Situating Cognition

All correspondence concerning this paper should be addressed to Wolff-Michael Roth, Lansdowne Professor, Applied Cognitive Science, Faculty of Education, University of Victoria, Victoria, BC, Canada V8W 3N4.
E-mail: mroth@uvic.ca.
Tel: 1-250-721-7885
FAX: 1-250-721-7767

In this article, I describe an epistemological framework and an associated research method that makes it useful to speak of cognition as a process situated in the material and social world. This framework takes the agent-in-setting as unit of analysis. The recurrent patterns or structure in the agent-in-setting that we associate with cognition occur at different time scales, and can be attributed somewhere between the poles of agent and setting (or across the spectrum). The analysis proposed therefore consists in zooming to identify structure that simultaneously occurs at multiple levels. Epistemological framework and research method are exemplified with data from a physics classroom where students learn to explain computer-animated microworld events in terms of Newtonian theory. These activities are analyzed using different units of analysis thereby giving rise to different aspects of cognition: interaction of gesture, material ground, and talk; co-construction of descriptions; constraints imposed by scientific resources (talk, objects); and interaction of physical arrangement, social configuration, and artifacts on discourse and participation. In the final analysis, cognition is situated not only because all cognitive activity occurs in some setting but because of the researcher's commitment to situated it there.

At least since Descartes' analysis of epistemology which separated body and mind, the source of intelligent action has been attributed to the gray matter of the brain. Despite periods of disinterest in the role of mind in human behavior (e.g., behaviorism), the rise of the computer age also led to a blossoming of the mind-as-computer metaphor (Baumgartner & Payr, 1995). In the traditional mind-as-computer metaphor, intelligence was modeled with systems of discrete physical symbols (tokens of declarative knowledge) that were processed according to fixed production rules (procedural knowledge). The resulting cognitive analyses and artificial intelligence systems exhibited considerable power in modeling intelligent behavior (e.g., Anderson, 1985; Larkin, McDermott, Simon, & Simon, 1980). However, this enormous power was limited to a perspective that physics is one of many so-called well-structured domains where one acquires knowledge made explicit based on algorithms and principles. These systems were virtually helpless when it came to more messy domains, in particular, when it came to modeling everyday competence (Dreyfus, 1992).
Recent research, sometimes referred to as "nouvelle" cognitive science and artificial intelligence, now realizes that cognition as rule-based reasoning is grounded in, and therefore made possible by, an here-to-fore unacknowledged, vast background of embodied and tacit knowledge (Agre, 1997; Brooks, 1995; Varela, Thompson, & Rosch, 1993).[1] This new way of understanding cognition--frequently commensurable with connectionist implementations--takes the human experience of "being-in-the-world" as its starting point and fundamental presupposition. This new wave of cognitive science makes the following central assumptions: representations are relational (e.g., Agre & Horswill, 1997), many activities do not require mental representation but use the world as its own representation (e.g., Kirsh, 1995; Lave, 1988), and embodied (physical) experiences are prerequisite for conceptual understanding (e.g., Hayward & Tarr, 1995; Johnson, 1987). Further, anthropological research showed that a considerable amount of what counts as knowing resides in the everyday practices in which people engage without being aware of their structures (e.g., Lave & Wenger, 1991); these practices frequently resist procedural specification, and only have a tenuous relationship to the plans that supposedly have "caused" them (Suchman, 1987). This flurry of research from many different domains (all concerned with knowledge and intelligent actions) clusters around the banner concepts of "situated" or "distributed cognition." From this perspective, it may not even make sense to speak of situated cognition. Rather, aspects of cognition are simply located differently along the agent-in-setting axis: Some structure in cognitive activity arises more from structure in the setting (cf., Scribner's [1986] analysis how structure of dairy cases and pallets shape calculations) and other structure more from structure in the agent (cf. Lave's [1988] of paper and pencil shopping problems). In this, the unit of agent-in-setting I have used in the past is semantically and syntactically equivalent to the recent conceptualization of a "dynamic unit" (Mandelblit & Zachar, 1998).
Recent work in (science and math) education has attempted to work out whether and how distributed cognition and situated learning are viable theoretical concepts in the context of schooling (e.g., Roth, 1996a, 1996b). Important during the development of new theoretical frameworks were the methodologies for collecting and interpreting data. Both activities presuppose that we understand what the appropriate units of analysis are to capture the different forms of cognition, and the recognition of the often radically different representations various cognitive agents use in their interaction with their (social and material) worlds.
I begin by outlining my epistemological frame which rests on the assumption that the agent-in-setting is the irreducible unit of analysis. To understand and study cognition, I develop a methodology that seeks to construct those structures that we can associate to cognitive activity. As a heuristic, I assume that structure may lie somewhere along the gradient between agent and setting and has different time scales. Learning is therefore understood as the change in this structure. By identifying individual ontologies and by zooming between concurrent levels of analysis, I identify both coarse and fine structure in cognitive activity. After describing the context of one study, I describe the multiple levels of cognition when high school physics students interact with each other and a modeling software tool.

Figure 2. Four maps that we might have to describe the trajectory of a fisherman in his boat. Although each frame of reference tells us a part of the story, it is the ontology of the indigenous map representing the world through the fisherman's eyes, that most plausibly explains the trajectory.

Central to the understanding of an unfolding activity, cognition, and actions is the ontology of the world as viewed by each agent. By ontology I mean the ensemble of salient elements each individual perceives, acts towards, and talks about. Most cognitive research presupposes a stable ontology of "problems," setting, utterances, etc. However, an absolute frame of reference, that is, the investigator's domain ontology may not be the most appropriate for understanding the actions of the research participants, nor does it permit to understand the representations that are associated with the ongoing activity, or the rationales for doing one thing over another. Rather, the representations are relational, historically-contingent, interactional and situation-specific and may therefore differ considerably across individuals. To make this further clear, consider the different representation of a fisherman's movements on a river (Figure 2). Whereas all four maps are suitable frames for representing his trajectory, only the fisherman's map related to traditional fishing spots provides an account that explains his activity; yet, when asked, he may not talk about his trajectory in terms of a map. Although the absolute referencing in terms of the Global Positioning System, the channel map, and the geographic map (all stops are at protruding points, but not all points are stops) are suitable external descriptors, they are inappropriate for describing the kinds of representation that the fisherman might have and that actually motivated his actions. (For an extensive discussion of representations in everyday activity see Agre [1997] and Chapman [1991].)
The importance of getting an individual's own ontology of the situation right was recently pointed out in two independent studies of children's reasoning on the balance beam (Metz, 1993; Roth, 1998b).[3] For many years, researchers assumed that children were acting on and towards "weight" and "distance-to-fulcrum." However, Metz and Roth pointed out that both "weight" and "distance" are emerging concepts in the sense that children developed them through their interaction with the materials and in the particular settings. That is, whereas researchers traditionally had assumed children to give incorrect responses, this research shows that children did not respond to weight and distance at all. The very structure of the focal phenomena and therefore children's ontologies were different from what had been assumed leading to a reinterpretation of what had been considered cognitive deficiencies in the past. Getting a sense for the ontology of others takes considerable familiarity with the people and the places they inhabit paired with a radical disbelief (questioning) of any ontology, however plausible it may seem (e.g., Bourdieu & Wacquant, 1992).

The data for the computer activities exist in a large context of other data collected during the same school year with the same three classes. These data include video records during students' experimental work, semantic networking activities, and during individual interviews about knowing and learning physics. Furthermore, hard copies of the results of laboratory work and student reflections on knowing and learning in diverse physics activities also entered the data base. For the group of three students presented here, the additional data base contextualizing the Interactive Physics^(TM) study includes 15 reports of independently-conducted laboratory investigations, 10 1-hour sessions of semantic networking, one exam and 3 tests per trimester, 13 essays on knowing and learning, and a series of interviews focusing on physics knowledge and epistemology.

Using an excerpt from the study explained in more detail below, the data turn into displays such as that featured in Figure 6.
 

A   video
B So like this arrow forces it (.) to a certain extent audio
C | [arrowup] | | [arrowup] marker
D 1 2 3 4 5 reference
E (0.80) (1.03) (0.10) (0.47) [Delta]t (time)

Figure 6. Analysis of gesture-discourse relationship during conversations over and about objects and events in a computer-based Newtonian microworld. Timing points are marked relative to utterances (vertical lines) and video frames (vertical arrows). For example, there is a 0.80 second delay between the onset of the utterance "like" [1] and the first of the two video frames [2].

This particular display is subsequently used to construct a relationship between gesture, talk, and their temporal development over shorter (minutes to one hour) and longer terms (4-6 weeks). The analysis proceeds as follows: The present excerpt shows that there is a 1.03 second delay between the deictic gesture (pointing) [2] and the utterance "this arrow" (further made salient by a little 0.10-second jerky movement of the pencil [3-4]); furthermore, the iconic gesture that simulates the force arrow's movement across the screen begins 0.47 seconds prior to the verbal description of the arrow's action on the object designated by the indexical utterance "it." From prior research on gestures and the relationship between gestures and utterances (e.g., McNeill, 1992) I know that these delays are significant; such delays between gestures and discourse are part of a developmental trajectory which ends in the eventual overlap between the two distinct of forms of representations embodied in kinesthetic and verbal coding (Roth, 1999). The context of the unfolding activity makes the interpretation of "it" as indexing the circular object very likely. Here, the student appropriately described the action of the arrow as "forcing" the object. Yet he came to consistently use the arrow as "force" only during the subsequent lesson two weeks later (and about 1 hour of further activity). These events are further notable, for they illustrate that even if students make utterances and gestures apparently consistent with the relevant science, there are no mechanisms inherent in the physical world that select these actions over others that may be scientifically incorrect.
 

velocity() force( )

At one point, Ryan accidentally detached the force arrow from the object; but the three decide to run an experiment in this new configuration. They discuss the resulting screen display in the following excerpt.
 

G: So when you don't run it with this arrow  POINTS[force]  it goes in the same velocity  TRACES[trajectory]  R: It just goes in the same direction  GESTURES[trajectory]

this arrow, like is initial (2.3)
POINTS[velocity]
the later direction
E: That means it's a constant
G: So like (2.8) this arrow forces it to a certain
POINTS[force]
extent
R: It changes direction after the start

Glen provided a first description in terms of [velocity] as moving "in the same velocity" while his deictic gesture first pick out the arrow, followed by an iconic gesture that traced and therefore made salient the trajectory. Ryan first followed up by describing the trajectory as being "in the same direction" and traced a straight line in the air, and then links [velocity] to a feature of the initial state in the experiment, but, overlapped by Elizabeth did not complete his statement about the final direction. Elizabeth's statement about something being constant can be read as confirming both Ryan and Glen's earlier utterances "same direction" and "same "velocity." Glen, followed by Ryan, describe the action of [force] as "forcing" and "changing direction after the start."
In this episode, the three students produce descriptions commensurable with Newtonian physics. They use gestures and utterances to pick out, and make salient, a limited number of objects ([force], [velocity]) and events (trajectory). These observation descriptions are assembled in a public space, and require both the inscription and the gesture. The gestures allow students to fix the referents of some words, though in this episode, the referent for the deictic term "it" fluctuates and its referents are never clarified. For example, in Glen's description, the [force] acts on "it," presumably the object. However, Ryan's "it changes direction" does not unambiguously pick out whether "it" is [force] that causes some change, [velocity] which changes, or the object which moves on a curvi-linear trajectory. It therefore needs to remain open whether students talked about the arrows or the objects. Furthermore, their observation sentences do not require internal representations (Quine, 1995). We can therefore take the unfolding conversation as something that exists in public space (thus somewhere other than the agent pole) leaving open what memory traces they leave (what students learned), or how this aspect constrains later developments of the conversation.
One might be tempted to infer from this transcript that the three have evolved mental representations consistent Newton physics. For example, Glen might be interpreted as having a representation of [velocity] ("this arrow") as indicating a constant velocity of "it," the circular object. As indicated, we need to radically question our own ontologies and how we attribute them to the agent. Later parts of the unfolding interaction shows that the discourse was not stable. In fact, as the conversation unfolds, there is a considerable variation in the designations used for [velocity] (little arrow, big arrow, initial speed, velocity, initial speed, velocity, force, effort, strength, speed, strength, speed, direction, speed & direction, velocity) and [force] (little arrow, big arrow, time set, time, direction, time & direction, velocity, redirection, gravity, force, gravity, gravity). The two lists show that the same labels were used to denote different arrows. In this sense, the above observation sentences constructed by the students were constructed in the context, contingent on the computer configuration, history of the emergent conversation, and students' perceptions. Existing family resemblances between scientific discourse and vernacular all too easily lead researchers to make assumptions about representations and conceptions are not viable representation of students' knowing.
The episode shows us how students coproduced a description of an event in the sense that all observation sentences highlighted something as being constant when one arrow ([force]) was disconnected from the object, and that there were changes in the direction when the same arrow was attached to the object. I know that students, out of these uncertain beginnings, developed a consistent way of describing and explaining the phenomena at hand (Roth, 1996c). However, this development did not occur independently of other events in the classroom. Rather, the interactions between myself (teacher) and students brought about changes in the way students perceived, and talked about the events.[5]
We might assume that conceptions drive what students say. Their talk is then considered as a medium of externalizing thoughts and conceptions from the computational hardware to the public forum. Such a view is inconsistent with the data presented here because of the considerable variations in the discourse which would have required to make the assumption that their "conceptions" constantly changed. Based on my epistemological frame, I make the less stringent assumption that students produce situated observation sentences out of their interactions in the setting. This does not necessitate representations, for the relevant elements (image to be described, language, gestures, etc.) can be picked from the setting. These descriptions are ephemeral and may be forgotten in the next instance so that subsequent sentences may in fact be incompatible when studied by the researcher. On the other hand, observation sentences can also be stabilize within the group and then become conversational results that students remember, and which therefore last beyond the immediate activity.

I: What if you had that point up (3.1)  GRABS[velocity] POINTS[up]  and this one would be pointed like this?  GRABS[force] POINTS[down]  G: It would go straight down  R: Yeah, it would go downward

Here, Ryan and Glen responded to my "What if. . .?" question by stating the hypothesis that the object would immediately descend ("straight"). They then stated that they saw the object going down before Elizabeth contradicted them:
 
 

I: But first?   E: I think it went backwards first though.  R: The initial velocity went the way the little arrow goes.  E: Didn't it go backwards first and then go forwards?  R: I think so.

Coincident with my questioning "But first?," Elizabeth contrastingly ("though") described the object as going backward first. This contrast, together with "first" can be read as an oppositive description to that provided by Glen and Ryan. Ryan responded by describing the initial movement in the direction of the "little arrow," to which Elizabeth reiterated her perception as a contrast ("didn't it...") to what she understood Ryan as saying. Subsequent to this interaction, I asked students to relate this phenomenon to something in their everyday life to which the three responded with descriptions of a returning hula hoop, yo-yo, and object thrown in the air. Out of this, Elizabeth suggested that [force] represented something like gravity.
Here then, Glen' and Ryan's ontologies were different from mine and that valued in the scientific community. That is, they perceived the object as moving straight down contrary to my own expectation and observation that the object should move upward before it descends. Not perceiving the initial upward motion is significant, for it does not allow an understanding of the relationship between velocity and forces in the early part of the trajectory, and therefore a more general theory of forces and the motion of objects. Therefore, despite setting up what I considered a crucial experiment, the move was unsuccessful in the very first instance. This changed with my question and Elizabeth's different observation description.
Whereas the students appear to have come to an agreement that the object moved up before it descended, the episode does not make clear whether they actually observed the [velocity] change: from being pointed upward it decreased to zero, and then increased again pointing downward. In fact, the subsequent episode shows that during the moment of teaching, I interpreted students' talk as not having made this observation and therefore ran the same experiment repeatedly in slow motion until the students' observation descriptions included not only the object but also the two vectors. We therefore note that despite the very small number of elements constituting the microworld--a circular object and two arrows at the focal point, and the remainder of the interface in the immediate background--students did not perceive it in the same way I, a physicist and physics teacher did perceive it. Students had a different ontology of the microworld: although it was experienced as real by the students, it was an ontology that was inconsistent with a Newtonian explanation. However, the constraints provided by different observations within a student group, and those provided in interactions with the teacher, are crucial for establishing the phenomenal backdrop to any correct understanding of the theory which students are to learn according to the curriculum.
Because of my familiarity with Interactive Physics^(TM), the microworld was a tool that permitted me to engage students in a situated inquiry. Particularly the slow-motion option afforded students to perceive the upward motion (and upwardly pointing [velocity]) during the initial phase of the trajectory.

Wouldn't the length of the  |(1.47)[arrowup] (2.00)	arrows (1.60) Since that arrow  [arrowup]	`s longer the velocity is higher  (1.47) [arrowup] (0.33) |(0.10)
that's  [arrowup] (0.20)	why:: it's  [arrowup] (0.53)	pushing it that'a way.  | (0.83) [arrowup]

Figure 7. Exerpt from the conversation in which Glen gestures pushing, parabolic trajectory, and changing velocity two weeks (lessons) prior to verbally expressing the same relationships.

At the moment of the episode, the three students still had no grasp of what the arrows stand for and how they related to the moving object. The students had previously affiliated them with time, energy, time step, and many other lexical items. Here, then, Glen attempted another description and explanation of what they just observed (still visible in the top left slide). His utterances (Figure 7) were accompanied by the gestures of both hands which enacted the arrows and their behavior as he had seen it previously. Glen had held his right hand with fingers parallel to the outline [force] arrow--in the form more clearly seen in the second frame--for 3.47 seconds prior to specifying its referent in Frame 2 ([1]-[3] in Figure 7). He then made another, 0.10-second circular gesture which marked the right hand while uttering "that arrow" that immediately preceded the causal meaning unit "that's why it is pushing it. . ." Before he voiced "the velocity" ([4]), his left hand can already be seen at the top left of the image, held parallel to [velocity]. In Frame 3, both hands are visible: the right parallel to [force], the left parallel to [velocity]. In the next frame, the right hand was already "pushing" against the left hand which is moving off the frame to the left. This movement continues to the end of the sentence and out of the video frame. This pushing motion of the right hand began 0.83 seconds before the associated lexical affiliate "pushing." Here, the shape of the object's trajectory (visible in Frame 1) which he attempted to explain was already completely described by the iconic gesture (and trajectory) of his left hand. The episode is complex because there is one word "arrow" but two arrows on the monitor, and the same indexicals "that" and "it" occur repeatedly but refer to different objects and have different functions.
"That" appears three times, and each time not only the referent but also the function is different. In the first instance, "that" ([3]) has a deictic function designating a particular arrow standing in opposition to the speaker (distal use). Coinciding with the utterance, the right hand which had moved to the right, came to a sudden stop. As can be seen from Frame 2, the fingers of the right hand stand parallel to [force]. This finger position, the noticeable (abrupt) stop of motion, and the coincident utterance "that arrow" makes it reasonable to assume that the right hand models [force]. The listener can draw further confirmation for this interpretation from the causal connection between "that arrow" and [force] because it is the one that the three students had previously manipulated, whereas [velocity] only changed as a function of their action.
In the second instance, "that" ([6]) introduces the causal consequence ("that's why") of the hand arrangement he had set up and described in the previous part of the utterance; "that" falls at the beginning of gestural trajectory which iconically re-represents the earlier visible trajectory (Frame 1). Finally, in the third instance, "that" is linked to "way," the immediately preceding trajectory ("way") enacted by the gesture. In vernacular, "that way" most frequently expresses a specific direction. Here, however, "that'a way" together with the curved motion of the hand, when read against the ground of the earlier curvi-linear motion of the object and the corresponding positioning of the arrows, highlights not only the existence of the trajectory but in particular its curvi-linear shape.
The indexical term "it" was used twice, but when the gesture is viewed against the microworld in the background, the two referents in "It's pushing it. . ." can be disambiguated. The utterance occurred while the right hand followed, fingers pointing to, the left; heard together with "It's pushing it," the right can be understood as literally pushing the left hand (Frames 3-6). Here, the first use has as referent the hand/ arrow which is pushing (enacted by the hand) and the second occurrence has as referent something that is being pushed which, in this case, could be the second arrow/left hand, or the object.
At the time of this episode, Glen (as his two peers) did not yet describe the arrows in scientific terms, that is, as force and velocity. He used the appropriate scientific (verbal) discourse only two weeks later during the subsequent lesson with the microworld. However, in the present episode, his gesture--when understood as a description of the relationship between the concepts of velocity and force--was consistent with scientific practice. He characterized the action of the outline arrow as "pushing," which is a vernacular form of describing forces. Finally, he associated the longer pushing arrow with a resulting higher velocity. Here, the referent of "velocity" is not completely clear and two readings are possible. Because the utterance coincided with the positioning of the left hand, "velocity" can be heard as the referent to the left hand: therefore, the longer force (right) arrow pushes more and therefore leads to a longer velocity (left) arrow. But the fragment "Since that arrow `s longer the velocity is higher" could also be interpreted such that the longer right arrow is equivalent to a higher velocity in which case "velocity" would have been anchored in the right arrow (incorrectly so from a scientific perspective). However, the nature of the referent for each of the two hands was disambiguated by their position in space in the course of the motion. The fingers of the right hand kept a constant direction just as the outline (force) arrow, whereas the left hand changed direction, though less rapid so, in the way the single-line arrow did previously. Thus, Glen's gestural description and explanation of the events was consistent with scientific practice long before his verbal explanations.
In this episode, gestures, animated diagrams, and words were deeply integrated. (Though there are some studies related to interaction of gesture and speech in a variety of non-motion domains, the role of gestures in scientific and mathematical discourse largely remains unexplored.) That is, the structure in the activities arise from structure in each of the levels so that we can view cognition as distributed across the agent-in-setting unit. Any one considered by itself does not help us understand or infer the moment of practice. Taken as a whole, gestures, words, and diagrams (both topic talk and background to gesture) make a lot of sense. Because we have to consider these elements together, it makes sense to speak of cognition as being situated. The structure and coordination of the actions make sense if considered in this particular setting.

Table 1. Machine and human perspectives on the unfolding activity
 
 

THE USERS

INTERACTIVE PHYSICS(TM)

 

Not available to Interactive Physics(TM)	Available to Interactive Physics(TM)	Available to the users	Design rationale
E: Pull it out
G: So pull it now go that way	R INCREASES [force] TURNS[force]		Object-oriented manipulation of physical variables.
R: Oh, what did I do?
G: Cancel, Oh there we go, leave it, yeah. Alright, now push it back, keep connected to the back, now run it.	R INCREASES [force] TURNS[force]		Object-oriented manipulation of physical variables.
G: OK run it, oh baby! yes	R STARTS[experiment]
			Given position, initial velocity, force, mass, calculate and display trajectory.
		DISPLAY PANEL: Object velocities are high for this simulation. Reduce time step for greater accuracy.	High object velocities along trajectory cause large position changes, cause inaccuracies in trajectory and recalculation of velocity, acceleration
E: Which one's the time step?
R: It's that big arrow
G: Oh yeah the big arrows time, OK. I comprehend.

 

Within the students' horizon, the display panel message followed their immediately preceding lengthening of one arrow (force). As a result of these actions, the software makes available a trajectory, followed by the panel message. Thus, within the students' interpretive horizon, the message was an immediate consequence of their previous action of lengthening the arrow. The word "reduce," when viewed in the context of the previous lengthening was used as a resource to relate the subsequent "time step" to the manipulated arrow. However, the interpretive frame of Interactive Physics^(TM) is different. It was designed to run and display the experiment given a particular specification of the relevant variables (mass, velocity, position, force). The message, based on the size of velocity somewhere along the trajectory, was designed to indicate an action that allowed greater accuracy given the user specifications. Thus, rather than basing its feedback on the history of the interaction or on the specified size of the variables, the system starts with an aspect of the simulation not directly available on the interface and often used in default mode and checks whether the simulation is possible with a modified time step. In the hand of a competent user who is familiar with the design rationale and simulation practices, however, the message is likely to be interpreted differently (e.g., in my own case).
Earlier analyses showed how students were enabled to use deictic and iconic gestures that grounded their utterances, and, when viewed against the interface as background, helped the speaker to make salient those aspects relevant to his explanation. However, as can be seen from Figure 4, when there are 3 or more individuals oriented toward the interface, there are space constraints on possible physical configurations. Whereas all members could see the gestures against background by those sitting close, the same affordance did not exist for other participants. Thus, while the physical setting did not preclude participation in the conversation, it did preclude anchoring functions of gestures. However, because gestures are central to scientific laboratory talk (Lemke, 1998; Suchman & Trigg, 1993), not having equal access to the representational medium actively interferes with learning (Roth, 1996d). The point here is that not being able to handle the computer input is far less important than the exclusion from the on-going conversation because of limited access to a different mode of communication. Those students who are excluded are likely to engage in activities unrelated to the task or subject, thus to engage in "off-task" activities.
Compared to the previous sections, the present analyses focused on learning in a broader frame considering how physical arrangements, (size of) social configurations, and nature of focal artifacts interact and affect conversational and participatory patterns. This broader focus then leads us to construct different aspects of cognition. Rather than focusing on mind alone, I find it useful to look at and describe events as they emerge from agent-setting ensembles. That is, I conduct my analyses from the perspective of an irreducible unit of analysis constituted by being-in-the-world which forces us to consider all events as acting-in-settings.

[1] In the framework developed here, all structural aspects of human agency that we recognize that contribute to cognitive activity are located (i.e., situated) somewhere along the agent-in-setting continuum. Some structures are embodied more on the agent side of this continuum, others more on the setting side. Where the most salient and significant structures lie along the continuum is, to me, an empirical matter rather than one to be decided a priori.
[2] As Jonna Kulikowich pointed out to me, my own perspectives on what the world of this classroom loooks like change with the setting: my perspectives of what is happening in the situations reported below depends on the time scale considered and therefore differ for Roth the teacher in situation, the physicist, and the cognitive analyst of videotapes. (See also the paper by Kulikovich and Young, this issue.)
[3] Research in social and discursive psychology (e.g., Edwards & Potter, 1992), sociology (e.g., Bourdieu, 1990), and sociology of science (e.g., Gilbert & Mulkay, 1984) showed that individuals, when asked, may describe and explain their ontologies in ways inconsistent with their actions.
[4] Here, for ease of reading, I use [velocity] and [force] to denote the respective arrows and . However, especially in the excerpts presented here, students do not perceive these arrows as denoting "velocity" or "force" or any thereby reified natural phenomena.
[5] Descriptions and theorizing of the dynamic and emergent aspects in my teaching from the same agent-in-setting perspective can be found elsewhere (Masciotra & Roth, 1999; Roth & Masciotra, in press; Roth, Masciotra, & Boyd, in press).
[6] In most general terms, constraints limit the possibilities of actions. In some situations, this limitation brings with it an affordance: handles limit the places where one might try to touch an object to carry it, but also allow for an easier way to actually do the carrying task. In other situations, a constraint prevents people from doing what they intend or are supposed to be doing: many newcomers to Macintosh computers were afraid to eject their diskettes, which required to move the diskette icon over the trash can icon, because they thought they would loose the diskette contents (Winograd, 1996).
[7] Talk about activity is a different kind of activity, with a different focus, and different context and properties (Bourdieu, 1980). Thus, it is not surprising that researchers frequently find little overlap between teachers' actions in the classroom and their descriptions and explanations of these actions (e.g., Tobin, Espinet, Byrd, & Adams, 1988).
[8] The notion of interactive stabilization has particular appeal because it is consistent with my computer models of interpretation formation. Here, the dynamic of a group with respect to the individual interpretations is modeled as constraint satisfaction among multiple interacting hypotheses in connectionist networks.