Notice Problem	Hypothesize Source	Test Hypothesis	Find Optimal Solution
Pick Problem	missed pick	wrong feeder pitch	check stopper block	change pitch to 8mm
Placement Problem	oversized socket	board design error	check computer program	skip socket placement

This chart is not meant to make claims about patterns in troubleshooting; rather, it is meant to demonstrate the workers' speed and efficiency in pinpointing and solving problems during this six-and-a-half-minute interval. Tran and Du have moved through these procedures with few words, most of which are a series of numbers that would appear quite literally indecipherable to an outside observer. In fact, the workers' language-in-action, taken alone, can be characterized as markedly sparse.[6] Certainly, their prior troubleshooting experiences and shared understandings may partially explain why there has been little call for extended discussions about which actions to take. The level of machine noise and pressure of customer deadlines may also have been contributing factors to this reduced speech. In the following sections, we deal with how the two men accomplished this kind of problem-solving so quickly and with minimal use of language. We pay particular attention to their skilled use of perceptions and representations. The analysis begins where they notice each problem and continues as they detect the source of the problem and search for an optimal solution.

Working with Perceptions

The workers draw on well-honed perceptions--auditory, visual, and kinesthetic--to notice the trouble and to find its source. These skills are especially evident during the first 44 seconds of the pick problem, during which they do the noticing and form a hypothesis. When set to high-speed, the SMT machine has an audible click-click--swish--click-clack rhythm from pick to placement. The arm can be seen to dance, as it were, to these sounds as it shifts from the feeder to the board and back again. In the following excerpt, we see that the rhythmic sound and movement, maintained up to the placement of the first socket, are broken when the arm tries and fails to pick the second socket. The pick sound is repeated (click-click, click-click) as the arm attempts twice to pick the socket. The men put one another on verbal notice just after the break in the rhythm:

10:15:14	Robotic Arm:		(first pick try: click-click sound)
10:17:03			(second pick try: click-click sound)
10:17:16	Du:	MISS PICK--
10:18:11	Robotic Arm:		(successful pick: click-click sound)
10:18:16	Tran:	Oo: : :h.
10:18:27	Robotic Arm:		(socket placed: swish--click-clack)

As the arm swings back for the third socket, Tran combines kinesic and visual action to stop and start the machine at key moments in the pick-placement sequence and to observe the movement of the robotic arm over the pick and place points. The broken rhythm continues--two miss picks before a successful one. As the arm begins to carry the third socket to its placement point, Tran presses "stop," counts the number of sockets already taken from the feeder, and observes for several seconds the position of the arm over the third place:

10:19:22	Robotic Arm:		(first pick try: click-click sound)
10:21:05			(second pick try: click-click sound)
10:22:13			(successful pick: click-click --swish sound)
10:22:17	Tran:		(presses "stop")
10:25:23		One, two, three.	(counting on fingers; gaze turned toward board)
10:29:19-30			(leans inside machine toward robotic arm poised over the board)

Then Tran presses "start." He observes the arm rapidly place the third socket and swing back to the feeder (click-clack--swish). As soon as it reaches the pick point again, he presses a button on the SMT operator panel, and the arm begins to pick in slow motion. Tran is able to see the downward movement of the arm (z-axis) together with the forward movement of the feeder (pitch). With slow motion, it is possible to see that the feeder pitches the tape forward three times before the socket is within the reach of the arm:

10:33:12	Tran:		(straightens up, hand on button, looks at screen)
10:35:04			(presses "start")
	Robotic Arm:		(placement: click-clack--swish)
	Tran:		(leans in to watch placement and arm's return to feeder)
10:35:21	Robotic Arm:		(first pick try: click-click sound)
10:35:25	Tran:		(presses button for slow motion)
10:36:18			(lifts plastic screen as arm goes down in slow motion to pick socket)
10:37:23-	Robotic Arm:		(second pick try: click-click sound in slow motion)
10:42:27
10:39:22	Tran:	Two (picks).
		Three (picks).
10:41:00			(directs gaze at pick point)

Du has been observing, too. He suggests adjusting the stopper block, and Tran agrees. This is the moment in which they are displaying to one another their hypothesis--the source of the problem may be an incorrect pitch point on the stopper block:

10:45:15	Du:	Is it necessary to fix the screw underneath?
10:47:28	Tran:	Yeah.

The second problem--the placement problem--is also detected perceptually. Tran has just changed the pitch value in the computer program from 12 to 24, twice the amount originally assigned to the socket. Before restarting the machine to finish the assembly, Tran takes the board from the conveyor belt, holds it in both hands, and looks down closely at the sockets placed on it. With this close look, he notices that something is wrong:

15:59:29	Tran	Oo: : :h, no, it doesn't work.
16:02:26		(xxxx) socket.
01:16:05:66		[It's] impossible to install the socket.

Then, still holding the board, he calls Du over to look. Tran displays the problem to Du by tracing a series of spatial boundaries over the board with his finger (see Figure 6).

He gestures toward the problem's effect on the entire board. His finger first swings from one socket to another in a large sweep, then marks spaces around the sockets in small loops, and finally traces another large sweep between the two sockets.

16:09:00	Tran:	Like this [it's] impossible	(pointing to spaces
		to install the socket.	around the 3 sockets)
16:09:23	Du:	Which?
16:11:02	Tran:		(sweeps finger right to
-16:11:20			left, socket to socket)
16:11:20		Forgot.
16:11:21			(makes small loops with
-16:13:08			finger around socket)
16:13:09		Something is wrong.
16:13:09			(makes small loops with
-16:16:29			finger around socket)
16:16:05	Du:	Three [of them], or how many?
16:17:07	Tran:	Up to eleven [of them].
16:17:04			(sweeps finger left to
-16:17:14			right, socket to socket)

Tran makes smaller, bouncing loops with his finger which coincide with and surround his verbal notification that a problem exists. That is, the fact that a problem exists is signaled, not identified verbally. The line above the utterances represents the time between the onset and completion of Tran's gestures. Tran's words are in boldface:

[-----------------------------------	Small loops with finger around socket	-----------------------------------]
16:11:21	16:13:09	16:16:05          16:16:29

"Something is wrong."   "Three, or how many?"

Also coinciding with these gestures is Du's question (in boldface) about the number of surrounding components:

[-----------------------------------	Small loops with finger around socket	-----------------------------------]
16:11:21	16:13:09	16:16:05          16:16:29

"Something is wrong."   "Three, or how many?"

Du's query, along with Tran's response ("eleven"), are evidence that they are both associating the gestures with the affected smaller components. It has already been noted that Tran's shorter gestures are further enclosed by the larger sweeps of his finger from socket to socket. With these distinguishing gestures, the two workers shift their concerted attention back and forth between three sockets and the 11 other components affected by their presence. That is, the layout of the sockets on the board's landscape provides more insight into the program error. The numerical information provided by the customer and programmed for the machine corresponded, not to the sockets, but to the smaller components that were to be plugged into them. In sum, these gestural practices serve conjoint problem-solving, enabling the workers to establish the scope of the placement problem and to hypothesize its source as a customer design error ("There's no program, it was not put in.").

Working with Representations

In this activity, the workers' sharing of perceptions is embedded in their work with a computerized machine. Thus, they both compare and convert the perceptual boundaries they have shaped to abstract representations of them. For comparison purposes, they may refer to ready-made inscriptions such as a blueprint of the board, the customer's bill of materials, data in the software, or the machine's instruction manual. Alternatively, they may create their own inscriptions by writing down calculations on scraps of paper or by measuring objects and modifying information in the program. Because number use pervades this activity, our analysis focuses on Du and Tran's work with numerical inscriptions, including computerized numerical controls.

The workers convert perceptions about distances (between components on the tape) and bounded spaces (on the board's landscape) to numerical values by taking various measurements. To test their hypothesis about the miss-pick trouble source, they use a metric ruler and digital calipers to verify their perceptual impressions. While they continue to draw on perceptual structures, these are given new representations with the help of the tools. They first determine the pitch value. Du places a ruler on the tape, while Tran looks on. At the moment the ruler and tape are aligned, Tran assesses what he sees:

11:12:23

Tran:

Strange.

His utterance makes it clear that the measured pitch length is going to be unusual. Then Du adjusts the ruler so that one end is placed directly at the center point of one component as shown in Figure 7.

Together, they calculate the distance between the two components. The centimeter markings on the ruler must be converted to the equivalent measure in millimeters. Tran calculates aloud and comes up with a 24mm pitch:

11:15:10	Tran:	Two.
11:16:10	Du:	(xxxx)
	Tran:	Two, three and a half.
11:19:06		Twenty and twenty-five
11:25:14		Twenty-four.

It's tempting to say that the workers turn to numerical inscriptions as hard evidence for their perceptual intuitions. Yet the verifications move in both directions during their conjoint problem-solving. Tracing the distance with his little finger, Du shifts from Tran's mathematical representation of distance back to a perceptual one as shown in Figure 8.

11:27:05-	Du:		(traces little finger of right handfrom midpoint of one component up
-11:27:29			to midpoint of other component)
11:27:22		From this point to this point.
	Tran:	Yeah.
11:31:05		Twenty-four.

The workers' skilled actions taken on machine parts and components have been pinpointed, along with their work with numerical representations of them. We gain a fuller understanding of the complexity of their actions by examining the workers' displays of understandings about how the SMT machine itself works. Their local actions on machine parts and corresponding numbers have global consequences at the level of the computerized machine. For example, having converted their perceptual assessments to numerical representations of them, they check the computerized database. The two men make sure that the information in the computer program conforms to component sizes and the movements of mechanical parts. In this case, the stopper blocks on the feeders have to be set at a pitch point that accommodates the large sockets on the reel, and the information displayed on the screen must correspond to the actual pitch for that reel. Recall that, when the men finish working with the stopper block, Du asks Tran to look for information about the component in the computer program:

14:51:00	Du:	Tran, Tran, look at the component ID.
14:51:00		Look, look, in there.
14:53:00	Tran:		(moves to display on machine)
14:59:00	Du:	See, see how much.	(getting up, pointing to screen)
15:04:00		Let it run through it again.	(standing next to Tran at machine)

To find information on the screen, Tran must be familiar with the screen organization, ways to select a menu, functions of the dialog boxes, setup commands, data entry, and data editing procedures, along with the functions of the keys and switches on the operator panel keyboard. In short, he must know all the basic operating procedures of the machine. At this point in the task, he uses a search command to find information about the socket in question, as seen in Figure 9. The component data appears on the screen, and the two men turn their attention to these numbers. They have now shifted their focus from machine parts and their own calculations to the on-screen data representing these elements; in doing so, they signal their understanding of the interdependence between the mechanical and computerized features of the machine.

To show one another that they are converging on the same unit of data--the feeder pitch value--, which is nestled among dozens of numbers on the screen in arrays of five to six columns, Tran reads the number from the screen, which Du repeats, signaling their alignment with the same information.[7] Because the on-screen numbers (representing computerized controls) do not conform to their own perceptions and calculations, the numbers must be corrected. Du directs Tran to change the pitch data to 24mm:

15:20:19	Tran:	Twelve.
	Du:	Twelve.
15:27:55		After this [we] must give it twenty-four.
		Twenty-four divided by the number eight is three.

In summary, numerical inscriptions have to be made so that the digital technology can function. Workers make sure that the information displayed on the screen conforms to the components' assigned place on the board. The x and y axis values are numerical controls directing the arm to place the component at a precise point; the z-axis value directs the arm to reach down the distance needed to pick and place a component having a given thickness. But this analysis has shown that their work is more than a straightforward matching function. As we have seen, these inscription-action relationships often do not hold. Newer components and boards may be outside the range of sizes for which the machine was designed. The workers have to know how to adapt the machine to these changes. In one worker's words, "You sometimes have to 'fool' the machine" by entering numerical controls that do not actually correspond to board or component sizes.

To summarize the analysis, circuit-board assemblage is all about goodness of fit. The robotic arm has to be in harmony with the components it picks and places; components have to be in harmony with one another on the board. Workers are, literally, sensitive to these connections and spaces, as eyes, ears, and hands detect relationships. The eye observes the relationships among the forward pitch of the feeder, the height of the component, and the downward movement of the robotic arm to pick and place it. The ear detects the rhythmic movements of the robotic arm amidst a cacophony of sounds. The eye observes and the hand signals the relationship between a given space on the board and the components assigned to that and surrounding spaces. The hand stops and starts the machine at precise moments. Workers also recognize and represent numerical versions of what they perceive. They assess perceptual structures with the use of other tools and inscriptions by measuring distances with precision instruments and converting them to numerical values and by matching spatial relations to numerical representations programmed in the software. This assessment is bidirectional; the use of numerical representations is inextricably tied to objects and actions in a work setting. Numbers uttered and entered are not disembodied entities; their meaning is co-constructed in the physical and social context in which they are embedded. In order to be accepted as valid by the workers, the representations must receive the final imprimatur of the workers' eyes, ears, and hands.

[4] Offsetting is a complex procedure in which the computerized machine must be programmed to "read" the size of the board and calculate the grid for placement of the components along an x/y axis.

[5] Whenever Vietnamese transcription appears, it is followed by an English translation.

[6] For a detailed analysis of how the two workers make parsimonious use of Vietnamese and English in this setting, see Kleifgen (1995).

[7] This use of repetition in interaction is congruent with studies of groups learning together at a computer, in which users signal to one another that they are focusing on the same on-screen referent either by pointing to areas on the screen or repeating aloud the information (Kleifgen, 1992; Pujol-Ferrán, 1993).