Shared prolepsis and intersubjectivity in open source development

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/262364640 Shared prolepsis and intersubjectivity in open source development: Expansive grounding in distributed work Conference Paper · February 2013 DOI: 10.1145/2441776.2441793 CITATIONS READS 2 107 3 authors, including: Leif CHRISTIAN Lahn Anders Mørch 16 PUBLICATIONS 97 CITATIONS 116 PUBLICATIONS 1,914 CITATIONS University of Oslo SEE PROFILE University of Oslo SEE PROFILE All content following this page was uploaded by Anders Mørch on 18 May 2014. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document and are linked to publications on ResearchGate, letting you access and read them immediately. Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA Shared Prolepsis and Intersubjectivity in Open Source Development: Expansive Grounding in Distributed Work Pål Fugelli Research Computing Services University of Oslo pal.fugelli@gmail.com +47-22852965 Leif C. Lahn Dept. of Educational Research University of Oslo l.c.lahn@ped.uio.no +47-22857076 ABSTRACT Anders I. Mørch InterMedia University of Oslo a.i.morch@intermedia.uio.no +47-22840713 the study of micro coordination among a group of colocated game players playing a collaborative problem solving game [18]. Here, micro coordination refers to how participants decide who plays next and other low-level decisions that might impact overall performance. The authors argue that learning and working are intertwined at this level. In this paper, we apply a related strategy for identifying sub-processes in distributed work, studying intersubjectivity by analyzing communication patterns and message postings in an open source development (OSD) community. Intersubjectivity is a term used to conceptualize the psychological relationship between people during conversation, e.g. for learning each other’s perspective and building a shared understanding. We consider this relevant for understanding aspects of communication and coordination in OSD; i.e., newcomers must learn how to get into a project and experts (core members) must learn how to sustain their understandings as the software evolves. Intersubjectivity is a term used to conceptualize the psychological relationship between people during conversation, e.g. for building a shared understanding. Ragnar Rommetveit, a Norwegian social psychologist, developed a conceptual framework for intersubjectivity, treating it as a social phenomenon and a dynamic process. One technique for increasing intersubjectivity according to Rommetveit is to issue ‘anticipatory cues,’ i.e. referring to common knowledge and indicating future situations where the knowledge will be relevant. This framework was adapted for online communication and applied to an analysis of the mod_perl module of the Apache Web server (an open source development project). Based on observations of 215 participants’ contributions to the project’s mailing list over a 6-month period, we explore how processes of intersubjectivity evolve across the developer network. We conclude with a discussion of how so-called proleptic instances in post-and-reply messages may be significant and trigger the co-construction of shared understanding. Intersubjectivity is perhaps most often associated with a theory of language use first proposed by the linguist Herbert Clark and referred to as the theory of common ground. Clark’s theory is comparable with the theory of intersubjectivity espoused in this paper, Ragnar Rommetveit’s theory of prolepsis-driven intersubjectivity. They share the assumption that individuals engaged in conversation must share knowledge in order to be understood and contribute to the conversation. Clark and Brennan said that grounding involves “mutual knowledge, mutual beliefs, and mutual assumptions” [6, p. 127], whereas Rommetveit defined intersubjectivity as a “temporarily shared social world” [25, p. 87], which is mutually accessible and partly based on so-called proleptic instances, a special type of utterance induced by the speaker. These proleptic instances function as anticipatory cues, expanding the space of intersubjectivity. Rommetveit’s position is that the social world in the outset is not the world of the private individual but a world of intersubjectivy shared by all, to a lesser or greater extent. We depict the differences between the two perspectives in Figure 1. We argue that both perspectives have strengths and weaknesses, and we use the latter one, which we refer to as prolepsis-driven intersubjectivy or expansive grounding, to develop a conceptual framework for analysis of distributed work in OSD. Author Keywords Collaboration, empirical study, grounding, intersubjectivity, prolepsis, shared understanding, software development. ACM Classification Keywords H.5.2 [Information Interfaces and Presentation]: Group and Organization Interfaces – Asynchronous interaction, evaluation/methodology. K.4.3 [Computers and Society]: Organizational Impacts – Computer-supported collaborative work. General Terms Design; Documentation; Theory. INTRODUCTION The complexity of CSCW calls for the identification of subprocesses of communication and coordination that may impact overall performance. A recent effort in this regard is Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CSCW ’13, February 23–27, 2013, San Antonio, Texas, USA. Copyright 2013 ACM 978-1-4503-1331-5/13/02...$15.00. 129 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA members of the community with the ultimate effect of expanding the group’s common ground or shared understanding. We have developed a theoretical framework to guide our efforts. We used a mixed methods approach for collecting and analyzing data. Regions of intersubjectivy were found using the social network analysis (SNA) method and included both peripheral and core members of the mod_perl development network. Our unit of qualitative analysis is conversations among developers, initiated by peripheral members and cultivated by core members who were perceived as both knowledgeable about the project (code and documentation) and good educators. We discuss the role of proleptic instances in these conversations and also how prolepses differ from ‘tasks’ and ‘problems’ in OSD. Figure 1. Two perspectives of common ground in interpersonal communication: shared cognition (left) and shared social reality (right). The picture has been simplified for illustration. Our object of study is the community around the mod_perl module of the Apache web server. The Apache project is described as “a collaborative software development effort aimed at creating a robust, commercial-grade, featureful, and freely-available source code implementation of an HTTP (Web) server” [1]. It is a worldwide network of more or less dedicated developers and users, and the software has continually evolved since the project was launched in April 1995 as a result of new patches submitted by developers and new upgrades released by the organization. Our work is theoretically informed but also motivated by informal examples, especially of the early stages of transitioning from natural to artificial (design) activity. We briefly describe three illustrative examples that have motivated of our own understanding of the role of proleptic instances in the early stages of cooperative work: The main aim of our research is to explore prolepsis-driven intersubjectivy as a dynamic, social phenomenon in distributed cooperative work. This has not been explored in previous work in CSCW. The second aim is to identify opportunities for learning and knowledge development. We surveyed related work to provide a rationale for our aims. For example, Lee, Tatar and Harrison [18] show in their study of micro-coordination that non-discursive behavior in decision-making needs to be accounted for in discourse analysis. Furthermore, in a case study of Second Life use, Shami, Erickson and Kellogg [29] found that implicit grounding, i.e. common ground that is established before conversation actually begins, could be supported by affordances built into in the collaboration environment. Moreover, Stahl, Koschmann and Suthers [30] studied collaborative knowledge building and suggest that the problem of intersubjectivity is of particular relevance for understanding how learning is produced through interaction, advocating for more in-depth interdisciplinary research and arguing that this issue has implications for research methods and for the design of collaboration software. • The Richmeister: The role played by Roy Schneider in a series of Saturday Night Live comedy sketches in the early 1990s. The character Richard Laymer (‘The Richmeister’) is an office worker in a copy room who annoys people by giving them nicknames based on knowing their real names, more or less randomly issued with the goal of making friends. He is not very successful in his numerous attempts, and his behavior is characteristic of what we refer to as pre-prolepsis [31]. The results we report in this paper can be summarized as follows: Intersubjectivity is the ‘object’ of a shared understanding that evolves in the OSD project. The ‘object’ in this sense is neither software nor documentation, but a parallel process of building this shared understanding over some time (in our case 5 days). We have traced the roots of these processes to a certain type of utterance we refer to as proleptic instance. Proleptic instances are explicit or indirect parts of postings that can be identified as preceding or otherwise in conjunction with questions, idea proposals, etc. They are peculiar in the sense that they draw attention to themselves by triggering a series of elaborations and clarifications over several days from more experienced • Cinema viewer capture technique: Prolepsis is a word that has been adopted in cinema production and used as technique to capture audience attention for a long time. It means flash-forward or foreshadowing, a technique used to reveal events that will occur in a future scene, often presented in discrete ways, leading the viewer to form her own expectations about what will happen next. An example is the notorious ‘kitchen knife as set decoration’ in early scenes in horror films, strategically placed for later use as a murder weapon. In the opposite direction, flashbacks are often used to recount events that happened before a story’s primary sequence of events or to fill in crucial backstory. • Gnu crossing: From July through September each year herds of wildebeest gather to cross two rivers in East Africa. The transition starts by more or less random and restless attempts at finding a place to cross safely, being aware of crocodiles lurking in the water but driven by the better feeding grounds on the other side. When a few lead animals (‘gunny pigs’) have found a satisficing route, a vigorous activity follows as the other animals follow quickly after in more or less the same path. It shows flock behavior triggered by a sudden act of movement by a few individuals [13]. 130 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA top contributors to the actual development effort” (p. 1). Furthermore they found that the communication network of the developers evolve to adapt to the modular structure of the source code, organized into communication hubs. Moreover, the authors report that in addition to skilled developers, also less experienced developers are drawn into the communication hubs if their work assignment leads them. Our analytic attention takes inspiration from these findings and focus on how core and peripheral participants take on different roles at different times in the construction of shared understandings. The rest of the paper is organized as follows: First, we provide a brief survey of related work in OSD. Next, we develop a conceptual framework for intersubjectivity based on a review of two theories of intersubjectivity as first proposed for face-to-face communication: Rommetveit’s [24, 25] notion of prolepsis-driven intersubjectivity and Clark and Brennan’s [6] extended contribution model. Two levels of analysis are employed for zooming in and out of the mod_perl developer community: macro (network structure) and micro (message content and interaction). We capture data at these two levels during a six-month logging of online communications among subscribers of the project’s mailing lists. The network data provide the context (visualize regions of low and high cohesion) for the selection of messages for in-depth interaction analysis. The framework is applied to the analysis of message exchanges among the contributors in high cohesion areas to identify characteristics of prolepsis-driven intersubjectivity. At the end, we compare our findings to related work in OSD and intersubjectivity and suggest some implications for design. Edwards [9] studied the transition from peripheral to core participation in OSD using “community of practice” as a theoretical framework. He found that situated learning was essential for newcomers to become core members. He also found that all contributors, but especially core members, relied on a common frame of reference to coordinate their joint effort. Much of this was not the result of a conscious effort but a consequence of maintaining membership in the community. The peripheral participants contributed to user support mailing lists (e.g. how to download and install a system), whereas the core members contributed to technical discussion lists and proposed patches to improve the software. The learning curve was steep for newcomers, and the motivation of core members to sustain membership was associated with the joy of solving software-related problems [9]. Our study reveals a separation of core and peripheral participants, but little transition between them. OPEN SOURCE SOFTWARE DEVELOPMENT There is little research that explicitly links OSD and intersubjectivity. We survey some of the related work below. Bergquist and Ljungberg [3] conducted an empirical study of open source developers for a period of two years and identified a phenomenon they labeled “gift giving” culture. They found that developers communally gave and appreciated gifts, and that the open source developer community relied on gift giving as a way of “getting new ideas and prototypes out into circulation.” Gifts were not physical commodities, but rather substances such as knowledge, source code and peer reviews (conceptual artifacts). They claim that examining the sharing of these resources provides a key for understanding the open source community’s cultural foundations, as well as its social organization [3]. There are studies of global online teams other than OSD projects that use organization structures comparative to those of OSD, like management teams of large scale engineering projects [14, 17]. They reveal that distributed teams have structures that may easily dissolve because of the high demands of coordination and ‘centrifugal forces’ that pull the teams apart. Hemetsberger and Reinhardt [14] contrasted these findings with the persisting cohesion typical of open source collaboration. They attributed the success of coordination in OSD to a number of technological artifacts that define the process; source code is the main artifact, and serves as a coordination mechanism by providing a modular structure for organizing mailing lists. The combination of a well-defined agenda (what to do) and an organization that facilitates the recruitment of newcomers ensures stability and progress [17]. Network-level studies of open source development projects have reported that interaction patterns among participants vary depending on their roles, most notably those of core developers vs. peripheral participants [8, 9]. In one study, Long and Siau [19] applied social network analysis to three projects accessible in SourceForge.net. They reported that the interaction pattern within these projects evolved from a single hub at the beginning of a software project’s life cycle into a core-periphery model as the project matured. In our study, another approach is proposed while using a similar model as a network-level contrast for analyzing message content targeting processes of intersubjectivity. These studies show how open source development projects are collaborative, knowledge intensive, and require learning, both for newcomers and core members. Newcomers must learn how to enter a project, while core and peripheral members must learn to sustain their understandings as software evolves. The participants’ interactions are organized around source code, tools for development (programming environments, programming languages, web browsers, debuggers, version control systems, and so on), releases for download, and mailing lists. These artifacts are different from the less tangible and In a study by Cataldo and Herbsleb on how software engineers communicate in geographically dispersed teams, a core group of developers emerged as the liaisons (or gatekeepers) between formally defined teams [5]. Their findings revealed that “individuals in the core not only perform a critical communication role but also they are the 131 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA sense of it. He called the tacitly induced elements shared prolepses [25], defined as a communicative act among participants in a conversation where the speaker presupposes something that has not yet been discussed, in effect creating a connection between a past situation and possible future situations. The term ‘prolepsis,’ meaning ‘preconception’ or ‘anticipation’ dates from ancient Greek writings. In Rommetveit’s framework, prolepsis is intimately connected to intersubjectivity as a mechanism for expansion. The ‘here and now’ of the present situation is connected with past events (before) and future events (afterwards) as depicted by the intersection of the x-axis with the curved lines in Figure 2. At a more general level the process is dynamically changing in three dimensions (x, y, and z-axes), which is triggered by issuing proleptic instances as communicative acts. transitory objects of knowledge that emerge during conversations in the mailing lists. These initially vague objects serve as cues and markers for evolving more durable objects of shared understanding to monitor and inform collaborative work in the OSD community. CONCEPTUAL FRAMEWORK We develop a conceptual framework for intersubjectivity based on a review of two theories of intersubjectivity as first proposed for face-to-face communication: Rommetveit’s notion of prolepsis-driven intersubjectivity and Clark and Brennan’s extended contribution model. Intersubjectivity and Prolepsis Nearly 40 years ago, Rommetveit proposed a phenomenological account of communication [24] that was partially inspired by the phenomenology of Husserl and social phenomenology [27] and partially by the sociology of knowledge [2]. In a nutshell, it states that the world of intersubjectivity is not the overlap of the worlds of the private individuals but a world of a shared social reality. Furthermore, it states that communicators’ perspectives of this world can only be partial, never complete because the world is a generalization of experiences and it is temporarily shared (see Figure 1, rightmost situation). It is an accumulation of shared presuppositions and historical information only partly accessible. Between the ‘I’ and ‘You’ depicted in Figure 2, potential states of intersubjectivity can emerge along a three-dimensional spatial-temporal-interpersonal space. Utterances are assigned meaning through a process of joint construction that operates in this space [25]. Shared prolepsis is also a mechanism for sustaining intersubjectivity when the actors continually fine-tune their assumptions and anticipations in reaction to others’ contributions and feedback [21]. Rommetveit [24] offers the following example of prolepsis in a personal letter received from a friend: “Today, I walked with one of the psychologists here past the Mayflower cinema in Eugene, where Bergman’s latest film movie is being shown. He asked me whether I had seen it. I said no, and asked if he had. He said yes, he had. I asked him how he liked it, and he said ‘I liked it very much, but Mary Ann did not’; without ever explicitly having ‘made known’ to me that he is married and that his wife’s name is Mary Ann, that they went to see the film together, and a lot of other things – and (if I am correct) without assuming that I knew all this. His utterance was proleptic in that it triggered a search on my part for a shared social reality, which in turn would provide a basis for understanding the sentence. Incidentally, it would have been barbaric and pedantic to say, ‘Oh, Mary Ann is your wife’. To be precise, prolepsis here served to establish a relationship between his wife and me as persons who should at some time get together. My comment would have been a crude rejection of that implication.” [24, p. 87-88] In this story the fact that the friend has a wife and her name is Mary Ann is presupposed but not directly mentioned. This may challenge the listener and “trigger anticipatory comprehension, and what is made known will hence necessarily transcend what is said” [24, p. 88]. It may also trigger the construction of new understandings of the situation. Prolepsis can thus serve as a catalyst for making sense of not-yet-provided information; in this case, to prepare for a future meeting of three individuals. Its role in intersubjectivity is to provide prompts or cues for building a special kind of common ground: referring to shared knowledge that has not yet been introduced and pointing towards a future situation where the knowledge is relevant. As such, it is a mechanism for constructing common Figure 2. The spatial-temporal-interpersonal coordinates of communicative acts defining the framework of intersubjectivity for face-to-face communication [25]. Rommetveit argued that this world is “in part based upon premises tacitly induced by the speaker” [25, p. 87], by which he means that some elements are left out of a conversation and the listener is subsequently invited to step into a partially shared space of intersubjectivity and make 132 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA understanding for use in practical situations (e.g. installing a web server) inspired by phenomenology [24, 25]. recent study of online communication in the virtual world Second Life, Shami, Erickson and Kellogg [29] found grounding to be different (less constrained) than predicted by Clark and Brennan’s [6] theory. They found that affordances of the virtual places and situational characteristics of the environment, such as knowledge of occurring events (in their case an online conference), might facilitate grounding. They proposed an adaptation of the extended contribution theory to virtual worlds with the notion of implicit grounding, which refers to actions in the beginning of a conversation that refer to elements of the immediate surroundings that allows for the creation of common ground, and in Second Life this information extends the contextual information found in face-to-face conversations [27]. The process of omitting information and, in its place, issuing more or less indirect cues of its aims, invites the listener to actively participate in the co-construction of an expanded intersubjective space [24]. The receiver of proleptic utterances ‘fills in’ the missing pieces and thus coconstructs the information needed to achieve a sufficient level of common understanding. Using proleptic instances as building blocks for conversational grounding is new in CSCW. We describe the most frequently cited work below before we present a conceptual framework for intersubjectivity in online communication, which we apply in the analysis of building common understanding in the mod_perl network. We propose another technique for facilitating grounding in online communication, which we apply in the empirical analysis of an OSD project, namely shared prolepsis, which we present next. The Extended Contribution Theory Clark and Schaefer [7] proposed a collaborative model for human communication based on grounding, referred to as the contribution model. This model revises a more traditional sender/receiver perspective on communication, broadening the analytic frame from single utterances to contributions developed through interaction. We visualized this model informally in Figure 1 (leftmost situation). Clark and Brennan [6] extended this model and postulated that, “all collective actions are built on a common ground and its accumulation” [6]. To coordinate the content and processes of human conduct, participants must update their common ground moment by moment in what Clark and Brennan refer to as ‘grounding,’ and these grounding processes vary according to the contexts of interactions. A Distributed Model of Shared Prolepsis Online communication introduces new dimensions to intersubjectivity that was not present in the framework proposed by Rommetveit. In the revised framework we present here [11], intersubjectivity emerges in the intersection of distributed work and the evolution of a shared knowledge object. This is depicted by the coordinate system shown in Figure 3, where the x-axis tracks the evolution of a shared knowledge object as a temporal process from fragmented (individual) knowledge towards a shared object of attention. The y-axis represents spatial displacement, which are supported by the features of collaboration software. The z-axis represents the interpersonal communication across the network (captured by the type of messages submitted and responded to). Clark and Brennan [6] claim that different media put different constraints on grounding processes. For example, real-time communication media like videoconferencing offer visibility and co-temporality, whereas e-mail or personal letters lack the same features, causing delays in turn taking for proper understanding. They suggest that, “people should ground with those techniques available in a medium that leads to the least collaborative effort” [6]. According to Rommetveit and his notion of prolepsis, such a shared understanding is only partially achieved and depends on access to historical information (common preunderstanding). Early research on applying the extended contribution theory of common ground in CSCW compared face to face situations with remote collaborations, finding that grounding was reduced in distributed settings unless participants had access to media channels that simulated the contextual information of co-presence (e.g. video, shared space for communication) [22] Arguably, this is less of a problem in today’s virtual collaboration environments, which have more sophisticated means for simulating copresence and capturing context (e.g. built-in cameras, desktop video, immersive worlds), but also because of new ways of communicating and grounding. For example in a Figure 3. Framework for intersubjectivity and shared prolepsis in knowledge-intensive distributed work [11]. 133 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA parameters like centrality, grouping, and density in a social network and the constitution of intersubjectivity in on-line communication We opted for a strategy to combine social network analysis and content analysis of messages between participants. The former approach computes structural parameters of interactions in terms of number of subgroups, centrality, and density of the network, whereas the latter approach follows an interpretative strategy with a focus on meaning making in the conversations between participants [16]. The structural analysis will produce network-level data and provide an empirical basis for selecting episodes of interaction among individuals, for example in highcohesive regions of the network. In a nutshell, the process goes as follows: 1) identify regions in the network that are interesting from the point of view of intersubjectivy, 2) identify meaning making processes in the selected regions, and 3) identify the mechanisms that trigger the meaning making process. The complexity of this model is partly a result of viewing it as an extension of the time-place matrix of groupware [15], with a dimension of evolving shared understanding as informed by Rommetveit’s theory of intersubjectivity. This is our main analytic tool for understanding intersubjectivity as a dynamic, social phenomenon in knowledge-intensive distributed work. The development of shared understanding within a group is a process represented as an emergent object of knowledge issued by an individual and moving along the x-axis, which may take a path on its own (z-axis) and involving other individuals if it is caught on, from an incomplete or vague utterance not yet fully understood, gradually leading to clarification and shared understanding based on feedback and follow up utterances from multiple individuals. This process may involve spatial, temporal, and interpersonal displacements as depicted in Figure 3. Furthermore, two aspects of time are represented in the coordinate system: time associated with duration of work and relative geographical location, i.e. clock time (y-axis), and time associated with past, present, and future (x-axis). It entails connecting past and future with the present by expanding from the ‘here and now’ towards the (predicable) future (here, afterwards) and the (remembered) past (here, before), i.e. expansive time. Finally, the coordinates ‘I’ and ‘You’ in Rommetveit’s original framework have been replaced by ‘Initial incomplete utterance’ and ‘More developed utterance,’ respectively (z-axis). This framework is later in the paper applied in the analysis of conversations among peripheral and core developers in the mod_perl mailing list. It is the temporal and interpersonal dimensions of the framework that we focus on. We did not collect data on the developers’ geographical location, so the spatial dimension is outside the scope of our work. We used UCINET [4] to carry out network-level calculations. E-mail conversations were content analyzed to highlight processes of intersubjectivity and shared prolepsis using thematic coding with Nvivo [23], a software package for qualitative data analysis. Data Selection and Strategies for Analysis Since Gisle Aas released the first version of mod_perl on March 25, 1996, thousands of individuals have contributed to its current state of development (http://perl.apache.org/about/history.html). All data used was obtained from the mod_perl project mailing lists related to the development of the mod_perl core, which are publically available as a text-based archive from http://mail-archives.apache.org/mod_mbox/perl-dev/. In order to strike a balance between a manageable data set and a rich data set, we decided to include all members active within a timeframe of six months. The study period lasted from 01.01.2008 to 30.05.2008 and included 215 individual contributors. The number of postings during this period was 1,154, with a mean of 6.2 postings per contributor. Research Questions Based on the above survey of previous work and the development of a new framework for intersubjectivy in online communication, we address the following questions regarding the intersection of OSD and intersubjectivity: The first author, then a PhD student, observed the activity in the mailing lists as it occurred, reading through previous postings and following the discussion threads without disturbing the ‘naturally occurring’ communication within the project. With respect to the transparency and public nature of the mod_perl project, we decided to send an informative letter introducing ourselves and explaining the study’s purposes [11]. None of the authors have any affiliation with mod_perl, though the first author has some knowledge of Perl programming and has installed an Apache web server. • How is intersubjectivity achieved in the mod_perl community and what role does prolepsis play? • How can SNA and content analysis complement one another in the analysis of prolepsis-driven intersubjectivity? • How is learning and knowledge development manifest in the OSD community? RESEARCH METHOD In the following section we will provide a rationale for our research design and discuss the strategies we used for collecting and analyzing our data. At the first descriptive level of analysis, formal measures associated with SNA [28, 32] were applied to render the overall communication among participants in the mod_perl project visible. The main focus at this level of analysis was to organize the message archive into sub-groups to serve as Rationale for the Mixed Methods Approach We used a mixed methods approach to address our research questions because of interdependencies between structural 134 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA Network-Level Data contrasts for a more detailed content analysis of individual email messages. We selected communicational ties among contributors on the mod_perl mailing list for users and developers as relation and the basic units for constructing the network. All communicational ties among the selected nodes were studied to provide a census for a full network analysis. The adjacency matrix generated from UCINET included information about each participant’s communicational ties with all other participants in the population. The rationale behind this selection was to collect information about all communicational dyads in the network and to construct an inclusive overview and provide parameters to calculate centrality and density in communication, which again determine cohesion. In this way, sub-groups of core and peripheral contributors were identified for further content analysis. The density score of the network is 0.0127, which indicates rather low overall cohesion (a value closer to 1 would be considered high). This means that relatively few participants in the community directly and frequently communicate with each other. However, network density provides only a rough estimate of cohesion in the overall network, and centrality measures provide additional information. The degree centrality [10] shows the overall network activity of individuals, as the number of ties incident upon a node or “the sum of each row in the adjacency matrix representing the network” [4]. Two of the participants with the highest degree centrality score are listed in Table 1 below. Member We constrained the space further by selecting among four possible core-periphery dyads: 1) periphery-periphery, periphery-core, 3) core-periphery, and 4) core-core. We chose periphery-core communication as our object of study as it raised the most interesting research questions for us (heterogeneous community, boundary crossing, multiple roles, and user participation in software development). Furthermore, we made a purposeful sampling of the total data set of 215 contributors, reducing the number to five core contributors and 10 peripheral contributors. The rationale for the reduction is to carry out a qualitative analysis to address our research questions within a reasonable time frame (PhD thesis). In the next section we explain in more detail the steps we went through in order to select representative sub-groups of core and peripheral contributors and the number of contributors in each group. OutDeg InDeg NrmOutDeg NrmInDeg dev-12 62 40 28.972 18.692 dev-29 29 19 13.551 8.879 Table 1: Freeman’s degree centrality measures [10] indicate the most active participants in the mod_perl community. The above degree centrality values indicate that dev-12 is the most central member of the mod_perl community. An Out/InDegree of 62/40 means that dev-12 has established 62 communicational ties (the number of postings) and received 40. In an asymmetric expert network such as the mod_perl community, a high InDegree may indicate prominence in the sense that other members consult the member more often. High OutDegree centralization could mean that the participant undertakes the majority of consulting in the project. The mailing lists consisted of post and reply messages. Our focus on interaction led us to ignore posts that were not responded to. The presence of interaction requires a minimum of one post and reply sequence, and this was coded as ‘1,’ whereas non-interacting dyads were coded ‘0.’ Consequently, messages that were not responded to were unaccounted for in our analysis. This left us with a standard asymmetric binary network for performing basic matrix operations. To obtain a visual impression of the prominence and influence of dev-12, the sociogram ego network of dev-12 depicts a relatively clear star structure, which indicates that this developer is one of the most active members in the community. From this diagram, one can find which other developers he interacts with frequently (e.g., dev-29 and dev-81 in the top-left region). The sociogram can also tell us what other developers dev-12 interacts with frequently, but it does not tell whether the nodes also cluster in subgroups (cliques) around the structural center, information we need to zoom in on potential regions of intersubjectivy. DATA AND RESULTS We organized our empirical results into two parts: 1) network-level data [28, 32] and 2) interaction data [16]. We computed density, centrality, ego networks and subgroups within the network to identify regions of high and low cohesion. We analyzed the processes of intersubjectivity within selected high cohesion sub-groups and the roles of shared prolepsis by zooming in on interactions among participants in selected sub-groups, focusing on the content of their messages and identifying proleptic instances in the message flow, and studied the impact the proleptic instances had on the conversation. We tested multiple values for group size to identify high cohesive sub-groups that could be used for content analyses. The following four sub-groups were found for N=5. This was the smallest group size where all members had ties to each other: 1: dev-12, dev-29, dev-40, dev-81, dev-92 2: dev-12, dev-29, dev-40, dev-81, dev-144 3: dev-12, dev-20, dev-22, dev-23, dev-65 4: dev-3, dev-12, dev-38, dev-39, dev-40 135 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA of 0.053 indicates relatively few ties between the periphery and core. These four sub-groups partially overlap, and their relationships are displayed in Figure 4. Group Members Nucleus N=4 dev-12, dev-29, dev40, dev-81 Peripheral N=10 dev-19, dev-30, dev-33, dev95, dev-131, dev-142, dev-153, dev169, dev-186, dev-207 Figure 4. Sub-group sets showing central participants’ comemberships. Characteristics High centrality and density. Densely knit around the structural center of the network. Frequent contributors to the project mailing list. Random participants with low centrality and density measures. Sparsely knit in the peripheral regions of the network. Infrequent contributors to the project mailing list. Table 2. Contrasting features of nucleus and periphery in the mod_perl network. dev-12 and dev-40 share membership in three of the four groups, and dev-12 and dev-81 in two of the four groups. On the other hand, dev-3, dev-20, dev-22, dev-23, dev-38, dev-39, dev-65, dev-92 and dev-144 are more isolated from the rest, since they belong to only one of the sub-groups. Given that the most central nodes in the network are comembers in clearly defined sub-groups, in particular dev12, dev-29, dev-40 and dev-81, we refer to them as the nucleus group in the content analysis that follows. Interaction Data In the following section, we analyze the content of selected messages submitted between nucleus group participants and peripheral group participants. They were selected on the basis of involving prolepsis-driven intersubjectivity (constructing shared understandings initiated by vague or incomplete utterances that catch on). The organization of this section is as follows: first, we present a sequence of “raw” interaction data, followed by a brief explanation in “common sense” terminology before we discuss it with the conceptual framework we developed in the theory portion of this paper. The extracts show a sequence of post and reply messages spanning five days. They start with a posting by a peripheral participant (dev-186), who describes difficulties installing an Apache server, proposes a possible solution, which is followed by a question. This triggers a “sense making” discussion involving two members of the nucleus group (dev-12 and dev-81). Extract 1 is displayed below: We adopted a similar technique (purposeful sampling) for identifying a peripheral group to serve as a contrast to the nucleus. However, as there were many more peripheral than core participants in the total population (202 vs. 13), we had to first reduce the data set. We opted for a practical solution and randomly extracted ten participants from the total population of peripheral participants. UCINET software was applied to categorize the participants in the two classes according to the core-periphery model [4]. The ten randomly selected participants in the peripheral group are nodes with low centrality and density measures (low cohesion). In comparison with the nucleus group, they contribute and interact infrequently to the project’s mailing lists. The two contrasting (nucleus and peripheral) groups’ distinguishing features are displayed in Table 2. In the core group we obtained an intragroup (core group with its own group) with a density score of 0.436 and an intergroup (core group with other groups) score of 0.053, whereas similar scores for the peripheral regions were much lower: 0.070 and 0.005, respectively. A density value of 0.005 indicates a very low prevalence of ties among peripheral members in the network (low cohesion), whereas a density value of 0.436 indicates frequent direct communication among core members (high cohesion). Almost half of the possible ties are present. Finally, a value 136 [Line] Message from dev-186, 13 Feb 2008 14:09 GMT 1 Hello @all, 2-4 we are just transferring our Intranet Server which used Apache 1.3 and mod_perl1 from SLES8 to SLES10 with Apache2 and mod_perl2. 5-7 The switch has to go quick and we can’t redesign all of our old mp1 programs to mp2, so we try to use the compat module. But I have no luck with it. 8-11 I tested my config with SELS10, Opensuse 10.3 and Ubuntu but i hit every time the same problem. I searched the Mail Archive and the Documentation, but no luck. Source Work: Social Factors in Software Development 12 [code] 13 I get the following output... 14 Software error: 15-16 Can't locate object method "request" via package "Apache" at /usr/lib/cgi-bin/test1.pl line 4. 17-19 For help, please send mail to the webmaster (webmaster@localhost), giving this error message and the time and date of the error. 20-22 It seems, he can’t find the requested methods... How can i validate that the compat module is loaded and working? February 23–27, 2013, San Antonio, TX, USA number of receivers. This proleptic communication may not provide the necessary pre-understanding for establishing intersubjectivity across the network, and a selective response should be expected either from one of the core members or perhaps from another peripheral member who has been in a similar situation before. A member of the nucleus group (dev-81) replies a few hours later, on the same day. It appears that he has experienced a similar problem with his own server installations. Extract 2 below is a reproduction of the reply: When switching to an upgraded intranet server, dev-186 experiences difficulties with the earlier mod_perl programs (mp1, line 6). Upon searching the project’s message archive and documentation (line 10), he found no adequate solution to the problem. In order to assist mod_perl peers in understanding the situation, a piece of code is posted as an attachment to the message (line 12) and the software error that appeared when he attempted to start the server is displayed in lines 15-19. He proposes a tentative solution (“It seems, he can’t find the requested methods,” line 20) before asking a question at the end (line 20-22). Following Rommetveit’s [25] framework for intersubjectivity, the sender of utterances is in a privileged position to control what features of the shared object will “enter the field of shared attention” [25, p. 95]. If we interpret source code as the main artifact attended to in the mailing list, the message sent by dev-186 points out a specific feature of the artifact in an effort to penetrate the shared space, namely the “compat module” (lines 7 and 21), which is used when transferring server programs written for mod_perl 1.0 to mod_perl 2.0. In the situation description, he also refers to several different server OS (SLES10, Opensuse 10.3 and Ubuntu) that are not part of the mod_perl bundle. However, a partially shared understanding of how these systems integrate with mod_perl may be important to make sense of dev-186’s posting. What shared background information can be assumed or presupposed by the receivers of this message? Dev-186 explains that prior to sending the message he searched the mail archive and the project documentation (line 8-11). The previous postings stored in the mail archive and project documentation can be seen as knowledge resources that are mutually accessible to the participants. [Line] Message from dev-81, 13 Feb 2008 21:49 GMT 1 Hi. 2 About your problem below… 3-5 I am not sure that this is going to help, nor even if it is really relevant to your specific problem. Just trying to give you ideas, because it reminds me of something. 6-16 Below are two configurations, of two of our systems, which have slightly different versions of mod_perl and the rest. I went through the same kind of problems you’re going through, a while ago when going from Apache 1.x to Apache 2.x, and again more recently when there were some changes in the mod_perl naming of modules between mod_perl 1.9xx and mod_perl 2.0, and then some additional things happened during a Linux Debian upgrade from Sarge to Etch, which changed the location where some of the Apache2 stuff was installed. 17-19 And there is still some differences to this day in some Perl scripts or modules between these two systems, which greatly bothers me, but it basically works. 20-23 The main difference is in the two first “use” statements in our “startup.pl” script (which I assume you are familiar with, as a way to pre-load some Perl stuff when the server starts). 24-26 I do not remember precisely which problems we had, but they were of the same “general gist” as yours, so maybe this helps. [code] Dev-81 indicates a shared understanding of dev-186’s description of a problem, but his understanding appears at first vague and suggestive (“it reminds me of something,” line 5). Dev-81 continues the reply by comparing two server configurations at his site (lines 6-16), referring to a problem in the mod_perl module when the server was upgraded to a newer version (lines 8-12), and eventually pinpointing the problem by comparing the startup script in the two systems (line 20). He refers to a script called “startup.pl,” (line 21) which dev-81 assumes is part of dev186’s pre-understanding (lines 21-23). However, there is no further specification of these differences and the question is left open (e.g. no response from the main recipient, dev186). This is an example of a proleptic instance; it is well formulated but lacks detail and begs for completion (lines 20-23, “use” statements in our “startup.pl” script). It caught the attention of the nucleus group, and the sense-making discussion that followed over the next days is shown in As the initial message is addressed to all mailing list subscribers (“Hello@all,” line 1), the potential ‘listeners’ or receivers of utterances are numerous. They are scattered around the entire developer network, ranging from peripheral to nucleus participants. It follows that ‘the listener’ invited into a partially shared intersubjective space is not a single individual, as in Rommetveit’s framework, but rather a relatively large group of peers (developers and web server administrators). This implies that the sender somehow has to adapt the message to a potentially large 137 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA 7-8). It appears that the function of the ’use’ statement is not sufficiently understood by the experts, and a temporary communication breakdown occurs. Another central nucleus group member (dev-12) attempts a repair sequence, suggesting that this is only a wording mistake by dev-81 (line 13), but dev-81 turns down dev-12s interpretation of his understanding. Dev-81 admits some “incomplete personal understanding” (lines 19-20) before elaborating using another example (lines 22-32), then proposing a tentative solution to the original problem raised by dev-186 (lines 33-38). At the end of the excerpt, dev-81’s explanation is approved by dev-12, who is one of the most central members of the nucleus group. At this stage of knowledge development the shared understanding seems to have been brought to a new level, elaborating upon the previous incomplete information from dev-81’s earlier attempts. The new understanding extends from what was partially known in advance and thus provides both an expansion of the intersubjective space and a clarification of the partly misunderstood feature [24, 25]. Extract 3. It involves two nucleus members (dev-12, dev81) and one experienced peripheral member (dev-19): [Line] Reply from dev-81, 13 Feb 2008 22:00:56 GMT 1-5 Of course, what I forgot to mention below - and sorry if you know that already - is that whichever perl modules you pre-load in your main Apache server config via the startup.pl script, you do not need to "use" anymore in all your perl scripts or Apache/mod_perl handlers. [..] 6 Reply from dev-19, 14 Feb 2008 14:13 7-11 This is the first time that NOT using "use" because it was preloaded is mentioned. In fact, how would the modules compile (while testing for example)? Wasn't preloading supposed to do the exact opposite? [..]I'm at a loss here. 12 Reply from dev-12, 14 Feb 2008 15:51:22 GMT: 13-16 [..] I think that was just a wording mistake. When the process forks, the loaded modules are shared by the operating system's copy-on-write feature. I believe that's what he was trying to say. 17 Reply from dev-81, 18 Feb 2008 09:40 18-21 Well, it was not exactly what I was trying to say, but undoubtedly this reflects my incomplete personal understanding of the process anyway, and I stand ready to be educated. 22-32 What I have until now believed is that perl "code" is in fact "data" for the perl interpreter, and that as such it cannot really be "shared". What I mean is that, as soon as some bit is changed in a "page" of any perl module, that "page" is dirty and must be copied and made private to the one child process. And since there is (in my understanding) not such a clear separation as to which parts in "perl code" are data and which are code, after a while one ends up with a full duplicate in each child anyway. Probably badly explained, but not so in the general sense? 33-38 P.S. What I really meant originally, is that if the speed to make it work was of the essence, it might be easier to (find/grep) and remove the various use Apache-x() from the multiple modules or cgi scripts, and put them all in the startup script. Then later one could go back and refine things, if it makes a difference. What we have described here is a process of constructing shared understanding for the run time configuration (installment) of a web server (Apache) using a logic that is different from programming and documentation (hierarchical organization), despite similarities for obvious reasons. It is a more dynamic and serendipitous process that involves the evolution of a shared object of attention, an ‘object of knowledge’ [11]. It starts with an uncompleted description of a difficult situation with a server upgrade as experienced by a peripheral contributor (involving problem formulations, knowledge sharing, and solution proposals), who describes the situation to the best of dev-186’s knowledge, which is responded to by a nucleus member (dev-81), who ‘fills in’ uncompleted information by suggesting a more appropriate description of the problem in terms of the startup script, followed by a critical remark by dev-19, which causes a temporary breakdown regarding their understanding of how Perl modules are included in the system during compilation, later resolved by the nucleus members, thus refining the object of knowledge and expanding intersubjectivity among the group of OSD practitioners. 39 Reply from dev-12, 18 Feb 2008 19:55:03 GMT In summary, the main findings from this analysis are: 40-43 Your technical understanding is correct, but in practice most pages remain shared. You can help this by using a tool like Apache::SizeLimit that kills off processes after a while. • Network-level findings: Large group of peripheral members compared with core members and relatively few interactions between periphery and core (compared with intragroup interactions in the core). In the above interaction sequence, the general and intended functionality of the startup.pl script’s ‘use’ statement is the object of shared attention. With reference to the follow-up posting by dev-81 to dev-186 (“”you do not need to ‘use’ anymore in all your Perl scripts,” lines 4-5), dev-19 disagrees with dev-81’s attempted explanation of how the ‘use’ statement works (“This is the first time that NOT using “use” because it was preloaded is mentioned," lines • Interaction findings: Two types of intersubjectivity: 1) among core members and 2) among peripheral and core members. We have assumed that core members draw heavily on a repertoire of common pre-understanding, whereas peripheral members enroll in the same common ground in a more rudimentary fashion. Previous studies have shown that peripheral members can enroll in the common ground by involvement in work tasks that draws 138 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA competent practitioners, the initial purpose of the communication thread seemed to change. In the second wave of replies, shown in Extract 3, the sense-making discussion moved beyond the initiation by dev-186. This is supported by the fact that dev-186 never returned to the thread with follow-up postings. We do not know whether dev-186 successfully installed the server, or if additional help was needed. them to the center [5]. Based on our findings we can add another path to enrollment: being immersed in practical matters in a local setting of shared concern for a developer community, and using the community as a resource for collaborative problem solving when difficulties arise. • Interaction findings: Proleptic instances appear twice: 1) An initiation phase where what is presupposed is surfaced but not necessarily picked up and acknowledged (extract 1), which may lead to partial understanding that is sufficiently clear to involve other more competent members in an 2) explication phase (extracts 2-3) that may create a new shared understanding within the group. Rommetveit argues that intersubjectivity should be approached as an expansive process of creating and maintaining a temporary shared social reality among a small group of communicators. This expansive process occurs along multiple dimensions (spatial, temporal and interpersonal), which we elaborate below based on our data (we do not have data on spatial expansion as we did not ask our informants about their whereabouts). GENERAL DISCUSSION In this section, we discuss our findings, organized according to our research questions, using our theoretical framework, and by comparing our findings to those reported in the literature. Temporal expansion Temporal expansion means to reach backward and forward in time (towards shared knowledge and presuppositions on one end and common goals and application of the knowledge on the other). How is Intersubjectivity Achieved by Shared Prolepsis Intersubjectivity is a term we use to conceptualize the psychological relationship between people during conversation, in order to build common understanding within a group and to reach goals. The goals may diverge across sub groups (e.g. improving server software vs. installing a web server) and the individuals may not know each other well in advance. We developed a framework for the analysis of empirical data based on Rommetveit’ theory of prolepsis-driven intersubjectivity, proposed for face-toface communication [25]. In Rommetveit’s original work on prolepsis-driven intersubjectivy, temporal expansion seems to be the main dimension for increasing intersubjectivy and shared prolepsis being the main mechanism. It is initiated by somewhat vague (not always identifiable) anticipatory cues inserted in a conversation. These cues must refer to common preunderstanding to succeed (be picked up by more experienced participants), and should point forward to a new event of common interest: the meeting of three people in Rommetveit’s example in the beginning of the paper, and successfully installing an upgraded Apache web server in our case. The common preunderstanding relates to previous postings in the mail archive and the mod_perl project documentation, general knowledge of server operating systems (SLES, Opensuse, and Ubuntu in extract 1), and the role of the ‘use’ statement in the ‘startup.pl’ script (extracts 2). This preunderstanding is not fully shared by dev-186 (the peripheral participant who triggered the conversation), and the subsequent elaboration (extract 3) indicates that information for understanding the situation was missing. Our findings indicate that many of the postings in the mod_perl mailing list are incomplete as problems, as they include problem descriptions, solution proposals, and ideas for understanding, i.e. question and answers, in the same formulation. In some situations they serve as proleptic instances, triggering prolonged discussions, and at other times they generate little or no response. This suggests a two-step process of shared prolepses: 1) initiation (sending a potentially interesting message, a pre-proleptic instance) and 2) explication (a proleptic instance vigorously followed up by joint problem solving and knowledge sharing). For explication to be possible, the pre-proleptic instances need to hit upon the ‘hot spots’ of shared attention in the community (at least one member in the nucleus group must find it interesting). Our analysis provides tentative evidence that peripheral participants reporting difficulties will not yet be carriers of the culturally taken-for-granted knowledge that makes prolepsis effective for enrolling outsiders in the project. Our findings indicate that the former have, first of all, a pragmatic role in providing pre-proleptic instances that serve as the ‘raw material’ or triggers for more competent practitioners to construct instances of shared prolepsis. The nucleus group members continue with a discussion of the ‘use’ statement, which dev-81 assumes is part of the community’s common pre-understanding (“which I assume you are familiar with,” line 21-22, extract 2). 11 minutes According to Clark and Brennan [6], two key factors shape grounding processes: purpose and medium. Purpose means that a sufficient condition for continuing the collective activity is that communicators mutually believe that the receivers “have understood what the contributor meant to a criterion sufficient for the current purpose” [6]. In the case of dev-186, reply messages provided by nucleus group members were perhaps sufficient to solve the problem of installing an upgraded Apache server. However, as the message thread continued into a discussion among more 139 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA last reply to dev-81 (lines 40-43 in excerpt 3), may perhaps be significant for sustaining participation in OSD projects like mod_perl. later (line 1-5, extract 3) dev-81 becomes unsure if dev-186 has understood what was meant, and perhaps is unsure about his own understanding, and follows up with more information and excuses himself for stating the obvious (“sorry if you know that already”, line 1-2, extract 2). Later on, dev-81 fine-tunes his understanding based on feedback from two experienced developers, dev-19 and dev-12 (clarifying their understanding of how the ‘use’ statement works and how Perl programs load). In the final reply message, dev-12 provides positive evidence that the procedure was understood sufficiently for the current purpose. The information can thus be added to the group members’ common ground. Our study would have benefited from interviews with the above-mentioned developers (dev-186, dev-81, dev-19 and dev-12) to verify our claims. Interviews can dig deeper into the meaning of individual postings and obtain valuable feedback for supporting and/or refuting our interpretations of informants’ text-based (public) conversations. This was not done in our case due to lack of time and for practical reasons (access to emails addresses). Future work ought to consider interview as a part of a mixed methods approach. Combining SNA and Content Analysis Interpersonal adaptation We approached the social phenomenon of the mod_perl community by choosing methods from the social sciences and applied mathematics (social network analysis and content analysis). These methods allowed us to identify high cohesion subgroups within the mod_perl community and study their interactions in some detail. We narrowed the scope further to zoom in on periphery-core communications, mainly for practical reasons (interest of the research team and constrained by the timeframe of a PhD thesis). Intersubjectivy was identified as an expansive process in central regions of the network in an effort to create common understanding of some topic of concern to the community of Perl developers and Apache web sever administrators. Our total data set covers a period of 6 months of mailing list discussions in the mod_perl community, whereas the data extracts chosen for content analysis had a much shorter time frame (5 ¼ days). Limitations of our work can be attributed to the short duration of the quantitative analysis and our purposeful sampling of the data set to reveal one detailed example. Further work ought to explore multiple data sets for comparison and include larger time spans for qualitative analysis (longitudinal study), to see if there are interdependencies of working, learning and knowledge development, and other characteristics of prolepsis-driven intersubjectivy that we have not yet identified. Interpersonal adaptation means to adapt to the role of the person one communicates with (sender, receiver, peripheral member, core member, etc.) and to the type of their utterances (vague, incomplete, elaborated, clarified, etc.). Our network data shows there were few interactions among periphery and core members, which can partly be explained by peripheral members using the mod_perl community as a resource for problem solving when difficulties arise with a web server installation. The interaction data indicates that the peripheral participant (dev-186) did not know the audience well (e.g. “Hello @all,” extract 1), whereas nucleus members and frequent contributors were more aware of whom they communicated with, and they adapted their messages differently across the conversation thread The contributors reveal attitude towards the others based on the tone of their utterances. Dev-81 and dev-12 (two of the most active members in the mod_perl community) makes the greatest effort of adapting responses to the perspective of the person they communicate with (dev-81  dev-186; dev-12  dev-81). Dev-19 is a somewhat at odds with the two nucleus members when he criticizes dev-81 for having an understanding different from his own (line 7-11, extract 3). Dev-19 seems to be knowledgeable about web server configurations, but he is only a peripheral contributor in the mod_perl community. His critical attitude created a temporary breakdown and helped resolve the situation by triggering a final round of ‘mutual education’ (“I stand ready to be educated,” dev-81 in line 20-21 in extract 3) and (“Your technical understanding is correct,” dev-12, line 40, extract 3). These message sequences show a somewhat bumpy road of proposing explanations, elaboration, friction, and clarification, leading eventually to success by increasing the group’s (and possible their numerous listeners’) intersubjectivity. The development of shared understanding is illustrated in the extracts as an emergent object of knowledge gradually capturing shared attention. It starts by dev-186 pointing out in extract 1 a specific feature of a server configuration that does not work as a result of an upgrade, in an effort to catch the attention of more experienced members of the community who can help, and penetrate the shared space. According to Matusov [20], processes of intersubjectivity transcend specific and time-limited joint problem solving and are “not only the basis and derivative of the joint activity but also the social glue of different sociocultural activities” [20, p. 30]. Praising peers, seeking education (line 20-21 in extract 3), and validating another person’s understanding of a shared object, as indicated by dev-12’s The original formulation issued by dev-186 is a preproleptic instance, which is responded to by dev-81 in extract 2 and turned into shared prolepsis by a meticulous explanation of certain parts of the system. The reformation by dev-81 turns out to be incomplete (partly “wording mistake,” line 13, extract 3, dev-12; and partly “incomplete personal understanding,” line 19-20, extract 3, dev-81). It Informal Learning and Knowledge Development 140 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA Previous organization studies in OSD report that artifacts (source code, patches, downloads, tools, documentation, mailing lists, etc.) are the main source of organizing tasks and joint work [17]. Looking ahead, we envision that a vast number of peripheral participants in the network will use social media and ‘feed’ the central core with pre-proleptic instances (potential problems, new ideas, etc.). As pragmatic elements in a ‘collective learning system,’ the swarm of actors on the outer edges monitors the evolution of the shared knowledge object and contribute ‘raw material’ that might trigger further development cycles in the central region. generates vigorous activity by three experienced developers, and further clarification becomes necessary. In the final stage of knowledge development the shared understanding seems to have been brought to a new level, elaborating upon the previous incomplete information from dev-81’s earlier attempts. The new understanding extends from what was partially known in advance and thus provides both an expansion of the intersubjective space and a clarification of the partly misunderstood feature. However, it is not clear from our data whether or not the shared understanding created within the nucleus group also reached out to dev-186 who first raised the concern as well as the other peripheral participants who listened in on the conversation. A follow up interview with dev-186 would be able to answer this question. This was outside the scope of our work. Schön’s notion of problem setting [26] is arguable the technical term that comes closest to capture the meaning of shared prolepsis. By problem setting it is meant essentially a non-technical process of deciding the boundaries to be investigated, naming the problem (if applicable), and framing the context in which it will be approached. They are not problems be solved in the outset, but problematic situations characterized by uncertainty, disorder and indeterminacy [26]. The literature on OSD describes paths for newcomers to become (peripheral and core) contributors to the activities taking place in the OSD communities. For example, Cataldo and Herbsleb [5] found that peripheral members’ involvement in work tasks that draws them to the center provides a path for enrolling newcomers in OSD projects [5]. Edwards [9] suggests that situated learning and apprenticeship (techniques from community of practice) do apply to OSD projects. Our own study corroborates previous findings of socialization, but we did find a clear demarcation between core and peripheral members in terms of participation, which could indicate a steep learning curve for newcomers or a different motivation for involvement, but further work ought to follow this up in more detail. Presenting a problem may be one possible path to enter into the community, but a problem-proposer also faces the risk of offending others as a result of a poorly formulated question. Implications for Design We can speculate on implication for design based on our empirical work by comparing (uttered) proleptic instances to (designed) prompts, and ‘sentence openers’ in Webbased inquiry environments to trigger reflections, discussions, and further development. Furberg [12] reports on a qualitative study of high school students’ engagement with a web-based inquiry environment system for science education aimed at prompting student reflections on processes of scientific inquiry. Her findings were that students often copied and pasted (e.g. verbatim reuses) the prompts provided by the system or by the teachers in their own work when they were told to rewrite the prompts in their own words. Her findings point to the fact that engaging in prolepsis-driven intersubjectivity is not straightforward for newcomers as it requires identifying cues, understanding their meaning and reformulating them in a language that resonates with shared knowledge within a given community. Prolepses vs. Problems The notion of shared prolepsis can be compared with the notion of problem as part of a problem solving process, but they are not the same. First, prolepsis appears in two facets, pre-proleptic and shared prolepsis. Shared prolepsis is closer to a problem description than a pre-proleptic instance, as the former is tuned into a path that may lead to cooperative problem solving. Second, proleptic instances should be thought of as embryonic solutions, capturing both problem setting and skeleton solution. A proleptic instance is better described as a ‘seed.’ A seed contains a ‘mini solution’ for its own development. When it is planted in fertile soil it will grow (expand) in two directions: downward (roots) and upward (stem, leaves, etc.). In the same fashion a proleptic instance stimulates growth of knowledge in two directions: backward in time (connecting personal knowledge with shared understanding) and forward in time (applying the knowledge to solve a problem or reach a goal). SUMMARY AND CONCLUSIONS We studied communication patterns and analyzed message exchanges between central and peripheral participants in the mod_perl open source software development project for a period of six months, showing how constructing a shared understanding is a subprocess of complex distributed work in this community. A framework for prolepsis-driven intersubjectivity constituted the theoretical perspective for understanding this process. We used social network analysis to identify and visualize high-cohesion sub-groups and interaction analysis to analyze conversations (contents of messages). 141 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA conference, Vol. 6, IEEE Computer Society (2006), 118125. A key mechanism in the framework is the notion of shared prolepsis, which serves as a catalyst (a ‘seed’) for joint meaning making among competent practitioners. However, for pre-proleptic instances to progress into proleptic instances that will expand the project’s shared understanding, the study suggests that the potential problems and solution ideas shared by peripheral participants need to be reformulated by the more competent, central developers, to serve as building blocks for a shared understanding that can be properly introduced into the intersubjective space (i.e. aligned with the community’s common pre-understanding. This implies that pre-proleptic utterances are often vague and sometimes offtrack, suggesting some lacking feature in the knowledge object that may invite reformulation as proleptic instances, which again may evolve into productive discussions among more competent peers. 9. Edwards, K. Epistemic Communities, Situated Learning and Open Source Software Development. In Proc. Workshop on Epistemic Cultures and the Practice of Interdisciplinarity. NTNU, Trondheim (2001), 11-12. 10. Freeman, L. C. Centrality in networks: Conceptual clarification. Social Networks 1, 3 (1979) 215–239. 11. Fugelli, P. (2010). Intersubjectivity and objects of knowledge: Making sense across sites in software development. PhD Thesis, Department of Educational Research, University of Oslo. 12. Furberg, A. Socio-cultural aspects of prompting student reflection in Web-based inquiry learning environments. Journal of Computer Assisted Learning 25 (2009), 397– 409. 13. Gnu crossing in East Africa, Battle at the Mara river. http://www.youtube.com/watch?NR=1&feature=endscre en&v=Mkg0E8c71IQ . ACKNOWLEDGMENTS We thank the Faculty of Education, University of Oslo for funding the PhD project of the first author, and we thank our colleagues in the ‘Professional Learning in a Changing Society’ (ProLearn) project for providing comments and advice along the way. The authors thank the anonymous CSCW’13 reviewers for constructive, elaborated comments on two previous versions of this paper. 14. Hemetsberger, A. and Reinhardt, C. Collective development in open-source communities: An activity theoretical perspective on successful online collaboration. Organization Studies 30, 9 (2009), 9871008. 15. Johansen, R. Groupware: Computer support for business teams. The Free Press, New York, 1988. REFERENCES 1. Apache Software Foundation. http://apache.org/. 16. Jordan, B., and Henderson, A. Interaction analysis: Foundations and practice. The Journal of the Learning Sciences 4, 1 (1995), 39-103. 2. Berger, P. L. and T. Luckmann. The Social Construction of Reality: A Treatise in the Sociology of Knowledge. Anchor Books, Garden City, 1966. 17. Lanzara, G.F. and Morner, M. Artifacts rule! How organizing happens in open source software projects, in B. Czarniawska and T. Hernes (eds.), Actor-Network Theory and Organizing. Liber & Copenhagen Business School Press, Copenhagen (2005), 67-90. 3. Bergquist, M. and J. Ljungberg (2001). The power of gifts: organizing social relationships in open source communities. Information Systems Journal 11, 4 (2001), 305-320. 4. Borgatti, P., M. G. Everett, et al. UCINET for Windows: Software for Social Network Analysis. Analytic Technologies, Harvard, MA, 2002. 18. Lee, J.S., Tatar, D. and Harrison, S. Micro-coordination: because we did not already learn everything we need to know about working with others in kindergarten. In Proc. CSCW 2012, ACM Press (2012), 1135-1144. 5. Cataldo, M. and J. D. Herbsleb. Communication networks in geographically distributed software development. In Proc. CSCW 2008. ACM Press (2008), 579-588. 19. Long, Y. and K. Siau. Social network structures in open source software development teams. Journal of Database Management 18, 2 (2007), 25-40. 6. Clark, H. and S. Brennan. Grounding in Communication, in L. B. Resnick, R. M. Levine and S. D. Teasley (eds.), Perspectives on socially shared cognition. American Psychological Association, Washington DC (1991), 127-149. 20. Matusov, E. Intersubjectivity without agreement. Mind, Culture, and Activity 3, 1 (1996), 25-45. 21. Matusov, E. Intersubjectivity as a way of informing teaching design for a community of learners’ classroom. Teaching and Teacher Education 17, 4 (2001), 383-402. 7. Clark, H. and E. F. Schaefer. Contributing to discourse. Cognitive Science 13 (1989), 259-294. 22. McCarthy, J.C, Miles, V.C, and Monk, A.F. An experimental study of common ground in text-based communication. In Proc. CHI 1991, ACM Press (1991), 209-215. 8. Crowston, K., K. Wei, et al. (2006). Core and Periphery in Free/Libre and Open Source Software Team Communications. In Proc. 39th Annual HICSS 23. Nvivo software package, www.scienceplus.nl/Nvivo 142 Source Work: Social Factors in Software Development February 23–27, 2013, San Antonio, TX, USA 29. Shami, N.S., Erickson, T. and Kellogg, W.A. Common Ground and Small Group Interaction in Large Virtual World Gatherings. In Proc. ECSCW 2011, Springer: London (2011), 393-404. 24. Rommetveit, R. On message structure: A framework for the study of language and communication. Wiley, London, 1974. 25. Rommetveit, R. On the architecture of intersubjectivity, in R. Rommetveit and R. M. Blakar (eds.), Studies of Language, Thought and Verbal Communication. Academic Press, New York (1979), 147-161. 30. Stahl, G., Koschmann, T., and Suthers, D. Computersupported collaborative learning: An historical perspective, in R. K. Sawyer (ed.), Cambridge handbook of the learning sciences. Cambridge University Press, Cambridge, UK (2006), 409-426. 26. Schön, D.A. The reflective practitioner: How professionals think in action. Basic Books, New York, 1983. 31. The Richmeister sketch, “Copy machine.” http://www.nbc.com/saturday-night-live/video/copymachine/1353535/. 27. Schütz, A. (1962). Collected papers I: The problem of social reality, with M.A. Natanson and H.L. van Breda (eds.). Martinus Nijhoff Publishers, Dordrecht, The Netherlands, 1962. 32. Wasserman, S. and Faust, K. Social network analysis methods and applications. Cambridge University Press, Cambridge, UK, 1994. 28. Scott, J. Social network analysis: A handbook. Sage, London, 2000. 143