The hippocampus supports both the recollection and the familiarity components of recognition memory

Abstract

The receiver operating characteristic (ROC) has been used to investigate the component processes of recognition memory. Some studies with this technique have been taken to indicate that the hippocampus selectively supports the process of recollection, whereas adjacent cortex in the parahippocampal gyrus supports the process of familiarity. We analyzed ROC data from young adults, memory-impaired patients with limited hippocampal lesions, and age-matched controls. The shape of the ROC changed in similar ways from asymmetric to symmetric, as a function of the strength of memory (strong to weak) in both the young adults and the patients. Moreover, once overall memory strength was similar, the shape of the patient ROC was asymmetric and matched the control ROC. These results suggest that the component processes that determine the shape of the ROC are operative in the absence of the hippocampus, and they argue against the idea that the hippocampus selectively supports the recollection process.

Figure 1. Hypothetical ROC Data Illustrating Symmetrical and Asymmetrical ROC Plots

The degree of asymmetry evident in an ROC is typically quantified by a ‘‘slope’’ parameter obtained by fitting the standard signal detection model (Macmillan and Creelman, 2005) to the data. A slope of 1.0 denotes a symmetrical ROC, whereas a slope less than 1.0 denotes an asymmetrical ROC. The dual process/detection model would yield a recollection parameter estimate of 0 for the symmetrical ROC at the top and an estimate greater than 0 for the asymmetrical ROC at the bottom.

Figure 2. Recognition Memory Performance of Young Adults Tested with 50-Item Lists at Retention Intervals of 1 hr, 1 day, 1 week, 2 weeks, and 8 weeks

Performance for 19–24 subjects/group was quantified by the standard, bias-free measure of recognition memory (d′), as derived from signal-detection theory, in which d′= z (Hit Rate) minus z (False Alarm Rate). The solid curve represents the least squares fit of a three-parameter power function that typically provides a good fit of forgetting data (Wixted, 2004).

Figure 3. ROC Data Produced by the Young Adults at Each Retention Interval

Following convention, the smooth curves represent the best fits from the standard signal-detection model (Macmillan and Creelman, 2005), and the slope (denoted by s) values represent one of the parameters that is estimated when performing those fits. The chi-square test comparing each slope value to 1.0 was significant for the 1 hr, 1 day, and the 2 week conditions (χ²[1] ≥6.21, p < 0.05) and was marginal for the 1 week condition, χ²(1) = 2.70, p = 0.10. For the 8 week condition, the slope of 1.03 did not differ significantly from 1.0. Previous work with rats suggested that the ROC might be linear after a long retention interval (Fortin et al., 2004). The degree of linearity in the ROC from the 8 week condition was assessed by comparing the fit of a one-parameter curvilinear signal-detection model with the slope fixed at 1.0 (to match the data that we obtained at the 8 week retention interval) and the fit of a one-parameter pure-recollection version of the dual-process/detection model with the familiarity parameter fixed at 0 (to match the linear plot reported for rats in Fortin et al., 2004). The chi-square goodness-of-fit statistic associated with the former (10.94) was much lower than the latter (22.03), indicating that the curvilinear function offered a better description of the data than the linear function even at the longest retention interval.

Figure 4. Recognition Memory Performance of the Hippocampal Patients and Controls

Patients were tested with 50-item lists in the H-50 condition and 10-item lists in the H-10 condition. Controls were tested with 50-item lists in the C-50 condition. A retention interval of 3 min was used in all cases. The mean score of the controls (C-50) was greater than that of the patients in the H-50 condition, t₍₁₂₎ = 5.23 (p < .01), but similar to the d′score obtained by the patients in the H-10 condition (p > 0.25). The d′ score in the H-10 condition was also greater than in the H-50 condition (t_[5] = 4.63, p < 0.01). The error bars represent standard errors.

Figure 5. ROC Data Produced by the Hippocampal Patients and Controls

Top shows the data for the hippocampal patients in the 50-item condition, middle shows the data for the hippocampal patients in the 10-item condition, and bottom shows the data for the controls in the 50-item condition. The slope of 1.14 for the H-50 ROC was not different from 1.0 (p < 0.10), indicating that the ROC was symmetric. The slope of 0.83 for the H-10 ROC and the slope of 0.83 for the C-50 ROC were both less than the slope of 1.14 for the H-50 ROC (χ²[1] ≥4.70, p < 0.05) and were significantly less than 1.0 by a one-tailed test (χ²[1] ≥2.70, p ≤0.05).