Memory

In 1984, Ronald Cotton was convicted for rape and burglary. He was sentenced to life + 50 years. In 1995, he was released having served over 10 years in prison. When Cotton was interrogated he provided a false alibi. Rather than report where he had been at the time of the crime, Cotton recalled where he had been the week before. A primary challenge for alibi generation research is establishing the ground truth of the real world events of interest. We used a smartphone app to record data on participants (N=57) for a month prior to a memory test. The app captured their accelerometry continuously and their GPS location and sound environment every ten minutes. After a week retention interval, we presented participants with a series of trials which asked them to identify where they were at a given time from among four alternatives. Participants were incorrect 36% of the time (SD = 16%). Furthermore, our forced choice procedure allowed us to conduct a conditional logit analysis to assess the relative importance of different aspects of the events to the decision process. The Cotton example suggests that participants might also confuse days across weeks and we found strong evidence of this kind of error. In addition, people often confused weeks in general and also hours across days. Similarity of location induced more errors than similarity of sound environments or movement types.

The method of loci is a powerful mnemonic technique for memorizing a list of unrelated items. With a pre-specified route in a familiar “memory palace”, one can encode material by attaching items to loci along this route, and later effectively recall them by mentally walking along the same route. Despite its efficacy, there is no existing model that explains why the method of loci promotes memory improvement during memory search. To fill this gap, we provide a rational account of why the method of loci improves memory. We define memory search as a task with the goal of minimizing retrieval cost, and demonstrate that the method of loci gives an optimal policy for this task. We discuss the implications of this result, and how it connects to the role of structural prior knowledge in facilitating new learning.

Technological advances are speeding up the pace of our lives. With this increasing pace, the rate at which the brain expects items should accelerate as well. How does technology affect memory retrieval? One theory suggests that human memory is adapted to the statistics of our environment (Anderson & Schooler, 1991). Environmental sources, such as newspaper headlines provide a reflection of the retrieval demands from human memory at different times. By analyzing changes in environmental statistics, such as the time-frequency of event occurrences, we tracked and analyzed how the environment affects memory. We focus on frequency and spacing effects, the latter of which is that the spacing between successive repetitions of an item affects how well the item is remembered at different times from the last occurrence. Working with headlines of The New York Times from 1919 to 2019, we captured changes in the spacing effect. We found that the recurring pattern is polarized between the most and least frequent words: popular words become more likely to recur, and uncommon words less likely. However, the overall recurring likelihood remains fairly constant. We fitted Hawkes’ self-exciting point processes well on the data and were able to predict word recurrences with high accuracy.

In a Linear Associative Net (LAN), all input settles to a single pattern, therefore Anderson, Silverstein, Ritz, and Jones (1977) introduced saturation to force the system to reach other steady-states in the Brain-State-in-a-Box (BSB). Unfortunately, the BSB is limited in its ability to generalize because its responses are restricted to previously stored patterns. We present simulations showing how a Dynamic-Eigen-Net (DEN), a LAN with Short-Term Plasticity (STP), overcomes the single-response limitation. Critically, a DEN also accommodates novel patterns by aligning them with encoded structure. We train a two-slot DEN on a text corpus, and provide an account of lexical decision and judgement-of-grammaticality (JOG) tasks showing how grammatical bi-grams yield stronger responses relative to ungrammatical bi-grams. Finally, we present a simulation showing how a DEN is sensitive to syntactic violations introduced in novel bi-grams. We propose DENs as associative nets with great.