Dialogical Reasoning in Patients with Schizophrenia
It is well known that reasoning in patients with schizophrenia is often considered atypical. This may contribute to the detrimental and untreatable social impairments that are central to this disorder. However, most research in the area has concerned how subjects individually solve various reasoning tasks, and few - if any - attempts have been made to find out how patients with schizophrenia reason in interaction.This project investigates how patients with schizophrenia reason in dialogue, taking their argumentation as a point of departure. In our analysis we assume that arguments in natural language are enthymematic, that is not strictly logical. Instead, enthymemes rely on accepted principles of reasoning - in rhetoric called topoi - which dialogue participants draw on to invent and interpret arguments. This study investigates how reasoning in dialogue diverge between patients and healthy controls, but also how having a patient in the group affects the reasoning of healthy controls. Our data set consists of dialogues where the participants have been asked to discuss and solve a moral dilemma. We will evaluate data quantitatively and qualitatively, and develop a precise model of the deviations and variations involved including the behavioural markers of social impairment in schizophrenia. This project will provide unique insights which will inform the development of targeted treatments in this area.
Final report
RJ projekt P16-0805 Dialogical Reasoning in patients with schizophrenia (DRiPS)
1. Purpose and development.
The purpose of DRiPS was to investigate the reasoning in patients diagnosed with schizophrenia based on an existing data set of triadic dialogues. The data consists of two parts, one of 20 dialogues each involving one patient and two healthy controls and one of 20 dialogues involving three healthy controls. This enabled us to look not only at the reasoning among the patients but also among the controls in the patient groups as compared to the control groups. In order to capture the non-logical reasoning of the corpus dialogues we used the theory introduced in Breitholtz (2020, 2014) where arguments are treated as enthymematic, that is relying on withheld information to be supplied by the listener. As this information, nor the principles of reasoning used to apply it, is not explicitly supplied by the speaker, the enthymeme can be a powerful rhetorical device in the sense that it provides deniability for the speaker. However, enthymemes are all around us in ordinary conversation, although we often do not think about the reasoning steps that are implicit as they are usually shared and commonplace. In the case of patients with schizophrenia this is not always the case. We hypothesised that part of the problem with reasoning reported in patients is related to this.
The following research questions have been addressed in the project:
1. In terms of natural language reasoning, how do patients with schizophrenia differ from their healthy interlocutors (patients' partners) and how do both of these groups differ from participants in dialogues without a patient (controls)?
a) How do the participants reason -- are there differences between the groups in terms of the arguments they use and how they express them?
b) Are there differences between the groups during reasoning sequences in terms of verbal dialogue behaviour (e.g. the use of repair, specific words and expressions)?
c) Does the use of head and hand gesture during reasoning sequences differ between patients, patients' partners and controls?
2. How do these factors interact and can we give a precise account of any differences?
An important and somewhat unexpected development of the project has been the emergence of a link between enthymematic arguments and social meaning. Social meaning refers to information that can be conveyed by means of linking linguistic cues to social information associated with particular groups. This led us to focus on aspects of social meaning and how it is linked to linguistic behaviour in our data set and the general mechanisms behind this. This unexpected insight resulted in several publications (Breitholtz & Cooper, 2018a; 2018b; 2018c; Noble et al, 2020) and a new project funded by the Wallenberg foundation. Another discovery outside the remit of the original research plan was that the turn-taking behaviour in patient interactions is different to that in the control groups, with measurable effects on the healthy controls interacting with a patient, despite them being unaware of the patient’s diagnosis (Lavelle and Howes, 2019, Howes and Lavelle, 2021, 2023).
2. Implementation
Annotation
We annotated a variety of factors in the dialogues. These included automatic annotation of disfluencies such as self-repair, filled pauses (e.g. “er”, “um”) and unfilled pauses (see Howes et al., 2017a), hand movement as a proxy for gesture (see Howes and Lavelle, 2023), turn-taking (Howes and Lavelle, 2023) and laughter (Lavelle et al., 2018), as well as manual annotation of reasons (Howes et al., 2021), rhetorical questions (Iossef et al., 2021) and topoi (in preparation). All these methods were developed during the project.
Analysis
Our analysis methods spanned cognitive science statistical methods (e.g. Howes et al., 2017a; Howes & Lavelle, 2023), qualitative analyses (e.g. Ioussef et al., 2021, Howes & Lavelle, 2023) and formal modelling (e.g. Breitholtz et al., 2021; Noble et al., 2020).
3. Result and conclusions
Well established difficulties in reasoning in patients diagnosed with schizophrenia manifest in the following ways in face-to-face interactions:
- Patients are more likely to “go outside” the remit of the conversational topic or task, providing a similar number of reasons per turn as non-patients but a greater proportion of reasons rejecting the constraints of the task. This might have to do with the fact that patients have greater difficulty in seeing the task as a kind of abstract game as opposed to an imagined real-life situation in which decisions have to be made, and is consistent with task based assessments showing that patients with schizophrenia have difficulty employing abstract thinking (Howes et al. 2021).
- Patients are also more likely to be consistent in their reasoning across a dialogue than their interlocutors or controls. Patients were shown to be more consistent than non-patients in that they produced fewer arguments both for and against throwing a particular person. This could indicate a lack of ability or willingness to weigh different arguments against each other. This pattern may stem from a cognitive rigidity, which has been identified in patients with schizophrenia using cognitive tests (Garc ´ia-Mieres et al., 2020). It may also be consistent with viewing the task as an imagined real-life situation rather than as an abstract game.
- We have developed a formal model that can account for the asymmetry that sometimes occurs in dialogue involving patients diagnosed with schizophrenia (Breitholtz et al. 2021; Cooper, 2023). This has also lead to formal analyses of topic shifts using an adjusted version of the same model (Decker, 2022; Decker, Amblard & Breitholtz, 2022; Decker, Breitholtz, Howes & Larsson, 2023)
- Patients tend to rank topoi differently to non-patients and do not accommodate or update rankings in the same way as non-patients and that this does have an effect on the contributions to the dialogue by non-patients (Breitholtz et al. 2021)
- Different patterns of disfluencies were found in the patient group and the control groups, with patients partners displaying different behaviour (Howes et al, 2017a). This led to our hypothesis that turn-taking was different in the patient and control groups, which was indeed the case with the interesting differences being between the healthy participants conversing with a patient and the control group rather than the patients themselves, suggesting that they are adapting to the patients’ behaviour (Lavelle and Howes, 2019; Howes and Lavelle, 2021, 2023). We also identified differences in patients’ non-verbal behaviours, with the correlation between gesture and repair seen in controls absent in the patient sample (Howes and Lavelle, 2023), and differences in laughter behaviours (Lavelle et al., 2018).
- In general patients were shown to be competent in the mechanics of reasoning but were less likely to take account of arguments presented by other participants in the task and to change their mind on the basis of other people’s contributions. They also tended to choose different patterns of reasoning (topoi) than participants without a diagnosis (Breitholtz et al., 2021)
4. New research questions
- Why are patients more consistent in the arguments they present compared to non-diagnosed participants? Our current analysis establishes this but does not distinguish between several possible causes. For example, it may be that patients are simply more defensive or less engaged with the task (which would be consistent with their on average shorter dialogue turns) or it might be that they find it hard to reason counterfactually.
- To what extent are the results we have obtained dependent on the fact that patients took the game more seriously and were more emotionally involved than participants without a diagnosis, who appeared to regard the task as a game not to be taken realistically? This problem would have to be investigated in a different kind of data than that used in this project.
- What are the implications of the results of differences in non-verbal behaviours such as gesture, laughter and turn-taking, and how does this impact on the dynamics of the interaction at a more fine-grained level?
5. Dissemination
Since the start of the project we have collaborated with a research group (SLAM) at the University of Lorraine, France, working on formal models of discourse involving patients diagnosed with schizophrenia. We have participated in workshops organised by them in Nancy, France and also hosted a workshop in Gothenburg to learn more about the work of this group and to discuss results and other issues with them. This has led to a joint PhD student (Decker, 2022) with the University of Lorraine working on topic shifts in conversations involving patients diagnosed with schizophrenia currently being co-supervised by Breitholtz. We have also presented our work and discussed it with the psychology group at Queen’s university, Belfast. We have also given several invited talks at universities and conferences around Europe (Breitholtz, 2023; Cooper, 2022a; Breitholtz & Howes, 2017) as well as in China (Cooper, 2022b) addressing aspects of DRiPS.
1. Purpose and development.
The purpose of DRiPS was to investigate the reasoning in patients diagnosed with schizophrenia based on an existing data set of triadic dialogues. The data consists of two parts, one of 20 dialogues each involving one patient and two healthy controls and one of 20 dialogues involving three healthy controls. This enabled us to look not only at the reasoning among the patients but also among the controls in the patient groups as compared to the control groups. In order to capture the non-logical reasoning of the corpus dialogues we used the theory introduced in Breitholtz (2020, 2014) where arguments are treated as enthymematic, that is relying on withheld information to be supplied by the listener. As this information, nor the principles of reasoning used to apply it, is not explicitly supplied by the speaker, the enthymeme can be a powerful rhetorical device in the sense that it provides deniability for the speaker. However, enthymemes are all around us in ordinary conversation, although we often do not think about the reasoning steps that are implicit as they are usually shared and commonplace. In the case of patients with schizophrenia this is not always the case. We hypothesised that part of the problem with reasoning reported in patients is related to this.
The following research questions have been addressed in the project:
1. In terms of natural language reasoning, how do patients with schizophrenia differ from their healthy interlocutors (patients' partners) and how do both of these groups differ from participants in dialogues without a patient (controls)?
a) How do the participants reason -- are there differences between the groups in terms of the arguments they use and how they express them?
b) Are there differences between the groups during reasoning sequences in terms of verbal dialogue behaviour (e.g. the use of repair, specific words and expressions)?
c) Does the use of head and hand gesture during reasoning sequences differ between patients, patients' partners and controls?
2. How do these factors interact and can we give a precise account of any differences?
An important and somewhat unexpected development of the project has been the emergence of a link between enthymematic arguments and social meaning. Social meaning refers to information that can be conveyed by means of linking linguistic cues to social information associated with particular groups. This led us to focus on aspects of social meaning and how it is linked to linguistic behaviour in our data set and the general mechanisms behind this. This unexpected insight resulted in several publications (Breitholtz & Cooper, 2018a; 2018b; 2018c; Noble et al, 2020) and a new project funded by the Wallenberg foundation. Another discovery outside the remit of the original research plan was that the turn-taking behaviour in patient interactions is different to that in the control groups, with measurable effects on the healthy controls interacting with a patient, despite them being unaware of the patient’s diagnosis (Lavelle and Howes, 2019, Howes and Lavelle, 2021, 2023).
2. Implementation
Annotation
We annotated a variety of factors in the dialogues. These included automatic annotation of disfluencies such as self-repair, filled pauses (e.g. “er”, “um”) and unfilled pauses (see Howes et al., 2017a), hand movement as a proxy for gesture (see Howes and Lavelle, 2023), turn-taking (Howes and Lavelle, 2023) and laughter (Lavelle et al., 2018), as well as manual annotation of reasons (Howes et al., 2021), rhetorical questions (Iossef et al., 2021) and topoi (in preparation). All these methods were developed during the project.
Analysis
Our analysis methods spanned cognitive science statistical methods (e.g. Howes et al., 2017a; Howes & Lavelle, 2023), qualitative analyses (e.g. Ioussef et al., 2021, Howes & Lavelle, 2023) and formal modelling (e.g. Breitholtz et al., 2021; Noble et al., 2020).
3. Result and conclusions
Well established difficulties in reasoning in patients diagnosed with schizophrenia manifest in the following ways in face-to-face interactions:
- Patients are more likely to “go outside” the remit of the conversational topic or task, providing a similar number of reasons per turn as non-patients but a greater proportion of reasons rejecting the constraints of the task. This might have to do with the fact that patients have greater difficulty in seeing the task as a kind of abstract game as opposed to an imagined real-life situation in which decisions have to be made, and is consistent with task based assessments showing that patients with schizophrenia have difficulty employing abstract thinking (Howes et al. 2021).
- Patients are also more likely to be consistent in their reasoning across a dialogue than their interlocutors or controls. Patients were shown to be more consistent than non-patients in that they produced fewer arguments both for and against throwing a particular person. This could indicate a lack of ability or willingness to weigh different arguments against each other. This pattern may stem from a cognitive rigidity, which has been identified in patients with schizophrenia using cognitive tests (Garc ´ia-Mieres et al., 2020). It may also be consistent with viewing the task as an imagined real-life situation rather than as an abstract game.
- We have developed a formal model that can account for the asymmetry that sometimes occurs in dialogue involving patients diagnosed with schizophrenia (Breitholtz et al. 2021; Cooper, 2023). This has also lead to formal analyses of topic shifts using an adjusted version of the same model (Decker, 2022; Decker, Amblard & Breitholtz, 2022; Decker, Breitholtz, Howes & Larsson, 2023)
- Patients tend to rank topoi differently to non-patients and do not accommodate or update rankings in the same way as non-patients and that this does have an effect on the contributions to the dialogue by non-patients (Breitholtz et al. 2021)
- Different patterns of disfluencies were found in the patient group and the control groups, with patients partners displaying different behaviour (Howes et al, 2017a). This led to our hypothesis that turn-taking was different in the patient and control groups, which was indeed the case with the interesting differences being between the healthy participants conversing with a patient and the control group rather than the patients themselves, suggesting that they are adapting to the patients’ behaviour (Lavelle and Howes, 2019; Howes and Lavelle, 2021, 2023). We also identified differences in patients’ non-verbal behaviours, with the correlation between gesture and repair seen in controls absent in the patient sample (Howes and Lavelle, 2023), and differences in laughter behaviours (Lavelle et al., 2018).
- In general patients were shown to be competent in the mechanics of reasoning but were less likely to take account of arguments presented by other participants in the task and to change their mind on the basis of other people’s contributions. They also tended to choose different patterns of reasoning (topoi) than participants without a diagnosis (Breitholtz et al., 2021)
4. New research questions
- Why are patients more consistent in the arguments they present compared to non-diagnosed participants? Our current analysis establishes this but does not distinguish between several possible causes. For example, it may be that patients are simply more defensive or less engaged with the task (which would be consistent with their on average shorter dialogue turns) or it might be that they find it hard to reason counterfactually.
- To what extent are the results we have obtained dependent on the fact that patients took the game more seriously and were more emotionally involved than participants without a diagnosis, who appeared to regard the task as a game not to be taken realistically? This problem would have to be investigated in a different kind of data than that used in this project.
- What are the implications of the results of differences in non-verbal behaviours such as gesture, laughter and turn-taking, and how does this impact on the dynamics of the interaction at a more fine-grained level?
5. Dissemination
Since the start of the project we have collaborated with a research group (SLAM) at the University of Lorraine, France, working on formal models of discourse involving patients diagnosed with schizophrenia. We have participated in workshops organised by them in Nancy, France and also hosted a workshop in Gothenburg to learn more about the work of this group and to discuss results and other issues with them. This has led to a joint PhD student (Decker, 2022) with the University of Lorraine working on topic shifts in conversations involving patients diagnosed with schizophrenia currently being co-supervised by Breitholtz. We have also presented our work and discussed it with the psychology group at Queen’s university, Belfast. We have also given several invited talks at universities and conferences around Europe (Breitholtz, 2023; Cooper, 2022a; Breitholtz & Howes, 2017) as well as in China (Cooper, 2022b) addressing aspects of DRiPS.