This thesis has been submitted in fulfilment of the requirements for a postgraduate degree (e.g. PhD, MPhil, DClinPsychol) at the University of Edinburgh. Please note the following terms and conditions of use: • This work is protected by copyright and other intellectual property rights, which are retained by the thesis author, unless otherwise stated. • A copy can be downloaded for personal non-commercial research or study, without prior permission or charge. • This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the author. • The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the author. • When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given. Concept Learning Challenged Richard St¨ockle-Schobel PhD (Philosophy) The University of Edinburgh 2013 Abstract In my thesis, I argue that the philosophical and psychological study of concept-learning mechanisms has failed to take the diversity of learning mechanisms into account, and that consequently researchers should embrace a new way of thinking about concept learning: `concept learning' as a class of psychological mechanisms is not a natural kind lending itself to unified study and should be eliminated. To arrive at this, I discuss several concept-learning models that attempt to overcome Jerry Fodor's challenge and base my judgment on the plurality of feasible concept-learning mechanisms and on criteria for theoretical notions from the philosophy of science. Chapter 1 serves as an introduction to the topic `concept learning' and highlights its importance as a research topic in the study of the mind. I argue that a mechanistic understanding of the shape of concept learning is best suited to explain the phenomena, in line with the recent resurgence of mechanism-based explanation in the philosophy of mind. As the main challenge to the idea that concepts can be learnt, I proceed to set up Fodor's challenge for concept learning in Chapter 2. This challenge is the idea that concepts cannot be learnt given the logically possible mechanisms of concept learning. I lay out the argumentative structure and background assumptions that support Fodor's argument, and propose to scrutinise his empirically based premise most closely in my thesis: this empirically based premise is that the only possible mechanism of concept learning is the process of forming and testing hypotheses. As replies to Fodor's challenge, I discuss Perceptual Learning (R. Goldstone), Per- ceptual Meaning Analysis (J. Mandler), Quinean Bootstrapping (S. Carey), pattern- governed learning (W. Sellars), joint-attentional learning (M. Tomasello), and the Syndrome-Based Sustaining Mechanism Model (E. Margolis and S. Laurence). I argue that almost every mechanism I discuss has some leverage against Fodors argument, suggesting that there may be a wide variety of non-hypothesis-based concept-learning mechanisms. The final chapter of my thesis, Chapter 7, takes a step back and reviews the fate of the notion of concept learning in light of the diverse set of learning mechanisms brought up in my thesis. My first and main worry is that it is questionable whether the previously discussed mechanisms of concept learning share many scientifically relevant properties that would justify seeing them as instances of the natural kind `concept learning mechanism'. I argue that the substantiation of this worry would necessitate the elimination of `concept learning' and `concept-learning mechanism' as terms of the cognitive sciences. The chapter lays out the argumentative structure on which Concept Learning Eliminativism (CLE) rests, along with a discussion of questions about natural kinds and pragmatics in theory construction. This is inspired by Edouard Machery's argument for the elimination of `concept', but independent of Machery's own project. With this in place, I go on to give a conclusive argument that supports CLE, based on the claims that `concept learning' is not a natural kind and that there are pragmatic advantages to eliminating `concept learning'. In this final chapter, I also raise pragmatic considerations that support the argument for CLE, and propose new research directions that could profit from the eliminativist position. Contents Declaration v Note on publications vi Acknowledgements vii 1 Introduction 1 1.1 The tale of the aardvark . .2 1.2 Challenging concept learning . .2 1.3 Mapping the territory . .3 1.3.1 What is a concept? . .3 1.3.2 Desiderata for a theory of concepts . .4 1.3.3 What is concept learning? . .6 1.3.4 Why should we look for mechanisms? . .7 1.4 The structure of the thesis . .7 2 Fodor's Challenge for theories of concept learning 10 2.1 Introduction . 11 2.2 Fodor's paradox . 11 2.3 Innateness in Empiricist and Rationalist theories . 13 2.4 A more detailed look at the steps of Fodor's argument for the paradox . 15 2.4.1 (4) The Hypothesis-Formation-and-Confirmation Model . 15 2.4.2 (7)+(8) Spelling out the paradox . 16 2.4.3 (3) The representational theory of mind (RTM) . 16 2.4.4 (9) The anti-learning conclusion . 19 2.5 The doorknob/DOORKNOB problem . 21 2.5.1 The metaphysical solution . 22 2.5.2 The neurological solution . 23 2.5.3 The upshot of Fodor's position . 26 2.6 Fodor's Challenge . 27 i 3 Developmental accounts of concept learning 29 3.1 The challenge for developmental psychology . 30 3.2 The `Evidence for Prelinguistic Concept Learning' hypothesis . 30 3.2.1 The dominant research paradigms in early infancy . 31 3.2.2 The argument against EPCL . 34 3.3 Jean Mandler . 40 3.3.1 Perceptual Meaning Analysis . 42 3.3.2 Empirical evidence for PMA . 43 3.3.3 Problems for PMA . 44 3.4 Susan Carey . 48 3.4.1 Foundations of Quinian Bootstrapping . 49 3.4.2 Analysis of Quinian Bootstrapping . 51 3.4.3 Evidence for Quinian Bootstrapping . 54 3.4.4 Problems for Quinian Bootstrapping . 54 3.5 Summary . 57 4 Perceptual Learning as a concept-learning mechanism 58 4.1 Perceptual Learning and feature-based approaches to concepts . 59 4.2 Fixed-feature approaches and Fodor's paradox . 59 4.3 Perceptual Learning as a reply to Fodor's Challenge . 61 4.3.1 Unitisation . 64 4.3.2 Differentiation . 65 4.4 Concept learning or conceptual change? . 67 4.4.1 Do unitisation and differentiation operate on concepts? . 67 4.4.2 Intermission: Kinds of learning . 68 4.4.3 Features as concepts { a related issue . 70 4.4.4 Landy and Goldstone's reply: the Pask device . 71 4.4.5 Intermediate conclusion . 72 4.5 The notions `feature' and `stimulus' . 72 4.5.1 The Compatibility criterion . 72 4.5.2 The proximity of stimuli . 73 4.5.3 Feature detectors and maturation . 75 4.5.4 Physical information . 75 4.6 Perspectives for an amendment of the Perceptual Learning approach . 76 4.6.1 Three kinds of percept-concept interactions in Perceptual Learning 76 4.6.2 Blurring a boundary: brute-causal and rational acquisition . 78 4.7 Conclusion . 81 5 Sellarsian concept learning 83 5.1 Introduction . 84 5.2 The Augustinian Conception . 84 ii 5.3 Sellars's argument against AC { a reconstruction . 87 5.3.1 The Awareness challenge . 88 5.3.2 The Holist challenge . 89 5.3.3 The Relational challenge . 90 5.4 Sellarsian concept learning . 92 5.4.1 Logical spaces . 92 5.4.2 Learning without `knowing the rules' . 95 5.4.3 The advantages and the problem of Sellars-type learning . 97 5.5 Gauker's view: imagistic cognition before conceptual thought . 99 5.5.1 Categorisation tasks in early infancy . 100 5.5.2 Are similarity judgments conceptual? . 101 5.6 Developmental evidence for prelinguistic concepts . 103 5.6.1 Person, agent, gaze-following . 103 5.6.2 Social referencing . 104 5.6.3 Pointing . 105 5.6.4 A cut-off point for speaking of concepts? . 106 5.7 A second kind of social concept learning mechanism: Tomasello's Social- Pragmatic Theory of Word Learning . 107 5.7.1 The Social-Pragmatic Theory of Word Learning . 108 5.7.2 Examining the fit of SPT and Sellars's model . 109 5.7.3 Standing up against Fodor's Challenge . 109 5.8 Conclusion . 110 6 Learning concepts with sustaining mechanisms 111 6.1 Introduction . 112 6.2 Margolis and Laurence's response to Fodor . 113 6.2.1 The simple Fodor puzzle . 113 6.2.2 Arguments for complex concept learning . 113 6.2.3 The argument for primitive concept learning . 114 6.2.4 The syndrome-based sustaining mechanism model . 117 6.2.5 Things to like about the sustaining mechanism model . 121 6.2.6 Problems for the SBSM . 123 6.3 Criteria for concept learning . 126 6.4 Between brute-causal and rational acquisition . 130 6.4.1 The HF model and SBSM learning . 131 6.4.2 The Pask device's learning . 131 6.4.3 Could triggering qualify as learning by this measure? . 134 6.4.4 Lessons from applying the learning criteria . 135 6.5 Conclusion: Defending the rationality of concept acquisition . 136 iii 7 Concept learning eliminated 138 7.1 The state of play . 139 7.2 The claim . 139 7.3 The Horizontal Argument for Scientific Eliminativism . 140 7.3.1 Types of eliminativist arguments . 141 7.3.2 Natural kinds . 144 7.3.3 Pragmatic grounds . 146 7.4 The argument for Concept Learning Eliminativism . 147 7.4.1 Against CL as a natural kind .