UNIVERSITÄT ZU KÖLN

The notion of predictive processing has garnered considerable attention in the literature on language comprehension. However, a substantial part of the discussion surrounding prediction has traditionally focused on the distinction between prediction and integration, and the difficulty of separating the two empirically. This has sometimes been taken to imply competing architectures, thus giving rise to the question of whether prediction should be viewed as an essential component of language processing. More recently, the notion that prediction need not be viewed as an "all-or-nothing" concept has gained more traction in the field, sparking discussion, for example, around whether the degree of prediction may differ depending on the circumstances in which language processing is occurring. In this talk, I will argue that variability in predictive language processing can be couched naturally within active inference as an overarching framework of information processing in the human brain. The active inference framework posits that the human brain actively generates explanations for its sensory input and strives to minimise the surprise associated with sensory observations. This is accomplished by combining (top-down) predictions generated by an internal model of the world with incoming (bottom-up) sensory input and updating the model in the case of a prediction error. I will show how variability in sentence comprehension can arise through shifts in the balance between top-down and bottom-up information during model updating and how this may lead to variability in predictive language processing between individuals, across the adult lifespan, in different contexts and between different languages. I will argue that, in addition to reflecting the varying information processing affordances of the language input, this variability may, in part, be linked to intrinsic differences in neural information processing.

Abstract following soon!

Our ability to successfully interact with other people depends on recognizing and understanding other persons in different contexts. Especially, when sensory signals are degraded, informative priors can improve speech perception but may also lead to deception. Therefore, unravelling how the human brain combines sensory input with context and prior knowledge is important. In my talk, I will present data from several studies investigating how prior expectations influence speech perception. Firstly, I will contrast two functionally distinct computational mechanisms by which prior expectations can influence sensory representations of speech. Expected features of the input could be enhanced or sharpened. Alternatively, in Prediction Error accounts, expected features are suppressed and unexpected signals are processed further. We aimed at distinguishing between these two accounts by combining behavioural, univariate, and multivariate fMRI measures with computational models. Secondly, I will address the question why we are better in understanding familiar speakers with data from behavioural online studies, in which listeners could use voice context to normalize vowel perception.

Short abstract following soon!

Most research on the neurobiology of language overlooks its relationship to consciousness (and vice versa). I have proposed theoretical models of the ‘natural organisation of language and the brain’ (the NOLB model) and a related model specifying the relationship between the neurobiology of language and consciousness (the HOLISTIC model). Both suggest a fundamental role of the ability of speech production systems to predict acoustic input in auditory cortices, in both speech production and perception. This is consistent with most models of the ‘neural correlates of consciousness’ which propose that consciousness is uniquely associated with reverberation, i.e., feedforward and back dynamics. This often takes the form of predictions and prediction error. I will give an overview of the NOLB and HOLISTIC models and provide a few examples of predictive processing in speech perception and language comprehension from our lab. I also provide some preliminary evidence for the relationship between speech and language prediction and consciousness, using perturbations of consciousness like the anaesthetic propofol and psychedelic drugs. Collectively, this work suggests that prediction is a core construct in a mechanistic understanding of the relationship between the neurobiology of language and consciousness, with implications for mental health and wellbeing.

Abstract following soon!

The N400 component of the event-related brain potential has aroused much interest because it is thought to provide an online measure of meaning processing in the brain. However, the underlying process remains incompletely understood and actively debated. In this talk, I present a computationally explicit account of this process and the emerging representation of sentence meaning. We simulate N400 amplitudes as the change induced by an incoming stimulus in an implicit and probabilistic representation of meaning captured by the hidden unit activation pattern in a neural network model of sentence comprehension, and we propose that the process underlying the N400 also drives implicit learning in the network. The model provides a unified account of 16 distinct findings from the N400 literature and connects human language comprehension with recent deep learning approaches to language processing. I will also give an overview over some recent and ongoing work in the lab scaling up the model to large scale language environments and testing various model predictions based on empirical EEG and MEG studies.

The predictive processing framework has found wide applications in cognitive science and philosophy. It is an attractive candidate for a unified account of the mind in which perception, cognition, and action fit together under common computational principles. For example, according to the standard barebones predictive processing, perception and imagination may differ merely by the role of the incoming sensory signal. While predictions that drive perception are tightly coupled to the sensory signal, imagination disregards it, which allegedly allows untying the imagined content from one’s experienced reality. However, decoupling one’s mind from the current experience may not be quite so simple. In my two-part talk, I discuss the challenges predictive-processing agents would face when thinking about past, future, or entirely fictional events and the necessary augmentations to the predictive-processing framework for accommodating these challenges. The first part of the talk focuses on counterfactual cognition, while the second on imagination and the role language may play in the systematicity of imaginative content in cases where imagination radically departs from our lived experience.