Modeling Acute and Compensated Language Disturbance in Schizophrenia
Uli Grasemann and Ralph Hoffman and Risto Miikkulainen · Cognitive Science · 2011
Modeling Acute and Compensated Language Disturbance in Schizophrenia Uli Grasemann 1 , Ralph Hoffman 2 , and Risto Miikkulainen 1 Department of Computer Sciences, The University of Texas at Austin Department of Psychiatry, Yale University School of Medicine Abstract No current laboratory test can reliably identify patients with schizophrenia. Instead, key symptoms are observed via lan- guage, including derailments, where patients cannot follow a coherent storyline, and delusions, where false beliefs are re- peated as fact. Brain processes underlying these and other symptoms remain unclear, and characterizing them would greatly enhance our understanding of schizophrenia. In this sit- uation, computational models can be valuable tools to formu- late testable hypotheses and to complement clinical research. This work aims to capture the link between biology and schizo- phrenic symptoms using DISCERN, a connectionist model of human story processing. Competing illness mechanisms pro- posed to underlie schizophrenia are simulated in DISCERN, and are evaluated at the level of narrative language, i.e. the same level used to diagnose patients. The result is the first simulation of abnormal storytelling in schizophrenia, both in acute psychotic and compensated stages of the disorder. Of all illness models tested, hyperlearning, a model of overly intense memory consolidation, produced the best fit to the language abnormalities of stable outpatients, as well as compelling mod- els of acute psychotic symptoms. If validated experimentally, the hyperlearning hypothesis could advance the current under- standing of schizophrenia, and provide a platform for develop- ing future treatments for this disorder. Introduction Stories are more than a useful social construct – they are a crucial part of who we are. We make sense of the world by fitting our experience into a coherent narrative. In schizo- phrenia, this ongoing narrative breaks down. Disturbances in the perception and expression of reality can be observed through the stories a patient tells. Indeed, narrative language is the primary diagnostic tool, and clinicians use it every day to observe and evaluate its manifestations. The purpose of clinical interviews, then, is to use narrative language as a window into the schizophrenic mind. The main idea behind this research is that neural network models of storytelling can provide mechnistic explanations of what is seen through that window. These explanations can then be evaluated through narrative language – at the same level used to diagnose real patients. The principal strength of neural network models lies in their ability to bridge the gap between complex mental states and behavior on the one hand and underlying neural informa- tion processing on the other. This ability is precisely what is needed in schizophrenia research, where the central chal- lenge for decades has been to explain how underlying illness mechanisms could cause altered the behavior. Consequently, the main goal of this research is to demon- strate that a neural network model can be used meaning- fully to simulate possible illness mechanisms in schizophre- nia. The different illness models will result in different lan- guage behavior, which can then be used to generate predic- tions about the underlying causes. Thus computational illness models have the potential to complement and guide future medical research. On the other hand, progress in understanding schizophre- nia is likely to lead to progress in basic cognitive science as well. Global, emergent faculties like understanding and telling stories, processing emotions, and forming long-term memories of real or imagined events are difficult to account for computationally, and mental illnesses where these facul- ties break down offer a unique opportunity to investigate how they emerge from their neural substrate. Developing meth- ods to model such high-level behavior computationally is a second major goal of the research presented in this paper. Schizophrenia Schizophrenia is a common and disabling psychiatric disor- der. Symptoms include hallucinations, bizarre behavior, delu- sions, and disorganized language that is hard for listeners to follow. These psychotic symptoms tend to wax and wane over time, and in later stages often give way to negative symp- toms, including blunted emotions and reduced language out- put. This paper focuses on symptoms that are observed di- rectly via language, most prominently 1. Delusions, which are pathological false beliefs. Delusions often share common themes, like being watched by the CIA or being controlled by outside forces. Patients with schizophrenia tend to insert themselves or persons they know into imaginary narratives. Such agent-slotting errors are thought to be the cause of the plots and conspiracies that characterize persecutory delusions. 2. Disorganized speech, which refers to fluent spoken lan- guage that fails to communicate effectively. It is believed to reflect impaired verbal thought (thought disorder). One of the most prominent signs are derailments, i.e. jumps from one topic or story to another without apparent cause. Symptoms in schizophrenia vary among patients. Clini- cal subtypes of schizophrenia include the paranoid type, where symptoms include delusions and hallucinations but not prominent language disorganization, and the disorganized type, where symptoms are dominated by disorganized lan- guage and behavior. Treatments for schizophrenia mainly rely on medication that can help manage psychotic symptoms. However, these drugs often have severe and dangerous side-effects, and do not help all patients or address all symptoms (Kapur and Mamo, 2003). These shortcomings make a better understand- ing of schizophrenia an important goal, because it would likely lead to more effective drugs, and might suggest new ways to treat or even prevent schizophrenia (Pearlson, 2000). However, our understanding of schizophrenia is far from complete. What is known is that schizophrenia is a physical disease. Structural brain abnormalities, genetics, and neu- rochemistry are key components, and virtually every brain area and major neurotransmitter system has been implicated (Pearlson and Marsh, 1999; Bogerts et al., 2009; Glenthoj