• Aging effects in selective attention

• Dynamic adaptation in decoding temporal information

Aging effects in selective attention

Selective attention to relevant information while ignoring distracting stimuli is crucial for goal-directed behavior. Negative priming (NP),a well-established paradigm (see Fig. 1) in the study of selective attention, is characterized by longer reaction times when responding to stimuli which have been actively ignored recently. Research has examined how the underlying processes change over the lifespan and revealed age-related deficits as well as different attentional processes that are not impaired by aging. Interestingly, our recent experimental results revealed NP together with neural correlates in simple tasks in older subjects, whereas young subjects did not show neither effect.

Although the NP effect is robust and has been obtained in a variety of tasks and for a number of different stimuli and response types, the results of NP studies with older subjects are very heterogeneous. A central problem of the interpretation of age-related differences in the NP effect is the lack of agreement about the underlying mechanisms. Over the past 20 years, various theoretical accounts have been developed to explain NP, e.g. distractor inhibition, response blocking, episodic retrieval, combination of inhibition and retrieval, temporal discrimination, adaptive thresholding and reaction retrieval. However, empirical evidence does not clearly favour one theory over the others. None of the approaches can consistently explain all priming phenomena. One of the reasons, why it is so hard to decide between the different theoretical accounts, is the lack of a concrete computational formulation (with the notable exception of the distractor inhibition theory) that could make cross-testing possible on a quantitative level.

A main goal of our project is the development of a general computational model that incorporates all mechanisms relevant to selective attention. The model should also be able to explain the hitherto diverging experimental findings of age-related effects. We are implementing such a model at three steps of realization and in a process of alternating computational modeling and experimental evaluation.

As a first step we designed a minimal computational model of NP based on the theoretical account provided by the imago-semantic action model. This dynamical-systems approach successfully reproduces effects of identity priming experiments not only qualitatively but even quantitatively, e.g. the dependency of NP on the response to stimulus intervals in a range of 500 to 2000 milliseconds. In addition, we simulated NP experiments with previously untested stimulus combinations, e.g. the absence of a distractor, to derive predictions that could be confirmed by recent experiments.

In the current stage we develop a general model of selective attention (see Fig. 2) based on a similar dynamical system. The general model is flexible enough to deal with all NP paradigms including forced-choice reaction tasks, flanker and Stroop tasks. It therefore contains all essential components of information processing (perception, attention, memory, semantic representation) as well as action selection. The concrete implementation incorporates building blocks for feature detection, feature binding, semantic representation, action planning and episodic memory as well as a control unit that keeps track of higher goals. Reaction time differences in various priming conditions emerge by an interplay of all model components. The model enables a quantitative comparison of various explanations of NP as their assumptions are built in such that single set-screws determine the impact of the assumptions of a single theory on the model behaviour.

In order to reveal the neurophysiological basis of the general model, we performed a model-based analysis of EEG data which have been obtained in two EEG-studies in cooperation with the Medical Psychology group and the Differential and Diagnostic Psychology Göttingen. We studied the causes for inter- and intraindividual specificities in the model as well as information about the time course of processing. The first study examined younger adults executing an identity priming task. Event related potential (ERP) correlates implicate, that positive and negative priming share early mechanisms, but their processing differs in later control stages. In a second EEG study we compared younger and older adults in a forced-choice NP task. During data analysis, we developed a method to determine a cleaner ERP average by reaction time correction with the aid of single trial ERP estimates. In earlier experiments we addressed the effect of experimental instructions on strategies of stimulus processing, the impact of arousal and alertness on reaction time differences and several constellations of distracting stimuli, which can be explained in the framework of the general model.

Dynamic adaptation in decoding temporal information

Thomas Rammsayer, J. Michael Herrmann, Stephan Blaschke and Joachim Hass

Time is a very important dimension of our life. However, the mechanisms that help us to deal with time are yet not well understood. In this project, we aim to develop a computational model that connects neurophysiological facts with simultaneously performed psychophysical experiments.

The two parallel approaches concentrate on timing errors that can be measured in tasks like interval discrimination or rhythm sensitivity. Theoretical models make predictions about how timing errors depend on certain characteristics of the task. For instance, it is known that errors increase with the duration of the estimated interval. Recent studies suggest that this increase is linear (Weber's law) only for a range of durations, while outside this range substantial deviations from linearity occur. We have constructed a neuronal model that can account both for Weber's law and its deviations. The model consists of a group of layered networks with feed-forward connectivity, synfire chains, which propagate a wave of neural activity along a sequence of pools of neurons. For stable synchronization of the activity of the neuronal pools the propagation progresses temporally linear. Therefore the system is able to translate temporal information into a spatial code at a precision in the range of milliseconds even in the presence of usual biological inaccuracies. For a single synfire chain the timing errors increase with the square root of the interval length as consequence of the law of large numbers. If, however, several synfire chains with different transmission speeds are allowed to compete, the experimentally observed error course could be shown to result as an optimal solution.

Another aspect of timing errors is their dependence on available context information. If a participant is presented with a sequence of intervals of constant length (CI), followed by an interval of a slightly varied length (VI), one might expect that the discrimination threshold to detect this deviating interval from the previous presented CIs decreases with an increasing number of CIs. In psychology, there are models that predict such an augmented discrimination, like the oscillator model, while others claim that there is no such effect, such as the standard neural-counter model. As a critical test of these two classes of models, we conducted a series of experiments presenting a VI at different positions within a sequence of CIs. We find that the performance to detect a VI improves with the position of the deviant interval, suggesting an adaptation (figure 3, left). However, the performance abruptly degrades if the VI is presented at the final position (“end effect”) and overall discrimination abilities are even worse than in a discrimination task with only two intervals. Tests at different CI lengths revealed that the timing errors are in full accordance with the predictions of the computational model.

Further experiments were performed in order to clarify why the discrimination performance varies with the position of the VI in the sequence. If, e.g., a one-cycle oscillation were used for timing, performance should be better if the intervals are presented in phase with the oscillation. We tested this assumption by matching the time between each CIs (ISI) to the duration of the CI, but did not find performance to improve in this case, so this mechanism could be ruled out. Furthermore, we made the surprising finding that the total number of CIs does not influence the performance. This excludes the “end effect” to be due to the “magical number seven memory effect”. This is interesting also in another respect: Sequences are apparently not processed as a whole. In this case, additional intervals would increase the total length of the stored sequence and decrease the performance, just as changing the CI length does. Thus, each interval must be processed individually and evaluation of the holistic features of the sequence must be performed in a higher processing stage. We propose that this stage averages the interval lengths using a stack-storage memory system working with the last-in-first-out principle. If the memory trace contains only two intervals at a time, the last interval has no following partner and so a comparison becomes more difficult. This serial memory system would also cause a processing bottleneck that explains the poor performance of the sequence task in comparison to a pairwise matching task. Currently, we are incorporating this processing stage into the computational model and plan experiments to quantitatively test its predictions.

In 2007, we started two experimental studies which are still in progress. In the first study, we test the influence of the neurotransmitter dopamine on the performance of humans to perform different timing tasks. Results from animal studies suggest that an increased level of dopamine leads to a systematic underestimation of time intervals. We aim to reproduce these findings in humans, complementing human studies with dopamine antagonists. The dopamine level of the participants is raised using a Rotigotin (Neupro®) patch, which is using in the treatment of Parkinson’s disease. The timing tasks include interval discrimination, time generalization, rhythm and temporal order tasks. With this study, we are able to assess whether different mechanisms for these classes of timing task can be pharmalogically dissected, or if we find evidence for a common mechanism.

The second study aims at the connection between time perception and motor actions. Using the haptic device, we guide the participants hand movements to follow an elliptic form. While passing different sections of the ellipse, they were asked to perform a timing task. Depending on the curvature and thus, the speed of the movement, we expect that the subjective time of the participants to contract or dilate.

Selected Publications: