Search

The aim of this thesis is to evaluate the effect of collaborative work on primary-age children's reasoning about shape, space and measure. Children's early development in these areas of mathematics has been well researched but the difficulties older children experience are less well charted. Method: An intervention study was carried out with two Year 5 classes in a South London primary school using a naturalistic experimental design with two parallel classes as the intervention and control groups. The intervention class undertook two iterations of group work while the control class worked individually across the two sessions. An innovative pre- to post-test measure was developed using the Rasch Analysis partial credit model. This consisted of two sets of six similar and overlapping think-aloud tasks. The design of the group work was informed by both sociocultural and socio-cognitive conflict theory: ground rules for talk were used in the intervention class and the children were grouped in triads with differing ideas (elicited from their pre-test responses). Novel features were included to increase the likelihood of all children encountering challenge to their initial ideas: a) the children were prompted to reflect on the certainty of their joint or individual answers at the end of the first session; b) the children all had the chance to consult written clues in the second session, at the cost of a point per clue. Analysis: Quantitative comparison of the children's progress from pre- to post-test was complemented with qualitative analysis of the children's use of language across the study. Transcripts of the children talking about the shape, space and measurement tasks were coded for imaginative expressions to explore whether imaginative forms were associated with more sophisticated reasoning about shape, space and measure. The children's group talk was coded for exploratory features of language to evaluate the impact on the children's reasoning of the prompt to reflect on the certainty of their joint responses. Findings: The children working in groups made significantly more progress than the children working alone from pre- to post-test, as hypothesised, and they were more effective in their decisions about whether to seek help in the form of written clues than children working alone. Following the prompt to reflect on their group answers, there were changes in the observed language of the intervention groups. While the overall use of exploratory features: "because", "I think", "I agree/disagree" and modal verbs remained similar from session one to session two, there was a significant increase in the use of modal verbs following the prompt (p < .05). Swift surface-level agreements were probed further after a group consensus had been reached. Using a grounded approach, the children's imaginative expressions across the study were coded and three categories identified: movement, manipulation or decomposition. Children working with peers made more progress and increased their use of imaginative expressions from pre- to post-test by comparison with the children working alone. Temporal analysis of the children's talk across the study revealed common examples of imaginative work that supported the development of more sophisticated reasoning. For example, imagining "splitting" and "halving" objects served as a heuristic bridge towards inferring length and finding the difference. Similarly, by imagining "turning" and "squashing" familiar shapes, children began to create axioms about the properties of shape beyond iconic examples and orientations. The partial credit model used to measure children's progress in this study increases our knowledge about the incremental steps that children make in their understanding about geometry and measurement in the upper primary years. By revealing the imaginative language tools the children used to probe their existing reasoning models, this study highlights the importance of children articulating their imaginative work in maths with peers.

A rapidly growing literature characterises reasoning as a social phenomenon. On the one hand, Mercier and Sperber (2011) argue that the main function of reasoning is for speakers to persuade group members, e.g., of a given belief or decision. On the other hand, Tomasello (2014) emphasises that reasoning is also critical in cooperative contexts, when groups need to find a good and justified solution benefiting all group members, rather than members competing for individual argumentative success. The literature on children's reasoning, however, focuses on their use of justifications in either conflictual/competitive or cooperative contexts without a systematic comparison of how young children's arguments change across contexts. Also, investigating how children judge arguments and use these judgements as informative input to further tasks would deepen our knowledge about their reasoning abilities. In my dissertation, I therefore investigated with whom children prefer to reason and how they reason with same-age peers as partners in cooperative and competitive contexts. In the first study, I investigated whether children who can choose between cooperative partners penalise candidates that accept poor reasons. 5- and 7-year-old children were presented with two candidates for solving a cooperative problem. In the experimental condition, one acceded to a good reason, the other to a poor reason. In the control condition, each agreed to a different good reason. Crucially, in both conditions, both arrived at a wrong conclusion. Results suggest that 7-year-olds, and 5-year-olds to a lesser degree, preferred the partner who followed the good reason in the experimental condition, but they showed no preference in the control condition. Thus, young children prefer partners who respond to good reasons over partners who comply more easily. This provides evidence that children evaluate reasons even in absence of one objectively correct informant, and use this evaluation to judge cooperation partners. In the second study, I used a more interactive paradigm and investigated how 5- and 7-year-old children reason with peer partners in cooperative and competitive contexts. Peer dyads were asked to place animals in various places in a zoo, which included unusual items such as an alarm clock. One child received training, playing the game with the experimenter before playing with his/her naive peer and learned a set of critical arguments (e.g., 'Bears hibernate so they need the alarm clock to wake up when winter is over'). When the trained child played the game with a naive peer, reproducing these arguments would help both of them in the cooperative condition but worsen the trained child's chances of winning in the competitive condition. The results suggest that 7-year-olds, but not 5-year-olds withheld the trained arguments more often in the competitive condition than in the cooperative condition. Thus, this study suggests that children's reasoning is less biased or less 'inhibited' in cooperative settings in which a joint decision benefits both parties. In the third study, I investigated whether combining cooperative and competitive contexts (namely, cooperating as a team against competitor) would facilitate children's reasoning further. Similar to the second study, 5- and 7-year-olds were asked to cooperate in a task (e.g., decorating a zoo). In the main condition, children were competing against another group, whereas in the control condition, they were not. The prediction was that children would show more productive reasoning and richer discussions in the main condition. However, this prediction did not bear out. Out-group competition did not trigger richer reasoning between partners. I discuss this result in addition to a set of theoretical remarks. As a whole, this set of studies suggests that starting in their sixth year, children start to identify and prefer to reason with people who submit to reason. They display less biased reasoning in cooperative contexts, in which the goal is to reach good joint decisions that benefit all group members. And out-group competition itself does not have a specific positive influence on children's cooperative reasoning.

Research suggests that reasoning biases play a key role in the development and maintenance of paranoia. A number of cognitive interventions have been developed to ameliorate these biases, with the view that this will improve symptoms, functioning and quality of life in individuals with psychosis. Numerous research studies have tested the efficacy of these therapies, and integrating the evidence has important implications for individual, family and service-wide outcomes. The two papers presented in this thesis, a systematic review and an empirical study, aim to contribute to research in the area. The final section, which aims to place the thesis in a wider context, discusses the overall findings, personal reflections on the research process, some limitations of the two studies, and outlines implications for clinical interventions and future research.

In computer science, an ontology is a machine-processable representation of knowledge about some domain. Ontologies are encoded in ontology languages, such as the Web Ontology Language (OWL) based on Description Logics (DLs). An ontology is a set of logical statements, called axioms. Some axioms make universal statements, e.g. all fathers are men, while others record data, i.e. facts about specific individuals, e.g. Bob is a father. A set of universal statements is called TBox, as it encodes terminology, i.e. schema-level conceptual relationships, and a set of facts is called ABox, as it encodes instance-level assertions. Ontologies are extensively developed and widely used in domains such as biology and medicine. Manual engineering of a TBox is a difficult task that includes modelling conceptual relationships of the domain and encoding those relationships in the ontology language, e.g. OWL. Hence, it requires the knowledge of domain experts and skills of ontology engineers combined together. In order to assist engineering of TBoxes and potentially automate it, acquisition (or induction) of axioms from data has attracted research attention and is usually called Ontology Learning (OL). This thesis investigates the problem of OL from general principles. We formulate it as General Terminology Induction that aims at acquiring general, expressive TBox axioms (called general terminology) from data. The thesis addresses and investigates in depth two main questions: how to rigorously evaluate the quality of general TBox axioms and how to efficiently construct them. We design an approach for General Terminology Induction and implement it in an algorithm called DL-Miner. We extensively evaluate DL-Miner, compare it with other approaches, and run case studies together with domain experts to gain insight into its potential applications. The thesis should be of interest to ontology developers seeking automated means to facilitate building or enriching ontologies. In addition, as our experiments show, DL-Miner can deliver valuable insights into the data, i.e. can be useful for data analysis and debugging.

Myth and Argument in Plato's Phaedrus, Republic, and Phaedo investigates the role played by eschatological myth in the arguments of Plato's Phaedrus, Republic and Phaedo. It argues that a reconsideration of the agenda followed by Socrates in each of these dialogues brings into view the contribution made by the mythological narrative to their argumentative line. Each of the three chapters of my thesis analyses the nature of this contribution. The first chapter argues that the myth occupying the central pages of the Phaedrus contributes to developing one of the themes addressed in the dialogue, namely a link between the divine realm and the activities thought by Phaedrus to be unrelated to the religious sphere. By showing that Eros fosters imitation of the gods, the palinode makes an important contribution to this topic. The second chapter proposes that the myth of Er and passage 608c2-621d3 in which it is included are an essential part of the line of argument of the Republic. I analyse the aims Socrates sets in Book 2 for his investigation into justice, and show that they include the description of the positive consequences of justice along with the benefits it causes in and by itself. By listing the rewards just people will receive from other people and the gods, passage 608c2-621d3 gives a description of the positive consequences of justice. The third chapter argues that the argumentative line followed in the Phaedo finds its culmination in the eschatological myth. Socrates expresses a hope for post-mortem justice in his defence of the philosophical life. To render it plausible to his interlocutors he needs to show that the soul is both immortal and intrinsically intelligent. After vindicating these notions, Socrates presents in the concluding myth the image of an afterlife governed by ethical principles.

Classification is a core reasoning service provided by most OWL reasoners. Classification in general is hard - up to 2NExptime for SROIQ(D), the Description Logic which is underpinning the Web Ontology Language (OWL). While it has been shown that classification is practical for a wide range of inputs, there are still ontologies for which classification takes an unreasonable amount of time for purposes such as ontology engineering (frequent classifications after updates). A natural optimisation strategy is divide and conquer, that is, to decompose the ontology into subsets which are hopefully easier to classify and whose classifications can be combined into a complete classification of the whole ontology. Unfortunately, an arbitrary subset may not be self-contained, i.e. it might be missing information that is needed to determine entailments over its signature. Moreover, such a subset can be potentially harder to classify than the whole ontology. In order to mitigate those problems, classification preserving decompositions (CPDs) must be designed with care that they support complete classification which is, in practice, more efficient than monolithic classification. Locality-based modules are subsets of an ontology that provide certain guarantees with respect to the entities (concepts, roles) in its signature - in particular, modules are self-contained. In this thesis we explore the use of syntactic locality-based modules for underpinning classification-preserving decompositions. In particular, we empirically explore their potential to avoid subsumption tests and reduce subsumption test hardness and weigh those benefits against detrimental effects such as overhead (for example the time it takes to compute the decomposition) and redundancy (a consequence of potentially overlapping chunks in the decomposition). The main contributions of this thesis are an in-depth empirical characterisation of these effects, an extensible framework for observing CPDs in action up until a granularity of individual subsumption tests, a large, public corpus of observations and its analysis and insights on experimental methodologies around OWL reasoning.

At first glance, the project of moral enhancement (making people more moral) may appear uncontroversial and obviously worth supporting; surely it is a good idea to make people better. However, as the recent literature on moral enhancement demonstrates, the situation is not so simple – there is significant disagreement over the content of moral norms as well as appropriate means by which to manipulate them. This disagreement seriously threatens many proposals to improve society via moral enhancement. In my dissertation, I develop an understanding of how, exactly, disagreement poses problems for moral enhancement. However, I also argue that there is a way forward. It is possible to bring about moral improvement without commitment to particular and controversial moral norms, but instead relying on relatively uncontroversial ideas concerning morally reliable processes. The upshot is that, while attempting to directly manipulate people’s moral ideas is objectionable, it is relatively unproblematic to focus on helping people reason better and avoid akrasia, with the justified expectation that this will generally lead to moral improvement. We should, therefore, focus not on how to bring people in line with what we take to be the right ideas, motives or behaviors. Rather, we should look to helping people determine for themselves what being moral consists in, as well as help ensure that they act on those judgments. Traditional, non-moral education, it turns out, is actually one of the best moral enhancers we have. In fact, the tools of philosophy (which is, in many aspects, concerned with proper reasoning) are central to the project of indirect moral enhancement. Ultimately, one of the best ways to make people morally better may well be to make them better philosophers.

DL-based ontologies have been widely used as knowledge infrastructures in knowledge management systems and on the Semantic Web. The development of efficient, sound and complete reasoning technologies has been a central topic in DL research. Recently, the paradigm shift from professional to novice users, and from standalone and static to inter-linked and dynamic applications raises new challenges: Can users build and evolve ontologies, both static and dynamic, with features provided by expressive DLs, while still enjoying e cient reasoning as in tractable DLs, without worrying too much about the quality (soundness and completeness) of results? To answer these challenges, this thesis investigates the problem of tractable and quality-guaranteed reasoning for ontologies in expressive DLs. The thesis develops syntactic approximation, a consequence-based reasoning procedure with worst-case PTime complexity, theoretically sound and empirically high-recall results, for ontologies constructed in DLs more expressive than any tractable DL. The thesis shows that a set of semantic completeness-guarantee conditions can be identifed to efficiently check if such a procedure is complete. Many ontologies tested in the thesis, including difficult ones for an off-the-shelf reasoner, satisfy such conditions. Furthermore, the thesis presents a stream reasoning mechanism to update reasoning results on dynamic ontologies without complete re-computation. Such a mechanism implements the Delete-and-Re-derive strategy with a truth maintenance system, and can help to reduce unnecessary over-deletion and re-derivation in stream reasoning and to improve its efficiency. As a whole, the thesis develops a worst-case tractable, guaranteed sound, conditionally complete and empirically high-recall reasoning solution for both static and dynamic ontologies in expressive DLs. Some techniques presented in the thesis can also be used to improve the performance and/or completeness of other existing reasoning solutions. The results can further be generalised and extended to support a wider range of knowledge representation formalisms, especially when a consequence-based algorithm is available.

The transitive inference (TI) problem (i.e., if A > B and B > C, then A > C) has traditionally been considered a hallmark of logical reasoning. However, considerable debate exists regarding the psychological processes involved when individuals perform TI tasks. The current thesis therefore sought to further explore this issue with adult humans as the population sample. Following a review of the literature, the first empirical chapter, Chapter 2, adopted a traditional TI task and exposed participants to training and testing with a simultaneous discrimination paradigm. In addition, the chapter sought to examine the potential facilitative effects of awareness and repeated exposure to training and test phases on the emergence of TI. Results broadly demonstrated that awareness led to more accurate responses at test, and that for a number of participants, repeated exposure to training and test phases, allowed the targeted performances to emerge over time. Chapter 3 developed and determined the utility of a novel behaviour-analytic account of TI as a form of derived comparative relational responding. For the most part, findings revealed that the model has the potential to generate arbitrarily applicable comparative responding in adults, comparable to TI. However, findings from Chapter 3 also revealed that despite the implementation of a number of interventions, response accuracy was still weak on a number of the targeted relations. Chapter 4 developed a variant of the Relational Completion Procedure (RCP) to examine derived comparative responding to 'More-than' and 'Less-than' relations, as an extension of the behavioural account of TI adopted in Chapter 3. Findings revealed that, for the most part, the protocol was effective in establishing the targeted relations, and that the linearity (e.g., A < B, B < C) of training pairs was not found to effect the emergence of this pattern of responding. Chapter 5 sought to explore the transformation of discriminative functions via a 5- member relational network of 'More-than' and 'Less-than' relations. Findings revealed that, across four experiments, approximately half of the participants displayed the predicted patterns of performance. That is, half of the participants responded 'less' to the stimuli ranked lower in the network (A and B) and 'more' to the stimuli ranked higher in the network (D and E), on the basis of training with stimulus C. The utility of the current behaviour-analytic approach to the study of TI is discussed.

This thesis investigates the Theory of Formal Discipline (TFD): the idea that studying mathematics develops general reasoning skills. This belief has been held since the time of Plato (2003/375B.C), and has been cited in recent policy reports (Smith, 2004; Walport, 2010) as an argument for why mathematics should hold a privileged place in the UK's National Curriculum. However, there is no rigorous research evidence that justifies the claim. The research presented in this thesis aims to address this shortcoming. Two questions are addressed in the investigation of the TFD: is studying advanced mathematics associated with development in reasoning skills, and if so, what might be the mechanism of this development? The primary type of reasoning measured is conditional inference validation (i.e. `if p then q; not p; therefore not q'). In two longitudinal studies it is shown that the conditional reasoning behaviour of mathematics students at AS level and undergraduate level does change over time, but that it does not become straightforwardly more normative. Instead, mathematics students reason more in line with the `defective' interpretation of the conditional, under which they assume p and reason about q. This leads to the assumption that not-p cases are irrelevant, which results in the rejection of two commonly-endorsed invalid inferences, but also in the rejection of the valid modus tollens inference. Mathematics students did not change in their reasoning behaviour on a thematic syllogisms task or a thematic version of the conditional inference task. Next, it is shown that mathematics students reason significantly less in line with a defective interpretation of the conditional when it is phrased `p only if q' compared to when it is phrased `if p then q', despite the two forms being logically equivalent. This suggests that their performance is determined by linguistic features rather than the underlying logic. The final two studies investigated the heuristic and algorithmic levels of Stanovich's (2009a) tri-process model of cognition as potential mechanisms of the change in conditional reasoning skills. It is shown that mathematicians' defective interpretation of the conditional stems in part from heuristic level processing and in part from effortful processing, and that the executive function skills of inhibition and shifting at the algorithmic level are correlated with its adoption. It is suggested that studying mathematics regularly exposes students to implicit `if then' statements where they are expected to assume p and reason about q, and that this encourages them to adopt a defective interpretation of conditionals. It is concluded that the TFD is not supported by the evidence; while mathematics does seem to develop abstract conditional reasoning skills, the result is not more normative reasoning.

There is a great deal of empirical evidence to support the efficacy and effectiveness of cognitive therapeutic approaches such as cognitive behaviour therapy (CBT) and motivational interviewing (MI). In addition to this, research has identified several mechanisms of change thought to underlie these approaches. However, there has been little exploration of the specific cognitive and emotional processes that may underlie change in CBT and MI, or indeed their interactions. This thesis specifically explored the role of reasoning and emotion in relation to MI and CBT.In the first paper, the initial sections provided a narrative review whereby the roles of reasoning and emotion in CBT and MI were critically evaluated in context of existing empirical evidence. Further to this, the literature examining the effects of emotion on deductive reasoning was systematically reviewed. A total of 19 papers were identified and the implications of the studies’ findings discussed in relation to the clinical practice of CBT and MI. The studies varied noticeably in terms of the quality of the methodologies employed. Overall the studies considered, suggested that the effects of emotion on reasoning were complex, and yet to be well understood. However, through extrapolating the findings from the experimental study of cognition and emotion to the clinical practice of CBT and MI, the review demonstrated the potential relevance of such findings, and indicated the need for research exploring the role of reasoning and emotion in these approaches.The second, empirical paper examined the extent to which statements reflecting informal reasoning and the emotional state of the client affected the generation of positive behaviour change statements and positive behaviour change in the context of MI with individuals with psychosis and alcohol use disorders. Participants (n=26) were clients selected from a large randomised controlled trial of integrated MI and CBT for psychosis and substance use. One audio recorded therapy session was selected for each participant. Reasoning and emotion statements were subsequently identified and categorised for each therapy session. It was found that the total number of reasoning statements generated positively predicted the total number of positive behaviour change statements generated. It was also found that negative emotional statements relating to the present, positively predicted the total number of reasoning statements generated. The results suggested that informal reasoning may be a specific cognitive process underpinning the production of behaviour change statements in MI with individuals with psychosis and alcohol use disorders, and provided support for the central role of discrepancy in producing change in MI. The final section provided a critical reflection of the research process. This included a rationale for the development of the literature review and the empirical paper, followed by a critical reflection on the study context and its’ implication to the study methodology. Finally the theoretical and future implications for research in this area were discussed along with the implications to clinical practice.

Theories of how individuals reason, and how they experience emotion abound in the psychological literature; yet, despite the common lay-theories of how emotions might affect a person’s reasoning, very little empirical work has been conducted on this relationship. The current thesis addresses this knowledge-gap by first distilling from the literature two classes of emotion theory; Information, and Load; and then systematically testing the explanatory power of these theories. A dual-process framework is employed in order to define low (Type One) and high effort (Type Two) strategies. Information theories predict that negative emotion cues more analytic processing relative to positive emotion, whereas load theories predict both positive and negative emotion to suppress use of high-effort strategies. Thus the two theories are compared by varying incidental and integral emotion across syllogistic reasoning, conditional reasoning, and the ratio-bias task, and assessing the engagement of Type One and Type Two processes across positive emotion, negative emotion, and control conditions. The findings suggest that emotion effects in syllogistic reasoning do not consistently support either Load or Information theories (Experiments 1-4). Emotion effects are found to be typically larger for integral than incidental emotion (Experiment 5), and most frequently serve as Information in verbal (Experiments 6 and 7) and visual conditional reasoning tasks (Experiment 8). Furthermore, these effects are to a large extent dependent on task properties such as the number of alternative antecedents (Experiments 9 and 10), and are greater on more difficult tasks (Experiments 11 and 12). These findings suggest that emotion has a greater impact on Type Two than Type One processes. A range of methodological and theoretical implications which will inform future work in this area are also discussed in the closing chapter.

Rationality is very widely regarded as a normative notion, which underwrites various everyday normative practices of evaluation, criticism, and advice. When some agent behaves irrationally, she is likely to be critically evaluated, and advised to change her ways. Such practices seem to presuppose that agents ought to behave as rationality requires. But some philosophers question this thought. They argue that at least some requirements of rationality cannot be ones that we ought to comply with. This thesis aims to dispel such sceptical doubts over the normativity of rationality; it defends the idea that the requirements of rationality are indeed normative, in the sense that if one is rationally required to F, one ought to F because rationality requires one to F. The normativity of three requirements of practical rationality in particular is the main target for defense in the following pages. They are: [ENKRASIA] Rationality requires of A that, if A believes she ought to F, then A intends to F. [MEANS-ENDS] Rationality requires of A that, if A intends to E, and believes that she will not E unless she intends to M, then A intends to M. [INTENTION CONSISTENCY] Rationality requires of A that, if A intends to F, and believes that she cannot both F and G, then A does not intend to G. After presenting some of the grounds for scepticism about the normativity of these three requirements in chapter 1, the thesis goes on in chapters 2 & 3 to critically examine several different accounts of why rationality is normative, concluding that they are all unsuccessful; a novel account is called for. An account of this kind is offered over the course of the two following chapters, 4 & 5. Each requirement is shown to be constituted by a certain kind of ought, while at the same time corresponding to a rule of correct reasoning. Chapter 6 is devoted to answering an objection to that account, according to which the rules of reasoning are given by permissions rather than requirements. Chapter 7 offers a digression into a related issue in action theory: it unfavorably explores the idea that reasoning is a factor that can be used to analyse not only rational action, but also intentional action more broadly; the chapter suggests that treating intentional action as irreducible is the more fruitful approach. Finally, chapter 8 summarizes the main conclusions of the thesis and comments on some remaining questions.

Selective attention is a mechanism used to allocate resources to information processing. Both mood states and emotionally salient stimuli can influence which information is selectively attended. This information is subsequently processed in a more elaborative manner and affects task performance. The experiments presented in this thesis explore the influence of mood and emotional stimuli on selective attention and consequently task performance. Mood induction procedures were used to induce transient neutral, sad, and happy mood states in healthy volunteers. A systematic review and meta-analysis of 41 studies using sad mood induction procedures showed cognitive impairments in performance in the context of task neutral stimuli. In contrast biases in attention towards mood-congruent negative stimuli led to improved task performance. A series of three behavioural experiments with 197 participants demonstrated that participants made decisions on the basis of less information when that information was preceded by emotional but not neutral stimuli. Induced mood state did not affect performance. The behavioural and neural correlates of visual attentional processing to emotional stimuli were explored using magnetoencephalography in 24 healthy participants following sad, happy, and neutral mood induction procedures. The M300, a component associated with selective attention, had greater amplitude following presentation of negative compared with positive stimuli, which was associated with improved task performance. Reduced M300 amplitude and impairments in performance occurred following sad mood induction procedures. The experiments presented in this thesis demonstrate prioritized processing of emotional information and provide some evidence for impaired performance following sad mood induction procedures.

In many settings, agents engage in problem-solving activities, which require them to share resources, act on each others behalf, coordinate individual acts, etc. If autonomous agents are to e ectively interact (or support interaction among humans) in situations such as deciding whom and how to approach the provision of a resource or the performance of an action, there are a number of important questions to address. Who do I choose to delegate a task to? What do I need to say to convince him/her to do something? Were similar requests granted from similar agents in similar circumstances? What arguments were most persuasive? What are the costs involved in putting certain arguments forward? Research in argumentation strategies has received signi cant attention in recent years, and a number of approaches has been proposed to enable agents to reason about arguments to present in order to persuade another. However, current approaches do not adequately address situations where agents may be operating under social constraints (e.g., policies) that regulate behaviour in a society. In this thesis, we propose a novel combination of techniques that takes into consideration the policies that others may be operating with. First, we present an approach where evidence derived from dialogue is utilised to learn the policies of others. We show that this approach enables agents to build more accurate and stable models of others more rapidly. Secondly, we present an agent decision-making mechanism where models of others are used to guide future argumentation strategy. This approach takes into account the learned policy constraints of others, the cost of revealing in- formation, and anticipated resource availability in deciding whom to approach. We empirically evaluate our approach within a simulated multi-agent frame- work, and demonstrate that through the use of informed strategies agents can improve their performance.

From a logical viewpoint, people must reason to as well as from interpretations in deductive reasoning tasks. There are two main interpretative stances (e.g., Stenning & van Lambalgen, 2004, 2005, 2008): credulous, the act of trying to infer the speaker's intended model; and sceptical, an adversarial strategy. A range of contextual factors in uence interpretation, but there are also differences between individuals across situations. Taking an individual differences approach, this thesis focuses on reasoning in relation to milder variants of the autism spectrum condition (ASC) phenotype in a typically developing (TD) population. Earlier work on discourse processing in ASC using the `suppression' task (van Lambalgen & Smid, 2004; Pijnacker et al., In press) shows that some aspects of reasoning to interpretations are different in the ASC population. Given that autistic traits involve impairment, e.g., in pragmatic language, and peaks of ability, e.g., in perceptual tasks, it was hypothesised that autistic traits would predict features of the inferences people in the TD population draw. Data were collected from university students on a range of reasoning tasks making it possible to investigate the extent to which interpretation is consistent across task within individuals. Tasks chosen were: conditional reasoning using the `suppression' task and Wason's selection task; one and two-premise Aristotelian quantifer reasoning; the Linda problem; and Raven's Advanced Progressive Matrices. Autistic traits were assessed using the Autism Spectrum Quotient (Baron-Cohen et al., 2001), used previously to study autistic traits in TD individuals, and the Broad Autism Phenotype Questionnaire (Hurley et al., 2007). Autistic traits predicted patterns of inference in many of the tasks. The earlier suppression task result in ASC was replicated and extended in our TD population. Different dimensions of autistic trait related differentially to features of the inferences drawn. Some of the inferences drawn were recognisably related to the credulous versus sceptical distinction and correlated cross-task whilst others were seemingly related to more general topdown versus bottom-up processing preferences. These results provide further evidence of the existence of qualitative individual differences in deductive reasoning. They also show the importance of seeking cross-task correlates to move beyond studies of individual tasks and study reasoning to and from interpretations in the same individual.

In the century or so that syllogisms have received the attention of psychologists, their interpretation both as and within particular types of discourse has been largely downplayed. A significant element in the guidance of interpretation is information structure as expressed in prosody. This thesis explores the role of intonation in syllogistic discourse and its effect on reasoning. A theoretical analysis of the information structures of syllogisms is presented which produces two classes of intonation patterns, the 'contextually concordant' (CC) and the 'contextually neutral' (CN), putatively corresponding to two discourse types. These are then investigated in a series of experiments. The initial observational study aims at confirming the use and significance of CC and CN patterns in a syllogism solving task. The remaining two experiments employ a purpose-built voice synthesiser to investigate the effects of imposing CC and CN contours on premises, first in a syllogism solving task and then in a syllogism evaluation task. The results show that both CC and CN intonation patterns are indeed used by participants and bear a systematic relationship to both the number and accuracy of conclusions they draw. When used in the presentation of syllogisms, however, these patterns do not influence the production of conclusions, only the evaluation of them. It is therefore argued that the discourse types to which they relate depend upon whether the syllogism is interpreted as a proof or as a problem. Further work based on these findings could aim to probe the informational links between conclusions and premises and thereby elucidate the coherence of arguments.

Knowledge graph (KG) reasoning has been studied extensively due to its wide applications, which is addressed by two lines of research, i.e., traditional symbolic reasoning and modern neural networks-based techniques. Symbolic reasoning has been the most studied approach toward KG reasoning since the earliest days. This approach represents expert knowledge as statements in a logic-based representation language and implements reasoning as logical inference, primarily addressing high-level reasoning and cognitive processes. For instance, consider the logical rule \emph{$\text{SpeakLanguage}(x, y) \leftarrow \text{LiveIn}(x, z) \wedge \text{OfficialLanguage}(z, y)$}, along with the observed facts \emph{``LiveIn(Mina Miller, USA)''} and \emph{``OfficialLanguage(USA, English)''}, we can infer the new fact \emph{``SpeakLanguage(Mina Miller, English)''}. Symbolic reasoning approaches have demonstrated strong interpretability and generalizability due to the power of logical rules. Nevertheless, they encounter certain limitations when it comes to scaling up to vast datasets, managing ambiguity, and capturing the correlations between entities and relations. To illustrate, consider the case where both \emph{``USA''} and \emph{``United States''} denote the same country. Symbolic reasoning approaches treat them as distinct entities, thereby failing to automatically transfer the knowledge from the existing triple \emph{``LiveIn(Mina Miller, USA)''} to deduce the new fact \emph{``LiveIn(Mina Miller, United States)''}.Neural networks-based techniques have recently emerged as state-of-the-art methods in KG reasoning, owing to their successful applications. Unlike symbolic reasoning, these techniques aim to predict unseen triples by exploring the graph structure in KGs to capture the similarity of entities. Neural networks-based techniques have demonstrated good scalability and a strong ability to capture correlations between entities and relations. However, they fall short in modeling higher-order dependencies of KG relations. For instance, while neural networks-based techniques can recognize that \emph{``USA''} and \emph{``United States''} refer to the same country based on their similarity in embeddings, they cannot infer that \emph{``Mary Stilwell''} and \emph{``Mina Miller''} speak \emph{``English''} without leveraging logical rules.While symbolic reasoning and neural networks-based techniques are typically seen as distinct approaches toward achieving the ultimate goal of KG reasoning, they possess a natural potential to complement and enhance each other. Symbolic approaches offer significant interpretability and generalizability, leveraging logical rules to provide insights into inferred results and enabling easy generalization to unobserved objects. However, they struggle to handle large datasets and fail to capture correlations between entities and relations. Conversely, neural networks-based techniques address the limitations of symbolic approaches by exploring intrinsic similarities among triples, but they lack interpretability and generalizability. By integrating both techniques into a unified framework, neural-symbolic reasoning provides a more efficient, generalizable, and interpretable way to perform KG reasoning. In this thesis, we provide a comprehensive overview of techniques in integrating symbolic logical rules and neural networks for KG reasoning, with a particular focus on two specific tasks, KG completion and logical rule learning and propose our own solutions to address the limitations of existing works.

When making decisions, people can either rely on a gut feeling or engage in reasoned analysis to make a choice. Past research has made competing claims on whether relying on gut feelings or reasoned analysis leads to better decisions. However, these competing claims may be due to the types of decisions being made. Relying on gut feelings has been demonstrated to be superior in judgments about attitudes and leads to greater post-decision satisfaction. However, prior research demonstrating the benefits of gut feelings has used subjective and mostly unverifiable criteria for which to judge the quality of a decision. On the other hand, reasoned analysis has largely been found to be beneficial in other tasks that have an objective, verifiable outcome or criterion for which to judge the quality of a decision. Here, we explore a potential exception to this, sports forecasting. Prior research has demonstrated that relying on gut feelings leads people to make more accurate forecasts about the outcome of sporting events. The first aim of the current research was to replicate this result. However, across the 5 experiments reported here, we see no evidence that relying on a gut feeling increases the accuracy of sports forecasts. The second aim of the current work was to extend prior research to examine how confidently people hold sports forecasts made based on gut feelings and reasoned analysis. We further extend prior work by examining how individual differences affect and interact with the experimental manipulation of how people make decisions. In these additional lines of inquiry, we first saw no evidence supporting the hypothesis that relying on gut feelings increases peoples’ confidence in their choices. Second, we observed that individual differences did not predict the accuracy of peoples’ forecasts or how confident they were about their forecasts. We conclude that relying on a gut feeling or reasoned analysis is unlikely to affect sports forecasting. Therefore, sports forecasts may be a boundary condition for which gut feelings and reasoned analysis arrive at the same decision and the benefits of relying on gut feelings may be restricted to subjective judgments and individual post-decision evaluations.

This thesis aims to provide a theory of human rationality which is consistent with, and supportive of, the naturalistic approach to philosophy of mind. I analyse why the common tendency not to furnish a systematic and rigorous account of rationality is philosophically inadequate - namely, because this overlooks the critical role which the concept of rationality plays in the 'principle of charity' (Quine 1960; Davidson 1984) in philosophy of mind, which is a central part of a unified naturalistic approach to philosophy of mind and rationality. I aim to contribute to the field by rectifying this omission. I identify the theoretical commitments of any naturalistic approach to the theory of rationality - in particular, I argue for the neglected fact that naturalism commits us to an instrumental theory of rationality. I defend the theory of instrumental rationality against its critics, by means of elaborating the implications of the notion - also frequently overlooked - that the concept of rationality is one which applies paradigmatically to agents, rather than to reasoning processes considered in isolation (for example). The resultant instrumental theory of personal rationality which I develop and defend is characterised by two main original (related) features: (1) Certain fundamental features of human rationality are identified as constitutive in nature and status - given that humans are agents, then it follows necessarily that our theory of rationality has these certain characteristics. (2) Full instrumental rationality demands rationality of the self, which introduces further rational constraints of consistency (for example). I build upon these two central features of my theory of rationality to demonstrate how this account can thereby rebut accusations of extreme relativism and inadequacy of scope. Thus I aim to show how my instrumental theory of personal rationality can provide a philosophically adequate account of the normative status of rationality within the remit of the naturalist approach, and thereby also furnish adequate philosophical support for a defensible naturalistic theory of the mind.

Studies of metaphors in teaching and learning have underlined the important role of metaphors in reasoning, but have sometimes failed to show the effect of metaphor on how scientific concepts are represented, and have sometimes overlooked hidden metaphors in their attempts to be explicit about how metaphor functions. This study investigates metaphor in the context of teaching environmental science. It does not assume any simple correlation between surface linguistic cues and the presence or kind of metaphor. Two theoretical approaches have been chosen, Systemic Functional Linguistics (M. Halliday) which sees language as a social construction of meaning, and Image Schema (M Johnson and G Lakoff) which has developed in cognitive science and cognitive linguistics. These two approaches are used to discuss examples of metaphors from a number of lessons which have been observed and video-recorded, and in a variety of textbooks used as resource materials in teaching environmental science. The choice of environmental science as the subject matter arises from two of its distinct characteristics. One is the fact that ideology triggers and shapes the interests, decisions and choices of materials, issues, arguments, reasons, etc. But there is nothing like one unique ideology, on the contrary conflicts of different ideologies raise differences about what will be selected and how it will be represented. At this point there is a special role taken on by metaphor. Metaphors provide the means for creating differences and similarities, thus bringing together or keeping apart ideologies. Second, the teaching of environmental science does not appear as the teaching of science only, bounded from anything else, but is a blend of accounts of scientific and commonsense knowledge. Metaphors appear at the overlapping points where this blending takes place. It is not the purpose of the thesis to question, or to contribute to, the theoretical perspectives adopted. Rather, its interest is in how these perspectives provide, each in their own way, insights into the nature of the discourse of teaching environmental science, and thus to raise questions about its effectiveness.

This thesis presents the RAP system: a Reasoner About Physiology. RAP consists of two modules: knowledge representation and reasoning. The knowledge representation module describes commonsense anatomy and physiology at various levels of abstraction and detail. This representation is broad (covers several physiological systems), dense (the number of relationships between anatomical and physiological elements is high) and uniform (the same kind of formalism is used to represent anatomy, physiology and their interrelationships). These features lead to a 'natural' representation of naive physiology which is, therefore, easy to understand and use. The reasoning module performs two tasks: 1) it infers the behaviour of a complex physiological process using the behaviours of its subprocesses and the relationships between them; 2) it reasons about the effect of introducing a fault in the model. In order to reason about the behaviour of a complex process, RAP uses a mechanism which consists of the following tasks: (i) understanding how subprocesses behave; (ii) comprehending how these subprocesses affect each others behaviours; (iii) "aggregating" these behaviours together to obtain the behaviour of the top level process; (iv) giving that process a temporal context in which to act. RAP uses limited commonsense knowledge about faults to reason about the effect of a fault in the model. It discovers new processes which originate as a consequence of a fault and detects processes which misbehave due to a fault. The effects of both newly generated and misbehaving processes are then propagated throughout the model to obtain the overall effect of the fault. RAP represents and reasons about naive physiology and is a step forward in the development of systems which use commonsense knowledge.

The demolition industry has been examined as a domain for knowledge-based systems, through two very different research investigations, both involving commercial companies. A small local demolition company has collaborated with an investigation into the use of knowledge-based systems to assist with, and model, the process of estimating arisings from demolition work. The majority of UK demolition companies do not have the resources or expertise to approach the quantity estimating task in the same way as the rest of the construction industry. The demolition estimator uses his expertise to arrive at quantity estimates that are as accurate as possible but are made in as short a time as possible, without requiring detailed measurements to be made. The research proposes a model of the reasoning processes that a demolition estimator has been observed to carry out on site. The model forms the basis of a reasoning strategy that ensures that the most accurate estimate is obtained from the available data. Missing data is handled by further inferencing, reference to a structure model, or a shift in the level of reasoning. A prototype knowledge-based system has been developed to estimate the mass of steel in a roof truss system, based upon this proposed model. The system has been developed on a PC using the Crystal shell. The above investigation identified the area of safety in demolition work as being suitable for further examination of the application of knowledge-based systems techniques. This examination was carried out in collaboration with a leading national demolition contractor, based in Rotherham. The demolition safety advisor research project involved the integration of knowledge about safe demolition practice from an expert Demolition Engineer, and from legislative and advisory documents. The methods, difficulties, and implications of this integration are described. Explanation facilities are discussed which combine a variety of information formats, including video and hypertext. A reasoning strategy is suggested which allows a suitable demolition method to be established, and which allows safety advice to be given regarding demolition operations and structure features. A prototype system has been developed on a Sun Workstation, which gives safety advice concerning the explosive demolition of brick chimneys, and reinforced concrete framed structures such as blocks of flats.

Human agents have to deal with a considerable amount of information from their environment and are also continuously faced with the need to take actions. As that information is largely of an uncertain nature, human agents have to decide whether, or how much, to believe individual pieces of information. To enable a reasoning system to deal in general with the demands of a real environment, and with information from human sources in particular, requires tools for uncertainty management and belief formation. This thesis presents a model for the management of uncertain information from human sources. Dealing, more specifically, with information which has been pre-processed by a natural language processor and transformed into an event-based representation, the model assesses information, forms beliefs and resolves conflicts between them in order to maintain a consistent world model. The approach is built on the fundamental principle that the uncertainty of information from people can, in the majority of situations, successfully be assessed through source models which record factors concerning the source's abilities and trustworthiness. These models are adjusted to reflect changes in the behaviour of the source. A mechanism is presented together with the underlying principles to reproduce such a behaviour. A high-level design is also given to make the proposed model reconstructible, and the successful operation of the model is demonstrated on two detailed examples.

Deep learning has achieved tremendous success in many fields when leveraging well-labeled large-scale datasets. However, for many challenging tasks, data are scarce, or labels are difficult to acquire. We propose different approaches to tackle such tasks. We consider two settings: 1) Deep unsupervised models for Pattern Demixing and for Traveling Salesman Problems (TSP). 1.1) We propose a general unsupervised framework (deep reasoning networks, DRNets) for demixing and inferring crystal structure. DRNets seamlessly integrate reasoning about prior scientific knowledge into neural network optimization using an interpretable latent space. As a result, DRNets require only modest amounts of (unlabeled) data. DRNets reach super-human performance for crystal-structure phase mapping, a core, long-standing challenge in materials science, solving the previously unsolved Bi–Cu–V oxide phase diagram and aiding in the discovery of solar-fuels materials. DRNets are a general framework, and we illustrate that by also demonstrating their effectiveness to demix two 9x9 complete overlapping handwritten Sudokus. We boost the performance of the state-of-the-art framework DRNets with curriculum learning with restart DRNets (CLR-DRNets) on a visual Sudoku task and a visual Sudoku demixing task. This is an extension of our early work on DRNets for unsupervised pattern demixing. Furthermore, we consider combinatorial optimization problems, where well-labeled datasets are hard to acquire. 1.2) We use zero training overhead portfolios (ZTop) to boost the performance of deep models for combinatorial optimization problems by leveraging the randomness of the optimization process. 1.3) We use unsupervised learning combined with local search to solve a large-scale Traveling Salesman Problem (TSP). Apart from the deep unsupervised models, we also study the unsupervised optimization model for the dam portfolio selection problem. 2) We have developed methods to harness AI to reduce the adverse impacts of the hydropower dam proliferation in the Amazon River basin on people and nature (e.g., river fragmentation, which affects fish migration; transportation; sediment production; increase of greenhouse gases emissions; displacement of indigenous populations, etc.). More specifically, we propose methods to efficiently approximate high-dimensional Pareto frontiers for tree-structured networks using expansion and compression methods.

This dissertation examines the nature of proving-related tasks included in textbooks from Korea, the United States, and Singapore. As an international comparative study, the nature of proving-related activities is analyzed in regard to how frequently proving-related tasks appear in textbook, what kinds of proving-related activities are expected from student, what prompts are used to solicit which of proving-related activities from student, and how consistently prompts are used. In this study, proving-related activities include identifying patterns, formulating generalized statements or conjectures, and justifying conjectures. The dissertation is composed of three distinct articles which are in equivalent form of articles seen in peer-reviewed journals. Using Herbel-Eisenmann’s (2007) framework, the first article examines the voice of the text to address the research questions: What prompts are used to solicit which of proving-related activities from student and how consistently prompts are used. The second article reports on lessons learned from the international comparative study on the nature of proving-related tasks discussing issues and challenges of decision making in the process of conducting the study. The last article discusses implications of the dissertation for teachers to implement proving-related tasks with attention to what teachers need to know and what might be helpful to support students when engaging in proving-related tasks. Findings from the dissertation indicate that textbooks tend to contribute a small fraction of the volume to proving-related tasks, that the vast majority of expected proving-related activities is testing the truth of a conjecture with the exception of the Korean textbook, that the select textbook series commonly use prompts, explain, tell, and what with relatively higher frequency than others, and that prompts are used with consistency indicating that prompts are associated with particular kinds of proving-related activities.

People have limited introspective access to many mental processes. In this dissertation, I test whether people can report self-awareness of motivational influences. Specifically, I investigate the phenomenon of motivated reasoning, where people’s decisions and evaluations of information are biased in favor or one’s attitudes, beliefs, and identities. In six empirical studies, people show limited ability to report the extent that their prior beliefs influence their reasoning. In Studies 1 – 3, I demonstrate that people update their beliefs consistent with their motivations, but do not report awareness of this influence consistently, or in a manner that aligns with a peer’s perception of them. In Studies 4-5, I show similar effects with a simulated political ballot counting task. Awareness was generally underreported relative to peers, but those who were greatly biased tended to report more perceived influence of political beliefs. Finally, in Study 6, I introduce an incentive-compatible reward task that maximizes the likelihood of honest responses. In this study, people evaluate the political ballots honestly, eliminating a possible counterexplanation about normative views of motivated evaluation. In sum, this dissertation advances our understanding of metacognitive processes related to motivated reasoning. I find that while people engage in motivated reasoning, they often underreport awareness of how their desires are influencing their decisions.

The study examined moral reasoning amongst Iranian men and women regarding gender inequality that is sanctioned by religious and legal laws in Iran. Through interviews with 32 recent Iranian immigrants to Australia aged 22 to 44 years, half of whom were male and half female, the study explored moral judgments on gender inequality and restrictions on women’s rights in familial contexts in Iran. The study investigated both the evaluations and justifications provided by participants regarding: (i) specific practices of gender inequality (by men) sanctioned by religious and legal laws, and (ii) transgressions (by women) of religious and legal laws that sanction gender inequality. The practices addressed involved restrictions on females of the rights to travel, work, maintain ownership of financial assets after divorce, and choose whether to wear a hijab. The study also investigated participant evaluations and justifications regarding the mutability of existing laws that sanction gender inequality, by rule abolition, authority dictates from religious and governmental leaders and God, and common practice in countries outside of Iran.It was found that Iranian men and women engage in critical evaluation of existing social arrangements pertaining to gender inequality, specifically practices that are religiously and legally sanctioned in Iran. The majority of participants of both sexes evaluated practices that enact gender inequality unacceptable. Most participants also approved of the dissolution of such laws, the prospect of support from religious and legal authority figures for the abolition of such practices, and the practices of other countries which denote gender equality in these areas. Reasoning provided by participants for their evaluations were predominantly moral justifications pertaining to human rights, equality and the well-being of persons. Concern with relationship harmony and honesty were also common concerns. Participants demonstrated a plurality of moral, conventional and personal concerns and engaged in coordination of concerns within and amongst these domains.While most participants evaluated practices that sanction gender inequality as unacceptable, and also condoned the cessation of such laws and practices, there were variegated evaluations regarding the acceptability of transgressions of existing gender restrictive laws. Justifications amongst male participants largely involved concern for honesty and disapproval of the deception involved in violating existing laws. Female participants also demonstrated concern with honesty in their justifications, but this was accompanied by concerns with the wellbeing of females who transgress existing laws and the imperative of challenging unjust social practices.

Human thinking relies on the ability to process relations between individuals, kinds, properties, and other relations. Explicit relation processing has been invoked to explain our ability to grasp ‘cross-domain’ analogies between situations whose similarity is driven by a shared relational structure, rather than any similarities among the relata populating each analog (e.g., between the solar system and an atom) and ‘cross-modal’ analogies between relata spanning different sensory modalities (e.g., sound and vision); generalize relational schemas, categories whose members share some canonical structure (e.g., things consisting of elements converging on a central location); or abstract rule-like sequences (e.g., an A-B-A sequence of syllables). At the same time, explicit relation processing requires that a reasoner simultaneously represent a set of individual relata and then bind them to a relational structure. This ability is slow to develop in childhood, and even among adults, it places high demands on working memory. Relations thus raise a tension between the expressive advantage they confer and the cognitive cost they impose, and this tension suggests that the human ability for relation processing does not imply its inevitable use, especially when less-demanding alternatives are available.The present dissertation confronts this tension and attempts to specify the computational mechanisms by which human reasoners process relations in the face of their cognitive demands. It presents novel research that clarifies how humans make use of explicitly relational thought instead of nonrelational alternatives. In Chapter 1, I start by examining the role of relations in comparison. Cognitive scientists researching analogy have generalized the processes governing analogical comparison, and the representations of relational structure that it operates on, to all comparison. A consequence of this view is that human reasoners make use of relations whenever they make any comparison. I test this claim and show that whereas relations do tend to underlie comparisons aimed at assessing similarity, they tend not to underlie assessments of difference. This asymmetry is consistent with recent accounts of a representational asymmetry between the relations same and different, in which different is represented as a negation of the relation same (i.e., different is represented as not-same). When judging difference, human reasoners are more likely to shift to simpler non-relational representations to ease working memory capacity. Having lent support to the claim that explicit similarity judgments do tend to incorporate relational information, I extend this claim to implicit similarity comparisons made during recognition in Chapter 2. When an agent attempts to assess whether they recognize a given stimulus, they make an implicit comparison between the perceptually available stimulus to a representation in memory. I show that when agents make this comparison, they tend to incorporate relational information; indeed, relations are available to serve as cues in human recognition memory. Finally, in Chapter 3, I examine a cognitive process, generative analogical inference, that integrates human reasoning and memory, investigated relatively independently in Chapters 1 and 2 respectively. I introduce a computational model of this process, in which a reasoner uses their prior knowledge of some familiar source domain to elaborate on some less-familiar target domain. This new model can reproduce human-like inference whether the relational structure that constrains inference is prespecified in the model input, as required by existing inference models, or are unspecified, unlike existing models. Across three simulations, I use comparisons between this model and a non-relational control model to clarify what relations contribute to the inference process. Specifically, relations promote far generalization across semantically distance analogs. My dissertation instantiates a framework for studying human relation processing that acknowledges both the expressive advantages that relations provide and the cognitive costs imposed by processing them.

This thesis examines the ethical implications of artificial intelligence through the lens of physical safety. It scrutinizes the various ways AI systems can instigate unsafe behavior in users, emphasizing the underexplored domain of covertly unsafe language. To improve the safety-related reasoning ability of large language models, we propose FARM to systematically generate rationales attributed to credible sources for physical safety scenarios. Lastly, we close with a broader discussion on AI transparency, delineating its differing research threads and associated considerations such as safety, and call toward a human-centered approach to evaluate future research, centering on the foundational debate of whether humans should trust intelligent systems.

A major goal of cognitive neuroscience is to understand the link between anatomical brain structure and cognition. In the work that follows, I characterize patterns of cortical folding in the prefrontal cortex and its relation to reasoning skill. A central thread running through this work is that by exploring the relationship between variability in anatomy and cognition, we can establish a link among cortical structure, function, and behavior.In Chapter 1, I test a classic hypothesis and ask whether small, variable structures known as tertiary sulci have cognitive relevance. By combining precise neuroanatomical techniques with a modern data-driven analysis pipeline, I show that these structures can be reliably identified in individuals across childhood and adolescence and that the depth of a subset of tertiary sulci explains additional variance in reasoning not accounted for by age. In Chapter 2, I further examine variability in the presence and prominence of cognitively relevant tertiary sulci in prefrontal cortex. I show that the variability in the presence of these structures is associated with variability in behavior and that these sulcal components have discernable functional connectivity profiles that relate to reasoning skill. Finally, in Chapter 3, I take a modern comparative anatomy approach and show that while these structures exist in chimpanzees, they are notably smaller, shallower, and more variable. This work lays out a methodological approach to compare structure, function, and behavior across individuals, age groups, and species. I show that sulci, particularly tertiary sulci, can be reliably identified across individuals and that variability in the presence and morphology of these structures can be linked to variability in cognition. The goal of these studies is not to claim that a single sulcus predicts a specific cognitive process, but rather to demonstrate that sulci are useful landmarks that can provide valuable insights into the relationship among cortical structure, function, and cognition.

This project is a critique of the concept of consensus and its relation to secular moral reasoning. Proponents of public deliberation argue that achieving consensus is crucial to informing moral norms in secular pluralist societies. Without a transcendental basis for morality, ascribing authority to moral norms requires a process of deliberation. Many bioethicists are concerned with formulating ways to ensure discourse is tolerant, non-coercive, mutually respectful, and grounded in intersubjective understanding. The problem is that secular discourse is fraught with varying conceptions of human rights, ethical principles, and what constitutes a morally authoritative consensus. Bioethicists acknowledge the tyranny of the majority problem, but secularism lacks a sufficient rationale to identify when a majority is wrong. Since competing visions of the good comprise bioethics and consensus does not necessarily indicate moral truth, moral uncertainty is the logical result of secular pluralism.Some moral scientists argue that science can inform moral norms, but a careful reading of their work suggests that what is being espoused is moral nihilism. From determinism to deep pragmatism, many scientists are inadvertently supporting a view of reality that obliterates the possibility of values. In secular pluralist societies, consensus is required to establish basic norms, but no account of consensus can indicate when moral truth is known. Consensus is necessary to fulfill the visions of moral scientists, but such scientists implicitly endorse nihilism. What secularists are discovering – by their own reasoning – is that moral truth is elusive, science cannot inform human values, and bioethical dilemmas are incapable of being resolved. The conclusion to this project offers an Engelhardtian solution. Not only is the principle of permission the only viable basis for secular pluralism – the principle coincides with moral scientists’ own account of human nature.

Variability in the brain’s functional network connectivity is associated with differences in cognition. The degree to which brain networks flexibly reconfigure, or alternatively remain stable, can differ across regions of cortex, across time, and across individuals. The goal of this dissertation was to investigate how the brain’s functional network architecture is reconfigured to support abstract reasoning and rule learning. I proposed that flexibility within frontoparietal cortex, combined with a stable network core, is beneficial for effective reasoning and rule learning. Experiment One investigated the activation patterns and dynamic community structure of brain networks associated with shifting task demands during abstract reasoning. Twenty-seven subjects underwent fMRI scanning during resting state and during a subsequent abstract reasoning task. When quantifying network reconfiguration between resting and task states, I found a stable system within default and somatomotor networks alongside a more flexible frontoparietal control network. The results motivated a novel understanding of how the brain performs reasoning tasks: an underlying stable functional network acts as a cognitive control mechanism, priming task-active nodes within frontoparietal cortex to variably activate for unique task conditions. Experiment Two used a dynamic network analysis to identify changes in functional brain networks that were associated with context-dependent rule learning. During fMRI scanning, twenty-nine naïve subjects were challenged to learn a set of context-dependent rules. Successful learners showed greater stability in ventral attention and somatomotor regions, increased assortative mixing of cognitive control regions as rules were learned, and greater segregation of attention networks throughout the entire task. The results suggested that a stable ventral attention network and a flexible frontoparietal control network support sustained attention and the formation of rule representations. In Experiment Three, I carried out a separate analysis of data from Experiment 2 to characterize the functional connectivity patterns with the hippocampus that emerged during successful rule learning. The results demonstrated that the hippocampal head became increasingly functionally connected to the lateral frontal pole and caudate in successful learners. Additionally, the entire hippocampus exhibited decreased functional connectivity with the mid-cingulate and precuneus in successful learners. These three experiments demonstrated that stable functional connectivity in somatomotor and ventral attention networks, combined with flexible reconfiguration of frontoparietal cortex, is advantageous for successful rule learning and abstract reasoning. Altogether, this dissertation demonstrated that individual differences in dynamic functional connectivity are associated with learning, and that stability of brain networks across time and tasks supports higher order cognition.

In the dissertation I explore the role of inconsistency in human reasoning as a way into broader questions about the structure of reasoning processes. In each of the next three chapters I defend the idea that certain types of inconsistency can be rational against a different pressure. These defenses fill out a picture of reasoning on which consistency plays an important role in managing specific modes of reasoning and in how we justify our beliefs. But inconsistency can arise in a well-functioning system, since it is in general better to draw on an inconsistent belief state than to force the system as a whole to be consistent or to only draw on a consistent subset of the total belief state in any given reasoning situation.We can helpfully divide discomfort with inconsistent belief states into different types: worries about how an agent could come to be in an inconsistent belief states and worries about how an agent could (or should) function once they are in an inconsistent belief state. The latter worry is addressed in chapter 2, where I address strengths and weaknesses in the fragmentation view. The former worry can be further divided into concerns about how an agent could end up in an inconsistent belief state full stop and how they could do so rationally. I discuss how a rational agent could end up in an inconsistent belief set in chapter 3, defending myself from a possible objection from Niko Kolodny (using a view of rationality as correct response to different pressures) along the way. I argue for the existence of inconsistent beliefs full stop in chapter 4; inconsistent beliefs are possible on an interpretationist framework despite some Davidsonian worries to the contrary. In chapter 5 I pause to explore what these results mean for our understanding of human reasoning, including: the value of consistency, integrating different methods of reasoning, social reconstructions of reasoning, and how we should be evaluating agents.

Artificial Intelligence (AI) aims to develop intelligent systems, inspired in part by human intelligence. AI systems are now performing at human and even superhuman levels on a range of tasks such as image identification, face, and speech recognition. These major recent AI achievements have been driven largely by advances in supervised deep learning, which requires large labeled datasets to supervise model training. In contrast, humans often solve complex tasks using far fewer data by amplifying intuitive pattern recognition with meticulous reasoning that uses prior knowledge, a hybrid strategy that is challenging for machines to emulate. In this thesis, we focus on integrating prior knowledge reasoning into deep learning via an interpretable latent space. When the prior knowledge is sufficiently rich, as is common in many scientific applications, we can supersede traditional example-based supervised learning and compensate for a dearth of labeled data by exploiting prior knowledge and magnifying it with logical and constraint reasoning seamlessly integrated into neural network optimization. We first illustrate this idea in the context of supervised learning on multi-label classification --- in particular, on joint species distribution modeling, where we propose the Deep Multivariate Probit Model (DMVP) to uncover species interactions and habitat associations via the interpretable latent space for the entire North American avifauna and accelerate the learning by an order of magnitude using prior knowledge of the low-rank structure of species interaction. Next, we demonstrate the capability of this approach on unsupervised tasks with rich prior knowledge via a novel framework called Deep Reasoning Networks (DRNets). For variants of visual Sudoku games, DRNets outperforms supervised state-of-the-art methods in an unsupervised manner. In materials science, DRNets surpasses previous approaches and the capability of experts on crystal-structure phase mapping, unraveling the Bi-Cu-V oxide phase diagram, and enabling the discovery of solar-fuels materials. In the future, we plan to further develop, adapt, and customize Deep Reasoning Networks to effectively solve a variety of tasks that require a combination of pattern recognition, reasoning, and learning, which are pervasive in science and other application domains.

The central topic of the thesis is the relationship between logic and the cognitive psychology of reasoning. This topic is treated in large part through a detailed examination of the recent work of P. N. Johnson-Laird, who has elaborated a widely-read and influential theory in the field. The thesis is divided into two parts, of which the first is a more general and philosophical coverage of some of the most central issues to be faced in relating psychology to logic, while the second draws upon this as introductory material for a critique of Johnson-Laird's `Mental Model' theory, particularly as it applies to syllogistic reasoning. An approach similar to Johnson-Laird's is taken to cognitive psychology, which centrally involves the notion of computation. On this view, a cognitive model presupposes an algorithm which can be seen as specifying the behaviour of a system in ideal conditions. Such behaviour is closely related to the notion of `competence' in reasoning, and this in turn is often described in terms of logic. Insofar as a logic is taken to specify the competence of reasoners in some domain, it forms a set of conditions on the 'input-output' behaviour of the system, to be accounted for by the algorithm. Cognitive models, however, must also be subjected to empirical test, and indeed are commonly built in a highly empirical manner. A strain can therefore develop between the empirical and the logical pressures on a theory of reasoning. Cognitive theories thus become entangled in a web of recently much-discussed issues concerning the rationality of human reasoners and the justification of a logic as a normative system. There has been an increased interest in the view that logic is subject to revision and development, in which there is a recognised place for the influence of psychological investigation. It is held, in this thesis, that logic and psychology are revealed by these considerations to be interdetermining in interesting ways, under the general a priori requirement that people are in an important and particular sense rational. Johnson-Laird's theory is a paradigm case of the sort of cognitive theory dealt with here. It is especially significant in view of the strong claims he makes about its relation to logic, and the role the latter plays in its justification and in its interpretation. The theory is claimed to be revealing about fundamental issues in semantics, and the nature of rationality. These claims are examined in detail, and several crucial ones refuted. Johnson- Laird's models are found to be wanting in the level of empirical support provided, and in their ability to found the considerable structure of explanation they are required to bear. They fail, most importantly, to be distinguishable from certain other kinds of models, at a level of theory where the putative differences are critical. The conclusion to be drawn is that the difficulties in this field are not yet properly appreciated. Psychological explantion requires a complexity which is hard to reconcile with the clarity and simplicity required for logical insights.

The role of "believes" is identified by finding a unique and unparaphraseable use of the term that is integral to the structure of important forms of language. This role can only be roughly indicated here. Roughly, "believes" identifies someone's reasons for doing or thinking something or what he counts as success in certain activities, without committing the speaker to accepting them; and so it allows the possibility of mistake. Standard objections to dispositional accounts are reviewed and found unconvincing. The central difficulty is found to be that dispositional accounts give no satisfactory explanation of the fact what it is (part of) an explanation of action to cite the agent's beliefs. It is argued that causal accounts have the same deficiency, that the role or point of "believes" in language is not satisfactorily explained. It is concluded that neglect of this central question vitlates even the real strengths of these accounts. The relation between different questions about belief is explored, and analogous questions about other cases, especially artefacts, are considered. Explanations of actions are discussed to bring out the role of "believes." This appears in explaining the relation between the "point or purpose" of an action and the actions of which it consists at a lower level of description. "Believes" identifies what governs this relation while allowing the possibility of mistake. Ryle's claims about thinking are considered, and an interpretation of the notion of an adverbial verb is suggested. The central difficulty concerns the cases of the thinking of le Penseur or of Eucild. Although it seems important that we should be able to say what the constituents of thinking are here, we cannot do so, since the relevant vocabulary is not adapted to giving the constituents of anything. These activities have their own "objects" as well as their own criteria of success. Activities, "objects", and criteria of success are bound up together, and do not connect in any simple way with the vocabulary that we try to connect them with in asking what their constituents are. This far from unique. Music provides another relevant example. The last chapters examine various contexts in which "believes" is important. Topics considered are reasons, evaluations, assertions, belief and the will, internationality, and speech-act accounts. In each case particular doctrines are critically examined.

The immediate object is a determinate resolution of Aristotle's position in Nicomachean Ethics 7.3. over the Socratic problem and in particular the possibility of last-ditch akrasia. My approach seeks interpretational constraints and illumination from considering the argument as a structured whole. Moreover, the passage is a point on which larger issues in Aristotle's philosophies of mind, action, and morals converge: the elucidation offered attempts both to frame and throw light on these. Chapter One makes preliminary moves on several fronts. Firstly it looks briefly at Aristotle's position over last-ditch akrasia in De Anima and Eudemian Ethics. Then it outlines the problems of Nicomachean Ethics 7.3. and classifies previous lines of solution. Thirdly an intuitive picture is given of Aristotle's method and basic stance. Finally some contrasts are drawn between Aristotle's and modern approaches to akrasia. Chapters Two to Five discuss the four Sections of Nicomachean Ethics 7.3's argument. Chapters Two and Three take Section 1 and 2 together and consider two major problems of interpretation. Chapter Two asks whether these Sections concern akrasia at all, and, if so, how. I argue that their concern is direct, general (i.e. not confined to some akratic species), and inclusive (i.e. embracing non-akratic phenomena). Chapter Three asks about the interpretation of "exercising knowledge". Firstly the results of Chapter Two are defended; an aporetic discussion of this difficult issue then follows. Chapter Four examines Section 3. After analysing its structure, it distinguishes three principal issues. I argue firstly that Section 3, like 1 and 2, concerns akrasia directly, generally and inclusively; secondly that the knowledge that the akratic is temporally unable to use (that is 'tied') is his (universal) knowledge of what is worthwhile; thirdly that this failure involves a cognitive failure (I suggest a distortion of the agent's situational appreciation) - and not, as some scholars have recently urged, merely a motivational failure. Chapter Five, perforce selective, tackles firstly various problems of Section 4's argumentative structure, and then the interpretation of 1147a26-31 (the 'normal case'). Finally 1147a31-5 (the 'akratic case') is examined and a case argued for its offering two syllogisms but only one practical syllogism.

This dissertation consists of four independent papers that address topics concerning the nature and function of reasoning. Together, they draw on a wide range of paleontological, archaeological, and ethnographic evidence, to constrain evolutionary theorizing about reasoning. Further, they draw on empirical results from literatures on persuasion and reasoning, to evaluate hypotheses about the characteristic features of reasoning. On these foundations, I argue against the increasingly influential view that the capacity for reasoning evolved in a social context, for public practices of persuasion and self-justification. The principal thesis of this dissertation is that the capacity for reasoning evolved for the private practice of generating true beliefs from rule-like cultural knowledge transmitted through stories, to solve novel problems in the natural environment.

This qualitative multiple case study examined the ways in which undergraduate and graduate students coordinate theory and evidence on the controversial topic of vitamin supplement use when faced with evidence that supports and evidence that contradicts their existing beliefs, as well as the roles of personal epistemology and prior knowledge in this process. Coordination of theory and evidence is an essential skill in both scientific and everyday reasoning, as it allows individuals to accurately evaluate conflicting information to arrive at an informed decision about important issues. Participants completed an online questionnaire in which they read three articles of differing stances on the topic of vitamin use and asked to describe the position and evidence used within each article and to state whether they agreed or disagreed with the article’s core position. Participants’ personal epistemologies and prior knowledge on the topic were also assessed. A majority of participants were largely successful in being able to articulate their existing theories as well as alternative theories, identify theory and evidence within the articles, and to integrate the information they read to develop, refine, and in some cases modify their theories. Many participants in this study held evaluativist epistemological beliefs which may have contributed to their success with these components of theory and evidence coordination. However, participants often struggled to fully differentiate between theory and evidence and rarely discussed the quantity and quality of sources cited as a factor when evaluating the articles. Participants who held at least some accurate knowledge about vitamins and nutrition were more likely to change their initial theories at the end of the task and to indicate a desire to do more personal research on the topic compared to those who lacked knowledge or held misconceptions.

This dissertation considers what it means to think differently, using naturalistic verbal evidence. This problem is inspired by a gap within the Wisdom of the Crowd (WoC) literature, but relevant to the study of language processes, mental models, and the vast emerging resource of social media data. I propose a methodological framework to characterize diversity of thought through the quantification of social media data. Four stages of research considered: a) the properties of a sample domain, b) how to identify and select diagnostic content using classification methods, c) how to quantify qualitative content in order to categorize and compare individuals, and d) how to assess the relative merits and challenges of content classification methods, including whether differences in thought actually affect outcomes. The emphasis is on pervasive issues pertinent the analysis of unstructured verbal data, rather than the specific, albeit largely successful solutions explored. Such issues were identified when defining and applying the methodological framework, and generally indicate the influence of sample domain on process measures, success at higher levels of abstraction, and a lack of continuity between all levels of analysis.

We formulate a domain-independent language for representing stochastic tasks, and show how this representation can be synthesized from few training observations. In one experiment, we study how a physical robot may learn to fold clothes from a small number of visual observations, in contrast to big data techniques. Under the same framework, we also show how a virtual chat-bot may learn dialogue policies from few example transcripts, resulting in an interpretable dialogue model, outperforming current statistical techniques. Central to both examples in the concept of utility, why it's essential for generalizability, and how to learn it from small data.