Abstract

Our behavior is predicated on mental models of the environment that must be updated to accommodate incoming information. We had 13 right-brain–damaged (RBD) patients and 10 left-brain–damaged (LBD) patients play the children's game “rock, paper, scissors” against a computer opponent that covertly altered its strategy. Healthy age-matched controls and LBD patients quickly detected extreme departures from uniform play (“paper” chosen on 80% of trials), but the RBD patient group did not. Seven RBD patients presented with neglect and although this was associated with greater impairment in strategy updating, there were exceptions: 2 of 7 neglect patients performed above the median of the patient group and 1 of the 6 nonneglect participants was severely impaired. Although speculative, lesion analyses contrasting high and low performing patients showed that severe impairments were associated with insula and putamen lesions. Interestingly, relative to the controls, the LBD group tended to “maximize” choices in the strongly biased condition (i.e., optimal strategy chosen on 100% of the trials), whereas controls “matched” the computer's strategy (i.e., optimal strategy chosen on 80% of the trials). We conclude that RBD leads to impaired updating of mental models to exploit environmental changes.

Introduction

Humans are constantly faced with an ongoing need to choose, from among an array of possible actions, the option that optimizes a desired outcome. Such choices may manifest as overt actions (e.g., stepping left to avoid another person in the mall) or as higher level decisions (e.g., investing in high- vs. low-risk stock). Inherent in both types of decisions is a reliance on some model of the environment that reflects prior experience. However, this reliance is not without risks; the world is in a state of flux and our environmental contingencies change. To make optimal choices, we need to update our mental models of the environment to reflect changed circumstances. On the other hand, as the world is also noisy, we should not “constantly” change our expectations or time, and resources will be wasted chasing our tails (Tenenbaum et al. 2011).

The cognitive mechanisms and brain structures that are important for decision making are under active study, but while impairments in updating may result in poor decisions, any such impairments would be distinct from a decision making impairment per se. If our mental models are a poor match to circumstance, we may decide on an action that in turn leads to a poor outcome, but the functions of the decision making process itself may be quite normal. By analogy, this would be like using logical reasoning to reach a faulty conclusion because we began with false premises. The objective of the current study was to examine how well individuals could update a mental model of their opponent's strategy in a competitive game. A priori, we expected that patients with right-brain damage would exhibit deficits on this task given the well-demonstrated impairments in spatial and temporal processes in these patients that we construe as updating deficits (see below for a more detailed discussion). Furthermore, we wanted to assess whether updating impairments were separable from the neglect syndrome, a common consequence of right-hemisphere injury with component deficits that could also be characterized as impaired updating of mental models.

Spatial Updating

The most common task used to explore spatial updating is the double-step saccade task. Participants are required to make sequential eye movements to rapidly extinguished targets (Duhamel, Goldberg, et al. 1992). Programming saccades based on retinal coordinates alone leads to an erroneous second saccade. Instead, participants must anticipate the outcome of the first saccade to update their spatial representation of the second target location (Duhamel, Goldberg, et al. 1992; Duhamel, Colby, et al. 1992; Heide et al. 1995; Vuilleumier et al. 2007). Patients with right-parietal lesions fail to acquire the second target when the first target is in ipsilesional space and the second appears contralesionally, consistent with a failure to update spatial representations based on the intended actions (Duhamel, Goldberg, et al. 1992; Heide et al. 1995).

Patients with right-parietal lesions also have poor spatial motor imagery (Sirigu et al. 1996; Danckert et al. 2002). In controls, imagined sequences of motor actions conform to Fitts' law (Fitts 1954; Decety et al. 1989; Jeannerod 1997). Put simply, movement times show a logarithmic relationship with changes in target distance and width. When transformed into an index of difficulty, this relationship can be expressed as a linear speed-accuracy trade-off such that, for example, movements to smaller targets are slower than movements to larger targets (Fitts 1954). In patients with parietal lesions, imagined movements do not conform to Fitts' law such that there is no relationship between the duration of the imagined movements and the factors that would normally modulate actual movements (e.g., target size and distance; Sirigu et al. 1996; Danckert et al. 2002). While impairments are evident in patients with either left- or right-parietal lesions, right-parietal patients show a greater deficit in “spatial” motor imagery (Danckert et al. 2002; Hermsdörfer et al. 2006). What the above results strongly suggest is that right-parietal injury impairs the ability to update mental models of the spatial layout of the environment.

Temporal Updating

Right-parietal lesions also severely disrupt temporal processing (Battelli et al. 2001, 2003, 2007, 2008; Danckert et al. 2007; Merrifield et al. 2010). For example, when presented with durations of 60 s, patients with right-parietal damage never reported an elapsed duration any greater than 10 s (Danckert et al. 2007). Such underestimations have been observed for both visual and auditory modalities (Merrifield et al. 2010). Battelli et al. (Battelli et al. 2001, 2003) also showed that the right-parietal patients were impaired at detecting rapid onsets and offsets of stimuli (i.e., their relative timing). Husain et al. (1997) showed an abnormally large attentional blink in patients with right-parietal lesions such that following a successful detection of one target, the patients (who also had neglect) took 3 times longer than controls to successfully detect a second target. Temporal processing of this kind represents a critical component of updating mental models given that the incoming information conveys constant spatiotemporal changes.

Updating Decisions

While there are a great many brain regions, notably the basal ganglia and frontal cortices, involved in the decision making (e.g., Paulus et al. 2004; Balleine et al. 2007; Bischoff-Grethe et al. 2009), there is also some evidence that parietal cortex is involved in the intention to act and in coding the expected gain of those actions. Although still the subject of a long-standing debate in the monkey neurophysiology literature, there is evidence that lateral intraparietal area (LIP) neurons code the intended action—saccade or point—of the animal (Snyder et al. 1997; Andersen and Cui 2009). In addition, activity in LIP has been shown to be modulated by expected gain (i.e., changes in the amount of juice reward; Platt and Glimcher 1999), such that LIP neurons increased their firing rate during cue periods when the expected gain was higher. Finally, behavioral studies with nonhuman primates using competitive games suggest that monkeys are capable of updating intentions based on prior play strategies implemented by an opponent (Lee et al. 2004, 2005). Monkeys were trained to play the classic children's game “rock, paper, scissors” (RPS) against a computer. During initial play, the computer chose each option uniformly, whereas the monkeys showed nonrandom choices (i.e., an initial bias favoring one particular option). When the computer exploited the biased play of the monkeys to maximize win rate, the monkeys quickly learned to alter their play indicative of an ability to update current decisions based on the changes to the incoming information. While this work does not speak directly to the neural bases of the ability to update play strategy, the work discussed above, exploring the role of parietal regions such as area LIP in coding expected gain and intention to act, suggests that this is a good candidate region for such a process.

An additional consideration relevant to the competitive games of the ilk of RPS, concerns the ability to infer an opponent's intentions. In other words, one needs a robust Theory of Mind (ToM) in order to first develop a model of an opponent's strategy (Aboulafia-Brakha et al. 2011). In adults, deficits in ToM are commonly associated with the frontal-lobe lesions and are more prominent following the right-cortical brain injury (Martín-Rodríguez and Carríon 2010). However, a recent review suggests that no single executive function (e.g., task shifting, inhibition, etc.) can adequately explain the deficits in ToM following neurological injury (Aboulafia-Brakha et al. 2011). In addition, a recent meta-analysis of functional magnetic resonance imaging (fMRI) studies covering a range of related processes, including the ToM, empathy, and a sense of agency (i.e., self as actor), were analyzed and the common region activated across all tasks was shown to be the right temporoparietal junction (Decety and Lamm 2007). Thus, the right-parietal cortex represents a reasonable candidate for generating a mental model of the intentions of an opponent in a competitive game scenario and, as we will argue, for updating that model based on the incoming information.

Defining a Generic “Updating” Function

The term “updating” has been used previously to refer to a range of functions. As evident from the discussion above, for a broad range of behaviors (e.g., directing saccades, motor imagery, time perception, and decisions), it is critical to be able to update mental models of the environment with respect to goal states. The term updating as used here does not simply reflect the ongoing “sampling” of the incoming sensory data. Instead, we are using the term to reflect the process by which a mental model of the environment is “tested” against the incoming data. In other words, the term updating implies both the detection of a mismatch between our mental model and collected evidence and the process of revising the mental model based on those mismatches. The ability to detect novel events or stimuli (i.e., “oddballs”), that would be necessary for prompting the updating of a mental model, has been shown to produce changes to event-related potentials (ERP) that are referred to as a “Novelty P3,” occurring around 300 ms after the event and localized to frontal cortex (Spencer et al. 1999). Importantly, a separate P300 component of the ERP, localized to the temporoparietal junction, has been associated with the need to update representations of the environment presumably based, in part, on the information provided by the novel events (Donchin and Coles 1988; Donchin et al. 1997; Dien et al. 2003). In other words, processing of the novel event itself may rely on the frontal cortex, but using that information to update mental models may rely more heavily on the parietal cortex (Donchin et al. 1997; Dien et al. 2003; for review, see also Herrmann and Knight 2001).

Mental models, as mental simulators of the environment and the sensory consequences of our actions within the environment, provide data useful to the attentional and executive control systems (Norman and Shallice 1986; Baddeley 2007). The orienting of attention determines the data available for testing the veracity of the mental model. Executive systems select actions that in turn alter the environment that our mental model is designed to fit. Thus, while conceptually distinct, there is a close functional linkage between the machinery for building and testing mental models of our environment, and the cognitive systems that bias perceptual inputs and select behavioral actions. How much these functions overlap at the neural level is an open issue.

Work on short-term or working memory, reinforcement learning, and processing reward contingencies also invoke the term updating. Working memory refers to the manipulation of information over a short interval for task-specific purposes (Bledowski et al. 2010). This is distinct from the updating mental models that operates over longer time scales and may be less task specific (e.g., recalling a phone number [working memory] does not require changes to a mental model of phone numbers—they all have 10 numerals). Updating has also been used to describe the use of feedback to change behavior (i.e., reinforcement learning). Reinforcement learning tasks examine the effect of frequent feedback on consistent repetitive behaviors. While such a process may represent a kind of updating, it fails to capture the full range of processes invoked here. The most important distinction being the fact that the need to update mental models is based on a less reliable or frequent feedback than is available in a reinforcement learning paradigm (i.e., mental model construction and updating is based on noisy incomplete data: Tolman 1948; Tenenbaum et al. 2011). Finally, updating is often used to refer to set-shifting capacities particularly in work with frontal patients (Barcelo et al. 2006; Eling et al. 2008). Set shifting of this kind should be construed not as updating but as a wholesale change from one model to another.

Much of the prior work indicative of an updating impairment implicates the parietal cortex (Donchin and Coles 1988; Duhamel, Goldberg, et al. 1992; Platt and Glimcher 1999; Battelli et al. 2001, 2003, 2007, 2008; Danckert et al. 2002; Dien et al. 2003; Leon and Shadlen 2003; Janssen and Shadlen 2005; Paulus et al. 2005; Danckert et al. 2007; Vuilleumier et al. 2007; Merrifield et al. 2010). Parietal cortex, and more specifically the inferior parietal lobule (IPL), is well situated to perform this integrative function. The parietal cortex sits at the hub of an extensive anatomical network (Hagmann et al. 2008), receiving multimodal sensory afferents and maintaining broad projections to the basal ganglia and frontal cortices. The inferior parietal cortex is well placed to detect discrepancies between the incoming sensory data and expectations based on mental models (Wolpert et al. 1998; Spencer et al. 1999; Blakemore and Frith 2003; Dien et al. 2003). To evaluate the hypothesis that updating impairments are common following the RBD and to assess the association of this impairment with the neglect syndrome, we need a simple task that requires the kind of updating described above. We chose the classic children's game RPS because it is known by most people, the rules are simple, performance is physically easy, the cognitive burden is small, and yet the dynamics of sequential plays are rich (Sato et al. 2002). To provoke the need for mental model updating, we exposed participants to a sequence of uniform random plays by a computer opponent and then covertly altered the computer program's choice probabilities. This change on the part of the computer would allow for the participants to increase their win percentage “only” if they noticed the alteration. Since the computer did not exploit predictabilities in the choices made by the participants, changes in the wins and losses of the participants was a consequence of their actions alone. We were interested in how players altered their play when their opponent's strategy changed and how this adaptation correlated with the presence or the absence of neglect. Although we had only a small group of participants in each patient group, we also conducted some preliminary lesion overlay analyses to generate hypotheses concerning the potential brain areas involved in the mental model updating.

Materials and Methods

The experimental protocol was approved by the University of Waterloo's Office of Research Ethics in accordance with the Declaration of Helsinki and all participants gave informed written consent prior to participation.

Participants

Three groups of participants were tested in this study; healthy age-matched controls (12 individuals; 10 females; mean age = 66.3 years, ±3.02), right-brain–damaged (RBD) patients (13 individuals; 8 females; mean age = 61.2 years, ±10.4), and left-brain–damaged (LBD) patients (10 individuals; 3 females; mean age = 57.67 years ±8.67). Healthy controls were recruited from the Waterloo Research in Aging Participant pool and were screened to exclude any history of past neurological or psychiatric illnesses. All had a Mini-Mental Status Examination (MMSE) score above 28 (Folstein et al. 1975). Patients were recruited from the Neurological Patient Database (funded through the Heart and Stroke Foundation of Ontario). The RBD group had a mean MMSE score of 27.8 (standard deviation [SD] = 1.93; Table 1 and Fig. 1), and the LBD group had a mean MMSE score of 27.6 (SD = 1.65). There was no difference in MMSE across all 3 groups. In addition, none of the LBD patients exhibited any difficulties on the language component of the MMSE indicating that aphasia was not a significant factor in their presentation. We screened all patients for spatial neglect using the line bisection, star cancellation and figure copying tasks from the Behavioral Inattention Test (Wilson et al. 1987). Neglect was deemed to be present if the average rightward bias on line bisection was >5% of the total line length, if left-sided omissions on the star cancellation task exceeded 10% or if neglect was evident on figure copying. Lesions for the RBD and LBD groups can be seen in Figures 1 and 2 (for demographics, see Tables 1 and 2).

Table 1

Patient demographics for the RBD group

Patient Age Sex MMSE Lesion Time since stroke (months) Line bisection (% bias) Star cancellation (% Left omissions) Figure copy 
58 28 Fr, Cereb, T −2 − 
56 26 BG, F − 
56 30 Fr 10 − 
65 NA P, T 17 
81 30 100 
66 NA 657 − 
67 24 Fr, P, T, BG 19 17 90 
49 27 Fr, P, T, BG 19 28 
42 30 Fr, P, T, BG 18 
10 69 28 Fr, P, T 40 
11 67 NA Fr, P, T, BG 33 
12 69 28 Fr, P, T, BG 16 − 
13 51 27 Fr, P, T, BG − 
Patient Age Sex MMSE Lesion Time since stroke (months) Line bisection (% bias) Star cancellation (% Left omissions) Figure copy 
58 28 Fr, Cereb, T −2 − 
56 26 BG, F − 
56 30 Fr 10 − 
65 NA P, T 17 
81 30 100 
66 NA 657 − 
67 24 Fr, P, T, BG 19 17 90 
49 27 Fr, P, T, BG 19 28 
42 30 Fr, P, T, BG 18 
10 69 28 Fr, P, T 40 
11 67 NA Fr, P, T, BG 33 
12 69 28 Fr, P, T, BG 16 − 
13 51 27 Fr, P, T, BG − 

Note: Sex: M = male, F = female; Fr = frontal; P = parietal; T = temporal; BG = basal ganglia; Cereb = cerebellum; NA = not available

Figure 1.

Lesion tracings for all 13 RBD patients superimposed on an MNI template (see Materials and Methods). Lesions are presented in a radiological convention (right hemisphere presented on the left).

Figure 1.

Lesion tracings for all 13 RBD patients superimposed on an MNI template (see Materials and Methods). Lesions are presented in a radiological convention (right hemisphere presented on the left).

Figure 2.

Lesion tracings for 7 of the LBD patients for whom scans were available superimposed on an MNI template (see Materials and Methods). Lesions are presented in a radiological convention (right hemisphere presented on the left).

Figure 2.

Lesion tracings for 7 of the LBD patients for whom scans were available superimposed on an MNI template (see Materials and Methods). Lesions are presented in a radiological convention (right hemisphere presented on the left).

Table 2

Patient demographics for the LBD patients

Patient Age Sex MMSE Lesion Time since stroke (months) Chi-square P value Area Neglect status 
14 66 28 Fr, BG, T 26 0.00 0.25 — 
15 58 29 Fr, P 267 0.94 0.04 — 
16 58 26 Fr, BGa 36 0.00 0.12 — 
17 46 28 O, P, T 38 0.01 0.08 — 
18 43 27 BGa 19 0.00 0.17 — 
19 70 29 Cereb 36 0.00 0.2 — 
20 57 24 Fr, T, P, BGa 61 0.00 0.19 — 
21 63 29 BG 34 0.00 0.14 — 
22 58 29 BG 23 0.00 0.17 — 
23 58 27 Fr, O, P, BG, T 23 0.00 0.19 — 
Patient Age Sex MMSE Lesion Time since stroke (months) Chi-square P value Area Neglect status 
14 66 28 Fr, BG, T 26 0.00 0.25 — 
15 58 29 Fr, P 267 0.94 0.04 — 
16 58 26 Fr, BGa 36 0.00 0.12 — 
17 46 28 O, P, T 38 0.01 0.08 — 
18 43 27 BGa 19 0.00 0.17 — 
19 70 29 Cereb 36 0.00 0.2 — 
20 57 24 Fr, T, P, BGa 61 0.00 0.19 — 
21 63 29 BG 34 0.00 0.14 — 
22 58 29 BG 23 0.00 0.17 — 
23 58 27 Fr, O, P, BG, T 23 0.00 0.19 — 

Note: Sex: M = male, F = female; Fr = frontal; P = parietal; T = temporal; O = occipital; BG = basal ganglia; Cereb = cerebellum.

a

Scans were not available for these 3 patients, lesion location is based on radiologisits reports.

Apparatus and Procedure

Berg Card Sorting Task

In order to differentiate model updating as instantiated via the RPS task from “set shifting,” we employed a version of the Wisconsin Card Sorting Task (Nelson 1976). That is, we are suggesting here that model updating involves processes by which new information is contrasted against the expectations generated by our mental models to determine whether or not there is a (mis)match. When there is a mismatch, the model needs to be updated. The term updating has also been used in the task switching literature where individuals are typically required to switch from one task to another or to switch modes of responding within a given task (e.g., the WCST; Barcelo et al. 2006). Tasks such as the WCST are often used to examine the integrity of executive functions, with frontal lesioned patients typically demonstrating perseverative response patterns (Joseph 1999). Therefore, we felt it was important to include a measure of this kind to determine the extent to which model updating could be differentiated from set shifting. We administered a brief computerized card-sorting task based on the original descriptions of Nelson (1976). This computerized version, the Berg Card Sorting Task (BCST), is downloadable from (http://pebl.sourceforge.net/battery.html) and has been validated (Piper et al. 2011). Four exemplar cards appear at the top of a computer screen and the participant uses a computer mouse to transfer a response card to his preferred pile. The exemplars differ in color, shape, and number. The participant is given a feedback as to whether their sorting was correct or incorrect according to an undivulged rule. After 10 sequentially correct responses, the rule is shifted. The participant is not informed of this shift and must simply use the feedback provided regarding correct responses to determine which criterion is now the correct one to sort the cards by (Nelson 1976; Piper et al. 2011). Patients completed 64 trials of this task. All participants saw the same cards presented in the same order. Dependent measures from this task included the number of categories acquired and attempted, the total number of errors and the number of perseverative responses. Three RBD patients (4, 6, and 11) were unavailable or unwilling to complete the BCST; all of the LBD patients completed the BCST.

The RPS Task

In the RPS game, 2 individuals normally face one another and, on the count of 3, reveal 1 of the 3 options by shaping their hand to mimic a rock (i.e., a clenched fist), paper (i.e., a flat downturned palm), or scissors (i.e., index and middle fingers shaped like a scissors). Rock beats scissors by smashing it, paper beats rock by wrapping it up and scissors beat paper by cutting it. Thus, this is a zero sum game with each option capable of beating (or being beaten by) one other option. Our version of the game was programmed in Python using the PsychoPy library (Peirce 2009). To minimize spatial aspects, all stimuli were presented at the screen center. Participants stated their choices aloud and responses were entered by a technician. To simplify the interpretation for our patients, we used pictures of the actual items: rock, paper, and scissors, rather than pictures of hand gestures (Fig. 3).

Figure 3.

Schematic representation of the RPS game. Panel A shows the stimuli used to represent each option and demonstrates the rules of the game (i.e., which choice wins in a given pair). Panel B shows a single trial of the task (for details see Materials and Methods).

Figure 3.

Schematic representation of the RPS game. Panel A shows the stimuli used to represent each option and demonstrates the rules of the game (i.e., which choice wins in a given pair). Panel B shows a single trial of the task (for details see Materials and Methods).

A trial began with 2 red squares, vertically aligned in the center of the screen (Fig. 3).

The top square represented the computer's choice, and the bottom square represented the participant's choice. Participants were told that once the computer had made its choice on a given trial, the square representing the computer would change from red to green. Furthermore, participants were told that this meant that the computer had “locked in” its choice and would not alter it at any stage afterward. Participants were then free to make their own choices and change their choices until they reported that their selection was finalized. The space bar was then pressed and the computer's choice was immediately revealed by replacing the colored squares with images of the choices (Fig. 3). When the participant was ready, another button press began a new trial. No explicit feedback was given regarding the outcome of a given trial, however, prior to commencing the task, all participants were given instructions regarding the game's rules and all understood the task. Periodically throughout the task, patients were asked to repeat the rules (i.e., which option beats which?) to ensure they still understood the nature of the task.

Participants completed 3 blocks of 200 trials. Participants were not informed that the strategy used by the computer opponent was fixed for a block of trials, and that it varied between blocks. All blocks were completed in the same sequence for all participants. For the first block, the computer's choices were random and uniform. Each option was selected with a probability equal to one-third (i.e., the “uniform” condition). Second, they completed 200 trials in which the computer played a “moderately biased” strategy (i.e., “rock” was chosen on 50% of trials with each of the other 2 options chosen on 25% of trials). Finally, participants completed 200 trials in which the computer played a “strongly biased” strategy (i.e., “paper” was chosen on 80% of trials with each of the other 2 options chosen on 10% of the trials).

Data Analysis

To visualize trends in participant choice probabilities, we calculated moving averages over 20 trials (the choice of window size was not critical as repeating the analyses with a window size of 10 produced similar results). If participants chose randomly, they would chose each option about one-third of the time and win one-third of the time. Deviations from this pattern of play represent a deviation from a uniform random strategy. Since we knew which choices participants “should” have favored, given the biases programmed into the computer, we display the proportion of choices for the option that would maximize wins for the participant.

By design, participants should change their preferred play in the biased conditions to increase their win probability. Therefore, participants who correctly updated their internal models of the computer's play would show variation in choice probability across conditions. (There are other reasons why a participant's choices might vary across conditions, but the participants with statistically significant variation in choice probability are a superset of the participants who are updating.) To identify these participants, we computed the Chi-square statistic and its likelihood for each participant across the 3 blocks of trials. Those with positive Chi-square tests were inferred to have changed their response probabilities.

To evaluate the dynamics of the changes in behavioral choice within the game, we created for each participant, an estimate of their mixed strategy by using the proportion of each choice in the preceding 20 trials. In a game theory sense (Gibbons 1992), a pure strategy is to pick one option 100% of the time. A mixed strategy can be viewed as the probability that each pure strategy will be chosen. By viewing how the mixed strategy evolves over time, we get some sense of the dynamics of the updating process. Although there are 3 possible pure strategies, since the probabilities must sum to one, there are only 2 degrees of freedom. For example, choosing one option on 100% of trials (i.e., a “pure” strategy) would cluster the exploration around 1.0:0.0 in a 2D vector with none of the remaining state space explored. In contrast, a uniform random strategy (i.e., choosing each option one-third of the time) would lead to a cluster centering on a 2D vector of 0.3:0.3. We can visualize the evolution of these mixed strategies by computing them for a small window of trials and then sliding that window along the data and using the components of the vector as the coordinates of a 2D plot (Fig. 6). Such a plot shows how much of the game space is explored and how systematic is that exploration. If the trajectory for each condition and all blocks of trials are similar then that individual is not updating their play strategy. Furthermore, other trajectories that indicate other response modes (e.g., perseverative responding), which may be expected following frontal lesions (Joseph 1999), can be easily visualized by a preponderance of points near the apices of the triangle representing the game's strategy space (e.g., Fig. 6, middle right panel).

To compactly quantify these graphical representations, we calculated the “area” of the game's strategy space explored by an individual. The data for each participant lie on a finite grid. Our area measurement was the number of unique grid points occupied by the participant across the entire experiment divided by the total number of possible grid points.

Lesion Tracing and Overlay Analysis

Lesions were traced from computed tomography (CT) or MRI images using the following protocol (Danckert et al. 2007): CT scans were transformed into digital 3D images that were then processed in an identical manner to MRIs. Lesions were manually traced on axial slices on a slice-by-slice basis, using Analyze AVW software (Biomedical Imaging Resource, Mayo Foundation, Rochester, MN). Lesions were defined as hypointense or hypodense signal compared with the surrounding parenchyma. Individual tracings were transformed to the International Consortium for Brain Mapping (ICBM152) template, which was created from 152 MRIs from healthy individuals from the Montreal Neurological Institute (http://www.bic.mni.mcgill.ca/ServicesAtlases/HomePage). Brain tissue was digitally extracted from the image (i.e., excluding skull, meninges, etc.) using Brain Extraction Tool software. A two-step transformation process was used (Automatic Image Registration version 5.2.5 software; http://bishopw.loni.ucla.edu/AIR5): first, a spatial normalization protocol was used that included a linear 12-parameter affine transformation (including aligning scans to Talairach space; Talairach and Tournoux 1988), and second, a nonlinear fourth order parameter warping model was used to make scans best fit the template. The resulting images have a voxel size of 1 mm3. Images were visually inspected to ensure a good fit between the raw and transformed lesion maps. Using the transformed lesion maps, the proportion of each anatomical region involved in each patient's lesion was estimated using the MRIcro single-subject Colin template (http://www.cabiatl.com/mricro/). Individual brain lesions were then superimposed using the same software allowing for lesion overlay analyses. Lesions are overlaid on the ICBM152 template from which Talairach coordinates for any region demonstrating overlap in all patients can be extracted. Talairach Daemon is then used to transform those coordinates into Brodman regions (http://www.talairach.org/daemon.html).

Results

The RPS Performance

When compared with RBD patients, controls and LBD patients show a much greater adaptation to the computer's covert strategy shifts. Figure 4 shows the probability of making the “optimal” choice for each participant group in each condition. A mixed design repeated measures ANOVA, with condition as the within subject variable (3 levels: uniformly, moderately, and strongly biased) and group (controls vs. RBD vs. LBD patients) as the between subject variable, demonstrated a significant group by condition interaction (F2,33 = 6.53, P = 0.004). Independent samples post hoc t-tests indicated that the RBD and control groups differed only in the strongly biased condition (t23 = 3.93; P < 0.01). In addition, there was a significant difference between the RBD and the LBD patients in the strongly biased condition (t22 = 3.82; P < 0.001). Controls and the LBD patients did not differ in their average choices in the strongly biased condition and began choosing the optimal strategy within the first 30 trials; the RBD patients did not (Fig. 4).

Figure 4.

Moving averages (i.e., across 20 trial windows) of choices for healthy controls (upper panels), the RBD patients (middle panels) and the LBD group (lower panels). Note that for the uniform condition, we plotted the percentage of rock choices; for the moderately biased condition, we plotted the percentage of paper choices as this should maximize win rate given that the computer opponent now chose rock on 50% of trials; and for the strongly biased condition, we plotted the percentage of scissors choices given that the computer opponent now chose paper on 80% of the trials. Error bars (shaded gray regions) represent standard error.

Figure 4.

Moving averages (i.e., across 20 trial windows) of choices for healthy controls (upper panels), the RBD patients (middle panels) and the LBD group (lower panels). Note that for the uniform condition, we plotted the percentage of rock choices; for the moderately biased condition, we plotted the percentage of paper choices as this should maximize win rate given that the computer opponent now chose rock on 50% of trials; and for the strongly biased condition, we plotted the percentage of scissors choices given that the computer opponent now chose paper on 80% of the trials. Error bars (shaded gray regions) represent standard error.

There was no statistical difference between the controls and the LBD patients in any condition. Both groups clearly adapted their play strategy in the strongly biased condition making the optimal choice (i.e., scissors) most often. Interestingly, by the end of this condition, healthy controls were choosing the optimal strategy on around 75% of trials, whereas LBD patients chose the optimal strategy on 89% of trials (Fig. 4). This difference is likely due to the fact that 8 of the 10 LBD patients adopted a “maximizing” strategy in this condition, choosing scissors 100% of the time for at least some portion of the task (note: patient 17 only adopted this strategy very late in the task; Fig. 5).

Figure 5.

Plots of the percentage of scissors choices made in the strongly biased condition for all patients (RBD patients are presented to the right and LBD patients are presented to the left). In each plot, trial number from 0 to 180 is presented on the x-axis with percentage of scissors choices from 0 to 100 presented on the y-axis. *indicates the patients with neglect.

Figure 5.

Plots of the percentage of scissors choices made in the strongly biased condition for all patients (RBD patients are presented to the right and LBD patients are presented to the left). In each plot, trial number from 0 to 180 is presented on the x-axis with percentage of scissors choices from 0 to 100 presented on the y-axis. *indicates the patients with neglect.

While some of our healthy controls (n = 7) also adopted this strategy, they did so for far shorter periods of time. That is, when the LBD patients maximized by choosing “scissors,” they did so on average for 62 (±43.97) consecutive trials. In contrast, the controls who chose this strategy only did so on average for 14 (±6.16) consecutive trials (t13 = 2.54, P < 0.05). As already noted, the LBD group adopted the optimal strategy very early on in the task, much like the healthy controls (Fig. 4). Even those patients in the RBD group who did alter their choices in the strongly biased condition took far longer than controls and LBD patients to do so (Fig. 5). In addition, only one of the LBD patients (patient 15) failed to demonstrate a significant Chi-square result, indicating that most of the LBD patients changed their proportion of choices over the course of the task indicative of updating their mental model of their opponent's play strategy (Table 2).

This preliminary analysis shows a group effect, but hides heterogeneity particularly within the RBD group. The Chi-square analyses demonstrated that many of the RBD patients did in fact change their choice probabilities from one condition to the next (Table 3; note, all controls and all but one LBD patient had significant Chi-square values—for LBD results, see Table 2). This fact, combined with the suboptimal win rates, suggests that some of the RBD patients were updating their models of the opponents play, but in a suboptimal fashion relative to the updating observed in the healthy controls and LBD patients.

Table 3

Chi-square results and area of state space explored for the RBD patients

Patient Chi-square P value Area Neglect status 
0.00 0.16 − 
0.48 0.06 − 
0.91 0.06 − 
0.87 0.10 
0.01 0.09 
0.00 0.12 − 
0.00 0.09 
0.03 0.11 
0.22 0.07 
10 0.00 0.18 
11 0.84 0.02 
12 0.04 0.12 − 
13 0.00 0.12 − 
Controls (mean ± SD) NA 0.17 (0.05) NA 
Patient Chi-square P value Area Neglect status 
0.00 0.16 − 
0.48 0.06 − 
0.91 0.06 − 
0.87 0.10 
0.01 0.09 
0.00 0.12 − 
0.00 0.09 
0.03 0.11 
0.22 0.07 
10 0.00 0.18 
11 0.84 0.02 
12 0.04 0.12 − 
13 0.00 0.12 − 
Controls (mean ± SD) NA 0.17 (0.05) NA 

Note: Sex: M = male, F = female; Fr = frontal; P = parietal; T = temporal; BG = basal ganglia; Cereb = cerebellum; NA = not available

To better determine the factors related to whether a patient participant would, or would not, update their play strategy, we constructed good and poor performance patient subgroups within the RBD group only (given that both controls and LBD patients clearly exhibited successful updating behavior it was not deemed necessary to split these groups). We did this by performing a median split using the measure of the area of the game's strategy space explored (Fig. 6). In addition to its common-sense appeal, this metric has face validity, as all but 2 controls and 2 LBD patients (Table 2) fell above the median split value for the RBD group (the 2 controls had values that were at the median value).

Figure 6.

Examples of state space exploration from a healthy control (upper panel), 3 RBD patients (middle panels) and an LBD patient (lower panel). The proportion of paper choices is represented on the x-axis with the proportion of rock choices represented on the y-axis averaged over 20 trial “blocks” in each of the 3 conditions. Performance in the uniform condition is represented in red, performance in the moderately biased condition is represented in green, with performance in the strongly biased condition represented in blue. As can be seen from the upper panel, the healthy control quickly finds the optimal strategy in the strongly biased condition (blue). That is, the optimal choice in the strongly biased condition would be to choose scissors on most trials (i.e., the computer opponent chooses paper 80% of the time), which would be represented as 0:0 in the 2D vector shown here (i.e., rarely if ever choosing paper or rock). The RBD patient represented in the middle left panel (patient 9) never changes his play across all 3 conditions. In contrast, the RBD patient in the middle right panel (patient 10) perseverates on choices in the moderately and strongly biased conditions although those choices are not always optimal (e.g., substantial time was spent perseverating on rock in the moderately nonrandom condition—represented in green—thus leading to a large proportion of ties). The RBD patient represented in the center middle panel (patient 7) eventually adopts the optimal strategy in the strongly biased condition despite taking far longer than the control to do so (this is evident in the area of state space overlapping across the 3 conditions). Finally, the lower panel shows a typical LBD patient (patient 19) who, like controls, quickly adopts the optimal strategy in the strongly biased condition and appears overall to explore more of the game's strategy space than do some controls.

Figure 6.

Examples of state space exploration from a healthy control (upper panel), 3 RBD patients (middle panels) and an LBD patient (lower panel). The proportion of paper choices is represented on the x-axis with the proportion of rock choices represented on the y-axis averaged over 20 trial “blocks” in each of the 3 conditions. Performance in the uniform condition is represented in red, performance in the moderately biased condition is represented in green, with performance in the strongly biased condition represented in blue. As can be seen from the upper panel, the healthy control quickly finds the optimal strategy in the strongly biased condition (blue). That is, the optimal choice in the strongly biased condition would be to choose scissors on most trials (i.e., the computer opponent chooses paper 80% of the time), which would be represented as 0:0 in the 2D vector shown here (i.e., rarely if ever choosing paper or rock). The RBD patient represented in the middle left panel (patient 9) never changes his play across all 3 conditions. In contrast, the RBD patient in the middle right panel (patient 10) perseverates on choices in the moderately and strongly biased conditions although those choices are not always optimal (e.g., substantial time was spent perseverating on rock in the moderately nonrandom condition—represented in green—thus leading to a large proportion of ties). The RBD patient represented in the center middle panel (patient 7) eventually adopts the optimal strategy in the strongly biased condition despite taking far longer than the control to do so (this is evident in the area of state space overlapping across the 3 conditions). Finally, the lower panel shows a typical LBD patient (patient 19) who, like controls, quickly adopts the optimal strategy in the strongly biased condition and appears overall to explore more of the game's strategy space than do some controls.

Differentiating RBD patients on the degree of the game's strategy space they explored provided a clear separation for how likely they were to update (Fig. 7). Furthermore, only 2 of the LBD patients exhibited area scores that were below the median cutoff for high performance based on healthy controls (Table 2). In fact, many of the LBD patients exhibited patterns of performance that suggested they explored “more” of the game's strategy space than did many of the controls (Figs 5 and 6). Independent samples t-tests revealed a significant difference between the LBD and the RBD groups on the area of the game's strategy space they explored (t21 = −2.53, P < 0.05) such that the LBD group explored more of the game's strategy space. There was no difference in this measure between the LBD patients and the healthy controls (t20 = −0.77, P = 0.45). Unsurprisingly, the RBD patients explored significantly less of the game's strategy space than did controls (t23 = −3.91, P < 0.01).

Figure 7.

Panel A. Percentage of scissors choices made in the strongly biased condition for patients in the low (orange) and high (purple) exploration groups over a moving average window of 20 trials. Panel B. Percentage of scissors choices made in the strongly biased condition for patients with (pink) and without (blue) neglect. Error bars are standard error of the mean.

Figure 7.

Panel A. Percentage of scissors choices made in the strongly biased condition for patients in the low (orange) and high (purple) exploration groups over a moving average window of 20 trials. Panel B. Percentage of scissors choices made in the strongly biased condition for patients with (pink) and without (blue) neglect. Error bars are standard error of the mean.

RBD participants with neglect clearly showed poorer performance on both the choice and the exploration measures of the RPS task. However, dividing the participants by neglect status did not yield as clear a partition of the RBD patient group, as did dividing by the exploration status (Table 1; Fig. 7). For the entire RBD group, patients who eventually did make optimal choices (i.e., choosing scissors in the strongly biased condition) took much longer to do so than did controls. Therefore, it seems that neglect and strategy updating are clearly associated as most of the neglect participants (5 of 7) performed poorly on the RPS task. In fact, if the poor performers on RPS were selected first, most of them would turn out to have neglect. However, the presence of some nonneglect patients in the low exploration area group, and the occasional neglect patient with RPS performance above the median area explored, indicates that we cannot assert that the deficits that lead to poor performance on the RPS task are identical to those that cause the neglect.

Lesion Analyses

Of the 7 RBD patients, 5 patients in the low exploration group exhibited some degree of overlap in their lesions. Lesion overlay analyses for these patients showed common involvement of the insula, putamen, and surrounding white matter (Table 4 and Fig. 8). Only 4 of the 6 RBD patients in the high exploration group showed any overlap in their lesions, with overlay analyses indicating that this group was more likely to have involvement of superficial cortical areas (e.g., primary motor and somatosensory cortices, premotor cortex, Broca's region, and the superior temporal gyrus (STG); Table 4 and Fig. 8). When the lesion overlays of the 2 groups were subtracted (i.e., low exploration group minus the high exploration group), the remaining region of overlap still included the insula and putamen (with a slight reduction in the size of the insula region relative to the lesion overlay performed within the low exploration group alone; Table 4). The region of overlap for the insular in the low exploration group encompassed parts of the anterior and middle insular cortex.

Table 4

Lesion overlay results

Region BA Number of voxels 
Low area (n = 5; patients 2, 4, 7, 9, 11) 
    Insula 13, 14 206 
    Putamen  146 
    White matter  333 
High area (n = 4; patients 8, 10, 12, 13) 
    Primary somatosensory 2, 3 14, 268 
    Primary motor 79 
    Premotor 171 
    STG 22 107 
    Parietal operculum 43 122 
    Broca's (pars opercularis) white matter 44 97 9004 
Low–High area 
    Insula 13, 14 113 
    Putamen  146 
Neglect (n = 6; patients 4, 7, 8, 9, 10, 11) 
    STG 22 387 
    Insula 13, 14 458 
    Supramarginal 40 425 
    Parietal operculum 43 865 
    Primary auditory cortex 41, 42 262 
    Primary somatosensory cortex 80 
    White matter  418 
Nonneglect (n = 4; patients 2, 3, 12, 13) 
    Insula 13, 14 475 
    Putamen  90 
    White matter  321 
Region BA Number of voxels 
Low area (n = 5; patients 2, 4, 7, 9, 11) 
    Insula 13, 14 206 
    Putamen  146 
    White matter  333 
High area (n = 4; patients 8, 10, 12, 13) 
    Primary somatosensory 2, 3 14, 268 
    Primary motor 79 
    Premotor 171 
    STG 22 107 
    Parietal operculum 43 122 
    Broca's (pars opercularis) white matter 44 97 9004 
Low–High area 
    Insula 13, 14 113 
    Putamen  146 
Neglect (n = 6; patients 4, 7, 8, 9, 10, 11) 
    STG 22 387 
    Insula 13, 14 458 
    Supramarginal 40 425 
    Parietal operculum 43 865 
    Primary auditory cortex 41, 42 262 
    Primary somatosensory cortex 80 
    White matter  418 
Nonneglect (n = 4; patients 2, 3, 12, 13) 
    Insula 13, 14 475 
    Putamen  90 
    White matter  321 

Note: BA = Brodmann area

Figure 8.

The upper section shows lesion overlay maps for 5 of the 7 patients in the low exploration group who showed some degree of overlap in their lesions and 4 of the 6 patients in the high exploration group who showed some degree of overlap in their lesions. The remaining patients had lesions that did not intersect at all with any other patient. The lower section shows lesion overlay maps for the 6 (of 7) neglect patients who demonstrated some degree of overlap in their lesions and the 4 (of 6) patients without neglect who also showed some degree of overlap with their lesions. The remaining patients had lesions that did not intersect at all with any other patient (see Table 3 for details).

Figure 8.

The upper section shows lesion overlay maps for 5 of the 7 patients in the low exploration group who showed some degree of overlap in their lesions and 4 of the 6 patients in the high exploration group who showed some degree of overlap in their lesions. The remaining patients had lesions that did not intersect at all with any other patient. The lower section shows lesion overlay maps for the 6 (of 7) neglect patients who demonstrated some degree of overlap in their lesions and the 4 (of 6) patients without neglect who also showed some degree of overlap with their lesions. The remaining patients had lesions that did not intersect at all with any other patient (see Table 3 for details).

Overlap was seen in 5 separate pairs of LBD patients. However, there was no overlap evident when 3 or more patients were included in the analysis in any combination. This is perhaps not surprising given that the scans were available for only 7 of the patients, 1 of whom had a lesion entirely constrained to the cerebellum (patient 19; Fig. 2). It is perhaps worth noting that 6 of the 10 patients had lesions that involved some portion of the insular cortex.

The BCST Performance

Table 5 presents the results of the BCST for both patient groups.

Table 5

Berg card sorting task results for both patient groups

 RBD LBD t P value 
Categories completed 1.3 (1.16) 2.10 (1.20) −1.52 0.15 
Correct responses 30.8 (11.4) 41.5 (7.56) −2.47 0.02 
Perseverative errors 8.9 (7.56) 13.8 (7.07) −1.50 0.15 
Nonperseverative errors 24.3 (16.94) 8.7 (5.12) 2.79 0.02 
Total errors 33.2 (11.4) 22.5 (7.56) 2.47 0.02 
Trials to complete first category 17.9 (19.58) 13.3 (5.66) 0.71 0.49 
 RBD LBD t P value 
Categories completed 1.3 (1.16) 2.10 (1.20) −1.52 0.15 
Correct responses 30.8 (11.4) 41.5 (7.56) −2.47 0.02 
Perseverative errors 8.9 (7.56) 13.8 (7.07) −1.50 0.15 
Nonperseverative errors 24.3 (16.94) 8.7 (5.12) 2.79 0.02 
Total errors 33.2 (11.4) 22.5 (7.56) 2.47 0.02 
Trials to complete first category 17.9 (19.58) 13.3 (5.66) 0.71 0.49 

Three of the RBD patients never successfully sorted a single category on the task. When only those patients who did successfully sort a single category were compared with one another, both patient groups took a comparable number of trials to complete their first category on the BCST. Furthermore, the number of trials taken to achieve the first category was very similar to that of a normative sample of healthy older controls (Piper et al. 2011). Both patient groups had fewer correct responses than a normative sample of healthy controls (Piper et al. 2011) with the RBD group having significantly fewer correct responses than the LBD group (Table 5). In addition, there was no difference between the patient groups in terms of the number of categories correctly sorted, and the 2 groups did not differ on the number of perseverative errors made (note: only those RBD who achieved at least one category (n = 7) were included in the latter analysis of perseverative errors as it is not possible to demonstrate a perseverative error without first correctly sorting 1 category; Table 5).

When we correlated the area measure for our RPS task with the performance metrics from the BCST for the RBD group, there was a significant negative correlation between RPS area and the number of trials taken to sort the first category in the BCST (r = −0.65; P < 0.05; note that only the 7 RBD patients who successfully achieved at least one category were included in this analysis). This indicates that the RBD patients who explored more of the game's strategy space in RPS also took fewer trials to successfully sort the first category in the BCST. There were no other significant correlations between any of the other metrics from the BCST and our measure of the area explored in the RPS task.

Lesion Volume and Time Since Stroke—Correlation to Behavior

For both RBD and LBD patient groups, lesion volume did not correlate with any of the performance measures from the RPS task. In particular, the area of state space explored and the proportion of optimal choices made in the strongly biased condition did not correlate with lesion volume. For the RBD group, time since stroke did not correlate with the area of state space explored (r = −0.28; P = 0.35), whereas for the LBD group there was a significant negative correlation between the area explored and the time since stroke (r = −0.67; P = 0.036). This result was probably driven by the one outlier (patient 15). With that patient removed from the analysis, the correlation was no longer significant (r = −0.15; it should be noted that when the outlier in terms of time since stroke in the RBD group—patient 6—was removed from the analysis, the correlation for that group was r = −0.09 which was nonsignificant). There was also no correlation in either group between MMSE scores and area of the game's state space explored.

Discussion

As reviewed in the Introduction, there are a variety of prior results that suggest RBD impairs the ability to update mental models. To pursue this notion further, we assessed the ability of 13 RBD participants, 10 LBD patients, and 12 healthy age-matched controls to successfully recognize and exploit the biased play of a computer opponent in a game of RPS. Direct inspection of their choice probabilities revealed a clear separation between the RBD patients and their LBD counterparts who did not differ from the healthy controls. All but one of our right-hemisphere injured patients demonstrated at least some difficulty in exploiting the strongly biased play of the computer opponent (patient 1 was the exception; Fig. 5). The most obvious and common impairment in the RBD group was the failure to make “any” change in strategy at all across the 3 conditions. In this case, play was not perseverative, as RBD patients varied their choices from trial to trial. What these participants did not do was vary the probability with which they selected each option (e.g., Fig. 6, middle left panel; this pattern was evident in 8 of the patients). For the RBD patients who did alter their choice probabilities across conditions, it was evident from inspecting win proportions that the choices made were suboptimal (i.e., worse than controls). Even for RBD patients who eventually “got it,” the “aha” moment, when they realized a change in choices would be advantageous, took far longer than for controls (RBD patients 6 and 7 took between 70 and 100 trials before they began to alter their play strategy—controls took around 30 trials; Fig. 5). Occasionally, we did observe other patterns (e.g., perseveration; RBD patients 10 and 13; Fig. 6 middle right panel). Even for the simple task we chose, there are multiple potential explanations for poor performance. Patients may fail to detect changes in their opponent's play, they may fail to accumulate a sufficiently large amount of prior evidence to enable them to adapt their own play (i.e., attending only to the last trial played), or they may detect changes in their opponent's play but fail to use that information to update the mental model that guides their selection. Each of these potential sources of failure implicates a different component of mental model updating and, potentially, may depend on distinct brain regions. However, what is clear is that within this heterogeneous collection of RBD participants there is a nearly ubiquitous inability to exploit a change in the environment (computer play probability) that is almost universally recognized and exploited by healthy people of the same age and LBD patients (although with a distinct pattern of choices relative to controls—see below).

To address the relationship between strategy updating and set shifting, we had our participants perform the BCST—a measure of set shifting (Nelson 1976; Eling et al. 2008; Piper et al. 2011). Both patient groups performed the BCST task poorly relative to published norms for healthy older controls (Piper et al. 2011). Given that there was no indication of increased perseverative responding in the RBD, we would argue that their failure on this task was not due to an inability to shift mental set, but instead reflects an inability to successfully develop a model for sorting the cards in the first instance. Given the deficit in updating mental models evident on the RPS tasks, this result is perhaps not surprising. That is, before a person can either perseverate or shift categories they must first successfully sort cards on an initial category. Many of our RBD patients failed to exploit the biased play of their opponent in the RPS task even after 200 trials. In some sense, the inability to detect biased play in RPS mirrors the inability of 3 of the RBD patients to successfully sort even one category on the BCST. This is also borne out by the negative correlation between the area of the game's strategy space explored and the number of trials taken to sort the first category of the BCST. Those patients who failed to explore much of strategy space in the RPS game also took far longer to sort the first category of the BCST (if indeed they sorted a single category at all). This may suggest that the primary deficit in the RBD patients lies in creating an accurate internal model of their opponent's play in the first instance. Without such a model, updating one's own play based on prior experience would be impossible. Importantly, the RPS performance for each patient group was much more distinct than their performance on the BCST. This suggests that each patient group failed the BCST for reasons different from their performance on the RPS task. Furthermore, a deficit in set shifting cannot explain the deficits we observed on the RPS task.

Clearly, damage to the right hemisphere dramatically disrupted the ability to update an internal representation of an opponent's strategy even when that opponent employed a strongly biased strategy. Patients with LBD exhibited no such difficulty in updating their own play strategy to exploit the biased play of their opponent. In fact, the LBD patients as a group tended to choose a maximizing strategy (i.e., uniformly implementing the optimal choice) much more so than did healthy controls who on an average “matched” their opponent's play (i.e., implementing the optimal choice on close to 80% of trials; Figs 4 and 5). Such hemispheric differences in employing either a matching or maximizing strategy are consistent with previous neuroimaging research, research on split brain patients, patients with focal left frontal lesions, and normal subjects using probability matching tasks (Wolford et al. 2000, 2004; Miller et al. 2005; Vickery and Jiang 2009; Roser et al. 2011). Miller et al. (2005) scanned participants while they performed a prediction task in which they had to guess whether a stimulus would appear at the top or bottom of a display, with targets biased to appear at the top location on 70% of trials. Participants could then be classified as employing either a probability matching or maximizing strategy based on their predictions. The vast majority of the participants (18 of 22) adopted a probability matching strategy with increased activation evident in a widespread network of frontal and parietal cortices, including the angular gyrus and IPL. Importantly, the volume of activity was larger on the right than on the left (Miller et al. 2005).

In a similar task, Vickery and Jiang (2009) used a version of the “matching pennies” game. In this game, 2 players independently choose 1 of 2 options. One player wins when the options match (e.g., heads and heads or tails and tails), and the other wins with mismatches. Just as in our study, participants here played against a computerized opponent. In their study, however, the computerized opponent punished biased strategies where evident in the participants. That is, if a participant tended to pick one option predictably, the computer used this information to beat them. Significant increases in functional activity were observed in the right IPL, which showed an attenuation of signal when attentional demands were increased. Such a decrease in activity in response to an increased “attentional” load suggests that the activity observed when participants made guesses in the game scenario reflects other functions indexing the need to adapt one's choices related to their own prior play and the inferred rules being used by the computerized opponent. In contrast, activity in the superior frontal gyrus, which was also increased in the matching pennies task, showed “increasing” activation with attentional load. Taken together, these imaging studies show that increased parietal activity, particularly in the right hemisphere, is associated with the need to make decisions in an uncertain environment (Miller et al. 2005; Vickery and Jiang 2009).

Distinct differences in play strategy on probability guessing games are also observed in split-brain patients. Wolford et al. (2000) used a spatial guessing game to probe the right and left hemisphere of 2 split-brain patients. Arrows directed the patient to attend to either the left or the right of a computer screen. The participant then guessed whether a colored figure would be near the top or the bottom of the screen and an imbalanced 80:20 ratio was used. For both participants, right hemisphere (left visual field) presentations led to responses that “maximized” their choice probability for the most likely location, whereas the opposite condition (left hemisphere processing related to right visual field presentation) led participants to engage in a probability matching strategy (for work in healthy individuals, see also Wolford et al. 2004). In the context of the current results, lesions to the left hemisphere may have disrupted the normal tendency of this hemisphere to adopt a probability matching strategy, leaving the intact maximizing right hemisphere unchecked (Gazzaniga 1995). It is somewhat more difficult to reconcile the results obtained from split-brain patients with those of the imaging results discussed above (Miller et al. 2005; Vickery and Jiang 2009). In those studies, probability matching is more consistently associated with increased right-hemisphere activity. It may be the case that the right hemisphere is critical for processing the statistical properties of the environment, whereas the left hemisphere overlays a conscious interpretation of those statistics (Gazzaniga 1995). For the LBD patients in our study then, the right hemisphere is able to find the optimal strategy (maximize one option) to ensure the highest win rate (note, matching is a suboptimal strategy) in part because the left hemisphere is not imposing any complex interpretations on the incoming data that would lead to the adoption of a more cautious strategy. In this context, the dominance of the right-hemisphere activity in the imaging studies is not reflective of the adoption of a maximizing strategy, but instead reflects the role the right hemisphere plays in developing and updating an accurate representation of the statistics of an uncertain environment. In support of this notion, Roser et al. (2011) recently showed that a split brain participant was only able to learn visual statistical relationships when the stimuli for familiarization and performance were both presented to the left visual field (right hemisphere), indicating a special role for the right hemisphere in unsupervised statistical learning. It is worth noting here that the apparent inconsistencies between patient and fMRI findings may simply reflect the differences inherent to the 2 techniques. More pointedly, it is not possible from fMRI data to determine whether the observed increases in activation represent increased neural excitation or inhibition, a nontrivial point when determining the relationship between maximizing or matching strategies in the right and left hemispheres. For example, increased excitation in the right hemisphere may be expected to lead to a maximizing strategy, whereas increased inhibition may be expected to lead to a matching strategy. Both circumstances would lead to an increased blood oxygen level–dependent (BOLD) signal in the same region.

In contrast to the LBD patients, the group with RBD had much greater difficulty exploiting the biased play of their opponent. Those patients who did eventually choose an optimal strategy did so after almost 100 trials, far longer than either the controls or the LBD group took to alter their strategies. While the presence of spatial neglect in the RBD patients clearly impaired mental model updating, it could not be considered the sole source of the impairment (Fig. 7). In our patient group, 2 of the 7 neglect patients demonstrated some capacity to update their play strategy, while others without neglect performed poorly. Nevertheless, lesion overlay analyses including 6 of the 7 neglect patients highlighted regions shown in other work to be commonly associated with the neglect syndrome including the STG and insula (Table 4; Karnath et al. 2001, 2004). Importantly, the insula and surrounding white matter were involved in both the neglect group and in the poorly performed group (Table 4). The STG and insula cortex have recently been shown to maintain strong white matter connections (Umarova et al. 2010). So while neglect is unlikely to represent the sole “cause” of updating deficits, it is still possible that the regions commonly involved in neglect play some role in updating the mental models (Table 4).

While findings from lesion overlay analyses in such small groups of patients are necessarily speculative, the areas highlighted in our RBD patient group provide some intriguing possibilities regarding the brain regions that may be important for updating mental models. Support for our suggestion that the cortical regions tentatively identified in the RBD patients, the insula, putamen, and the STG, may be nodes of an updating network comes from their demonstrated participation in updating internal representations of intention. Furthermore, the STG and insula represent multimodal associations cortices with extensive links to the basal ganglia, parietal and frontal cortex (Karnath et al. 2001, 2004; Craig 2009). These regions are multimodal in nature and maintain extensive connections with the rest of the brain making them ideal candidate regions for creating and updating mental models of the external environment. Below, we review some relevant literature regarding the functions of these regions that may provide some insights into the neural basis of updating that will in turn provide the impetus for future work. We are not making strong claims at this stage about the role each region plays, but merely speculating about their possible involvement in representational updating based on previous work.

Insular Cortex

A recent review of the insula cortex suggests a graded representation from primary interoceptive sensations in the posterior insula cortex to more complex representations of subjective feelings of the current moment in the anterior insula (Craig 2009). fMRI studies have shown activation in the insula for a range of tasks from interoceptive perceptions of pain and heartbeat, to time perception, maternal affection, musical perception, perceptual decision making, and risk processing (Brooks et al. 2002; Critchley et al. 2004; Leibenluft et al. 2004; Seymour et al. 2004; Koelsch et al. 2006; Preuschoff et al. 2006, 2008; Livesey et al. 2007; Thielscher and Pessoa 2007). Thus, the insula may be important for representing salient information (including perceptual, motor and interoceptive information) and switching between large-scale neural networks alternatively responsible for executive functions and the so-called default mode (Menon and Uddin 2010). One suggestion is that the insula functions as a comparator similar to our description of updating, such that the insula detects mismatches between predicted and actual sensory feedback (Spinazzola et al. 2008; Jones et al. 2010). Furthermore, when making decisions under different conditions of risk, patients with insula damage bet at a consistently higher level than controls even in circumstances where the odds of winning are poor (Clark et al. 2008). Somewhat contradictory findings have also shown that patients with insula lesions behave more cautiously than controls (Weller et al. 2009). These finding can be reconciled with each other and with our data if insula lesions render people less sensitive to risk-reward relationships (Preuschoff et al. 2008).

Putamen

The putamen has been repeatedly shown to be involved in representing the reward value of decisions under reinforcement learning paradigms (e.g., Bischoff-Grethe et al. 2009; Seger et al. 2010). As already mentioned in the Introduction, although reinforcement learning requires modifying the associations between the stimuli and the actions, this modification is likely not the same as mental model updating. Most importantly, rewards are by necessity made explicit for reinforcement learning, when in reality, we often develop expectations about the “behavior” of our mental models without such explicit feedback, as is the case in the RPS task used here. Also, in most reinforcement learning tasks there is no explicit need to infer the strategies of an opponent. This need is paramount in the setting of competitive games. Despite these differences, the association of the putamen with reward systems makes it a candidate for consideration as a node in an updating network.

Many of the studies demonstrating putamen activation are human neuroimaging studies that use decision-making scenarios similar to the RPS game we used. Increased activity has been observed in the putamen in response to immediate versus delayed rewards, with higher activity evident for the former (Wittmann et al. 2007; Luo et al. 2009; for work in nonhuman primates, see also Hori et al. 2009). In addition, Ino et al. (2010) found increased activity bilaterally in the putamen and nucleus accumbens when participants made choices associated with large gains or losses relative to the choices that would deliver only small gains or losses. The right putamen was more active for choices related to gains than losses (Ino et al. 2010). As there was no difference in activity in the putamen at the outcome stage of their task, the putamen may be primarily involved in calculating “expected” risk/reward relationships (Wittmann et al. 2007; Ino et al. 2010; for review, see also Brooks 2006).

Superior Temporal Gyrus

The STG, which was involved in the group of neglect patients who also showed impaired updating processes (although to a lesser extent than the low exploration group; Fig. 7) has also been implicated in the studies of normal human participants playing RPS. Paulus et al. (2004, 2005) measured the BOLD activity in the response to a version of the RPS, somewhat different than what we used. In their study, they fixed the proportion of wins, ties, and losses for particular choices and then switched these associations every 20 trials. Participants were not told which option was preferred or the structure of the task. Even when participants consistently chose the preferred over the nonpreferred option and adjusted their play to the regular switches, very few were explicitly aware of the nonuniform play of their opponent (Paulus et al. 2004). Contrasting early versus late trials in a block showed greater activation in the inferior and medial frontal gyri and in the pallidum (Paulus et al. 2004). Subsequent analyses correlated BOLD activity with a model of trend detection to determine which brain regions modulated activity relative to the (presumably implicit) detection of the preferred option for obtaining wins. Several areas showed a pattern indicative of detection of the opponent's play including the inferior frontal gyrus, the STG, and the posterior parietal cortex. Bilateral increases in caudate activation were observed relative to wins vs. losses suggesting that this region coded the outcome of choices (Paulus et al. 2004). In a follow-up study, the STG and insula bilaterally were found to be more active during response selection (Paulus et al. 2005). Finally, a recent fMRI study of visuospatial attention which also employed diffusion tensor imaging demonstrated a ventral white matter pathway connecting the inferior parietal and superior temporal cortices with the anterior insula (Umarova et al. 2010). Such a pathway may prove integral for representing the current focus of attention relative to the salient visual events—the so-called “ventral” attentional network outlined by Corbetta and Shulman (2002). This connection also highlights the possibility that these 2 structures, in conjunction with the putamen, form nodes in a neural network for updating mental models of the environment.

The description of updating we have offered is of a high level and is abstract, but we have given a concrete example of it in the need to detect and update a model of the opponent's play in a game of RPS. People benefit from exploiting regularities in their environment, and they need to detect changes in those regularities (Tenenbaum et al. 2011). This updating may need to occur in the absence of salient signals of change and without explicit variation in overt reward. Our experience with RBD participants and the RPS game suggests a pervasive impairment in this updating function. The behavioral impairment overlaps with, but appears distinct from neglect. Regions of the brain that may be associated with updating are hinted at by our lesion overlay analyses and include the insula, putamen, and superior temporal region. As our description of updating does not have an explicit component addressing the calculation of risk or reward, we feel it more broadly captures behavior in the everyday world. Future work should consider an interaction between the environmental assessment and reward calculation with experiments that use combined manipulations. If this updating function is as ubiquitous an aspect of behavior as we suggest, its impairment in RBD has important practical consequences, and further study of these patients may inform our understanding of this important function.

Funding

Natural Sciences and Engineering Research Council (http://www.nserc-crsng.gc.ca/index_eng.asp) of Canada Discovery (#261628-07); Canada Research Chair grants; Heart and Stroke Foundation of Ontario (http://www.heartandstroke.on.ca; #NA 6999 to J.D.); Canadian Institutes of Health Research (http://www.cihr-irsc.gc.ca/e/193.html; #219972 operating grant to J.D. and B.A.). The above-mentioned funding agencies had no role in the study design, data collection and analysis, decision to publish, or the preparation of the manuscript.

We would like to thank Dr Susanne Ferber for helpful comments on an earlier version of this manuscript. The authors would like to thank all the patients for their time and effort in this experiment. Conflict of Interest : None declared.

References

Aboulafia-Brakha
T
Christe
B
Martory
MD
Annoni
JM
Theory of mine tasks and executive functions: a systematic review of group studies in neurology
J Neuropsychol
 , 
2011
, vol. 
5
 (pg. 
39
-
55
)
Andersen
RR
Cui
H
Intention, action planning, and decision making in parietal-frontal circuits
Neuron
 , 
2009
, vol. 
63
 (pg. 
568
-
583
)
Baddeley
AD
Working memory, thought and action
 , 
2007
Oxford
Oxford University Press
Balleine
BW
Delgado
MR
Hikosaka
O
The role of the dorsal striatum in reward and decision- making
J Neurosci
 , 
2007
, vol. 
27
 (pg. 
8161
-
8165
)
Barcelo
F
Escera
C
Corral
M
Periáñez
J
Task switching and novelty processing activate a common neural network for cognitive control
J Cogn Neurosci
 , 
2006
, vol. 
18
 (pg. 
1734
-
1748
)
Battelli
L
Cavanagh
P
Intriligator
J
Tramo
MJ
Hénaff
MA
Michèl
F
Barton
JJ
Unilateral right parietal damage leads to bilateral deficit for high-level motion
Neuron
 , 
2001
, vol. 
32
 (pg. 
985
-
995
)
Battelli
L
Cavanagh
P
Martini
P
Barton
JJS
Bilateral deficits of transient visual attention in right parietal patients
Brain
 , 
2003
, vol. 
126
 (pg. 
2164
-
2174
)
Battelli
L
Pascual-Leone
A
Cavanagh
P
The ‘when’ pathway of the right parietal lobe
Trends Cogn Sci
 , 
2007
, vol. 
11
 (pg. 
204
-
210
)
Battelli
L
Walsh
V
Pascual-Leone
A
Cavanagh
P
The ‘when’ parietal pathway explored by lesion studies
Curr Opin Neurobiol
 , 
2008
, vol. 
18
 (pg. 
120
-
126
)
Bischoff-Grethe
A
Hazeltine
E
Bergren
L
Ivry
RB
Grafton
ST
The influence of feedback valence in associative learning
Neuroimage
 , 
2009
, vol. 
44
 (pg. 
243
-
251
)
Blakemore
S-J
Frith
CD
Self-awareness and action
Curr Opin Neurobiol
 , 
2003
, vol. 
13
 (pg. 
219
-
224
)
Bledowski
C
Kaiser
J
Rahm
B
Basic operations in working memory: contributions from functional imaging studies
Behav Brain Res
 , 
2010
, vol. 
214
 (pg. 
172
-
179
)
Brooks
JC
Nurmikko
TJ
Bimson
WE
Singh
KD
Roberts
N
fMRI of thermal pain: effects of stimulus laterality and attention
Neuroimage
 , 
2002
, vol. 
15
 (pg. 
293
-
301
)
Brooks
DJ
Dopaminergic action beyond its effects on motor function: imaging studies
J Neurol
 , 
2006
, vol. 
253
 
Suppl 4
(pg. 
8
-
15
)
Clark
L
Bechera
A
Damasio
H
Aitken
MR
Sahakian
BJ
Robbins
TW
Differential effects of insular and ventromedial prefrontal lesions on risky decision-making
Brain
 , 
2008
, vol. 
131
 (pg. 
1311
-
1322
)
Corbetta
M
Shulman
GL
Control of goal-directed and stimulus-driven attention in the brain
Nat Rev Neurosci
 , 
2002
, vol. 
3
 (pg. 
201
-
215
)
Craig
AD
How do you feel—now? The anterior insula and human awareness
Nat Rev Neurosci
 , 
2009
, vol. 
10
 (pg. 
59
-
70
)
Critchley
HD
Wiens
S
Rotshtein
P
Ohman
A
Dolan
RJ
Neural systems supporting interoceptive awareness
Nat Neurosci
 , 
2004
, vol. 
7
 (pg. 
189
-
195
)
Danckert
J
Ferber
S
Pun
C
Broderick
C
Striemer
C
Rock
S
Stewart
D
Neglected time: impaired temporal perception of multisecond intervals in unilateral neglect
J Cogn Neurosci
 , 
2007
, vol. 
19
 (pg. 
1706
-
1720
)
Danckert
J
Ferber
S
Doherty
T
Steinmetz
H
Nicolle
D
Goodale
MA
Selective, non-lateralized impairment of motor imagery following right parietal damage
Neurocase
 , 
2002
, vol. 
8
 (pg. 
194
-
204
)
Decety
J
Lamm
C
The role of the right temporoparietal junction in social interaction: how low- level computational processes contribute to meta-cognition
Neuroscientist
 , 
2007
, vol. 
13
 (pg. 
580
-
593
)
Decety
J
Jeannerod
M
Prablanc
C
The timing of mentally represented actions
Behav Brain Res
 , 
1989
, vol. 
34
 (pg. 
35
-
42
)
Dien
J
Spencer
KM
Donchin
E
Localization of the event-related potential novelty response as defined by principal components analysis
Brain Res Cogn Brain Res
 , 
2003
, vol. 
17
 (pg. 
637
-
650
)
Donchin
E
Coles
MGH
Is the P300 component a manifestation of context updating?
Behav Brain Sci
 , 
1988
, vol. 
11
 (pg. 
357
-
374
)
Donchin
E
Spencer
KM
Dien
J
van Boxtel
GJM
Bocker
KBE
Brain and behavior: past, present, and future
 , 
1997
Tilburg (the Netherland)
Tilburg University Press
Duhamel
JR
Goldberg
ME
Fitzgibbon
EJ
Sirigu
A
Grafman
J
Saccadic dysmetria in a patient with a right frontoparietal lesion. The importance of corollary discharge for accurate spatial behaviour
Brain
 , 
1992
, vol. 
115
 (pg. 
1387
-
1402
)
Duhamel
JR
Colby
CL
Goldberg
ME
The updating of the representation of visual space in parietal cortex by intended eye movements
Science
 , 
1992
, vol. 
255
 (pg. 
90
-
92
)
Eling
P
Derckx
K
Maes
R
On the historical and conceptual background of the wisconsin card sorting test
Brain Cogn
 , 
2008
, vol. 
67
 (pg. 
247
-
253
)
Fitts
PM
The informed capacity of the human motor system in controlling the amplitude of movements
J Exp Psych
 , 
1954
, vol. 
47
 (pg. 
381
-
391
)
Folstein
MF
Folstein
SE
McHugh
PR
Mini-mental state. A practical method for grading the cognitive state of patients for the clinician
J Psychiatric Res
 , 
1975
, vol. 
12
 (pg. 
189
-
198
)
Gazzaniga
MS
Principles of human brain organization derived from split-brain studies
Neuron
 , 
1995
, vol. 
14
 (pg. 
217
-
228
)
Gibbons
R
Game theory for applied economics
 , 
1992
Princeton (NJ)
Princeton University Press
Hagmann
P
Cammoun
L
Gigandet
X
Meuli
R
Honey
CJ
Wedeen
VJ
Sporns
O
Mapping the structural core of human cerebral cortex
PLoS Biol
 , 
2008
, vol. 
6
 (pg. 
1
-
15
)
Heide
W
Blankenburg
M
Zimmermann
E
Kömpf
D
Cortical control of double-step saccades: implications for spatial orientation
Ann Neurol
 , 
1995
, vol. 
38
 (pg. 
739
-
748
)
Hermsdörfer
J
Hentze
S
Goldenberg
G
Spatial and kinematic features of apraxic movement depend on the mode of execution
Neuropsychologia
 , 
2006
, vol. 
44
 (pg. 
1642
-
1652
)
Herrmann
CS
Knight
RT
Mechanisms of human attention: event-related potential and oscillations
Neurosci Biobehav Rev
 , 
2001
, vol. 
25
 (pg. 
465
-
476
)
Hori
Y
Minamimoto
T
Kimura
M
Neuronal encoding of reward value and direction of actions in the primate putamen
J Neurophys
 , 
2009
, vol. 
102
 (pg. 
3530
-
3543
)
Husain
M
Shapiro
K
Martin
J
Kennard
C
Abnormal temporal dynamics of visual attention in spatial neglect patients
Nature
 , 
1997
, vol. 
385
 (pg. 
154
-
156
)
Ino
T
Nakai
R
Azuma
T
Kimura
T
Fukuyama
H
Differential activation of the striatum for decision making and outcomes in a monetary task with gain and loss
Cortex
 , 
2010
, vol. 
46
 (pg. 
2
-
14
)
Janssen
P
Shadlen
MN
A representation of the hazard rate of elapsed time in macaque area LIP
Nat Neurosci
 , 
2005
, vol. 
8
 (pg. 
234
-
241
)
Jeannerod
M
The cognitive neuroscience of action
 , 
1997
Cambridge (MA)
Blackwell
Jones
CL
Ward
J
Critchley
HD
The neuropsychological impact of insular cortex lesions
J Neurol Neurosurg Psychiatry
 , 
2010
, vol. 
81
 (pg. 
611
-
618
)
Joseph
R
Frontal lobe psychopathology: mania, depression, confabulation, catatonia, perseveration, obsessive compulsions, and schizophrenia
Psychiatry
 , 
1999
, vol. 
62
 (pg. 
138
-
172
)
Karnath
H-O
Fruhmann-Berger
M
Kuker
W
Rorden
C
The anatomy of spatial neglect based on voxelwise statistical analysis: a study of 140 patients
Cereb Cortex
 , 
2004
, vol. 
14
 (pg. 
1164
-
1172
)
Karnath
H-O
Ferber
S
Himmelbach
M
Spatial awareness is a function of the temporal not the posterior parietal lobe
Nature
 , 
2001
, vol. 
411
 (pg. 
950
-
953
)
Koelsch
S
Fritz
T
Cramon
DY
Muller
K
Friederici
AD
Investigating emotion with music: an fMRI study
Human Brain Map
 , 
2006
, vol. 
27
 (pg. 
239
-
250
)
Lee
D
McGreevy
BP
Barraclough
DJ
Learning and decision making in monkeys during a rock paper-scissors game
Cogn Brain Res
 , 
2005
, vol. 
25
 (pg. 
416
-
430
)
Lee
D
Conroy
ML
McGreevy
BP
Barraclough
DJ
Reinforcement learning and decision making in monkeys during a competitive game
Cogn Brain Res
 , 
2004
, vol. 
22
 (pg. 
45
-
58
)
Leibenluft
E
Gobbini
MI
Harrison
T
Haxby
JV
Mothers' neural activation in response to pictures of their children and other children
Biol Psychiatry
 , 
2004
, vol. 
56
 (pg. 
225
-
232
)
Leon
MI
Shadlen
MN
Representation of time by neurons in the posterior parietal cortex of the macaque
Neuron
 , 
2003
, vol. 
38
 (pg. 
317
-
327
)
Livesey
AC
Wall
MB
Smith
AT
Time perception: manipulation of task difficulty dissociates clock functions from other cognitive demands
Neuropsychologia
 , 
2007
, vol. 
45
 (pg. 
321
-
331
)
Luo
S
Ainslie
G
Giragosian
L
Monterosso
JR
Behavioural and neural evidence of incentive bias for immediate rewards relative to preference-matched delayed rewards
J Neurosci
 , 
2009
, vol. 
29
 (pg. 
14820
-
14827
)
Martín-Rodríguez
JF
Carríon
J
Theory of mind deficits in patients with acquired brain injury: a quantitative review
Neuropsychologia
 , 
2010
, vol. 
48
 (pg. 
1181
-
1191
)
Menon
V
Uddin
LQ
Saliency, switching, attention and control: a network model of insula function
Brain Struct Funct
 , 
2010
, vol. 
214
 (pg. 
655
-
667
)
Merrifield
C
Hurwitz
M
Danckert
J
Multimodal temporal perception deficits in a patient with left spatial neglect
Cogn Neurosci
 , 
2010
, vol. 
1
 (pg. 
244
-
253
)
Miller
MB
Valsangkar-Smith
M
Newman
S
Dumont
H
Wolford
G
Brain activations associated with probability matching
Neuropsychologia
 , 
2005
, vol. 
43
 (pg. 
1598
-
1608
)
Nelson
HE
A modified card sorting test sensitive to frontal lobe defects
Cortex
 , 
1976
, vol. 
12
 (pg. 
313
-
324
)
Norman
DA
Shallice
T
Davidson
RJ
Schwartz
GE
Shapiro
D
Attention to action: willed and automatic control of behaviour
Consciousness and self-regulation: advances in research and theory
 , 
1986
New York (NY)
Plenum Press
Paulus
MP
Feinstein
JS
leland
D
Simmons
AN
Superior temporal gyrus and insula provide response outcome-dependent information during assessment and action selection in a decision-making situation
Neuroimage
 , 
2005
, vol. 
25
 (pg. 
607
-
615
)
Paulus
MP
Feinstein
JS
Tapert
SF
Liu
TT
Trend detection via temporal difference model predicts inferior prefrontal cortex activation during acquisition of advantageous action selection
Neuroimage
 , 
2004
, vol. 
21
 (pg. 
733
-
743
)
Peirce
JW
Generating stimuli for neuroscience using PsychoPy
Front Neuroinform
 , 
2009
, vol. 
2
 (pg. 
1
-
8
)
Piper
BJ
Li
V
Eiwaz
MA
Kobel
YV
Benice
TS
Chu
AM
Olsen
RHJ
Rice
DZ
Gray
HM
Mueller
ST
Executive function on the psychology experiment building language tests
Behav Res Methods
 , 
2011
 
doi: 10.3758/s13428-011-0096-6
Platt
ML
Glimcher
PW
Neural correlates of decision variables in parietal cortex
Nature
 , 
1999
, vol. 
400
 (pg. 
233
-
238
)
Preuschoff
K
Quartz
SR
Bossaerts
P
Human insula activation reflects risk prediction errors as well as risk
J Neurosci
 , 
2008
, vol. 
28
 (pg. 
2745
-
2752
)
Preuschoff
K
Bossaerts
P
Quartz
SR
Neural differentiation of expected reward and risk in human subcortical structures
Neuron
 , 
2006
, vol. 
51
 (pg. 
381
-
390
)
Roser
ME
Fiser
J
Aslin
RN
Gazzaniga
MS
Right hemisphere dominance in visual statistical learning
J Cogn Neurosci
 , 
2011
, vol. 
23
 (pg. 
1088
-
1099
)
Sato
Y
Akiyama
E
Farmer
J
Chaos in learning a simple two-person game
Proc Natl Acad Sci U S A
 , 
2002
, vol. 
99
 (pg. 
4748
-
4751
)
Seger
CA
Peterson
EJ
Cincotta
CM
Lopez-Paniagua
D
Anderson
CW
Dissociating the contributions of independent corticostriatal systems to visual categorization learning through the use of reinforcement learning modeling and Granger causality modeling
Neuroimage
 , 
2010
, vol. 
50
 (pg. 
644
-
656
)
Seymour
B
O'Doherty
JP
Dayan
P
Koltzenburg
M
Jones
AK
Dolan
RJ
Friston
KJ
Frackowiak
RS
Temporal difference models describe higher-order learning in humans
Nature
 , 
2004
, vol. 
429
 (pg. 
664
-
667
)
Sirigu
A
Duhamel
JR
Cohen
L
Pillon
B
Dubois
B
Agid
Y
The mental representation of hand movements after parietal cortex damage
Science
 , 
1996
, vol. 
273
 (pg. 
1564
-
1568
)
Snyder
LH
Batista
AP
Andersen
RA
Coding of intention in the posterior parietal cortex
Nature
 , 
1997
, vol. 
386
 (pg. 
167
-
170
)
Spencer
KM
Dien
J
Donchin
E
A componential analysis of the ERP elicited by novel events using a dense electrode array
Psychophysiol
 , 
1999
, vol. 
36
 (pg. 
409
-
414
)
Spinazzola
L
Pia
L
Folegatti
A
Marchetti
C
Berti
A
Modular structure of awareness for sensorimotor disorders: evidence from anosagnosia for hemiplegia and anosagnosia for hemianaesthesia
Neuropsychologia
 , 
2008
, vol. 
46
 (pg. 
915
-
926
)
Talairach
J
Tournoux
P
Co-planar stereotaxic atlas of the human brain
 , 
1988
New York
Thieme Medical Publishers
Tenenbaum
JB
Kemp
C
Griffiths
TL
Goodman
ND
How to grow a mind: statistics, structure, and abstraction
Science
 , 
2011
, vol. 
331
 (pg. 
1279
-
1285
)
Thielscher
A
Pessoa
L
Neural correlates of perceptual choice and decision making during fear/disgust discrimination
J Neurosci
 , 
2007
, vol. 
27
 (pg. 
2908
-
2917
)
Tolman
EC
Cognitive maps in rats and men
Psych Rev
 , 
1948
, vol. 
55
 (pg. 
189
-
208
)
Umarova
R
Saur
D
Schnell
S
Kaller
C
Vry
M
Glauche
V
Rijntjes
M
Hennig
J
Kiselev
V
Weiller
C
Structural connectivity for visuospatial attention: significance of ventral pathways
Cereb Cortex
 , 
2010
, vol. 
20
 (pg. 
121
-
129
)
Vickery
TJ
Jiang
YV
Inferior parietal lobule supports decision making under uncertainty in humans
Cereb Cortex
 , 
2009
, vol. 
19
 (pg. 
916
-
925
)
Vuilleumier
P
Sergent
C
Schwartz
S
Valenza
N
Girardi
M
Husain
M
Driver
J
Impaired perceptual memory of locations across gaze-shifts in patients with unilateral spatial neglect
J Cogn Neurosci
 , 
2007
, vol. 
19
 (pg. 
1388
-
1406
)
Weller
JA
Levin
IP
Shiv
B
Bechera
A
The effects of insula damage on decision-making for risky gains and losses
Social Neurosci
 , 
2009
, vol. 
4
 (pg. 
347
-
358
)
Wilson
B
Cockburn
J
Halligan
P
Development of a behavioural test of visuospatial neglect
Arch Phys Med Rehab
 , 
1987
, vol. 
68
 (pg. 
98
-
102
)
Wittmann
M
Leland
DS
Paulus
MP
Time and decision making: differential contribution of the posterior insular cortex and the striatum during a delay discounting task
Exp Brain Res
 , 
2007
, vol. 
179
 (pg. 
643
-
653
)
Wolford
G
Miller
MB
Gazzaniga
M
The left hemisphere's role in hypothesis formation
J Neurosci
 , 
2000
, vol. 
20
 pg. 
RC64
 
Wolford
G
Newman
SE
Miller
MB
Wig
GS
Searching for patterns in random sequences
Can J Exp Psych
 , 
2004
, vol. 
58
 (pg. 
221
-
228
)
Wolpert
D
Goodbody
SJ
Husain
M
Maintaining internal representations: the role of the human superior parietal lobe
Nat Neurosci
 , 
1998
, vol. 
1
 (pg. 
529
-
533
)