Perceptual decisions between multiple directions of visual motion

Abstract

Previous studies and models of perceptual decision making have largely focused on binary choices. However, we often have to choose from multiple alternatives. To study the neural mechanisms underlying multialternative decision making, we have asked human subjects to make perceptual decisions between multiple possible directions of visual motion. Using a multicomponent version of the random-dot stimulus, we were able to control experimentally how much sensory evidence we wanted to provide for each of the possible alternatives. We demonstrate that this task provides a rich quantitative dataset for multialternative decision making, spanning a wide range of accuracy levels and mean response times. We further present a computational model that can explain the structure of our behavioral dataset. It is based on the idea of a race between multiple integrators to a decision threshold. Each of these integrators accumulates net sensory evidence for a particular choice, provided by linear combinations of the activities of decision-relevant pools of sensory neurons.

Figure 1.

Experimental paradigm. Human observers were asked to make a judgment about the strongest direction of motion in a random-dot pattern with multiple motion components. They were free to watch the stimulus as long as they wanted to and responded with a goal-directed eye movement to one of three choice targets. Choices and RTs were measured.

Figure 2.

Experimental results. A, Relative frequency of choice as a function of the motion strength of the strongest component. The symbols represent the data points, and the error bars represent 95% confidence intervals for the estimated probabilities (see text for details). The dotted lines connect neighboring data points for the same trial type (different combinations of motion strengths of the two weaker components). The trial type is indicated by the color (see B for the legend). The solid lines show the function fits used for quantifying the slope of the psychometric functions (see text). The shape of the symbol indicates the choice [circle, correct choice (strongest component has been chosen); square, target associated with the intermediate component has been chosen; diamond, target associated with the weakest component has been chosen]. The dashed line indicates chance level. The symbols (and lines) have been shifted horizontally to avoid overlap. This is indicated by the light gray areas. All symbols in such an area would normally be located on the central vertical line. B, Mean RT as a function of the motion strength of the strongest component. The symbols represent the data points, and the error bars represent the mean ± 1 and 2 SEs. The dotted lines connect neighboring data points for the same trial type. The color code is identical to the one used in A (see legend). The light gray areas again indicate a horizontal shift (see A for details).

Figure 3.

Computational model. A, Structure of the model. Three integrators (each associated with one of the three alternatives) race against each other. The integrator output signal (i₁, i₂, or i₃) reaching a decision threshold first determines the choice and terminates the decision process. The integrator input signals (e₁, e₂, and e₃) are net evidence signals, which are linear combinations of the three relevant sensory signals (s₁, s₂, and s₃). Solid arrows indicate positive weights (excitatory connections), and dashed arrows indicate negative weights (inhibitory connections). B, Linear response model for the sensory pools. The piecewise linear response function (purple dashed line) is modeled as the sum of two linear response components (gray and blue solid lines; see text for details). C, Model implementation as a two-dimensional diffusion process with three boundaries (see text for details). Each trial starts at (0, 0). Which of the three thresholds (solid, dashed, and dotted lines) is crossed first determines the choice. The time of the threshold crossing determines the decision time.

Figure 4.

Comparison between the pooled dataset and the model predictions. A, The model was fitted to the mean RTs. The symbols represent the data points as in Figure 2, but the solid lines now connect the model results. Color conventions and horizontal shifts (light gray areas) are as in Figure 2. B, Comparison between the probabilities of the particular choices predicted by the model (connected by solid lines) and the relative frequencies of the choices in the data (symbols; not used for the model fit). Shape conventions are as in Figure 2. C, Comparison between the RT distributions predicted by the model (solid blue lines) and the RT distributions observed in the experiment (gray histograms). The distributions are shown for the four trial types with the largest numbers of observations.

Figure 5.

Comparison between the individual datasets and the model predictions. Each row represents one experimental subject. Otherwise, this figure follows the conventions of Figure 4.