How tufted capuchin monkeys (Sapajus spp.) and humans (Homo sapiens) handle a jointed tool

Abstract

The embodied theory of tooling predicts that when using a grasped object as a tool, individuals accommodate their actions to manage the altered degrees of freedom (DoFs) in the body-plus-object system. We tested predictions from this theory by studying how three tufted capuchin monkeys (Sapajus spp.) and six humans (Homo sapiens) used a hoe to retrieve a token. The hoe’s handle was rigid, had two segments with one planar joint, or had three segments with two (orthogonal) planar joints. When jointed, rotating the handle could render it rigid. The monkeys used more actions to retrieve the token when the handle had one joint than when it had no joints or two joints. They did not use exploratory actions frequently nor in a directed manner in any condition. Although they sometimes rotated the handle of the hoe, they did not make it rigid. In a follow-up study, we explored whether humans would rotate the handle to use a two-jointed hoe in a conventional manner, as predicted both by the embodied theory and theories of functional fixedness in humans. Two people rotated the handle to use the hoe conventionally, but four people did not; instead, they used the hoe as it was presented, as did the monkeys. These results confirm some predictions but also highlight shortcomings of the embodied theory with respect to specifying the consequences of adding multiple DoFs. The study of species’ perceptual sensitivity to jointed object’s inertial properties could help to refine the embodied theory of tooling.

Using objects as hand tools (hereafter, tooling) to accomplish otherwise challenging goals is one foundation of human technology. Understanding how tooling varies (or not) across species remains a challenge for behavioral science. Progress in this area will be aided by the development and refinement of theories that allow prospective experimentation to complement the largely atheoretical approach that has characterized much of the work in this area (Shumaker et al., 2011). Fragaszy and Mangalam (2018; Mangalam & Fragaszy, 2016) proposed an embodied theory of tooling that explains tooling as an activity of the individual-environment system (Turvey & Shaw, 1999). We defined tooling as

… deliberately producing a mechanical effect upon a target object/surface by first grasping an object, thus transforming the body into the body-plus-object system, and then using the body-plus-object system to manage (at least one) spatial relation(s) between a grasped object and a target object/surface, creating a mechanical interface between the two

(Fragaszy & Mangalam, 2018, p. 194).

According to the embodied theory of tooling (Fragaszy & Mangalam, 2018; Mangalam & Fragaszy, 2016), the difficulty of tooling increases as the number of degrees of freedom (DoFs) in the body-plus-object system increases. A DoF refers to a direction in which movement can occur. Compared to a rigid object, a jointed object possesses one additional DoF for each planar joint added to the object. For example, a hoe with one or more planar joints in its handle has one or more additional DoFs than a hoe with a conventional rigid handle. The actor moving the hoe with a joint in the handle toward a target must control two segments moving in relation to each other as well as in relation to the target, rather than one. The inertial properties of a jointed object (how it resists motion) are quite different from those of the same object in a rigid state. These features should make the movement of a jointed hoe more challenging to control than an equivalent hoe with a rigid handle. The present work provides experimental evaluation of this hypothesis in a nonhuman genus, tufted capuchin monkeys (Sapajus spp.). We presented monkeys with a hoe with a rigid handle or with one or two joints in the handle. When jointed, the user could easily make the handle rigid by rotating the handle to a particular orientation.

Working from the hypothesis that added DoFs increase the difficulty of tooling with a given object, we predicted the monkeys would use the hoe with a rigid handle more efficiently than one with a jointed handle, that they would explore the jointed hoe more often than the rigid hoe, and that their performance would improve with experience to a greater extent for the jointed hoe than the rigid hoe, reflecting learning how to manage the extra DoFs present in the jointed hoe. We expected the effect of adding a DoF to the system to be additive — that the double-jointed hoe would be more difficult to use than a single-jointed hoe.

To our knowledge, two studies have examined how nonhuman animals use objects with multiple DoFs, or with pliable instead of rigid objects, as tools after much experience with a rigid object for a given task. Povinelli & Frey (2016) found that when given a choice between a rake where the hinged blade of the rake moved away from the handle when it was pulled and a conventional rigid rake, captive chimpanzees experienced at using rigid (conventional) rakes did not act to discover which rake was hinged. Instead, they chose randomly between the two rakes. The chimpanzees did so even after experiencing failures when trying to use the hinged rake. Sabbatini et al. (2014) showed that capuchin monkeys used pliable and rigid objects in effective sequences to solve a problem. The monkeys used a pliable object to retrieve yoghurt from a tube; when necessary they used a rigid object to pull in a pliable object from out of reach. Neither study addressed how the subjects accommodated their action to an object that deformed when they used it.

Among nonhuman primates, tufted capuchin monkeys exhibit a relatively high propensity for tooling in both free-ranging and captive conditions (Falótico & Ottoni, 2016; Fragaszy, Visalberghi, & Fedigan, 2004; Fragaszy, Izar, Visalberghi, et al. 2004; Mannu & Ottoni, 2009; Painter et al., 2019; Souto et al., 2011), and they readily use hoes and hooks to pull things towards themselves in the lab (Cummins-Sebree & Fragaszy, 2005; Fujita et al., 2003). Capuchin monkeys use manual exploratory actions when retrieving small objects hidden out of view (Lacreuse & Fragaszy, 1997); they use exploratory manual actions to identify if a nut is empty or full (Visalberghi & Neel, 2003), and to position nuts in anvils in preparation to crack them with a stone (Fragaszy et al., 2013). They use exploratory manual actions with stones prior to selecting one to transport (Visalberghi et al., 2009) and prior to lifting them to crack nuts placed on anvils (Fragaszy, Morrow, et al., 2019). Capuchin monkeys reposition canes, altering the spatial relation between the canes and a piece of food so that the canes can be used to retrieve the food (Cummins-Sebree & Fragaszy, 2005). Thus capuchin monkeys are good candidates for a study in which manual exploration could play an important role in guiding goal-directed behavior with an object (of which tooling is just one example).

Humans occasionally tool with objects that require managing multiple DoFs external to the body (e.g., a pivoting screwdriver), but to our knowledge, only one previous study has investigated how humans handle objects with multiple DoFs in a context related to tooling. Mangalam et al. (2019) showed that humans discover affordances of a jointed object through effortful touch (i.e., by wielding the object in the hand). They gave blindfolded participants a handled aluminum hoe with the handle variously configured as two segments with one planar joint, or three segments with two (orthogonal) planar joints. They instructed the participants to manage the two joints to obtain a single rigid segment such that the object could be used conventionally as a hoe. All the participants could do this. The participants changed the object’s configuration many more times than necessary to obtain a single rigid segment but the number of changes they made per trial declined with experience. Moreover, they changed the configuration of the jointed object more rapidly with experience. The authors interpreted these findings as evidence that the participants learned to detect the inertial properties of the object affecting its respective tooling affordances, and that they became increasingly accurate in the perception of such properties. These findings provide support for some predictions of the embodied theory of tooling, but Mangalam et al.’s (2019) study did not address other predictions because the participants did not use the object; they did not tool with it. The present study addresses that lacuna.

In sum, in an experiment with capuchin monkeys, we examined three predictions drawn from the embodied theory of tooling, which considers tooling with a hand tool as an activity of the body-plus-object system. The theory leads to the general hypothesis that one challenge in tooling is managing altered degrees of freedom of the body-plus-object system. Here, we examine specific hypotheses that follow from this general hypothesis. Using a jointed object with one or more additional DoFs than a conventional object for a task previously practiced with the conventional object should (1) result in a less efficient performance, reflecting the greater difficulty in using the jointed tool; (2) be accompanied by greater changes in performance with practice than evident with the conventional object, reflecting learning to control the initially unfamiliar jointed object, and (3) result in more exploratory behaviors, reflecting efforts to discover how or if the jointed object could be used. We predicted (1) the monkeys would use the rigid hoe more efficiently than the jointed hoe, (2) that their performance would improve with experience more for the jointed hoe than the rigid hoe, and 3) they would perform more exploratory actions with the jointed hoe than the rigid hoe. We used a hoe with no, one, or two joints in the handle in the study with monkeys. In a follow-up study with humans, we examined the prediction that (4) people would preferentially act to make the handle of a jointed hoe rigid, to use the hoe in a conventional manner, in accord with reducing DoFs of the body-plus-object system, and in accord with a cognitive bias to use objects in conventional ways —“performatory knowledge” (Osiurak & Badets, 2016). We used an identical hoe with humans (scaled to human size), and presented it in what we anticipated was the most difficult condition: the condition with two joints in the handle. We present the study with monkeys first.

Experiment 1. Monkeys

Method

Subjects

Three male tufted capuchin monkeys (20 – 25 years old) housed in the Primate Cognition and Behavior Laboratory at the University of Georgia (Athens, GA) participated in the present study. They were unfamiliar with jointed objects at the start of the experiment, although they had used hoes with conventional rigid handles to pull objects across surfaces and around barriers in several studies carried out in 1998 – 2005 (Fragaszy & Cummins-Sebree, 2005; Fragaszy, Visalberghi, & Fedigan, 2004) and other studies involving handling objects (e.g., Hoy Kennedy & Fragaszy, 2008; Fragaszy et al., 2011; Rosengart & Fragaszy, 2005). Data were collected for this study in 2012 – 2013. The monkeys had a consistent diet of monkey chow and fruit provided twice a day, and water was available ad libitum. The Institutional Animal Care and Use Committee at the University of Georgia approved the study (IACUC, A2013 03-001-Y3-A2).

Materials

We designed and manufactured an aluminum hoe (62 g) consisting of a blade (7.5×7.5 cm) and a handle (14.0 cm long, .5 cm diameter; Figure 1a). The handle scaled 1.2:1 the average length of the forearm and 1:6.4 to the average palm span of an adult male capuchin monkey (cf. Fragaszy et al., 1989). By using a locking metal sleeve, the hoe’s handle could be configured by the experimenter as a single rigid segment, two segments with one planar joint (at the proximal joint; Figure 2a) or three segments with two (orthogonal) planar joints (Figure 2c). When the handle had one planar joint, at the “zero” position with that joint oriented directly downward and 150 – 160° to either side of that position, the blade of the hoe hinged down or sideways when the blade of the hoe was lifted above the surface. When the hoe was turned 180° from the zero position and 20 – 30 ° to either side, the blade remained vertical when the hoe was lifted. Finally, when the handle had two planar (orthogonal) joints, the blade hinged down and/or or sideways unless held at the zero position. Rendering the hoe to keep the blade in a vertical position could be accomplished in both jointed conditions by rotating the handle to the zero position and keeping it there. Reflective tape was placed on three corners of the blade and at points along the handle to mark the position of the blade and joints of the handle in video replay. The monkeys used the hoe to retrieve a token (1.6×1.6×1.6 cm aluminum cube) covered with reflective tape to be maximally visible in video replay.

Figure 1.

The jointed hoes used to retrieve a token, shown in the two-joints condition, with the handle in a straight position. (a) Smaller/lighter hoe used by the monkeys in Study 1. (b) Larger/heavier hoe used by the humans in Study 2. The joints each allowed 90° of movement in one plane, and the planes were orthogonal to each other.

Figure 2.

The jointed hoes illustrating possible positions when both planar joints were free to move. The hoe could be in one of the three positions at any time while being used by the monkeys or humans. (a) Proximally jointed smaller hoe. (b) Distally jointed smaller hoe. (c) Proximally and distally jointed smaller hoe. (d) Proximally jointed larger hoe. (e) Distally jointed larger hoe. (f) Proximally and distally jointed larger hoe. The red circles surround the position of the joint in the handle of the hoe.

The monkeys were tested in a cage (60 × 75 × 75 cm) with acrylic sides and a wire mesh floor and top. The sides and back of the cage were draped on the outside with black cloth to reduce reflections. A rectangular aperture (4 cm × 12cm) centered 8 cm above the floor at the front panel allowed the monkey to reach one arm out of the cage at a time. The hoe and the token were presented to each monkey on a flat, horizontal PVC platform (30×45 cm) covered in black felt. The platform was positioned in front of and in contact with the test cage with its surface leveled with the lower edge of the aperture in the front panel.

Each monkey’s actions with the hoe were recorded using Canon GL2^™ and Canon FX100^™ digital video cameras (30 fps, Canon Inc., Tokyo, Japan) positioned at right angles with respect to the front of the cage in which the monkey was tested. A toggled light signal allowed synchronization of the video replay from the two cameras. Two additional video cameras placed at angled frontal views provided supplementary recordings.

Experimental Procedure

We anticipated that the monkeys would face difficulty (and potential failure) when they tried to use the jointed hoe, particularly when it had two joints. To instill strong familiarity with the rigid handle, and to allow the monkeys first to master what we expected would be an easy task, we presented the rigid handle condition first (100 trials). To ease the monkeys into working with the most difficult condition (the hoe with two joints) we presented the single jointed condition first (100 trials), then the two joints condition (100 trials). Each monkey was assigned to receive the hoe with one specific side of the four-sided rectangular blade resting on the surface at the initiation of each trial, which determines in which directions the blade can hinge when lifted off the surface in the jointed trials. Monkeys 1 and 2 were assigned the same side (illustrated in Figure 2a); the third monkey was assigned a different side (illustrated in Figure 2e).

After completing Part 1, and following a temporal break of a few months, the monkeys completed Part 2, a ‘refresher’ series of 10 trials for each condition, in random order and with the same positioning rules for the blade as in Part 1. Immediately after completing Part 2, the monkeys completed Part 3, intended to evaluate the generality of the monkeys’ mastery of the hoe following equal practice with each condition, and with the blade presented in all four positions. Part 3 consisted of 40 trials of each condition, with the order of the conditions and position of the blade randomized for each monkey. Thus, in Part 3, the monkeys completed 30 trials with the blade in a familiar orientation (10 for each condition) and 90 trials (30 for each condition) with the blade in unfamiliar orientations, with condition and blade position both randomized.

Each monkey completed 20 trials per daily testing session or until he lost interest in the task, whichever came first. Monkeys rarely failed to complete 20 trials per session in any condition. To start a trial, the experimenter placed the token on the tray at a marked point where the monkey could move the blade of the hoe behind the token while holding the handle of the hoe. The Experimenter then placed the head of the hoe between the monkey and the token such that the monkey had to move the blade of the hoe laterally and/or forward at least 2 cm to place it behind the token. The Experimenter held the handle in a straight position pointing toward the aperture in the front panel of the test cage, with the blade resting vertically on the platform’s surface. While holding the hoe in this position, the Experimenter toggled a light to signal the start of the trial on the video recording, held the hoe until the monkey grasped the handle. The monkey was allowed to handle the hoe until he retrieved the token using the hoe, or, if he knocked the token out of reach, the experimenter replaced the token at the marked point for that trial. When the monkey handed the token to the experimenter, the experimenter gave the monkey a small food reward (dried fruit or nut). Once the monkey had consumed the food, the next trial was prepared until the session ended. If the monkey moved away from the task for longer than a minute, we considered the trial ended, and the experimenter tried to engage the monkey’s interest again. In the jointed conditions, we did this by offering a trial in which the handle had no joints. If the monkey remained disinterested, we ended the testing session. Each monkey completed the task (i.e., retrieved the token and exchanged it for a food reward) in 100 trials per condition in Part 1, 30 trials in Part 2, and 120 trials in Part 3.

In Part 1 we placed the token in one of five locations, randomly assigned within sessions. The five locations were at 2.5 cm intervals along a curve so that all the points were the same distance from the midpoint of the aperture in the front panel of the test cage, and the middle position was centered with the midline of the aperture. Each position was marked with a small strip of reflective tape. This procedure ensured that the monkey learned to move the hoe’s blade to varied positions relative to the surface. For Parts 2 and 3, the token was placed in the middle position, directly in front of the aperture. Supplemental video S1 shows a monkey completing a trial in Part 3 from a frontal view.

Video Coding

Video recordings from the cameras were synchronized in Adobe Premiere Pro^™ (Adobe Inc., San Francisco, CA) using the flashing light as a synchronization point. We coded each monkey’s actions frame-by-frame using Observer XT^™ software (Noldus Inc., Leesburg, VA) while he contacted, positioned, and manipulated the hoe, the position of the blade and the movement of the hoe, and the movement of the token towards or away from the monkey. A list of events and behaviors coded is provided in Table 1. Only those trials were coded in which we could view the full trial in video replay. Among the trials which met this criterion, we coded the first 20 and last 20 trials per condition from Part 1 (120 trials coded per monkey in Part 1) and the last 20 trials (out of 40) per condition from Part 3 (60 trials coded per monkey in Part 3). We used Observer XT^™ software (Noldus Inc., Leesburg, VA) for video coding.

Table 1.

Events and actions coded from video while monkeys used a jointed hoe to retrieve a token and exchange it for a food reward

Events

New position for the blade (which corners of the blade are oriented toward the platform)

Final position for the blade when the token is retrieved (horizontal or vertical)

Performatory actions

Reposition the hand– the hand releases, loses contact or slides over the handle, and re-grasps the handle at a new position within 2 s

Move the hoe– the hoe moves at least 2 cm above the testing platform

Swipe the hoe– the hoe swipes at least 2 cm across the testing platform

Pull the hoe toward oneself – the hoe shifts toward the monkey at least 2 cm

Exploratory actions

Push the handle down – the handle moves visibly downward (and the blade tilts on the testing platform)

Lift the handle up – the handle moves visibly upward (but leaves the blade on the testing platform)

Lift the whole hoe (but not move it above the testing platform)

Strike the testing platform with the hoe

Rotate the blade by turning the handle (followed by a new coding of the position for the blade, if rotated far enough)

The reliability of the coding scheme was initially established by J.D.L. and J.E.R.C. independently coding sets of 12 trials from Part 3 (four trials for each monkey randomly selected from each of the three conditions) until inter-observer agreement the occurrence of each variable exceeded 90% on two successive sets of trials. Subsequent coders for Part 3 (S.V., M.C.K., and A.H.) established reliability with J.D.L and J.E.R.C. in the same manner by coding sets of 12 trials from Part 3. Coders working with Part 1 trials (S.S., M.Q., and M.C.S.) established reliability by coding sets of 12 trials from Part 3, and subsequently with J.D.L. by coding sets of five trials randomly selected across monkeys and conditions from Part 1 (as we anticipated more variable behavior in Part 1 than Part 3). Following each replication, the two coders reviewed discrepancies to arrive at a consensual agreement about the discrepantly coded events. During routine coding, notes of unusual events allowed subsequent joint review.

Analysis

We analyzed the frequency of each monkey’s actions separately. The data used in this report are presented in Supplemental Table 1. We pooled the following actions into the class of Performatory Actions: (i) reposition the hand on the hoe, (ii) move the hoe above the testing platform, (ii) swipe the hoe across the platform, (iv) and pull the hoe towards oneself (Table 1). We pooled other actions into the class of Exploratory Actions (Table 1). To test our first prediction, that the monkeys would use the hoe with a rigid handle more efficiently than one with a jointed handle, we compared their frequencies of performatory actions across conditions in Part 1, when the monkeys first encountered each condition, and in Part 3, terminal performance, using χ² goodness of fit tests. A reduction in the frequency of performatory actions reflects improved efficiency of performance. To examine our second prediction, that the monkeys’ performance would improve with practice more for the jointed hoes than the hoe with no joint, we used χ² goodness of fit tests to compare the frequencies of performatory actions between the first 20 trials per condition in Part 1 and last 20 trials of the same conditions in Part 3. Finally, to examine the prediction that the monkeys in the present study would perform more exploratory actions with the jointed hoes than the hoe with no joint, we used χ² goodness of fit tests to compare the frequencies of exploratory actions across conditions in Part 1 and again in Part 3.

Results

Effect of Number of Joints on Performatory Actions

Part 1: Initial performance

The monkeys used 426 to 470 performatory actions each over the first 20 trials of each condition (60 trials total per monkey; 7.1 – 7.8 per trial) in Part 1. Across the first 20 trials, each monkey used more performatory actions when the handle had one joint (M = 216.0) than when it had no joint (M = 151.3) or two joints (M = 121.7) (χ² (2) = 5.32, W = .11, p < .05 Monkey 1; χ² (2) = 59.49, W = .37, p < .001, Monkey 2, and χ² (2) = 127.1, W = .52, p < .001, Monkey 3; see Figure). Thus, adding one joint to the hoe impacted efficiency of performance, but adding two joints did not.

Part 3: Terminal Performance

The monkeys exhibited effective mastery of the task in all conditions in Part 3 (Figure 2 and Table S1). There was no discernible difference in performance for any monkey on any variable on trials with familiar positions of the blade compared to trials with novel positions of the blade, and we do not address these comparisons further. Across all conditions, the monkeys used 2.4, 2.7, and 2.7 performatory actions per trial, Monkey 1, Monkey 2, and Monkey 3, respectively, with small deviations from the expected values across conditions for each monkey (all p > .10). Thus, by the end of the experiment, the monkeys used the jointed hoes as efficiently as the non-jointed hoe.

The blade’s final position when the monkey retrieved the cube could be vertical (indicating that the monkey slid the cube across the surface using the hoe conventionally) or horizontal (indicating that the monkey rested the blade on the top of the cube to draw the cube toward itself). The monkeys always used the blade in a vertical position when the handle was rigid, but rarely did so when it was jointed (one to six times across 40 trials per monkey).

Magnitude of Learning with the Nonjointed and Jointed Hoes

In each condition, each monkey exhibited a significant reduction in the number of performatory actions from the first 20 trials in Part 1 to the last 20 trials in Part 3 (Monkey 1: χ²(2) = 9.72, W = .62, p < .010; Monkey 2: χ²(2) = 63.62, W = .28; p <.001; and Monkey 3: χ²(2) = 59.07, W = .26, p < .001), indicating improvement in performance with practice in all three conditions. In the first 20 trials in Part 1, each monkey used on average 151, 216, and 122 performatory actions when using the hoe with no joint, one joint, and two joints, respectively. The number of performatory actions in the last 20 trials in Part 3 clustered tightly between 42 and 59 per condition (see Figure 3). The largest reductions in performatory actions for each monkey occurred in the one-joint condition—the values declined from 180 to 50 for Monkey 1 (70% reduction), from 203 to 52 to Monkey 2 (70% reduction), and from 265 to 58 for Monkey 3 (78% reduction; all χ²(1) > 73.48; W > .32; p < .001). Reductions in the number of performatory actions in the no-joint condition averaged 48% (range 27%–67%), and in the two joint condition, 51% (range 20% to 68%). Monkeys 1 and 2 exhibited a significant reduction in the frequency of performatory actions in the two-joint condition between the first 20 trials in Part 1 and the last 20 trials in Part 3 (Monkey 1: from 142 to 45, χ²(1) = 50.32; W = .52; p < .001; Monkey 2: from 149 to 53, χ²(1) = 45.62; W = .48; p < .001). Monkeys 1 and 3 produced fewer performatory actions in Part 3 than in Part 1 in the no-joint condition (Monkey 1: from 148 to 49, χ²(1) = 49.75, W = .25, p < .001; Monkey 3: from 84 to 42, χ²(1) = 14.00, W = .11, p < .001).

Figure 3.

The number of Performatory Actions performed by each monkey (Left to Right) with the hoe in each phase of testing when the hoe had 0, 1, and 2 joints.

Frequencies of Exploratory Actions

The monkeys explored the hoes by rotating the handle, lifting the whole hoe, and striking it on the testing platform. Across all conditions, the monkeys produced 81, 180, and 143 exploratory actions with the hoe (monkeys 1, 2, and 3, respectively) in the first 20 trials of Part 1. While the distribution of exploratory actions across conditions for each monkey differed strongly from expected (χ² (2) =19.83, 43.25, and 8.57; W = .44, .48, and .24; p’s < .001, .001, and .05; Monkey 1, 2, 3, respectively), only Monkey 2 performed more exploratory actions in the jointed conditions than in the unjointed condition (see Figure 3). Notice the difference in scale on the y axis between Figures 3 and 4.

Figure 4.

The number of Exploratory Actions (excluding “other”) performed by each monkey (Left to Right) with the hoe in each phase of testing when the hoe had 0, 1, and 2 joints.

All monkeys produced many more exploratory actions in the first 20 trials of Part 1 than in the last 20 trials of Part 3 (M = 2.25 per trial in Part 1, vs. .28/per trial in Part 3; all χ²(1) ≥ 36.84; W > .36; p < .001; see Figure 4 and Table S1). In Part 3 each monkey produced a single exploratory action in the rigid condition but several exploratory actions in the one-joint condition (11, 9, and 4) and the two-joints condition (8, 7, and 9; Monkey 1, Monkey 2, and Monkey 3, respectively). This distribution deviated from the expected values for Monkey 1 (χ²(2) = 7.90, W = .62, p < .05) and Monkey 2 (χ²(2) = 6.11, W = .60, p < .05). Expected values of less than five for two cells rendered a goodness of fit test inappropriate for these data for Monkey 3.

Discussion

Fragaszy & Mangalam’s (2018; Mangalam & Fragaszy, 2016) embodied theory of tooling frames using grasped objects as tools (tooling) as an activity of the body-plus-object system, set within a given task setting. The actor must coordinate movements of their body and the tool as a system to achieve a given mechanical goal (to strike, to scrape, to pierce, etc.) as part of reaching a functional goal (to open something, to move something, etc.). Conditions that increase the difficulty of coordinating movements of the body-plus-object system (such as adding DoF’s to the system) should increase the difficulty of tooling, evident as decreased efficiency or delayed mastery. In the present study, we presented a simple tooling problem, retrieving a token using a hoe, to tufted capuchin monkeys (Sapajus spp.)—a taxon noted for its manual dexterity and affinity for using tools (Falotico & Ottoni, 2016; Fragaszy, Visalberghi, & Fedigan, 2004). The hoe featured a handle with no joint (like a conventional hoe) or a handle with one or two joints.

Our first prediction was that adding one or two joints in the handle of a hoe would increase the difficulty of using the hoe to pull in an object from out of reach, in accord with the proposed effects of an increased number of DoFs in the body-plus-object system. Our results partially supported this prediction. The monkeys produced significantly more performatory actions (that contributed to moving the hoe into the position to retrieve the token) in the one-joint condition than the no-joint condition in the first 20 trials of Part 1. Adding a joint to the handle made it more difficult for the monkeys to use the hoe when it was first presented in that novel form. However, the monkeys produced an equivalent number of performatory actions with the rigid hoe and the hoe with two joints, suggesting that the additional DoFs in the hoe with two joints led to the monkeys acting differently with this object than they acted with the hoe with one joint.

Our second prediction was that the monkeys would show greater performance changes with practice when using the jointed hoes than the non-jointed (conventional) hoe. We assessed learning in each condition as the decrease in the number of performatory actions from the start (the first 20 trials in Part 1) to the end of testing (the last 20 trials in Part 3). The magnitude of improvement over testing was greatest for the hoe with one joint. The changes evident in the conditions presenting the hoe with no joints or two joints were smaller and equivalent in the two conditions. This outcome partially supports the second prediction.

Our final prediction was that the monkeys would more frequently manipulate the jointed hoes than the nonjointed hoe in exploratory ways, to discover affordances of acting with the novel configuration, including, perhaps, returning the jointed hoe to the familiar rigid condition (which the monkey could do by rotating the handle of the hoe). The results partially supported this prediction as well. Two monkeys explored the hoe equivalently often in each condition when they first encountered them (in the first 20 trials of Part 1). One monkey explored the jointed hoes more frequently than the non-jointed hoe. However, at the end of testing, when they had used each of the hoes more than 100 times, all the monkeys produced more exploratory behaviors in the jointed conditions than in the no-joint condition. These findings suggest that the jointed conditions continued to provoke the monkeys’ interest. However, the monkeys used exploratory actions at relatively low rates across all testing (less than two actions per trial in Part 1; many fewer in Part 3). The monkeys occasionally acted in ways that could enable them to perceive the situation’s relevant affordances (i.e., how they could act with the hoes to achieve their goal), but they did not seem to do so intentionally. Persistent but low rates of haptic exploration characterize capuchin monkeys’ actions in other settings as well—as when searching for sunflower seeds on an unseen, uneven surface (Lacreuse & Fragaszy, 1997) and preparing to use stones to crack nuts (Fragaszy et al., 2019).

Capuchin monkeys have been noted to explore objects manually in an intentional manner (i.e., to guide future action). For example, in the study by Visalberghi et al. (2009), wild tufted capuchin monkeys touched, tapped and lifted stones of variable mass before choosing one (of two or three presented) to transport to an anvil where they used it to crack a nut. Phillips et al. (2003) and Gunst et al. (2010) reported that wild tufted capuchin monkeys tapped bamboo stalks prior to biting them open to extract invertebrate prey from the interior. Tapping supports the detection of a stable property of an object (most likely density, in the case of bamboo stalks and stones, which reflects value for biting open, in the case of bamboo, and value for cracking, in the case of a stone). Perhaps tufted capuchin monkeys explore objects to guide direct action on the object (e.g., to locate prey, and thus, where to bite the bamboo, or to select, grasp and transport a stone) but not to explore the allocentric DoFs associated with the object, as in the case of working with a jointed object. If this hypothesis is confirmed, it will provide one reason why capuchin monkeys do not master instrumental skills or solve instrumental problems that incorporate managing one or more allocentric spatial or force relations as readily as do humans. Humans manifestly can manage these kinds of problems (such as shaping a clay vessel using a potting wheel, using sticks and strings to control the limbs of a puppet, and controlling a fish hook at the end of a line on a fishing pole, to provide a few familiar examples).

Overall, the findings of this study partially supported the predictions drawn from the embodied theory of tooling. The study’s small sample size renders our findings preliminary, and we encourage others to replicate this work. However, the main finding that one joint makes a big difference in performance compared to no joint, but that two joints did not, is robust. Adding one degree of freedom to the system, as a planar joint in the handle of the hoe used to pull in a token, compare to the non-jointed hoe, initially reduced the efficiency of performance, led to greater change in performance with practice (as the monkeys became as efficient with the jointed hoes as with the unjointed hoe), and was associated with more frequent exploration after extended experience with all the hoes. These findings were predicted. However, the findings with the hoe with two joints did not confirm the predictions. Two monkeys acted similarly with the hoe with no or two joints, both when they were first presented and after extended practice with all three kinds of hoes used in the present study. This result may reflect capuchin monkeys’ capacity to detect inertial properties of an object, their motivation to alter their actions, or the particular affordances of using the double-jointed hoe in this experiment. When using the hoe with a jointed handle, the monkeys relied on an unanticipated solution. They allowed the hoe’s blade to flop horizontally and placed it on top of the token to pull it in. Although this solution is unlikely to work when retrieving a heavy object, it did work with the small token used in this study. The monkeys accommodated their action to the hoe’s properties as they encountered it, instead of rendering the handle to be rigid, which they could have done simply by rotating it.

In part to explore this unexpected result, and more generally to determine if humans manage the body-plus-object with a jointed hoe differently than the monkeys, we conducted Study 2. We presented people with a human-scaled hoe identical in structure to the hoe presented to the monkeys but scaled for human body proportions. We presented the hoe with two joints specifically to evaluate if humans would spontaneously act to render the hoe’s handle rigid, as people had demonstrated they do rather easily when asked to do so in Mangalam et al.’s (2019) study using the same hoe.

Experiment 2. People

Method

Participants

Participants were recruited from the Research Participation Pool in the Psychology Department at the University of Georgia. One man and five women (M±SD age = 23.0±4.3 years, 20–24 years), all self-identified as right-handed, voluntarily participated in the present study after providing informed consent. The study was approved by the Institutional Review Board (IRB, STUDY00002073) at the University of Georgia.

Materials

We designed and manufactured an aluminum hoe (464 g) consisting of a blade (7.5×7.5 cm) and a handle (25.5 cm long, 1 cm diameter; Figure 1b) with two (orthogonal) planar joints. Reflective tape placed on three corners of the blade and at points along the handle marked the position of the joints in video replay (Figure 2 d–f). The participants used the hoe to retrieve a token (1.6×1.6×1.6 cm aluminum cube) covered with reflective tape to be maximally visible in video replay.

Experimental Procedure

Each participant sat in an armless chair in the front of a table. The hoe was placed with the end of the handle about 25 cm in the front of the participant’s midline, and the token was placed about 30 cm in front of the hoe’s blade in the same line. We instructed the participant to pick up the hoe, retrieve the token using the hoe, and hand the token to the experimenter. A new trial began as soon as the token and hoe were repositioned, and the experimenter gave a verbal ‘start’ signal. We video-recorded the participant’s behavior in his/her sagittal plane at 30 fps using a Casio EXILIM EX-ZR700^™ (Casio America, Inc., Dover, NJ) camera. The participants completed all trials in a single session.

In Part 1, the hoe was presented to each participant with the blade in three randomly selected orientations relative to the surface. Each participant completed 20 consecutive trials per orientation (60 trials in all). The order of the orientations was randomized across participants. To evaluate the generality of the findings in Part 1, in Part 2, the hoe was presented in all four orientations, so that each participant experienced one novel orientation. Each participant completed 40 trials, ten trials per orientation with the order of the orientations randomized across trials for each participant.

Video coding and Analysis

M.M. coded the video of each trial frame by frame. Participants always retrieved the token, so we do not address that variable further. For each trial, we determined whether the participant manipulated the hoe to render its handle rigid and used it in that form to pull in the token (i.e., used it conventionally).

Results

In Part 1, only two participants manipulated the hoe to render its handle rigid and used the hoe conventionally (P1 and P4, all 40 and 37 out of 40 trials, respectively, across all conditions). The other four participants did not do so in any trial (see Table 2). These four participants, like the monkeys in Study 1, allowed the blade of the hoe to fall and placed it horizontally on the top of the token to pull it in.

Table 2.

The number of trials in which the human participants manipulated manipulate the hoe with two joints to transform it into a rigid, conventional hoe

Participant	Part 1: Training				Part 2: Testing
	Orientation 1 N = 20	Orientation 2 N = 20	Orientation 3 n = 20	Orientation 4 N = 20	Orientation 1 N = 10	Orientation 2 N = 10	Orientation 3 N = 10	Orientation 4 N = 10
P1		20	20	20 (19)	10	10 (20)	10 (4)	10 (8)
P2	0	0	20 (20)		0	1	10 (10)	10 (10)
P3	0	0	20 (20)		0	0	8 (8)	10 (10)
P4	20	17	20 (13)		10	10	10	10 (10)
P5	0	0		13 (13)	0	0	0	10 (10)
P6		0	20 (20)	20 (20)	0	0	10 (10)	10 (10)

Note: The values in parenthesis are the number of trials in which the handle was rigid when the object was lifted off the table.

Unlike the hoe used in Study 1, in certain orientations of the blade, the heavier hoe used in this study retained a rigid handle when lifted off the table unless the handle was rotated. When this happened in Part 1, all six participants used the hoe conventionally. Thus, each participant experienced using the hoe conventionally.

In Part 2, the same two participants (P1 and P4) who manipulated the hoe to render its handle rigid and use it conventionally in Part 1 did so again, for all 40 trials. The other participants used the hoe conventionally only when the handle remained rigid when they picked it up (Table 2). In other trials, as in Part 1, they placed the hoe’s blade horizontally on top of the token to pull it in. In summary, the participants behaved similarly in Parts 1 and 2, and a majority of participants did not alter the properties of the hoe’s handle or use the hoe conventionally. The inclusion of a novel position of the hoe in Part 2 did not affect the participants’ actions with the hoe.

Discussion

We evaluated how human participants would use a hoe to retrieve a token when the hoe featured a handle with two joints (and thus two DoFs more than a conventional hoe with a rigid handle). We anticipated that the participants would discover how, by rotating the handle, they could render the handle rigid and the blade to have a stable vertical position and that they would prefer to use the hoe in this conventional configuration. Our observations do not support either hypothesis. Note that we included just six participants in our study, enough to identify variability within normal adult humans, but clearly our study should be replicated to confirm our findings.

Mangalam et al. (2019) presented this same hoe with one joint and with two joints to twelve blindfolded participants and asked them to reconfigure the hoe to have a rigid handle. People produced many more than the minimum number of changes needed in the configuration of the hoe prior to determining that it had a rigid handle, but this number declined by half across 18 trials within a single testing session. These findings suggest that humans readily learn to perceive the affordances of a handled jointed object and that under specific instructions, they learn to control the DoFs of such objects in the absence of vision. These findings are similar to those of Wagman and Carello (2001), who report that humans readily detected whether a rigid object could be used to poke or to hammer after wielding it in the absence of vision. Thus we were surprised that a minority of participants did so in the present study, where they could see the hoe’s configuration and how it changed when they moved it. Instead, the majority of participants, like the capuchin monkeys in Study 1, did not manipulate the double-jointed hoe to use it conventionally, even when they could have done so relatively easily. If people can tool with an object unconventionally but apparently equally effectively, then adoption of conventional use might require explicit verbal instruction, as was provided in Mangalam et al.’s (2019) study, or other forms of social mediation. We found no evidence for reliance on functional knowledge (Osiurak & Badets, 2016) in this context, which would have been evident as a bias to use the hoe in a conventional way, with a rigid handle, or that humans tried to reduce or to control extra DoFs in the object, as predicted by the embodied theory of tooling. In these respects, the humans’ performance was similar to the monkeys’ performance. Especially given the small N of the present study (six people), our findings should be considered preliminary. We hope others will follow up on this work to confirm our findings.

General Discussion

The embodied theory of tooling posits that individuals act to discover the affordances of actions with particular objects, that they learn to manage the DoFs of movement of their bodies and a grasped object when they use the object as a tool, and that the challenge of doing so increases with the number of DoFs present in the body-plus-object system (Fragaszy & Mangalam, 2018; Mangalam & Fragaszy, 2016). Our goals in the present study were to examine several predictions drawn from this theory relating to how individuals would handle an unconventional object to tool with it. We presented tufted capuchin monkeys and humans with a hoe to pull in a token across a flat surface. The hoe’s handle was a single rigid segment, two segments with one joint, or three segments with two joints. In the jointed conditions, the hoe’s blade could pivot by its own mass 90° in one plane at each joint, and thus could fall parallel to the surface, parallel to the hoe’s handle, or both.

The monkeys displayed most of the patterns of behavior that we predicted. When they first encountered the hoe with a jointed handle, they used more actions to bring in the token than they used with the hoe with a rigid handle. When the task was well-learned, the monkeys pulled in the token as efficiently when the handle was jointed as when it was rigid. Improvement in performance was particularly marked for the hoe with a single joint in the handle. When the handle was rigid, they placed the blade of the hoe behind the token and perpendicular to the surface to pull in the token. However, when the handle was jointed, they predominantly placed the blade on top of the token to pull it in. Instead of working to alter the handle of the hoe, they accommodated their actions to each hoe as it was presented to them. This was the case in initial trials and remained the case when the task was well-learned. They occasionally acted in exploratory ways with the hoe, more often early in testing than in later testing, and in later testing, more often with the jointed hoes than with the nonjointed hoe. However, overall they explored the hoes infrequently and not in an apparently directed way.

These findings support some aspects of the embodied theory. Adding one DoF to the object the monkeys used as a tool impacted their performance in the predicted ways. However, the monkeys’ low rates of exploration suggested that they do not intentionally explore objects they will use as tools. Their lack of efforts to alter the condition of the handle of the hoe suggests that, at least when they generate a suitable solution using an object as they find it, they are not particularly sensitive to perceptual information about affordances for action with jointed objects, or that they do not act on the perceptual information that they generate by their actions. One could say they rapidly learned an alternate way of using the jointed hoes. Further investigation of this question will require a differently-designed study.

Povinelli & Frey (2016) investigated a similar question by offering chimpanzees two visually similar rakes, where the head of one rake deformed (due to the action of a spring connecting the head to the handle) when pulled against a reward object placed on a table. The authors reported that even after experience attempting to use a spring-loaded rake to pull in a reward (and failing to get the reward in those attempts), when presented with two rakes, chimpanzees did not act with “prospective diagnostic interventions” with either rake to determine if it had a rigid connection between head and handle. Instead, they indiscriminately picked up one rake and attempted to use it. Even when holding a rake that was deforming in their hands as they held it, they transported that rake to the work area and attempted to use it to pull in the reward. The authors concluded that, despite chimpanzees’ considerable expertise at using rakes in other situations, they do not act in an intentional way to discover how a selected object will perform when they attempt to use it for a given task. In Povinelli & Frey’s (2016) experiment, the chimpanzees had no option to modify the rake through action, so it is unknown if they would have altered the rake by handling it, as was possible in the present study (by rotating the handle of the hoe). Perhaps capuchins or chimpanzees, when presented with a task posing a greater penalty than presented in the present study for using a rake “as is”, would act prospectively.

In addition to probing the boundaries of various species’ prospective actions with objects they could use as tools, it may be fruitful to probe the nature of dynamic haptic perception (effortful touch) in exploration and control of object movement, outside of a tooling setting (Mangalam et al., 2017, 2018; Wagman & Carello, 2001; Wagman & Taylor, 2005). Wagman and Carello (2001) note that adult humans need only minimal experience to “tune” their perception to the inertial properties of a given object to determine if it is useful for poking or for hammering. Psychophysical studies would be a starting point to explore how sensitive nonhuman animals are in similar situations.

Counter to our prediction, only two of six human participants using a hoe with two joints in the handle routinely rotated the handle to use the hoe conventionally. The other four participants, like the monkeys, accommodated their actions to the hoe as they picked it up, putting the blade of the hoe horizontally on top of the token to move. Our findings suggest that most individuals in our study did not exhibit functional fixedness, a bias to use an object in a conventional way (German & Barrett, 2005; Munoz-Rubke et al., 2018. Munoz-Rubke et al. (2018) noted that functional fixedness impacted how people approached simpler mechanical problems such as balancing a bar by sliding a cylinder beneath it, but did not play a role in more difficult mechanical problems that they presented, such as pushing a marble down a hole from the side by using a flexible rod. Using a hoe to rake in an object is simpler than all these puzzles, and thus, a good candidate to elicit a bias to adopt a conventional solution. Nonetheless, this bias was not evident in the present study. Adopting a conventional solution when using an unfamiliar hoe to retrieve something might follow consistently from verbal instruction (as given to participants in Mangalam et al.’s (2019) study, where participants readily moved the jointed object to achieve a rigid handle when asked to do so) or other forms of social mediation. In any case, our findings suggest that humans do not automatically try to use an object conventionally when an unconventional use is adequate. The findings remind us that behavioral flexibility can produce simpler as well as more complicated behavior in our species as in others.

Our findings indicate aspects of the embodied theory of tooling that fared well and others that need refinement. Adding one degree of freedom to the system affected monkeys’ behavior more or less as predicted. However, exploration contributed to the organization of behavior with the jointed object less than we predicted, and the addition of DoFs to the object did not impact the coordination of movement with the object in monkeys or humans. Improved understanding of how wielding an object allows the user to determine its usability is needed. Developments in understanding haptic perception— an active research topic—will help in this direction (Cabe, 2019; Carello & Turvey, 2017; Mangalam et al., 2020; Mangalam & Kelty-Stephen, 2020; Turvey & Fonseca, 2014; Wagman & Hajnal, 2014). We look forward to building upon this developing body of research to improve the theory of tooling.