "vast and dizzy silence" or a "fast and furious" "pace of conversation" are if not consensual or essential truths at least identifiable enough not to be challenged by co-participants. On the Xlists, where the average mailflow has increased substantially since the late 80s (see Fig.1) these perceptions have calibrated along with the growing traffic: What was referred to as a "fast and furious" pace in 1988 was a mailflow averaging 1.7 msgs/day over the preceding two weeks, which included five days without messages. In 1996 a period of three weeks with an average mailflow of 4.6 msgs/day, also including five days without mail, gave rise to several episodes of contributors worrying about the channel or commenting on the "eerie" quiet. Times and habits have been changing...

In addition to the heterochronicity resulting from distribution across time zones and differences in access and mailing habits, a server lag of an hour or more is also a standing feature of the Xlist setting. However, Xlist contributors do their best to make the spatiotemporal gaps in their shared setting invisible by addressing others as present in spirit if not in body, by explaining their absences from the shared virtual space due to travel or "end of term madness", by apologizing for "late" contributions, and other time-honoured practices for maintaining presence at a distance observed by researchers of pre-electronic letter writing (Decker, 1998; Mulkay, 1985). Thus, from the vantage point of a participant it makes sense to construe mailinglist activity as a continuity of its own, an internally coherent turn-taking sequence, produced in acts of reading and maintained in acts of responding.

Nevertheless, when looking at the activity system from the outside, the flow of messages on a mailinglist is evidently the resulting product of the aggregate behavior of many independent participants reacting in parallel to events within the system which is a hallmark of self-organization (see Dooley & Van de Ven, 1997, p. 19). The time available for subscribers to participate in the activity of reading and writing mailinglist contributions is a likely candidate for a simple self-regulating model: the more mail there is to read, the more there will be to respond to, but there will be a point where all the available time for mailinglist participation is consumed simply by reading the contributions of others. Then the next response will not be written - which in turn means that others will have one message less to read and respond to, and so may have time to produce a message of their own, etc. An episode of intensifying mailflow may thus lead to its own collapse, but if there is enough stability in the community of list participants the ensuing period of mail sparsity prepares the ground (the screen) for a second growth of multilogue. As suggested in the companion paper a self-regulating model of mailinglist activity may be achieved by regarding the mailinglist as an ecology with time as the limiting resource "consumed" by messages and "produced" by the readers/writers. As a scholarly mailinglist will typically be the channel not just for scholarly multilogues but also for networking activity in the scholarly community of its subscribers and for the maintenance of the communication channel itself (Ekeblad, 1998) there will always be a share of the mailflow that neither originates from the internal flow of events nor has any effect on the subsequent mailflow. Items of academic information value like conference calls and job announcements are usually floated into the mailstream according to external timetables and without spawning further activity over the mailinglist channel. The scattered input of academic and other information into the network will appear within the list at random intervals, being the output of processes external to the mailinglist. On the other hand, multilogical episodes - once they get started - will run their course as a process internal to the mailinglist ecology. They can thus be expected to follow a dynamics of self-organized criticality, which would result in contribution patterns much reminiscent of the avalanches caused by the single grain of sand dropped on to a sandpile in a state of criticality - a macro-level episode is generated by a single micro-level event which, because of coupling, cascades its effect across many interrelated entities (Dooley & Van de Ven, 1997, p. 16; Bak, 1996). If this is more than a superficial analogy, there are only very limited possibilities for predicting and controlling the self-regulated flow of activity on a scholarly mailinglist.

Three outstanding multilogues: Tools, Goals and Settings

The other two multilogues, the 1988 Tools Multilogue and the 1993 Goals Multilogue, were chosen from among the celebrated episodes of multilogue that have been remembered for years in the list community, and brought back by long-term participants to collective attention as legendary events. The Tools Multilogue of 1988 took place in the single-list phase of XLCHC, and was the first major archived Xlist discussion on the recurring theme of the mediating role of tools in human psychological activity. Just like the Settings Multlogue, this episode emerged from a single seed message: a posting in the second week of October by a Japanese visiting scholar, reflecting on the nature of tools in modern society, as an afterthought to a UCSD seminar. Like the seed message of the Settings Multlogue this posting brings fresh external material into the mailinglist ecology. The development of the Tools Multilogue is considerably slower across calendar time than the Settings Multilogue, corresponding to the fact that the nine years passing between the two episodes have brought with them an almost ninefold increase in the mailflow (see Fig.1.). The Tools Multilogue lasts through November and December, moving through three rounds, first introducing the theme of mediating tools in general, taking computers as one of the examples, the second round shifting the focus more towards computers as tools, and the third one moving into the related topic of the decontextualizing effects of computer mediation. There are all in all 74 postings over a period of 78 days, debating whether the power of computer tools is an illusion or a real benefit, celebrating the relation of the carpenter to his tools, and discussing the artifact mediated nature of human cognition.

The 1993 Goals Multilogue occurred in the complex situation of the middle years of Xlist history, when there were a number of subconferences with overlapping but far from identical subscriber collectives. Its topic is the activity-theoretical system of concepts centering around action-goal and activity-motive (originating with Leont'ev (1978, 1982)), which has been a regularly recurring object of multilogue on the Xlists. The main part of the evemt took place in November on the XACT list - the subconference designated to activity theory - but there is a complex pre-history to this episode spanning over other parts of the Xlist space, illustrating Syverson's observation (1994, 1999) of how hard it can be to draw the line on where a discussion topic first is brought into the mailinglist ecology of readers, writers and texts. First, there were weekly summaries of CHAT readings posted to XACT through October by graduate students following a UCSD seminar on Activity, Mind and Communication. While these postings did not evoke many responses, they did bring Leont'ev and the theme of goals and agency into the shared space of the mailstream. Then, in the third week of October, a couple of new subscribers to the XLCHC ask for information about collaborative learning, which results in a discussion routed into the XCLASS subconference, where the question about learning strategies, and about who sets the goal for learning, comes up. A couple of postings take up the theme of goals - or strategies - on the XACT in the end of October. One of them is from a contributor asking for a non-circular definition of 'goals', the other a cross-posting between XACT and XCLASS on the topic of goals and strategies. It takes a few days, into the beginning of November, before there is a third long message in response. It is with these three messages that the collective of participants is coordinated into a full-blown multilogical episode - it is from there that the sequence of responses starts expanding to a densely interrelated cluster, and it is these messages that are referred to later in the episode as the ones initiating the discussion. What ensues is perceived by participants as a multilogical event out of the ordinary - "the best one I have ever seen in any 10 days that I can recall". Including a smaller spinoff discussion on the related topic of internalization there were 85 postings in the Goals Multilogue over a period of 34 days.The archived version of the Goals Multilogue contains material that if trimmed into shape would have made a useful FAQ for the list. There are useful explanations of the central concepts of Leont'ev's activity theory (activity/ /motive, action/ /goal, and Object) given in response to questions from relative beginners, and there is discussion of how to relate Leont'ev's treatment to other cultural-historical formulations, including suggestions for further development. The enthusiastic proposal, at a late stage in the discussion, to prepare a collective publication resulted in considerable activity, but no published outcome.

Cascading patterns

	sample start	sample end	msgs	days	mult msgs	mult days	multilogue start
Tools	88-07-01	88-12-31	186	184	72	78	88-10-12
Goals	93-09-01	93-12-31	202	121	75	34	93-10-26
Settings	97-08-28	97-10-08	473	35	68	11	97-09-20

Table 2. Description of the coded samples.

Multitopicality of messages and multistrandedness of the mailflow are well-known characteristica of many forms of asynchronous electronic communication (Black et al., 1983; Levin et al. 1990; Gruber, 1998; McElhearn, 1996; Herring, 1999). Episodes of intense multilogue on the Xlists typically emerge in an unpredictable and ungovernable fashion from the more mundane background fabric of Internet troubleshooting, exchanges of academic information, greetings between participants and scattered offerings of discussion topics that generate few responses and no extended discussion. This is not to say that it is only the scholarly multilogue on a list that generates self-organizing clusters of postings: items from activities of channel maintenance and community building (Ekeblad, 1998) may also provide objects coordinating the actions of list participants in the same fashion. So, for example, the Big Xlist Brownout of November 1998 was an event of channel failure that triggered an avalanche of troubleshooting, which in its turn spawned an outburst of community-building self-presentations. My preference for investigating multilogues on scholarly topics stems from my educational research interest in episodes of intensified topical focus which touch many readers with the urges to "jump into the fray", "steal some time to write", "break" their self-imposed "e-mail rule" etc. - episodes of multilogue that have been subjectively perceived - and socially constructed - as more intense activity than usual. These jointly maintained (inter)subjective impressions of how "the pace of conversation has been fast and furious" or "the flow of postings is back to a pace where it is hard to keep up" have adapted themselves to the increasing mailflow over the years: when multilogues get going, participants regularly find it hard to keep up with the pace of arriving mail - whether the total harvest of the year is 370 postings, as in 1988, or 3300 as in 1996. A simple measure of the number of messages per day seems to confirm the impression that it is the major episodes of densely interconnected multilogue that produce the above-average parts of the mailflow. This impression, however, does not provide enough support for telling a dynamics of self-organization apart from random fluctuations, overlayered on weekly cycles and the cycle of the acedemic year. A firmer support must be built on analysis of references between messages. The production of a link map, i.e. a graphical representation of intermessage references (cf. Levin et al. 1990; Donath et al. 1999) has been an instructive part of this process.

Links between messages cannot be coded just by trusting that contributors have correctly used the response function of the mailer software. It often happens that the response function is used merely as a convenience for getting the message addressed to the list, which can be totally misleading for the purpose of analyzing intermessage references. Thus neither an inherited subject line, nor the inclusion of the message responded to are reliable indicators of relations between messages. Similarly, when the collective activity is triggered by a single message, it's subject line may be inherited for many generations, so that it does not indicate which of several followups that a given message is responding to. The coding of reference links between messages also must allow for many-to-many relations: a posting may refer to several previous postings, as well as it may be referred to by several later postings. For these reasons my method for coding links has been holistic and interpretative: I have read messages carefully for their references to previous postings, then called up pairs of messages in parallel on the screen for simultaneous scrutiny, coding justified links by means of a checklist of confirmatory indications - lack of any confirmation being the only counterindication of a link.

In the next step after links were coded, graphical representations - link maps - were constructed in a flowcharter. Messages are represented by rectangles ordered sequentially from left to right in the link map, each message constituting a step in a system-internal temporality that compresses conventional calendar time in periods of low tide in the mailflow and expands it in periods of high tide. Vertically the messages have been arranged in the link maps so as to disentangle clusters as much as possible, minimizing crossing links. The resulting link map provides a useful visualization of message relations, where different types of event may be discerned, and the development of specific episodes may be followed. In Fig. 5 - 7 the link maps of the 1988 sample and the 1993 sample are articulated week by week, while the link map of the 1997 sample is articulated day by day.

As for similarities, the three samples all contain a sprinkling of isolates - messages that neither respond to another message nor get responded to - and they all contain a measure of small and simple message clusters, representing simple question-answer or topic-comment exchanges. Then there are the larger, cascading multilogical clusters where the links between messages form a tangled web of dense interconnections. While any large cluster will contain a few messages with outside links, the densely linked regions of intermessage references will typically be separated by regions that are traversed only by a few stray links. There is also a certain visual similarity between these clusters in all the maps, showing the cascade structure of a concentrated beginning and a dissipating end, which is so characteristic of the phases of multilogical emergence and decay.

Fig. 8., showing the initial growth of the Settings Multilogue, highlights how a cascading cluster forms when a posting evokes more than one response, and responses in their turn evoke multiple followers. The list-internal temporality expands to accommodate an increasing number of postings within a relatively short span of calendar time - writers spend more time producing messages and the mailflow demands more time from the conscientious reader. A growing cascade soon starts getting tangled: when contributors post messages that refer back to more than one message and relationships get mixed over message generations. Characteristically what makes for the emergence of a densely interconnected multilogue is when there are a number of postings that not only receive many quick responses, but also turn out to be remembered for a comparatively long time by contributors. Fig. 9., showing the whole Settings cascade, illustrates how a substantial part of the visual density of is produced at a fairly late stage of a multilogical episode, as contributors keep referring back to some of the early postings. Looking to message content, this remembering occurs both because these initial postings are those that have initiated the convergence on a shared boundary object, and because of the tendency to collective self-observation when contributors start braiding metacomments into their text about how the particular multilogue began.

In the spatial representation of the link maps a message cascade necessarily spreads vertically as the number of active messages in it increases. This serves well as a metaphor for how the conceptual chains drift apart. As the multilogue unfolds in time it is simply not possible for each contributor to explicitly take account of all previous postings in the episode, much less to preserve topical coherence with them.

The workload of maintaining an overview of the whole discussion grows past individual and collective capacity: reading carefully and fleshing out all the implications would take more time than anyone has to spare. The divergence of the relations between strands of a cascading multilogue may be made more visible in the link map by looking at the daily harvest of messages from points in the development of the cascade after the phase of initial growth, and retaining only their immediate ancestors in the link map. Fig.10. shows two selections of this kind, from day 5 and day 8 in the Settings multilogue.The reduction illustrates quite nicely how common origin may be forgotten as a multilogical episode starts to decay. The spatial drifting apart of branches corresponds, at least metaphorically, with the way different strands of a multilogue take up different aspects of the topic and bring them in very different directions, in the fashion of Vygotsky's chains and complexes. So the link map shows how the network of intermessage references gets looser in the decay phase of a cascade. The frequency of posting goes down, and unless other clusters have started growing the list-internal timeflow shrinks. While there have certainly been expiring branches throughout the development of a multilogical cascade, soon there is no other kind. This is, of course, just a description in words of what the link map makes visible as the nature of endings: in the final phase few postings evoke further response, and these final postings often have little connection with each other. However, there are also postings even late in the life of an episode that cross over and tie together nodes widely apart in the net of past messages - whether by seriously attempted integration, witty application of a phrase transported from one context to the other, or just a friendly nudge to neighbours discussing another topic: there are many possibilities. The exemplar in day 8 of the Settings Multilogue happens to be of the kind where a latecomer seriously attempts an integration and reformulation of questions without receiving any further response. This happens with a certain regularity - individual actions are not enough to turn the tide of a decaying topic. Visually, in terms of the link map, these postings close the local structure, but as the genre of summary adopted on the Xlists seems oriented towards bringing the discussion further, they usually contribute questions and challenges, which, when left without uptake, contributes to the unfinished and frustrating quality of multilogical endings. This may be a point where participants might adapt their practices to the contingencies of the medium, and learn to produce "closing turns" to threads that are "running out" - an idea with some connection to practices of joint reading and other applied forms of structured events on scholarly mailinglists.

Some indicators of self-organization

Distribution of message production over contributors

  

With larger samples the slopes of the logarithmic plots tend to get slightly steeper, while still remaining within the range of "pink noise": for the samples including the target multilogues slopes are -0.87, -1.04 and -1.02, and the slopes for whole years range from -0.99 (for 1988) to -1.38 (for 1996). In other words, the pattern of a relatively small core of highly productive Xlist contributors, a wider pool of moderate contributors, and a large number of minor contributors recurs over the years, as can be seen from Fig. 14. As Fig. 15 shows, the main part of the mailflow is produced by the heavier contributors.

  

The inverse power law pattern appears pervasively over months, quarters, and years, as well as over the whole decade of Xlist existence. As long as there is a sufficient number of postings within the chosen period, the pattern will also appear on the smaller scales of weeks and days. There is thus a self-similarity in this phenomenon over a wide range of timescales. However, it is not the case that it is always the same contributor playing the part of "enormously" productive, "moderately" productive, or "minimally" productive. Contributing a burst of postings in a single multilogue may rank an otherwise infrequent contributor very highly for a particular week or month. A high or moderate ranking for a longer period of time is rather a case of how persistently a contributor keeps posting across weeks, months and years. Another twist to the story is the fact that a contributor playing the part of facilitator, drawing newcomers into the conversation, and in other ways attending to channel maintenance, community building, and the broadening of participation in multilogue, will by necessity come to occupy a prominent rank in a raw count of postings.

The pattern of strongly asymmetric distribution of the message production over contributors is not unique to the Xlists, but appears over and over in data from scholarly mailinglists and other similar forms of asynchronous CMC. Indeed, the asymmetry, although contradicting romantic notions of the broadening of opportunities for getting a turn at writing in computer mediated multilogue, would seem to be the natural outcome of the properties of a virtual community as expressed in Jones' (1997) criteria of a minumum level of interactivity between a variety of communicators in a shared communication space and with a minimum level of sustained stable membership. Together these criteria describe the conditions for an activity system that maintains itself as a system through the "cascading of micro level couplings across many interrelated entities" (see companion paper). The increasing strength of interrelatedness between "entities" in the activity system for longer timespans, that is indicated by the increasing negative slope, corresponds to the dynamic stability in the contributor collective. As observed by Jones (1997) the mailinglist with its relatively low density of the message flow (compared to synchronous forms of CMC) requires a higher level of stability in membership to produce interactive discussions. It seems reasonable to think that the interactive quality of the discussionsalso contributes to the stability of mailinglist membership. The practice of carrying on the mailinglist activity as a written conversation (as opposed to parallel monologues) produces the micro level couplings between postings. Links between postings translate into interrelations between participants, and produce a dynamic network of responsibilities, like the obligation to answer to whoever has commented your last posting - evidenced by the frequent apologies for delay - and all those other community building practices of conversational civility (cf. Baym, 1996; Galegher et al., 1998; Herrmann, 1998). Without the responsive activity there would be no community, no virtual meeting ground, no activity system, just an inert list of subscriber addresses at the server - or a one-way channel of news and announcements. With the semiotically self-regulating responsive activity we get asymmetric posting rates, but we also have virtual community.

Message lifetime

How long, then, do messages stay active in the conversation? The conditions of the mailinglist setting do not allow us to actually see messages expire - there is no telling when the last reader will lay eyes on a message - but for the purpose of multilogue in progress, a posting vanishes into obscurity (merges into the general past and the archival sediments) when it no longer is explicitly referred to in new postings. Thus it is reasonable to regard messages as active in the system from the point in time where they are posted to the point where they are last cited. As long as the written conversation is still in progress, there is, of course, no knowing if and when a posting has been collectively forgotten. There is always the possibility that a contributor busy with her teaching will respond to a message "waaay back" from a "discussion of ten days ago" (which is "almost another era in list life"), or a traveller returned from abroad produce "a really late response to your interesting message" after the passing of four weeks. Indeed, the text as sedimented into the electronic archive at the listserver or on any number of other computer systems may always be picked up from its petrified state and rekindled into life on the list by being re-posted or drawn upon in discussions long after its original production. However, collective memory is usually fairly short, and it is a vanishing minority of the messages that are made present in the mailinglist ecology after months or years.

While all messages must be counted as active on the day they are posted, one half (or more) of all messages do not survive their first 24-hour cycle. They either receive no further response (whether they are isolates or end points of a strand) or they are responded to only within the same day. On their third day only some 25% of the messages are still active. However, it appears from the the three coded samples from 1988, 1993 and 1997 that the collective memory of the Xlists has deteriorated over time. In 1988 messages quite frequently evoked responses ten days or a forthnight after being posted - the estimate of ten days as the boundary of collective short-term memory returns more than once in Xlist metacommunicative discourse:

Things that aren't in the (say) ten-days-back stream of the xlchc 
exchanges that moves incessantly forward at Californian pace (:-) are 
*collectively unconscious*. (A.R. 95-03-26)

In order to further investigate this impression of an overall decrease in message lifespan, batches of 90 postings were drawn from the three coded and link mapped samples (leaving a margin of time/ /postings not used at the end of each sample). For each batch of 90 postings the number of messages still active on their second day, their third day etc. was calculated. After 30 days all messages in all the batches had expired. The markedly steeper slope of the logarithmic plots of the batches from the 1997 sample (Fig. 18) as compared to the slopes derived from the two older samples (Fig. 17) corresponds to a considerably faster rate of decay - measured in calendar time - in 1997. The difference can be illustrated in more concrete everyday terms by mentioning the point where a batch has decayed so that less than 10% of the messages are still active. In the four batches from the 1997 sample this occurs after 4, 3, 4 and 4 days, while in both batches from the 1988 sample it takes 10 days and in the two batches from the 1993 sample it takes 9 and 8 days respectively. This confirms the impression of a shorter collective "attention span" on the Xlists in later years.

  

However, it has to be remembered that the faster expiration of messages in 1997 compared to earlier years occurs together with a much heavier mailflow. Even if participants visit the shared setting more often, it will also demand more time from them each time they visit. Whether they comply with the demand or not is another matter...

For me as one of the moderators it is quite a nasty experience to ask such 
echo-less questions. Sometimes I think that only about ten persons out 
there are reading this stuff, while all other participants just scan 
and stack it somewhere "for later use". 
(A.R.  92-04-03)

  

As we can see in Figs. 17 and 18. it is now the older samples that exhibit the steeper slope: Messages in 1997 typically survive over a larger number of subsequent messages than messages in 1993 or 1988. So the decreasing lifespan in calendar time is not just due to the faster occlusion of old messages by the increasing mailflow. The variation in how long it takes for 90% of the messages to expire in the four batches from the 1997 sample is rather suggestive of the opposite: it takes 31, 21, 40 and 74 succeeding messages for the batches to be reduced to 10% of their initial size, which corresponds quite well to what the result of a current standard four-day span of collective attention would yield under the varying mailflow conditions prevailing in these four batches. The two batches from 1988 are very similar both over days and over intervening messages. It takes 20 or 21 subsequent postings (or 10 days) for the batches to be reduced to 10% of their initial size. In 1993 it takes 31 and 25 subsequent messages respectively (9 and 8 days) for 90% of the messages in a batch to expire.

Although without data on mailing habits it can only be conjectured, the shorter lifetime of messages in later years would seem to be, after all, an effect of more frequent Net access. Xlist participants visit the shared setting more often - or, rather, receive visitations from it more often - and usually there are several new turns to ongoing conversations calling for attention each time. Indeed, the fact that with increasing mailflow, and more frequent mail access, messages typically survive a shorter time in terms of the calendar, but longer in terms of list-internal time (i.e. across a larger number of subsequent messages) suggests that participants, in order to comfortably inhabit the virtual neighborhood, will apply an increasing selectivity in which threads of the multilogue they spend their time on. The rest of the mailflow will be the buzzing noise of overheard conversation - like a "noisy party" - or the vast flowing masses of unread mail leaving participants "wading through" their backlog of messages, and catching them "in a whirl" of mail, confirming participants in their perceptions of the online environment as just as hectic as their busy offline lives. In the years of parallel subconferences, the structure of the setting provided some pre-selective shielding between thematic fields within the common grounds of the Xlist settlement, and it would be very interesting to compare this more sheltered era further with the subsequent single-list period in terms of multilogical coherence and community maintenance.

Four episodes of joint paper reading: Lotman, Wells, Bateson and Engeström

Concerns for cumulativity and development on the collective level have prompted core participants to explore ways of harnessing the multilogue by inserting planned activities into the Xlist setting: exchanges between local seminar groups, Xlist channeling of communication between similar courses run at separate locations, setting up of special-purpose subconferences, joint readings of papers or entire books etc. Here I have chosen to briefly examine four episodes of joint reading of paper-size publications, all highly relevant to recurring concerns in the repertory of Xlist multilogues.

• A reading of Lotman (1988) in the autumn of 1990
• A reading of two pre-publication papers by Wells (Wells, 1993 and Wells & Chang, 1997) from a methodological perspective in the autumn of 1992
• A reading of Bateson (1972) in the winter quarter of 1993
• A reading of Engeström (1996) in the autumn of 1996

The two pre-publication papers by Wells (Wells, 1993 and Wells & Chang, 1997) chosen for a joint reading in the autumn of 1992 both deal with the co-construction of meaning through discourse in the classroom, and the implications of classroom discourse for learning and development. These are educationally central questions that Xlist multilogues return to over and over again. However, in the case of this particular event of joint reading, the papers were not chosen as a target with the purpose of discussing the educational implications of their results, but for the purpose of providing a common ground for a suggested discussion of methodological issues of Activity Theory and data gathering methods like classroom observation. Issues of methodology and of ideological influence on research have been around in the Xlist mailstream throughout October 1992, when one contributor posts a request for literature on the topic of research methodology, and also suggests a number of questions for discussion on the mailinglist. There are a number of postings giving bibliographic references over the next few days, and there are also a few that take up the issue of methodology in response to the questions. Then it is suggested that it would be a good idea to ground the discussion in particulars by using some research paper, preferably one with extended excerpts from classroom discourse so that the process of analysis can be made visible. The Wells papers are nominated (and agreement obtained from the authors), and there is an exchange of postings about how the event should be organized. It is decided that the discussion is to be channeled into the XCLASS, and the papers to be distributed electronically for the asking. The initial questioner volunteers to serve as facilitator of the discussion. Like in the Lotman reading, this episode results in very little discussion of the intended nature - after the putative start of the reading event the first week in November there are 11 postings that recognizably belong to this thread, 7 in the first week and another small batch of 4 postings after a gap of two weeks, when one contributor asks when the discussion is supposed to start. In this episode there are three times as many messages of an administrative character: discussion of the organization of the event, requests for the papers and for being subscribed to the XCLASS subconference. Of course there is no knowing how much of a multilogue on methodology there would have been without the transformation of the event into a joint reading, but the delay caused by having a paper distributed and perhaps the very process of organizing the event over the mailinglist channel, may have diverted the collectively available time. Limited time for reading the paper carefully enough to have opinions on its methodology will also have had its effect.

The essay by Bateson, "Form, Substance, and Difference", included in Steps to an Ecology of Mind (1972), was chosen for a joint reading in the winter quarter of 1993. Bateson argues for the theorizing of the simplest unit of mind as an elementary cybernetic system with its messages in circuit, where the transform of a difference traveling in the circuit is the elementary unit of ideas. This paper contains two illustrative examples of such systemic circuits - mind extending outside the boundary of the skin - that have been recurring objects for Xlist discussion: the blind man with his stick and the man cutting down a tree with an axe. It is one of these Batesonian occasions that prompts the suggestion for a joint reading: Bateson has been invoked several times in the Xlist mailstream of December 1992, when a joint reading of this particular paper is suggested. The suggestion is taken up by other participants, but the matter of organization is quickly channeled outside the public forum, directly to the contributor who initiated the topic, and who has volunteered to facilitate the discussion. The second week of January 1993 this participant announces that the Bateson reading will start on February first, with the first pages of the essay - but it is not until a few days later that he starts the discussion by forwarding a commentary on the essay from an external contributor, who soon thereafter joins the list. After a delay of almost a week, a multilogue does get off the ground: an initial round of roughly a week, with several participants and quick responses, followed by a three-week "panel" discussion in slow but steady pace between three skilled and knowledgeable contributors, posting long and reflective messages, which often meticulously respond not to the latest message, but to the last issue left dangling. A file of the accumulated text from this multilogue has been circulated on the Xlists on at least one later occasion where Batesonian questions were in the mailstream. After a month, one week into March, the facilitator opens the discussion of the final part of the essay, expressing hopes that the content of this section will appeal to a wider range of contributors. The next posting praises this invitation, but also turns out to be the final posting in the episode - the Bateson reading is alluded to as unfinished business a month afterwards. Nevertheless, in this episode of joint reading there is not just a larger number of discussion postings than in the two previous ones (28 in all), but these postings also form an interconnected multilogical cascade, which cannot be said about the two events described previously. The successful character of the Bateson episode is presumably formed by the concise planning of the event, the ample time given for preparatory reading (or re-reading) of a classical and easily available essay, and the presence of a "special guest contributor".

In the fourth case, the Engeström paper, "Development as breaking away and opening up" (Engeström, 1996), chosen for joint reading in the autumn of 1996, the initial phases of the joint reading event develop more rapidly than in any of the other cases: the episode passes from proposal to discussion in four days. The paper is selected for joint reading just a little over a week after being presented at the Second Conference for Sociocultural Research, held in Geneva, September 11-15, 1996. Some of the Xlist participants who were not going to Geneva had expressed their interest in post-conference discussion on the list even before the conference, and the intention was to take up other papers after the first one had been treated. Engeström's paper argues against theories of development that look only to the peaceful and stepwise individual progress along a universal, pre-established developmental path. It suggests that individual transformation may depend on collective transformation, that development may be viewed as partially destructive rejection of the old - rather than simply as benign achievement of mastery - and that instead of just being a movement across vertical levels, development may also involve movement across horizontal borders. The theoretical arguments of the paper are illustrated by means of examples from Peter Høeg's novel "The Borderliners" (1994), and its re-formulations of the Vygotskian Zone of Proximal Development are well in line with Xlist debates over the years. When the paper is first suggested for joint reading there is briefly some organisational discussion: other papers are mentioned, the author posts his agreement, and, following a suggestion that it would be a good idea to make the target paper available over the MCA Website, there is discussion about the best procedures for putting papers on the Web. However, on the third day after the initial suggestion an electronic copy of the paper is sent over the list, and the next day there are already responses from contributors who have read it. This event is different from the earlier occasions of joint reading in being a lot less pre-planned and in having no designated chairperson. In keeping with its occurrence in the single-list, high-mailflow era of local history discussion part of it contains a larger number of contributions than any of the other three joint reading episodes - 53 postings in the course of three weeks - and there are many crossreferences between strands of the cluster, although late in the episode an independent thread on the use of fiction as data develops, which is related to the paper, but not to the other strands of the multilogue. When the multilogue is far into its phase of decay there is an announcement that the paper is now available through the Web page. One of the two last postings in the episode that arrive after this point is an attempt to bring together some of the issues brought up and formulate new questions. There is, however, no uptake to this. The only difference between this episode of joint reading and a spontaneously emerging multilogue is the fact that the discussion takes off from the reading of the Engeström paper and keeps returning to this text. Evidently the strategy of broadcasting the paper on very short notice worked well as a way of riding the waves of a self-organizing system. On the other hand, as with other attempts at initiating a new multilogue by introducing external material, this strategy is a gamble - although probably less so when the paper is requested by list participants than when it is first offered up by the author. (There are many instances of this in the Xlist archives, and, especially in the early years, paper-length texts sent over the list were likely to be taken as insults by at least some subscribers, and responded to accordingly. However, with increasing Internet access, the floating of bulky material into the shared setting is less commonly seen as an offense.)

The two cases of successfully carried out joint reading shared the characteristic lack of closure with spontaneously emerging multilogues. Trying to rekindle a topic that has started to decay may be no more feasible than leaving the popcorn in the microwave until every single corn has popped. However, when the pops get few and far between, the late contributor might win the day by rhetorically tying up loose ends - it would seem as if the wily writer could produce a fair simulacrum of consensus with no great risk of being gainsaid. Perhaps the Xlist custom of leaving always some questions in the air is to be preferred, even to the price of the frustration that may arise from the lack of closure.

Concluding remarks: tuning in to mailinglist dynamics

Graphical link maps of intermessage references were introduced as a tool and used for making the temporal structure of multilogical events and the relations between messages visible. Three fairly large samples chosen because of containing interesting multilogical episodes were presented, and graphically represented by link maps, bridging between the perceptions of subscribers participating in multilogical activity in its expanding and dissipating phases and an analytical perspective on the events. The perception of accelerating intensity in the written conversation that characterizes the emergence of a multilogical cascade corresponds to an expansion of the system-internal temporality - when the discussion gets hot the mailinglist takes more reading and writing time from its active participants each time they visit the virtual setting - and at the same time there is an increasing density of interlinkage between postings. However, the expansion contains the seed of its own dissolution, as with an increasing number of active messages in the system (relevant postings, old and new) the collective capacity of participants to hold together the whole semiotic complexity of the multilogue will reach its limit. The perception of thematic chaining and dissolution into unrelated branches of conversation correspond to a spatial drifting apart and loss of connections in the link map. What has been presented here from this conceptually generative phase of research naturally demands further testing of its viability.

However, my observations of the several ways in which the activity on a scholarly mailinglist can be said to be self-regulating warrant some concluding remarks in the direction of application. For those of us who wish to capture the learning potential of written conversation in the mailinglist medium it is important to be well aware of the self-regulating nature of scholarly mailinglists, in all senses of the word. If nothing else it may spare us from some disappointments: a self-organizing system is not really amenable to control or planned change - but it is, nevertheless, possible to get in tune with the internal dynamics of a mailinglist, and learn to recognize the moments when small interventions have a fair chance of triggering noticeable effects. The four case studies of more or less planned and successful events of joint reading on the Xlists illustrate both how plans may go wrong in this distributed and loosely coordinated medium, and how organized episodes may be carried through, with the right timing in relation to the current dynamics in the activity system. The success of a scholarly mailinglist as a multilogical activity system, then, depends on the apt performance of semiotic self-regulation in the subscriber community, involving for each and every responsible participant a coordination of the moments of centered participation with a certain amount of de-centering towards a view of the emergent states of the mailinglist ecology. On the Xlists these kinds of facilitation skills have always been regarded as a collective responsibility for maintenance of the cultural practices of multilogical discussion. The combination of the centered, participatory appropriation of these practices with occasional de-centered knowledge-building about the emergent nature of mailinglist dynamics at a systemic level would seem to me a promising road towards maintaining practices of mailinglist re-centering.

I would like to acknowledge my gratitude to the whole Xlist community for collegial support, scholarship and friendship during my years as a virtual neighbor (starting in November 1993). Mike Cole and Arne Raeithel provided me with the encouragement and inspiration to take up mailinglist dynamics as an object of research. My quotations of their messages are an expression of my indebtedness: they "said it all" before I even got there. Any mistakes I have made while filling in the blanks between their wizardly intuitions and the cyber-archaeological material are, of course, my own.

This work was supported by The Swedish Council for Research in the Humanities and Social Sciences (HSFR).
Opinions are those of the author and are not necessarily those of the Council.

References

Barbatsis, Gretchen, Michael Fegan and Kenneth Hansen. 1999. The Performance of Cyberspace: an Exploration inti Computer Mediated Reality. Journal of Computer Mediated Communication [On-line], 5 (1). Available at URL = http://www.ascusc.org/jcmc/vol5/issue1/barbatsis.html

Barowy, William. 1999. A Model for Temporal Self-Regulation in Electronic Learning Communities. Paper presented in the symposium: Time and coordination in a virtual community of learners. EARLI (European Association for Research on Learning and Instruction), Göteborg, Sweden, August 24-28 1999.

Bateson, Gregory. 1972. Form, Substance, and Difference. In Steps to an Ecology of Mind. pp. 448-466. New York: Ballantine Books.

Baym, Nancy. 1996. "Agreements and Disagreements in a Computer-Mediated Discussion", Research on Language and Social Interaction 29(4): 315-345.

Black, Steven. D. James A. Levin, Hugh Mehan and Clark N. Quinn (1983). Real and non-real time interaction: Unraveling the multiple threads of discourse. Discourse Processes, 6, 59-75.

Cole, Michael. 1997. From The Creation of Settings to the Sustaining of Institutions. Mind, Culture, and Activity. 4(3): 183-190

Cole, Michael, and Yrjö Engeström. 1993. A cultural-historical approach to distributed cognition. In G. Salomon (ed) Distributed Cognitions. Psychological and educational considerations. Cambridge MA: Cambridge University press.

Decker, William M. 1998. Epistolary Practices. Letter Writing in America before Telecommunications. Chapel Hill: University of North Carolina Press

Donath, Judith, Karrie Karahalios, and Fernanda Viégas. 1999. Visualizing Conversation. Journal of Computer Mediated Communication [On-line], 4 (4). Available at URL = http://www.ascusc.org/jcmc/vol4/issue4/donath.html

Dooley, Kevin J., and Andrew H. Van de Ven. 1997. A Primer on Diagnosing Dynamic Organizational Processes. Available at URL: http://www.eas.asu.edu/~kdooley/os.pdf

Ekeblad, Eva. 1998. Contact, Community and Multilogue: Electronic communication in the practice of scholarship. Paper presented at The Fourth Congress of the International Society for Cultural Research and Activity Theory, ISCRAT 1998, Aarhus University, Denmark, June 7-11, 1998. Available at URL: http://cite.ped.gu.se/Eklanda/Papers/ISCRAT/CoCoMu.html

Engeström, Yrjö. 1996. Development as breaking away and opening up: a challenge to Vygotsky and Piaget. Paper presented at at the IInd Conference for Sociocultural Research, Geneva, 11-15 September 1996.

Galegher, Jolene, Lee Sproull, and Sara Kiesler. 1998. Legitimacy, authority, and community in electronic support groups. Written Communication 15(4), 493-530.

Gruber, Helmut. 1998. Computer-mediated communication and scholarly discourse: forms of topic-initiation and thematic development. Pragmatics: 8(1): 21-45

Herring, Susan. 1999. Interactional coherence in CMC. Journal of Computer Mediated Communication [On-line], 4 (4). Available at URL = http://www.ascusc.org/jcmc/vol4/issue4/herring.html

Herrmann, Françoise. 1998. Building on-line communities of practice: an example and implications. Educational Technology, 38 (1), 16-23.

Høeg, Peter. 1994. Borderliners. New York: Farrar, Straus and Giroux.

Jones, Quentin. 1997. Virtual-Communities, Virtual Settlements & Cyber-Archaeology: A Theoretical Outline. Journal of Computer Mediated Communication [On-line], 3 (3). Available at URL = http://www.ascusc.org/jcmc/vol3/issue3/jones.html

Leont'ev, A. N. 1978. Activity, Consciousness and Personality. Englewood Cliffs, NJ: Prentice-Hall.

Leont'ev, A. N. 1982. Problems of the Development of the Mind. Moscow: Progress Publishers.

Lewenstein, Bruce. 1995. Do Public Electronic Bulletin Boards Help Create Scientific Knowledge? The Cold Fusion Case. Science, Technology, & Human Values 20(2): 123-149

Levin, James A., Haesun Kim, and Margaret M. Riel. 1990. Analyzing instructional interactions on electronic message networks. In: L.M. Harasim (Ed) Online Education. New York: Praeger.

Lotman, Yuri M. 1988. Text within Text. Soviet Psychology, 1988, Vol. 26, No. 3 pp 32-51

McElhearn, Kirk. 1996. Writing Conversation: An Analysis of Speech Events in E-mail Mailing Lists. Available at URL: http://www.mcelhearn.com/cmc.html

Moll, Luis. 1997. The Creation of Mediating Settings. Mind, Culture, and Activity. 4(3): 191-199

Mulkay, Michael J. 1985. The word and the world: explorations in the form of sociological thought. Winchester: Allen & Unwin.

Palme, Jacob. 1989. Elektronisk post. Lund: Studentlitteratur.

Palme, Jacob. 1995. The Optimal Group Size in Computer Mediated Communication. In J. Palme. Electronic Mail. London UK: Artech House Publishers. The relevant excerpt is available at URL: http://www.dsv.su.se/~jpalme/e-mail-book/group-size.html

Raeithel, Arne. 1992. Semiotic self-regulation and work. An activity-theoretical foundation for design. In C.Floyd, H. Züllighoven, R. Budde & R. Keil-Slawik (Eds.), Software development and reality construction (pp 391-415). Berlin: Springer

Raeithel, Arne. 1996. On the ethnography of cooperative work. In: Y. Engeström and D. Middleton (Eds.) Communication and Cognition at Work. New York: Cambridge University Press

Rafaeli, Sheizaf and Fay Sudweeks. 1996. Networked Interactivity. Journal of Computer Mediated Communication [On-line], 2 (4). Available at URL = http://www.ascusc.org/jcmc/vol2/issue4/rafaeli.sudweeks.html

Rojo, Alejandra. 1995. Participation in scholarly electronic forums. A University of Toronto Ph.D. thesis. Electronically available at URL: http://www.oise.on.ca/~arojo/tabcont.html

Sandbothe, Mike. 1998. Media Temporalities in the Internet: Philosophy of Time and Media with Derrida and RortyJournal of Computer Mediated Communication [On-line], 4 (2). Available at URL = http://www.ascusc.org/jcmc/vol4/issue2/sandbothe.html Sarason, Seymour, 1997. Revisiting The Creation of Settings. Mind, Culture, and Activity 4(3): 170-182.

Shank, Gary. 1993. Abductive multiloguing. The semiotic dynamics of navigating the Net. Arachnet Electronic Journal on Virtual Culture v1n01 (March 22, 1993) URL = ftp://ftp.lib.ncsu.edu/pub/stacks/aejvc/aejvc-v1n01-shank-abductive

Star, Susan Leigh, and James R. Griesemer. 1989. Institutional ecology, 'translations' and boundary objects: amateurs and professionals in Berkley's museum of vertebrate zoology, 1907-1939. Social Studies of Science, 19: 387-420.

Syverson, Margaret A. 1994. The wealth of reality: an ecology of composition. Ph.D Dissertation, UCSD.

Syverson, Margaret A. 1999. The wealth of reality: an ecology of composition. Southern Illinois Univ Press

Wells, Gordon and Gen-Ling Chang. 1997. "What have you learned?": Co-constructing the meaning of time. In J. Flood, S. Brice-Heath and D. Lapp (Eds.) A Handbook for Literacy Educators: Research on Teaching the Communicative and Visual Arts. New York: Macmillan.

Wells, Gordon. 1993. Reevaluating the IRF sequence: A proposal for the articulation of theories of activity and discourse for the analysis of teaching and learning in the classroom. Linguistics and Education 5: 1-37

Vygotsky, Lev.S. 1934/1987. Thinking and speech. In R.W. Rieber and A.S. Carton (Eds.) The collected works of L.S. Vygotsky, Volume 1: Problems of general psychology. (Trans. N. Minick). New York: Plenum.