ABSTRACT. We present a computational evaluation of three hypotheses about sources of deficit in sentence comprehension in aphasia: slowed processing, intermittent deficiency, and resource reduction. The ACT-R based Lewis and Vasishth (2005) model is used to implement these three proposals. Slowed processing is implemented as slowed default production-rule firing time; intermittent deficiency as increased random noise in activation of chunks in memory; and resource reduction as reduced goal activation. As data, we considered subject vs. object rela- tives whose matrix clause contained either an NP or a reflexive, presented in a self-paced listening modality to 56 individuals with aphasia (IWA) and 46 matched controls. The participants heard the sentences and carried out a picture verification task to decide on an interpretation of the sentence. These response accuracies are used to identify the best parameters (for each participant) that correspond to the three hypotheses mentioned above. We show that controls have more tightly clustered (less variable) parameter values than IWA; specifically, compared to controls, among IWA there are more individuals with low goal activations, high noise, and slow default action times. This suggests that (i) individual patients show differential amounts of deficit along the three dimensions of slowed processing, intermittent deficient, and resource reduction, (ii) overall, there is evidence for all three sources of deficit playing a role, and (iii) IWA have a more variable range of parameter values than controls. In sum, this study contributes a proof of concept of a quantitative implementation of, and evidence for, these three accounts of comprehension deficits in aphasia.

ABSTRACT. The Lucia comprehension system attempts to model human comprehension by using the Soar cognitive architecture, Embodied Construction Grammar (ECG), and an incremental, word-by-word approach to grounded processing. Traditional approaches use techniques such as parallel paths and global optimization to resolve ambiguities. Here we describe how Lucia deals with lexical, grammatical, structural, and semantic ambiguities by using knowledge from the surrounding linguistic and environmental context. It uses a local repair mechanism to maintain a single path, and shows a garden path effect when local repair breaks down. Data on adding new linguistic knowledge shows that the ECG grammar grows faster than the knowledge for handling context, and that low-level grammar items grow faster than more general ones.

ABSTRACT. The ungrammatical sentence "The key to the cabinets are on the table" is known to lead to an illusion of grammaticality. As discussed in the meta-analysis by Jaeger et al., 2017, faster reading times are observed at the verb are in the agreement-attraction sentence above compared to the equally ungrammatical sentence "The key to the cabinet are on the table". One explanation for this facilitation effect is the feature percolation account: the plural feature on cabinets percolates up to the head noun key, leading to the illusion. An alternative account is in terms of cue-based retrieval (Lewis & Vasishth, 2005), which assumes that the non-subject noun cabinets is misretrieved due to a partial feature-match when a dependency completion process at the auxiliary initiates a memory access for a subject with plural marking. We present evidence for yet another explanation for the observed facilitation. Because the second sentence has two nouns with identical number, it is possible that these are, in some proportion of trials, more difficult to keep distinct, leading to slower reading times at the verb in the first sentence above; this is the feature overwriting account of Nairne, 1990. We show that the feature overwriting proposal can be implemented as a finite mixture process. We reanalysed ten published data-sets, fitting hierarchical Bayesian mixture models to these data assuming a two-mixture distribution. We show that in nine out of the ten studies, a mixture distribution corresponding to feature overwriting furnishes a superior fit over both the feature percolation and the cue-based retrieval accounts.

ABSTRACT. This paper presents a simulation study focusing on implicit memory in the formation of a new communication system. In the models presented here, two agents aim to achieve their common goal by exchanging messages composed of two figures, whose meanings are not defined in advance. The effect of implicit memory has been studied with two different symbolic processes, implemented in ACT-R. Our results indicate that the difference caused by symbolic processes reduces when implicit memory is incorporated into the model. We have also found the effect of implicit memory on the creation of an isomorphic communication system, shared among agents. These findings suggest that implicit memory has some roles in the formation of a human communication system.

ABSTRACT. Rumination is a process of uncontrolled, narrowly-foused negative thinking that is often self-referential, and that is a hallmark of depression. Despite its importance, little is known about its cognitive mechanisms. Rumination can be thought of as a specific, constrained form of mind-wandering. Here, we introduce a cognitive model of rumination that we developed on the basis of our existing model of mind-wandering. The rumination model implements the hypothesis that rumination is caused by maladaptive habits of thought. These habits of thought are modelled by adjusting the number of memory chunks and their associative structure, which changes the sequence of memories that are retrieved during mind-wandering, such that during rumination the same set of negative memories is retrieved repeatedly. The implementation of habits of thought was guided by empirical data from an experience sampling study in healthy and depressed participants. On the basis of this empirically-dervied memory structure, our model naturally predicts the declines in cognitive task performance that are typically observed in depressed patients. This study shows how we can use cognitive models to better understand the cognitive mechanisms underlying rumination and depression.

ABSTRACT. How do affective processes interact with cognitive processes to modulate our behavior? Understanding the processes that influence the interactions between affective stimuli and human decision-making behavior is important for predicting typical behavior under a variety of circumstances, from purchasing behavior to deciding when to enact certain rules of engagement in battle scenarios. Though some computational process models have been proposed in the past, they typically focus on higher-level phenomena and are less focused on the particular architectural mechanisms related to the behavior explored. This, in turn, can make it very difficult to combine the proposed model with existing related work (i.e., the models can’t be tractably combined). We used a modified version of the Iowa Gambling Task to explore the effects of subliminal affective (visual) stimuli on decision-making behavior. We developed a model that runs within the ACT-R/Φ architecture that completes the same task completed by participants. In addition to the affective and cognitive memory components particularly important to the discussion, the model also uses perceptual and motor components within the architecture to complete the task. The architecture has representations of primitive affect that interact with cognitive memory components mainly through an affective-associations module (meant to capture behavior typically ascribed to several amygdalar substructures). The model and affective architectural mechanisms provide a process-oriented explanation for the ways affect may interact with higher-level cognition to mediate human behavior during daily-life.

ABSTRACT. Attachment modelling is an emerging field at the intersection of research in Attachment Theory and computational modelling of emotion. Existing attachment models vary from very abstract models to simulations of specific experimental protocols, such as the Strange Situation Procedure. This paper argues for the benefits in broadening attachment modelling of infants and young children to also include simulating attachment Q set descriptors. These descriptors provide a broader and more rounded challenge for attachment modelling because they can be observed in naturalistic contexts and are less dependent on the specific details of laboratory settings. A computational model is presented which simulates a selection of attachment Q set descriptors. This is an extension of a model designed to simulate the Strange Situation Procedure. A `route map' for future developments towards capturing all Q sort descriptors is discussed.

ABSTRACT. Although meditation and mindfulness practices are widely discussed and studied more and more in the scientific literature, there is little theory about the cognitive mechanisms that comprise it. Here we begin to develop such a theory by creating a computational cognitive model of a particular type of meditation: focused attention mediation. This model was created within Prims, a cognitive architecture similar to and based on ACT-R, which enables us to make predictions about the cognitive tasks that meditation experience may affect. We implemented a model based on an extensive literature review of how the meditation experience unfolds over time. We then subjected the Prims model to a session of the Sustained Reaction to Response Task, a task typically used to study sustained attention, a faculty that may be trained with meditation practice. Analyses revealed that the model was significantly more sensitive to detecting targets and non-targets after the meditation practice than before. These results agree with empirical findings of a longitudinal study conducted in 2010. These results suggest that our approach to modeling meditation and its effects of cognition is feasible.

ABSTRACT. ACT-R has been successfully used in psycholinguistics to model experimental processing data. In this paper, I show how it could be further scaled up to model a much larger set of data, eye-tracking corpus data. It is shown that the resulting model has a good fit to the data in the considered (low-level) processes. Furthermore, it is argued that free parameters of ACT-R could and should be estimated using the well-established methods in other fields, rather than by manually searching through parameter space. The latter option is simply impossible to use once we hit the amount of data considered here. The latter option also makes it hard, if not impossible, to compare parameters across different (ACT-R) models since manual search is subjective and usually not well documented in research papers.

ABSTRACT. In this talk, I will discuss recent developments in data-driven accumulation-to-bound process models (e.g. sequential sampling models), and demonstrate how we have advanced findings in our domain through incorporating brain-lesion patient data, in which these lesions were localized to the left pre-frontal cortex through brain-imaging work (fMRI). Particularly, patients with lesions in the left prefrontal cortex (PFC) have been shown to be impaired in lexical selection, especially when interference between semantically-related alternatives is increased. Current theory has considered that task inference is handled by cognitive control, which works have identified in the brain with frontoparietal activity (e.g. Badre, 2008; Koechlin et al., 2003). Recent model-based neuroscience work involving accumulation models with healthy participants, consider such control to be handled by decision threshold modulation (e.g. Domenech & Dreher 2010; Forstmann et. al 2010). Herein by analyzing patients specifically with PFC lesions, and comparing them to healthy speakers on lexical selection, we indeed find deficits in appropriate modulation of the decision threshold as task interference is introduced. The modeling also provides accounts for how such mechanisms may be modulated by other task factors such as repetition and trial lag.

ABSTRACT. Many cognitive modeling efforts are concerned with when cognitive events occur in time and many cognitive neuroscience efforts are concerned with where things are happening in the brain. We have combined hidden semi-Markov models (HSMM) and multivariate pattern analysis (MVPA) to merge the information from both sources. I will describe how we have used HSMM-MVPA to both discover and test models of cognitive processes.

ABSTRACT. To gain a better understanding of the cognitive mechanisms underlying choice it is often useful to explicate theories with formal cognitive models. In the domain of perceptual choice, this approach has a relatively long tradition, and has helped in identifying the role of certain brain regions in determining decision boundaries. In particular, based on accumulator models of decision making, it has been shown that individual differences in average BOLD responses in the striatum reflect the difference between decision boundary estimates for speed-stressed and accuracy-stressed conditions (Forstmann et al., 2008), and that trial-to-trial fluctuations in striatal BOLD prior to stimulus onset reflect trial-to-trial fluctuations in the decision boundary (Van Maanen et al., 2011). In the current work, we extend these results and show that also time-variant changes in the decision boundary (i.e., within a single-trial) are reflected by striatal BOLD responses, both in terms of individual differences (Van Maanen et al., 2016) and in terms of trial-to-trial variability. These results are in support of models of decision making that propose time-variant decision boundaries (e.g., Cisek et al., 2009, Drugowitsch et al., 2012, Frazier & Yu, 2008).

ABSTRACT. Computational models have often been used to interpret neuroimaging data. For example, both the transient activation of groups or layers of neurons in a neural network and the periods of resource utilization of difference modules in symbolic architectures can predict regional differences in observed brain activity. While this approach is useful and profound, it stills falls short of taking advantage of the full range of predictions that computational models can make. For example, comparisons between neuroimaging data and model activity overlook the directionality of information processing. Models often make strong assumptions on how information passes through different components at different times. Indeed, this directionality is often necessary for the model to work properly, but is rarely tested. While directionality cannot be directly estimated from neuroimaging recordings, it can be inferred using more complex analysis techniques, such as Granger causality and dynamic causal modeling. As an example, we present the case of an ACT-R model that solves Raven’s Advanced Progressive Matrices, a non-verbal test of fluid intelligence. Two versions of the model make identical predictions in terms of behavioral regional fMRI activity, but entail opposite views about the role of the basal ganglia in mediating the connectivity between prefrontal cortex and higher-level visual areas. An analysis of effective connectivity between the two regions reveals that the basal ganglia act by reducing, rather than increasing, the connectivity between the two regions, thus providing support for one version of the model and disproving the alternative.

ABSTRACT. Intertemporal choice requires a mixture of valuation and self-control processes, and previous studies have implicated the contribution of several key brain areas in these types of tasks, yet their precise role has yet to be unraveled. Here we propose a model based on decision field theory where attention is allocated to separate attribute dimensions in a stochastic manner. In particular, the model takes the value of the rewards and delays of the two options as inputs, and probabilistically samples information on a moment-by-moment basis. This integration process creates a subjective representation of the “smaller sooner” and “larger later” options, where the relative contributions of reward and temporal delay within the subjective representations is determined by an attribute bias parameter. On top of the integration process are mechanisms such as lateral inhibition and leakage, which operate asymmetrically across the two options. In the model, these mechanisms mimic concepts such as self control and impulsivity when operating in a goal-directed manner. By fitting a hierarchical version of the model to data from an intertemporal choice experiment, we found that the model provides a good account of choice and response time behavior. A whole-brain general linear model analysis revealed separate patterns of correlations between brain areas putatively associated with valuation or self control and the model parameters.

ABSTRACT. I will discuss how we used large-scale spiking neural networks to simulate associative recognition. Associative recognition is the important ability to learn that two items co-occur. For example, in the current experiment participants first studied word pairs. In a subsequent test phase, they had to distinguish between target pairs, re-paired foils, and foils consisting of entirely new words. To make this distinction, participants did not only need to remember the words they learned, but also which words occurred together as a pair. Although detailed symbolic models exist that account for behavior, fMRI, and EEG data, it remains unclear how associative recognition is performed at the neural level. To investigate this, we used the Neural Engineering Framework to simulate associative recognition with spiking neural networks that can process symbols (using a vector representation known as semantic pointers) and coordinate cognition through the basal ganglia (e.g., Eliasmith et al., 2012). The model goes through the established stages of a recall-to-reject model of associative recognition: perceiving the word pair on the screen, determining whether the encoded words are familiar, recollecting and representing the most similar word pair from memory, deciding whether this is the same pair as presented on the screen, and issuing a response. Because the resulting neural network model is very complex (> 500,000 neurons) we use magnetoencephalographic (MEG) data to constrain the model (Borst et al., 2016). The model matches data in occipital, temporal, prefrontal, and motor cortices, and shows how the associative recognition process could be implemented in the human brain.