Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Time in mind : the cognitive science of temporal representation Viera, Gerardo 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2017_february_viera_gerardo.pdf [ 2.12MB ]
JSON: 24-1.0340322.json
JSON-LD: 24-1.0340322-ld.json
RDF/XML (Pretty): 24-1.0340322-rdf.xml
RDF/JSON: 24-1.0340322-rdf.json
Turtle: 24-1.0340322-turtle.txt
N-Triples: 24-1.0340322-rdf-ntriples.txt
Original Record: 24-1.0340322-source.json
Full Text

Full Text

    Time in Mind:  The Cognitive Science of Temporal Representation  by Gerardo Viera   M.A., The University of Houston, 2009 B.A., New York University, 2006    A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (PHILOSOPHY)  THE UNIVERSITY OF BRITISH COLUMBIA (VANCOUVER)  December 2016   © Gerardo Viera, 2016ii  Abstract Philosophers and cognitive scientists have always been interested in how people come to mentally represent time. Surprisingly though, contemporary philosophers have largely neglected the wealth of relevant empirical research coming from neuroscience, computational psychology, zoology and related fields. My dissertation is meant to remedy this neglect by bringing together major strands in the philosophical and empirical literatures on temporal representation in order to show how both fields can mutually benefit one another.  Chapter 1 describes what I call the temporal coordination problem and provides the needed philosophical background on mental representation that frames the majority of the thesis. Chapter 2 provides a taxonomy of the general approaches to explaining how animals coordinate their behaviors with the temporal structure of the world around them.  Chapter 3 argues that part of the explanation for how animals come to mentally represent time is through the operation of a genuine sense of time centered on the circadian systems that provides animals with information about the approximate time of day.  Chapters 4 and 5 argue for what I call the fragmentary model of temporal perception – temporal perception is not a unified capacity but is importantly fragmented. Chapter 4 argues that the fragmentary model undermines the central debate in the philosophical literature over the mirroring constraint. I conclude that there simply is no single story to be told about how the temporal structure of experience itself relates to the temporal content of experience.   While chapter 4 emphasizes the fragmentary nature of temporal perception, chapter 5 emphasizes the way in which time appears unified in perception and cognition and proposes an explanation of how this apparent unity comes about. Here I highlight how literature more commonly found in the history and philosophy of science on the unitization of measurement actually informs current understanding of the iii  mind. In particular, I argue that the brain comes to integrate the temporal information encoded in various time keeping devices by unitizing time in a manner that parallels how our cultural time keeping practices have unitized time.   Finally, chapter 6 concludes by recapping many of the major conclusions of the thesis.    iv  Preface All content is the original work of the author. Versions of chapter 3 have been presented at the Canadian Philosophical Association, American Philosophical Association, and the Society for Philosophy and Psychology. Versions of chapter 4 have been presented at the Philosophy of Time Society and the Southern Society for Philosophy and Psychology. While it was the aim of the author to reduce redundancies across chapters, the body of this thesis, chapters 2-5, was written as independent research articles, and therefore some redundancies exist. Furthermore, there will be some minor variations in terminology as the terminology chosen in each chapter was shaped by existing literature. However, as each chapter was designed as an independent research paper, any variations in terminology should remain clear within the individual chapters.    v  Table of Contents  Abstract ......................................................................................................................................................... ii Preface ......................................................................................................................................................... iv Table of Contents .......................................................................................................................................... v List of Figures ............................................................................................................................................ viii Acknowledgements ...................................................................................................................................... ix Chapter 1: Introduction ................................................................................................................................. 1 1.1. The representational theory of mind .................................................................................................. 5 1.2. Theories of content .......................................................................................................................... 17 1.3. The temporal coordination problem ................................................................................................. 20 1.4. The mental representation of time ................................................................................................... 26 1.5. Moving forward ............................................................................................................................... 32 Chapter 2: Empirical Timing Models and the Explicit / Implicit Distinction ............................................. 35 2.1. The explicit / implicit distinction in cognitive science .................................................................... 36 2.2. Minimal semantic interpretations and the implicit / explicit distinction .......................................... 41 2.3. Explicit and dedicated timing mechanisms – internal clocks .......................................................... 49 2.4. No dedicating timing mechanism approaches.................................................................................. 58 2.5. Summing up ..................................................................................................................................... 70 Chapter 3: The Sense of Time..................................................................................................................... 71 vi  3.1. The senses in an information processing architecture ...................................................................... 73 3.2. Against the sense of time ................................................................................................................. 95 3.3. Towards a sense of time ................................................................................................................. 100 3.4. Circadian rhythms and internal clocks ........................................................................................... 105 3.5. The semantics of (internal) clocks ................................................................................................. 119 3.6. An information-theoretic account .................................................................................................. 125 3.7. Conclusion ..................................................................................................................................... 131 Chapter 4: The Temporal Structure of Experience: Against the Atomism / Extensionalism Debate ....... 133 4.1. What is temporal experience? ........................................................................................................ 136 4.2. The classic dialectic and the mirroring constraint .......................................................................... 139 4.3. The case for extensionalism: introspection .................................................................................... 145 4.4. The case for atomism ..................................................................................................................... 158 4.5. The fragmentary model of temporal perception ............................................................................. 170 4.6. Conclusion ..................................................................................................................................... 196 Chapter 5: The Units of Temporal Perception and the Unity of Time ...................................................... 198 5.1. The (purportedly) unit-free nature of temporal perception ............................................................ 200 5.2. The cultural unitization of time ...................................................................................................... 216 5.3. Returning to Beck and Peacocke ................................................................................................... 226 5.4. The fragmentary model of temporal perception ............................................................................. 227 5.5. Event perception ............................................................................................................................ 230 5.6. The units of temporal perception ................................................................................................... 236 vii  5.7. Conclusion ..................................................................................................................................... 242 Chapter 6: Conclusion ............................................................................................................................... 244 Bibliography ............................................................................................................................................. 251    viii  List of Figures Figure 1: The simplified feedforward architecture. .................................................................................... 80 Figure 2: The complex interactive architecture. ......................................................................................... 82 Figure 3: Common-cause information channel. ........................................................................................ 103 Figure 4: Mirroring and non-mirroring experiences. ................................................................................ 145    ix  Acknowledgements  I would like to begin by thanking my supervisor, Eric Margolis, for the hours that he spent going over my work and advising me in how to tackle this dissertation and how to navigate the world of philosophy in general. I can’t imagine having finished this dissertation without his constant support and guidance. I would also like to thank my committee members, Christopher Mole and Murat Aydede. Their comments, questions, and objections pushed me to significantly improve my work.  The graduate students at UBC deserve quite a bit of acknowledgment as well. Over the years, many of them have had a huge impact on my work. Tyler DesRoches and Jiwon Byun deserve special thanks for all of our conversations over the years. I also want to single out Max Weiss for always being available to talk about the most esoteric philosophical concerns and to help me make sense of my confusions.   A very special thanks goes to Emma Esmaili. Every word in this dissertation was improved by her influence. The dedication and high standards that she set for her own work, and the feedback that she gave me on mine, pushed me to try and do better at every step of the dissertation. Her support and friendship through the years got me through some of the toughest parts of this dissertation and I could not have done it without her.  Parts of chapters 3 and 4 were presented at a number of conferences. I would like to thank my audiences at the Eastern APA, Society for Philosophy and Psychology, Southern Society for Philosophy and Psychology, Canadian Philosophical Association, the Philosophy of Time Society, and the UBC Graduate Colloquia for your questions and comments. I would also like to thank David Sackris, Kathy Fazekas, and Daniel Booth for comments at the APA, Philosophy of Time Society, and CPA. I would also like to thank Raja Rosenhagen, Andre Sant’Anna, and Mark Fortney for comments on my work as part of the Virtual Dissertation Group. x   Finally, I want to thank my family for supporting me throughout every step of this process. 1  Chapter 1: Introduction  Time has an important role in our mental lives. In fact, it has two important roles. The more general of these two roles is that our mental lives exist or unfold in time. We have a series of experiences that follow one another in time. Our beliefs and desires change as time passes. For the more reductively minded among us the point can be put at the level of the biological and neurological processes that underpin the mind. As time progresses, the various metabolic and chemical processes that drive the mind will themselves develop through time. This is all to say that our mental lives have their own complex temporal structure.   The second role that time has in our mental lives is as part of the content of a number of different mental states and processes. From navigating a busy sidewalk, interpreting speech, feeling the pain of an awkward pause in a conversation, listening to music, planning your day, wondering about the future, or even smoothly tying your own shoes, it is vital that we be able to keep track of the timing of events. So, it’s not just the case that our mental lives have their own temporal structure, but much of our mental lives concern the temporal structure of the world around us. Our minds exist in time and they are also directed at or represent time.  That time plays these two roles in our mental lives hasn’t gone unnoticed. In fact, throughout the history of philosophy, from antiquity through current discussions in philosophy and cognitive science, there has been a tendency to connect these two ways in which time is wrapped up with the mind. Something about how the mind exists in time is supposed to explain how the mind comes to be directed at or about time. Whether this is the case will be a recurring theme throughout the dissertation that we’ll return to shortly.  Despite the fact that philosophers and cognitive scientists have been interested in understanding how time relates to the mind, it is surprising how little these two disciplines have interacted with one 2  another. Yes, philosophers will often discuss empirical results, particularly the existence of temporal illusions, when they are writing about temporal experience. But their engagement with the empirical literature tends to end there.1 On the flip side, cognitive scientists may give a cursory acknowledgment of general philosophical issues that arise when they discuss experience or mental representation, but again, there simply isn’t the deep engagement that you would expect given that both literatures are often investigating the very same phenomena.  This thesis aims to alleviate this lack of true engagement between these literatures by bringing together dominant strands in both the philosophical and empirical investigation of the role time has in the mind. The goal is that by bringing together these literatures we will gain a more complete understanding of the way in which the mind comes to acquire its temporal content and what role, if any, the temporal structure of our mental lives has in explaining the acquisition of that content. To put the point differently, it is by bringing together this literature that we’ll gain a better understanding how the mind’s being in time relates to the mind’s being about time.  The main body of this thesis consists of four distinct chapters. Chapter 2 provides an overview of the different empirical approaches to understanding how minds might come to produce behavior that is coordinated with the temporal structure of the events in the environment. In laying out this taxonomy of possible explanatory approaches, I also articulate a form of the implicit / explicit representation distinction that is useful for distinguishing between these various explanatory approaches. While chapter 2 provides some of the background for the thesis, chapters 3, 4, and 5 deal more specifically with particular aspects of how animals come to represent the temporal structure of the world around them.  Chapter 3 focuses on the operation of the circadian system and its role in coordinating behavior with the daily pattern of events. In that chapter I discuss the challenges that people have raised for a                                                      1 Of course there are some notable exceptions to this general rule. For instance, see (Grush 2005, 2006, 2016, Lee 2014a, 2014b, Phillips 2010, 2012; Thompson 2007). 3  sensory based account of temporal representation and argue that those arguments fail. In fact, I further argue, that given a proper understanding of what sensory systems are we have good evidence for thinking that many animals have a genuine sense of time that is in many ways analogous to the classic senses like vision or olfaction, which provides them with information about the approximate time of day.  In chapters 4 and 5 I engage with what is typically picked out when philosophers and cognitive scientists talk about temporal perception or temporal experience. That is, in these chapters I focus on the way that animals come to acquire information from various sensory modalities about time at scales from around 10ms to several seconds. In chapter 4, I argue for what I call the fragmentary model of temporal perception. According to this model, temporal perception is not a single unified phenomenon, but is instead, a motley assortment of various capacities that allow animals to keep track of the temporal structure of their environment. With this model in hand I argue that the central debate in the philosophical literature on temporal perception, the debate over the mirroring constraint, rests on a faulty assumption. There is simply no single answer to whether the temporal content of experience mirrors the temporal structure of experience itself. By looking closely at the various sorts of mechanisms included in the fragmentary model we’ll see that there is simply no single answer to be told as to how the temporal structure of perception relates to the temporal structure that is represented by perception. As a result, the central debate in the philosophical literature on temporal perception rests on a mistake. Temporal perception is not a single unified phenomenon and as a result will not admit of a single answer.  While the emphasis of chapter 4 is the fragmentary nature of temporal perception, chapter 5 focuses on the way in which time is unified in our perceptual and cognitive lives. We perceive the temporal structure of our world as though time is a single, seamless, and unified framework within which individual events occur, despite the fact that the temporal information we acquire from the world is given to us through a variety of sometimes radically distinct time keeping mechanisms and processes. Somehow, the temporal information encoded by these various peripheral time keeping mechanisms needs 4  to be integrated in order to produce this unified temporal order that we are aware of in perception and cognition and that is required to underpin our complex ways of physically interacting with the world. While the empirical work on this topic is still in its infancy, I argue that a fruitful way of understanding how the brain might overcome this integration problem is as a process of unitizing time. The integration problem that the perceptual system faces in many ways parallels the problem that society faced and that motivated the adoption of a system of temporal units for our cultural time keeping practices. By drawing out the parallels we see that the brain has all of the required resources to unitize time in a manner that parallels how we as a society unitized time, and in fact, this unitization would provide the framework for understanding how the brain overcomes its integration problem.  Chapter 3, 4, and 5 were written and designed as independent research papers. However, there are certain themes that run throughout each of the chapters. As mentioned already, one theme that runs throughout the thesis is the question of how the temporal structure of experience, or of our mental life more generally, relates to the temporal content of our mental life. Since, it will be argued, there is no single way in which animals, including humans, come to mentally represent time, there is no a priori reason for believing that once we figure out how we mentally represent some aspect of the temporal structure of our world that we will thereby have an explanation for how we mentally represent other temporal aspects of the world. We have to take on the various components of our capacity to represent time one at a time.  In trying to tease apart the role of the temporal structure of our mental lives in the explanation of the temporal content of mental representation, we will see over and over again appeals to the philosophical literature on mental representation. The question of how the physical stuff that we are made of comes to have states that have semantic properties – i.e. how the physical stuff we are made of comes to realize mental representations that are about the world – is arguably one of the most foundational questions in the philosophy of mind and philosophy of cognitive science. It is right up there with the 5  question of how consciousness can fit into the natural world. The importance of this literature on mental representation cuts two ways. First, by appealing to the literature on mental representation, and the developed theories that have appeared in that literature, we get the resources we need to understand how the mind’s being in time comes to (or fails to) explain the mind’s being about time. Second, by looking at the various sorts of mechanisms that seem to underpin our capacity for temporal representation, we will get a clearer understanding of how empirical, and in particularly neuroscientific, findings can inform long standing philosophical debates. That is, by looking at the relevant scientific literature, we see how scientific theories constrain both our attributions of representational states and our explanations for how representations come to have their contents.  In the remainder of this introduction, I’ll do two things. First, provide some of the background from the philosophical literature on mental representation that will be appealed to throughout the thesis. Second, I will lay out a neutral way of characterizing how the mind can be about or is concerned with the temporal structure of its environment, in order to provide us with an adequate target for later discussion.  1.1. The representational theory of mind  As so many authors have put it, if we know anything, then we know our own minds.2 Yet, despite the fact that we are supposedly so familiar with our own minds, it is mental phenomena that has posed the most difficulty in being integrated into the naturalistic understanding of the world. In particular, there are two aspects of the mind that Descartes pinpointed in his Discourse on Methods (1637/1960) that are thought to cause significant problems for a naturalistic understanding of the mind.                                                       2 Trying to give an exhaustive list of people who’ve said this would be impossible. It’s simply that widespread. However, the classic statement of this sort of claim is typically attributed to Descartes in his Meditations (1641/1993), while the more recent revival of this sort of position can be found in Chalmers (1996). 6  First, there is the problem of qualia or phenomenal consciousness. We interact with the world, we gather information about our environment, and in many cases, there is something that it is like for the subject to undergo this sort of interaction with the environment. The problem of qualia asks for an explanation of why there is anything at all that it is like to go about the world as opposed to there simply being nothing. Why can’t we all just go about our business like unconscious zombies?  The second problem for providing a naturalistic understanding of the mind, and the one that will be the primary focus of this dissertation, is the problem of providing a naturalistic account of intentionality or the aboutness of mental states. Our mental states exhibit intentionality when they are semantically evaluable (e.g. for accuracy, truth, or satisfaction) with regards to the way the world is. Most saliently our beliefs are readily evaluable for truth or falsity depending on whether they accurately depict the way the world is. Our desires are evaluable for whether or not they are satisfied depending on whether what we desire is the case or not.  But not only are our mental states often about the world, but our reasoning and thinking is such that it preserves the semantic relations between our mental states. I can go from the belief that it is raining outside, and the desire that I don’t want to get wet, to the belief that I should grab my rain jacket. The question is then, how can a purely physical system operate in such a way that it respects these semantic / rational relations? Now, many have ultimately come to see these two forms of the mind-body problem as being closely related. Some have seen these problems as so intertwined that solving one will solve (or put us within a short stone’s throw from solving) the other.3 But we can focus on the approach to solving the                                                      3 Two prime examples of this come from opposite directions. Strong representationalists (e.g. Drestke (1995), Lycan (1996), and Tye (1995, 2000)) argue that once we have an understanding of how the mind comes to have representational states, then we will have all we need to understand how a mind could be phenomenally conscious. From the other direction, phenomenal intentionality theorists (e.g. Mendelovici & Bourget (2014)) argue that intentionality will be borne out of an account of phenomenal consciousness. 7  problem of aboutness as the machinery involved in that debate will be of far more importance for our later discussions of temporal representation.  The standard theoretical framework that has dominated the philosophical and scientific literature since the 70’s has been the representational theory of mind (RTM).4 According to RTM, the problem of aboutness and how a physical system could respect the appropriate semantic relations between ideas is to be solved by appealing to a system of internal representations. To avoid making the position sound trivial, it will take a few steps to outline the proposed theoretical framework. The primary domain for which RTM was developed was for our propositional attitudes. According to RTM, states like beliefs, desires, hopes, etc, the so called ‘propositional attitudes’, amount to a subject bearing a relation to some semantic entity, a proposition. The attitudinal component of the phrase ‘propositional attitude’ simply picks out the different ways that someone might be related to a proposition (e.g. that attitude one takes to a proposition will differ between belief and desire). The insight of RTM was how they account for someone’s being related to a proposition. According to RTM, we are related to a proposition through some internal representational state that bears the proposition as its semantic content, and the appropriate attitude is fixed through the functional profile of the particular instantiation of the internal representation. For instance, to have the belief that it’s raining right now, is to have an internal representation whose semantic content is the proposition <it’s raining right now> and for this representation to be functionally related to other states in a manner that is characteristic of belief (for instance, it might lead you to infer other beliefs such as it’s probably cloudy). To desire that it be raining right now5 is to also have an internal representational state whose semantic content is again the proposition <it’s raining right now>, but this time the representation                                                      4 For classic statements of RTM see Field (1978) and Fodor (1975). 5 The grammar of the propositional clause needs to be changed slightly to fit English grammar. However, to be exact I should have said desire that it’s raining right now.  8  will have a distinct functional profile that is typical of desire (for instance, it might drive me to curse the bright, blue, sunny sky I see out my window).  What proponents of RTM need to provide is an account of what it is to have an internal representational state that could take on a proposition as its semantic content and what it is for a physical system to instantiate a representational state of this sort with the appropriate functional profile. Let’s take each of these in turn. As Fodor (1975, 1987, 2008) argued, we cannot account for the full richness of the propositional attitudes if we were to take each proposition that can be entertained by a person as involving its own unique and unanalyzable representational state. The primary reason being that we are finite physical beings, so do not have an unlimited storage space to house a unique individual symbol for each proposition we can entertain. Furthermore, our propositional attitudes exhibit two properties, systematicity and productivity. Systematicity is the property of our propositional attitudes whereby the ability to grasp one proposition is systematically related to the ability to grasp other propositions. For example, being able to entertain the belief that Sally loves dogs has a systematic relation to the ability to entertain other beliefs, such as dogs love Sally. Productivity is the property of our propositional attitudes to entertain an in principle unlimited number of propositions. Once I can entertain the thought it’s not the case that it’s raining I can also entertain the thought it’s not the case that it’s not the case that it’s raining. Simply by reiterating it’s not the case that I can produce an endless number of novel thoughts. However, beyond these trivial cases of novel thoughts produced through reiteration, I can also entertain novel thoughts that I have never confronted before. With the right resources I can perfectly well understand the sentence “the piebald scorpion rode the L-train to Eastern lands” even though this sentence is one that I have never confronted before in my life.  Fodor argued that in order to account for the systematicity and productivity of thought it must be the case that our propositional attitudes involve complex structured representations that can be combined 9  in a way that accords with a compositional semantics. All that we need, according to Fodor’s version of RTM, The Language of Thought (LOT) (1975, 1998, 2008), is that we possess concepts (i.e. the mental representations that are the building blocks of propositional representations) that individually have semantic content, and can be combined in a manner that allows for their individual semantic values to contribute to the semantic value of the larger complex representation.  The major explanatory burden that’s then placed on this form of RTM is to provide an explanation for how some neurally implemented symbol, i.e. state of a neural system, could come to have its semantic content. In other words, if the mental representations of propositions are composed of concepts, and the concepts are the primary bearers of meaning, then we need an account of what determines the content of the individual concepts.  In the literature, the theory that would provide this sort of explanation goes under the title of a theory of content or psychosemantics. Since the aim is to provide a naturalistic understanding of the mind, one that is broadly in line with our other scientific commitments, then whatever theory of content we provide must be one that accords with what we know about science. And specifically, it cannot appeal to other semantically significant psychological states, as any appeals to other semantically significant psychological states would send us down the path towards a problematic regress. Any semantically significant states that are appealed to by a theory of content would themselves require some sort of theory of content that explains how those representational states acquire their content. The very same problem would simply arise again. At some point we need an account of how our mental states come to be about the world that does not rely on there already being mental states that are about the world. So, once we have an account of how the individual symbols of RTM have their content, then it’s a matter of designing a physical system whose transitions respect these semantic contents.  The guiding analogy in the construction of RTM has always been a linguistic one. In particular, RTM is nicely characterized by analogy to formal languages. Formal languages possess a distinct 10  semantics and syntax. The semantics provides a mapping of the individual basic terms of the language to a domain of objects that are the semantic values of the interpreted formal language. This specifies “the meaning” of the symbols. The syntax operates over the non-semantic aspects of the symbols and dictates what inferential rules are allowed within the system and what combinations of the individual symbols compose a well-formed formula or well-formed complex symbol. If it can be shown that the syntax is such that it preserves the semantic properties of the system, e.g. preserves truth, then we will have shown the soundness of the system. Taking the analogy on board, then for RTM to explain how we can not only have semantically evaluable thought but reason in a way that respects these semantic relations, what we need is an account of a physical system’s mechanical operations, i.e. its syntax, for which soundness can be shown. We know from the advent of computer design that systems of this sort are possible, just look at how your calculator’s operations respect rules like addition, but what we need is an account of how the brain (or whatever the relevant physical system is) can implement a system of this sort. Similarly, the attitudes will be given the same gloss as the syntax of the system. We simply need to find a way in which the brain implements a functional architecture that preserves the relevant attitudes.  Now, RTM and specifically Fodor’s LOT are specific proposal for providing a naturalistic account of our propositional attitudes. And they were designed with the propositional attitudes specifically in mind. However, researchers in various fields have extended the general theoretical framework of RTM to explain other aspects of the mind. Perhaps the most successful (or at least most influential) extensions of RTM to other aspects of the mind can be found in the theories of visual perception given by Biederman (1987), Marr (1982), and Treisman (Treisman & Gelade, 1980). According to them, vision itself exhibits something that parallels the productivity of thought in that we can in principle perceive an incredibly large range of novel visual scenes and recognize an incredibly large number of objects in the world.6 They argue, that in order to account for this sort of productivity in                                                      6 I describe this form of productivity in terms of an incredibly large range of visual scenes and not an infinite number since depending on how we individuate visual scenes there may in fact no be an infinite number of possible 11  vision we need to posit the existence of a complex system of internal representations that can be combined in various ways to construct complex visual representations.  However, in both the case of the propositional attitudes and in the case of perception, appeals to representational theories have not gone unchallenged. While the various criticisms differ depending on which critic’s argument we are considering and whether we are considering the propositional attitudes or perception, there is nevertheless a common core to their arguments. The basic strategy is to provide another explanation of the phenomena in question that does not appeal to any representational states in the production of the phenomena in question. Since positing representations places you in a position where you acquire particular explanatory burdens, i.e. needing to explain how those representations acquire their content, then a theory that does not posit representations will be more parsimonious. Further, many attacks on RTM come from theories that are claimed to provide a deeper understanding or more widespread predictive power. For instance, the attack on RTM from dynamical systems theory argues that without any appeal to representational states, and by only using the mathematical tools of dynamical systems theory that are used to understand the operation of a number of physical phenomena, one can provide a deeper understanding of the origin of behavior (e.g. see Chemero (2000), van Gelder & Port (1995)).   Whether or not dynamical systems theory is able to live up to its own hype is still a matter of debate.7 What is important for our discussion is the general approach that the anti-representationalist argument takes. Take some psychological phenomenon and then see if you can provide a non-representationalist explanation of that psychological phenomenon. If you can, then, the inference goes,                                                      scenes. For instance, if we falsely understood visual scenes on analogy to computer displays where there is an array of pixel locations that can take on one of a fixed number of values, then the visual field containing a finite number of pixel locations would only be capable of producing a finite number of distinct visual representations. However, the pull of productivity is still there. The incredibly large number of different visual states is enough to motivate the need for some system of structured representations. 7 Although see Eliasmith (1996) for a criticism that dynamical systems theory fails to provide the sort of revolutionary alternative to RTM that proponents of the view suggest. 12  representationalism is shown to be false. However, again following this sort of reasoning, if, and only if, you cannot provide a non-representationalist explanation, then representationalism is vindicated. Now, in some cases, we may be able to nicely specify some psychological phenomenon and clearly determine whether a non-representationalist theory provides us with an adequate explanation of that theory (i.e. allows us to make reliable predictions). However, there are cases in which it simply isn’t clear which theory provides us with a better (or the true) understanding of some phenomenon as different theories might allow us to make different types of inferences, predictions, or generalizations that all are empirically useful. Let’s take an example from Fodor that he takes as vindicating RTM over a non-representational neuroscientific theory for conceptual thought (1987). Take my behavior that upon speaking with you several weeks ago, I empty my car’s trunk and go to the airport today at a specific time. Now consider the non-representational neuroscientific story of my behavior. As Fodor puts it, the neuroscientific explanation would appeal to various neural mechanisms that cause my muscles to contract in just the right way that I turn on the car, drive it across town, at just the right time of day. Since the specific neural mechanisms that underpin my movement would be different than the neural mechanisms that underpin your movements in a similar situation, given that our brains likely have significant differences between them, then a neuroscientific explanation of why I went to the airport and why you went to the airport would be different. We would require different explanations for each instance of this sort of behavior, and as Fodor puts it, the similarity in the reasons for why I performed an action and why you performed a similar action would be lost.   However, Fodor continues, a representational story makes both your and my behavior intelligible and allows us to make certain generalizations that the purely non-representational neuroscientific explanation does not allow for. If we allow for representational states, then we can explain our airport directed movement as the result of our having a belief that we should go to the airport at such and such a 13  time ready to pick you and your luggage up. While the particular neural implementation of these representational states might differ between individuals, it is nevertheless the case that we may have internal representations that have the same content (in much the same way that ‘the sky is blue’ and ‘the sky is blue’ differ in their implementation, that is, differ in the sorts of marks on the page, but still share a single content). And this content will have a particular relation to actions, i.e. we will believe these propositions. Further, we can generalize from these representational states to conclude that for any individual if they believe that they should be at the airport at such and such a time, then without any outside hindrances they will go to the airport at that time.  This sort of generalization, that appeals to mental states that are individuated or characterized by appeals to their semantic content, allows us to understand something about human behavior that simply could not be captured by the non-representational explanation. If we characterize mental states in non-representational terms, then we will invariably appeal to either the means by which those mental states are realized (i.e. the neural machinery involved) or by their functional profiles (i.e. how those mental states, characterized in non-representational terms, relate to other mental states again characterized in non-representational terms). However, it’s unclear that everyone who has the belief that they should be at the airport at such and such a time will have the very same neural machinery realizing that state, since not only are there interpersonal differences in the neural wiring found in brains at gross anatomical levels but at the finer grain level the same occurs (Friston et al., 2004). Similarly, as Fodor (1987) has famously argued, there will also be functional differences in how these beliefs are realized in individual people. In one person the belief that one should go to the airport will have the functional properties that it cues that person to constantly check the time. In another person, the same representational state will have a distinct functional profile that disposes them to call a taxi cab. While the functional profiles differ, and thus the mental states characterized at the level of functional roles will differ, we nevertheless treat both token mental states as being beliefs that one should be at the airport at such and such a time. However, this 14  characterization of the mental state is one that appeals to the representational properties of the belief and not merely the local physical or functional properties of the belief.   Now Fodor makes the same inference that his critics do. He finds that representationalism allows for certain generalization that he privileges, and which the opposing theory does not capture, therefore, he takes the representational theory as being the account of the mind. But, if our goal in theory choice is to adopt those theories that provide us with the best predictive power, then why should we presume that only one theory is correct?8 Insisting on the idea that one theory is getting at the world correctly requires that we have some means of evaluating these claims. But any evidence we acquire to test whether a theory is getting at the real structure of the world would simply be further empirical data that is either predicted by or not predicted by our theories. Ultimately, the only epistemic grounds by which we have to evaluate our theories is the predictive power of these theories. The representational story allows for certain predictions, but the non-representational neuroscientific explanation allows for predictions that occur at a smaller scale (and even in some cases that will manifest at behavioral levels).9 Of course, if the theories in competition are in direct conflict, then we cannot adopt both theories. However, if the various theories can be reconciled, then there should be no barring adopting both sorts of theories. The inference from the existence of a non-representational theory of some psychological capacity, to the conclusion that representationalism is false, simply doesn’t                                                      8 The spectre of Carnap (1937) looms large here. Carnap argued that we have a freedom in choosing our theories that is not restricted simply by the data. The principle of tolerance that he argued for was developed primarily in response to the choice of a formal / mathematical system for describing the world given to us in modern relativistic physics. There is no true formalization in describing any aspect of the world. Instead, we can adopt any number of distinct formalizations to describe the world (i.e. we can describe the physical structure of the world given either Euclidean or Non-Euclidean geometry). Instead, the only constraint in our choice of theory or formalization is due to the pragmatic factors in theory choice. Some formalization will simply be more useful than others in that they allow us to formalize empirical data. 9 For example, explanations given in terms of the neural implementation of some capacity can make novel predictions about how some capacity would degenerate due to particular sorts of neurological intervention. 15  hold. Multiple types of explanations are acceptable, provided that each one brings with it some additional explanatory powers and that the theories be compatible with each other.  Now, the situation becomes more complicated when one of the theories is thought to apply to a level of explanation that is supposedly implemented by what the other theory is talking about. In this case, the representational properties of the mind are not thought to be free-floating but are thought to be implemented by physical stuff that makes up the body (presumably, what makes up the brain). If the representational theory is supposed to be implemented by the mechanisms described by the neuroscientific theory, then it better be the case that the representational capacities we attribute to the organism be realizable in the neural machinery. This much should be obvious. However, for some, namely Fodor (1991), the idea that a representational or psychological theory should be beholden to the underlying neuroscientific theory is simply mistaken. Our psychological theories, or more generally, the special sciences (i.e. anything other than basic physics), are supposed to be independent of the theories of the underlying physical mechanisms. Instead of arguing directly against his position, we can take some of the content of the later chapters as direct counterexamples to Fodor’s conclusion. The representational states that we attribute in the explanation of behavior are not independent of our theories of the underlying neural machinery. One reason to notice this, and that will be appealed to in the later chapters, is that independent of any commitments as to what the underlying neural machinery involved in some process might be, we can actually attribute radically different representational states to explain the very same behavior. However, by looking at the underlying machinery, we are given further data points that constrain the attribution of semantic content.10 The representational story might still be underdetermined,                                                      10 It should be emphasized, however, that nothing in these data points logically require a different attribution of representational content. As was argued by Quine (1960) in his famous gavagai example, our attributions of semantic content are logically underdetermined by any observable behavior. As Quine argued, by merely noting a foreign language speaker pointing to what we call ‘a rabbit’ and hearing the person say ‘gavagai’ as they point, we cannot know for sure if the person intends to pick out with their exclamation the rabbit, a collection of undetached rabbit parts, a particular time-slice of a rabbit, etc. Instead, the point is that the information we learn about the neural implementation of representational states provides us with pragmatic grounds for adopting a more parsimonious attribution of representational states. 16  like all theories are, but taking into account the mechanisms that do the implementation, we place further restrictions on the viable representational stories.  So, the upshot of this discussion so far is that when we are evaluating whether any given psychological phenomena should be given a representational or non-representational explanation we need to look to whether the representational explanation provides us with explanatory power that is over and above the neuroscientific or biological explanations that must be admitted due to the biological facts that we all must accept. And further, whatever representational story we do provide must be such that it fits with the constraints imposed by the underlying neural machinery.  One more qualification needs to be made. There’s a way of putting the debate over representational theories that overly simplifies the situation. And that way of putting the debate is as follows: take some phenomenon in question, provide the characterization of that phenomenon, and then see what explains it. However, as Rey (1997) points out, a significant amount of begging the question can be done by how one characterizes the relevant phenomena. For instance, if I specify some psychological phenomenon as the ability to express the structured thought about the rains in Vancouver being too frequent, then this sort of phenomena might seem to require a representational explanation, since the very characterization of the relevant phenomena described the phenomena in representational terms. However, if I specify the phenomena as the ability to produce the utterance ‘the rains in Vancouver are too frequent’, then it’s possible to provide a non-representational account of this phenomena, since the characterization does not beg the question in favor of a representational account. This later characterization of what needs to be explained simply describes the production of certain sounds, but doesn’t say that those sounds bear any semantic content.  17  1.2. Theories of content  We just outlined one way in which the debate between representational and non-representational theories proceeds. However, in this section we turn to a slightly different tactic that is often appealed to in tandem with the approach we just discussed. In the last section we saw that the debate is often evaluated in terms of the explanatory power of the competing theories. However, in addition to that approach, some anti-representationalists attempt to show that adopting representationalism introduces theoretical burdens that cannot be maintained. Specifically, a number of anti-representationalists have argued for their non-representationalist theories by showing that representationalists cannot explain how the individual representations they posit come to have their semantic content.11  The way in which these arguments go is that the critic chooses some particular theory of content, typically the theory that the anti-representationalist thinks is strongest12, and shows that that chosen theory cannot account for the representations posited by a representational theory.13 If the theory of content cannot account for the particular representational content that is in question, then representationalism is discarded. Now, this sort of argument actually has similarities to arguments that we find within the representationalist camp over what is the correct theory of content. Amongst representationalists we find theorists proposing counterexamples to particular theories of content by providing cases where it appears that some theory of content gets things wrong in that the theory of content would predict some content that is not what the particular representational theory requires. As a result of these sorts of counterexamples, the particular theory of content in question is generally taken to be false across the board. What the representationalist infighting and the anti-representationalist arguments have in common is the idea that there will be a theory of content, and if some theory fails to explain how some                                                      11 Examples of this can be found in Chemero (2000), Hutto & Myin (2013), Ramsey (2007), and Stich (1982). 12 Chemero (2000) focuses on the teleological theory of Millikan (1989). While Hutto and Myin (2013) focus on the information theoretic approach of Dretske (1981, 1995). 13 We’ll see this sort of argument in many of the later chapters. I’ll leave off the explanation of the various types of theory of content until then 18  representation gets its content, then that theory will simply be false. Perhaps a concrete example should help.  Consider the theory of content given by Jesse Prinz in his book Furnishing the Mind (2002). In that book Prinz proposes a theory of content according to which a representation’s content is determined by the class of objects (or states) in the world that the representation is causally sensitive to and whatever actual object originally caused the production of the initial representational state. According to Prinz, I may have a representational state that gets triggered by both ducks and geese, however, this representation does not have the disjunctive content of duck or geese but rather represents ducks just in case the original cause of the representation being tokened, or activated, in my mind was a duck and not a goose.  Now, clear counterexamples for this theory are easy to come up with. For instance, I may acquire the concept DUCK through exposure to wooden ducks in a hunting store or at a museum. If Prinz’s theory of content were correct, then my concept DUCK would not actually have ducks (i.e. the water loving birds that quack) as its content, but instead would represent replica dummy ducks. This seems to clearly be getting the content of this sort of representational state wrong. But I would caution us from thereby inferring that the particular theory of content should be abandoned. Instead, it may simply be the case that in order to understand how the various representational states of the mind come to have their content that we must appeal to a variety of different types of theories of content. That is, to make sense of our attributions of representational content it may simply be the case that we require different theories of content. However, this is not to conclude that there are distinct types of content. Instead, the point is just that there may simply be multiple explanations for how some representation comes to have its content. Let me explain through an analogy.  Consider the property of being a parent. We can give a characterization of what it is to be a parent that fits across the various instances of being a parent. Namely, it is to have a parental relationship 19  to a child, where parental relationship can be cashed out in terms of a long list (likely disjunctive) of the sorts of responsibilities that a parent has. However, there may be various ways in which someone comes to be a parent. You might become a parent through childbirth, through adoption, or even through some other means whereby you simply take on the parenting role. In all of these cases we have rather different stories or explanations for how someone comes to be a parent, but there’s no sense in which any of these ways of becoming a parent are in conflict with one another. The fact that one person became a parent through adoption doesn’t thereby bar another person from becoming a parent through childbirth.   Now, let’s turn back to the matter of representation. What is it to be a mental representation? Well, it takes two things. First, there must be a representational vehicle. That is, some physical state of the brain (or some other physical system) that is the bearer of the semantic content. In this way, the representational vehicle is like the linguistic symbol that is just a physical mark on the page. Then, in addition to the representational vehicle, a semantic value must be attached to the vehicle. Take this characterization of what it is to be a mental representation as being analogous to the characterization of being a parent. Now, the distinct theories of content are proposals for how something may come to have the properties that are attributed to mental representations. However, in just the same way that there needn’t be a single story about how someone might become a parent, there needn’t be any single story about how representations come to have their content.   It should be kept in mind though that we have to be careful here as not any proposed theory of content should be taken as a viable one. Once again, as will become clear in later chapters, what theory of content we adopt is in part constrained by what we know about the underlying neural machinery and the way this underlying neural machinery is related to the world via the behaviors and other mechanisms possessed by the animal. A theory of content should be able to explain how the relevant components of the underlying neural machinery may come to bear certain contents that are attributed to that machinery in the service of providing a predictive and explanatory theory. 20   1.3. The temporal coordination problem  With all of that ground clearing out of the way, let’s turn to the main event. My initial characterization of the topic of interest has been put, and will be put, in terms of how animals come to mentally represent time. But from the previous sections, we saw that this is a question begging way of putting the issue. This isn’t an accident but a choice since it provides us with a much clearer initial characterization of what we want to explain. However, at the end of this section I will provide a more neutral reformulation of the problem in terms of how animals coordinate their behaviors with a temporally structured environment.  Most discussions of temporal perception and cognition, at least within philosophy, begin with an appeal to introspection. Simply by reflecting on your experience of the world it should be clear that we are conscious of and / or perceive the temporal structure of the world around us. As you sit there at a sidewalk table, you hear and see the cars passing by at different speeds, some waiting longer at stop signs than others. Pedestrians walk by; some even jog. You hear the music coming from the nearby café. First one song, and then as that song fades into the past, another song comes on. You might even manage to react quickly enough to catch a pen as it rolls off the table. In all of these ways, you simply need to reflect on your experience to notice that you are being presented with a dynamically rich world of temporally structured events.  Moving beyond these perceptual cases, our cognitive lives are also replete with temporal content. You sit there and wonder whether you’ll be able to leave work before rush hour begins or whether it will be quicker if you take one route home or another. You might rethink your weekend plans once you see that the weatherman says to expect rain. You might even start reliving old experiences recalling past situations. Again, in all of these cases, reflection reveals the temporal content involved in your mental life.  21   But notice that if we were to limit our discussion of temporal representation to just those cases in which introspective reflection on our own mental lives reveals temporal content, then we would be doing a great disservice to the role that the mental representation of time plays. As we’ll see even in humans, let alone other animals, there is an incredible range of time keeping capacities that simply aren’t available to us through introspection.   Let’s begin by sticking to two human cases. First, take speech perception. When someone produces an utterance we break down the speech pattern into discrete phonemes. In many cases we take the particular pitches and stops as what would delimit the distinct phonemes that we perceive. However, much of our phoneme perception is the result of a fine-grained sensitivity to the temporal characteristics of the sounds being produced. Buonomano and Karmarkar (2002) have a wonderful example to show this. Take the famous Hendrix lyric “kiss the sky”. By simply modifying the temporal pattern of the speech sounds, and without changing any of the other characteristics of the sound, we can change the perceived lyric from “kiss the sky” to what would have made for a very different song, “kiss this guy”! On reflection, at least for the naïve amongst us, we probably would have explained the perceived difference as there simply being a phoneme difference. However, in order to make this difference our perceptual system must be keeping track of the timing of the speech sound at the scale of milliseconds. The temporal content of speech perception simply isn’t worn on the sleeve of speech perception itself.  As another example consider auditory localization. Humans have a fairly decent capacity to judge the location of a sound source based purely on auditory cues. Chief amongst the auditory cues used in spatial localization is a sensitivity to the interaural timing difference brought about by the delay between a sound wave impacting upon on ear versus the other. By keeping track of these very slight temporal differences (along with a host of other differences that are in part due to the shape of the ears) we are able determine the location of a sound source (Wightman & Kistler 1992). Now, when you reflect on this sort of capacity, it is spatial content that is made available – you localize the sound. However, it is only 22  through empirical investigation that we find that the ability to localize sound in this manner requires the fine grained temporal sensitivity at the scale of microseconds.  We can continue this process of empirically uncovering cases where individuals must keep track of time to cases involving non-human animals. It’s when we start to investigate how animals coordinate their behaviors with their environment that we uncover a genuine treasure trove of temporal sensitivities. Chapter 3 will discuss the ability of animals to keep track of the time of day in the service of a discussion of the sense of time and the circadian systems, but here we can discuss other cases of time keeping capacities.  When we look at the animal kingdom we find that all animals that need to interact with and successfully navigate their environments must be capable of coordinating their behaviors with the temporal structure of the events around them.14 Merely being able to act in accordance with the what and the where of the world simply isn’t enough. For animals to move about a dynamic world, to forage properly, to hunt properly, to migrate appropriately, to do a number of different tasks, animals cannot leave it up to chance that they will be at the right place at the right time. They must have some sensitivity to the temporal structure of the world.  Now, the truly stunning fact about animal time keeping capacities is that in the animal kingdom we find animals that exhibit a sensitivity to the temporal structure of their environment over a time scale that spans 10 orders of magnitude (Buhusi & Meck, 2005). To give you an idea of the sheer range of this sensitivity consider what it would mean to have a spatial sensitivity over the same scale. That would require that an individual creature be able to discriminate distances as short as the length of an individual                                                      14 This point extends to all living things. Even plants and bacteria must be able to coordinate their activities with the temporal structure of the world around them. For example, plants must open and close their stoma to coordinate their gas exchange with environmental conditions. If the stomata are opened at the incorrect time, whether this ill-timed opening is due to temporary moisture levels or temporary carbon dioxide levels, then the plant will not survive for long. In this way, the plant itself must have some mechanism that coordinates its activities with the temporal occurrence of events in the environment. Temporal coordination, in this sense, is ubiquitous throughout not just the animal kingdom but throughout all of biological life. 23  flea to distances as long as the circumference of the planet Earth! All without the aid of any technological tools or written mathematics.  The sensori-motor sensitivities that rely on time keeping capacities should be fairly clear by this point. Animals, just like humans, move about in a dynamic and changing world where the changes that occur are entirely contingent15. In order to not constantly be bumping into things, animals, again like humans, need some way of keeping track of or coordinating their activities with these goings on. But, animal temporal sensitivities do not end with sensori-motor sensitivities.  Take for instance the feeding and caching behavior of the Western Scrub-Jay (Dally et al., 2006; Gallistel, 1990; Raby et al., 2007; Thom & Clayton, 2013). Scrub-jays exhibit not only the sensori-motor temporal sensitivities that we expect of any creature that physically navigates its environment, but they also exhibit rather surprising capacities for temporal reasoning and prospective planning. Scrub-jays eat a variety of different foods. However, these foods are typically available in cycles. At certain points some food sources, nuts or worm larvae, might be more abundant than other food sources. What the scrub-jay will do in these situations is harvest more than they can eat in one sitting and store the remainder in a food cache somewhere in its territory, thereby exhibiting some future directed concerns. If withholding food from their competitors were their main concern, then they could accomplish this task by simply destroying the food that they do not eat. The scrub-jay seems to genuinely plan for its future well-being.  But, their abilities begin to appear even more impressive when we consider how scrub-jays harvest their food caches. They do not simply feed at whichever food cache is closest or according to which is their favorite source of food, as would be the case if they simply stored the information concerning the spatial location of the caches or what type of food was stored. Instead, the scrub-jay will harvest the food that is most likely to spoil quickest (Clayton & Dickinson 1998). Clayton & Dickinson                                                      15 Whether the world is a deterministic one or not doesn’t matter for this point. The events in the world are epistemically contingent in that no animal has Laplacian knowledge of their environment, therefore the events going on in the environment are typically not ones that individual creatures could predict. 24  (1998) showed that scrub-jays will avoid harvesting worm larvae (their preferred food source in the experiment) from their caches, and instead go for the less desirable, but longer lasting, nuts, provided that the larvae were cached too long ago (between several hours and a day). Importantly, it was shown that scrub-jays were not responding to the smell of rotten larvae but were tracking the length of time different foods were stored. Scrub-jays not only keep track of how long certain food caches will remain usable, but they will also keep track of when the food was placed in those caches. We’re lucky in that our cell phones and computers tell us the date and our jugs of milk have their expirations dates stamped on their label. Imagine trying to keep track of these things without a calendar and without a best by date. Put in those terms, the temporal sensitivity of the scrub jay is genuinely stunning.  It’s here where we run up against a precaution that was raised in the previous section. In all of these cases, the human and the animal cases, we see that animals exhibit a sensitivity to the temporal structure of the world around them. What then is the phenomena that we need to explain? Well, if we want to leave it open as to whether these phenomena require a representational story or not, then we cannot characterize the question we want to answer as how do animals come to represent the temporal structure of the world around them. That would be to give a representationally loaded characterization of the phenomena that would likely beg the question in favor of a representationalist story. However, a more general characterization of the phenomena can be given (and in fact, we just saw this briefly). The world we live in is a temporally dynamic one. At any given moment certain states of affairs obtain at that time and at other times different states of affairs obtain.16 In order for an animal to successfully navigate its environment, it must have some way of coordinating its behaviors with the temporal structure of the world around it. Call this the temporal coordination task. All animals must have some way of overcoming this task. This coordination is what we want to explain. And notice, there is no                                                      16 This is all I mean by dynamic. Surely this wouldn’t satisfy someone with sympathies for McTaggart (1908), however, for the purposes of understanding something about the mind, this form of dynamicism is enough to set up the problem that I am laying out.  25  begging the question in favor of a representationalist story. Instead, it is left open whether positing mental representations of time for the explanation of this sort of coordination is required.  Now, the interest in noticing the extreme time range over which animals are able to coordinate their behavior with their environment was not just to produce a wow! response. Instead, the added purpose of showing that wide sensitivity was to make it plausible that the temporal coordination task, even within a single organism, is not a single task but instead breaks down into a number of distinct capacities to coordinate behavior with different aspects of the temporal world. From a pure engineering standpoint, it’s unlikely that any neural mechanism, made of noisy biological materials, would be precise enough to explain how animals can track microsecond timing differences as well as differences that span several days or weeks. Furthermore, there are a number of different empirical cases that show that different timing capacities come apart. At very coarse grained levels, Buhusi and Meck (2005) provide an overview of how sensori-motor timing comes apart from circadian timing. Destruction of the system responsible for many circadian behaviors in mammals, the suprachiasmatic nucleus, leaves animals unable to produce regular circadian behaviors but leaves more fine-grained sensori-motor capabilities undisturbed (Lewis et al., 2003). Craver et al (2014) provide evidence that sensori-motor capacities also come apart from the timing capacities involved in memory. The famous case of Mr. B, a patient who suffered almost complete loss of the ability to store new memories could successfully navigate his environment, and even engage in complex sensori-motor tasks of drinking a beer (Craver et al., 2014), despite the fact that he could not retain any information for more than approximately 2 seconds. And in chapter 4 and 5 I provide arguments that even at the scale of millseconds to seconds, the timing capacities of humans are due to a motley assortment of different types of time keeping devices.  The temporal coordination task is many, not one.  26  1.4. The mental representation of time  We’re leaving it open that the explanations for how animals overcome the various temporal coordination tasks they face should involve an appeal to the mental representation of time. In fact, the majority of the next chapter is devoted to discussing the various sorts of approaches that can be found in the literature to explaining how animals overcome the temporal coordination task. It is there that we’ll lay out the various representationalist and non-representationalist approaches. However, in this section we’ll lay out why some authors have thought that the mental representation of time poses particularly damning problems. Recall that from previous sections it was mentioned that one of the significant hurdles that a representationalist theory of some psychological capacity faces is an explanation for how the purported representations come to have their content. That is, the problem comes up as providing a theory of content for the representations in question.   It is in trying to provide a theory of content for temporal representation that the temporal coordination task becomes the temporal coordination problem. The worry can be put bluntly. Time is an odd aspect of our environment, so we’ll need a correspondingly peculiar account of how we come to represent time. But, for a statement of the problem, put with somewhat more finesse, we can turn to a quote from the psychologist Lera Boroditsky. She says, All of our experience of the world is physical, accomplished through sensory perception and motor action. And yet our internal mental lives go far beyond those things observable through physical experience: we invent sophisticated notions of number and time […] so how is it possible that physical organisms who collect photons through their eyes, respond to physical pressure in their ears, and bend their knees and flex their toes in just the right amount to defy gravity are able to invent and reason about the unperceivable and abstract? (Boroditsky 2011, p. 333) 27  Time, it would appear, is simply an odd aspect of our world. Unlike the medium sized objects that we interact with and their spatial locations, time isn’t something that we can readily point to. Yes, we use hand gestures when we speak about time. An impatient parent might point down into their palm when they want something done right now. Or a Westerner might point behind them when they talk about the past being way back then while an Easterner might point in front of them when they speak about the past being known. However, in none of these cases where physical gestures are used alongside speaking about time do we take the gestures to literally be pointing to the temporal features of the world we have in mind. The worry, also, isn’t just one where time is abstract and objects and locations are not. We can readily point to concrete instantiations of abstract categories. I can point to a pair of people and take myself to be pointing to an instance of friendship. But again, we simply can’t point to time in this way. The temporally extended properties that events possess in our world simply aren’t made available to us at a time to be able to point to them.17 Time is also something that we cannot readily manipulate. We can grasp objects and move them around thereby changing their spatial relations to other objects. We can figure out where to hang a picture by putting it here, and then there, and then back to the first place. We can move ourselves around the room to get different perspective on the layout. We can even control the speed at which we move though space, moving slowly when we want and faster when there’s the need. In so far as we seem to be able to control anything, we seem to be able to control at least some of the spatial properties of the world around us.  However, our ability to manipulate the temporal relations of events is much more limited. In some cases, we can delay events or bring them about quicker. I can choose to send an email now or later.                                                      17 This is in part one of the problems that arises with temporal representation. Many other cases of representation, including pointing, involve a representational vehicle and a represented content that are both entirely present at the moment of representation. However, with temporally extended phenomena, the thing being represented does not exist at the moment of being represented. Rather, we only get access to one part of what we are pointing to. 28  But this manipulation of the temporal order of events only occurs in the prospective case – we can change the expected or planned temporal order of events. But once an action is performed, we cannot undo it and change its temporal relations to other events. Once events are in the past, there is nothing that we can do to change their location in the past. The analogy to space would require that once we locate an object somewhere, then we cannot move that object to another location. However, nothing like that happens with our ability to manipulate the spatial location of objects. Insofar as an object can be moved, there is no limitation on moving an object back and forth in space, but with time this back and forth simply isn’t possible. We also can’t change the speed at which time passes.18 Finally, time is also something that seems to lack any causal influence on the world around us. Time isn’t some form of energy that can be transmitted and impact our sensory transducers. Time isn’t the sort of thing that we might bump up against as we’re walking down the stairs. Time simply doesn’t seem to be the right sort of thing to have any causal influence on us. Now, we do sometimes speak as though time is causing certain effects in the world. We might say that the milk went bad because of how long it was out on the counter. Or, we might say that the joke failed because of its being delivered too quickly. However, a number of philosophers working on the metaphysics of time would caution us from taking these expression as genuine evidence that time exerts causal influence on the world (Lewis, 1973; Maudlin, 2002; Newton-Smith, 1980). According to these philosophers we should avoid attributing to time genuine causal influences and instead understand time as being a requirement for causal influences to occur at all. In the case of the milk, the genuine causal story, they would suggest, is one that appeals to the metabolic processes of the bacteria found in the milk and the warm temperatures. While these processes unfold in time that does not make time a cause of the milk’s going bad. Whether these sorts of                                                      18 An initial reaction to all of this might be to say, well, we can do these things, modern physics allows for all sorts of temporal distortion and varieties of time travel. However, once we look at that literature we find that the ability to change the past is not like how it’s depicted in movies like Back to the Future or The Terminator. Instead, discussions of time travel still involve the acceptance of the idea that what’s done is done, however, in cases of time travel the stories just become more complicated. For an example of this see Lewis (1976). 29  arguments that time is causally inefficacious go through, it’s nevertheless the case that if time is to be causally efficacious thing in our world, then it does so in a way that is rather different than how other physical things in our world cause things to happen. There would seem to be no medium or vehicle, over and above the physical things in the world that we already believe possess causal powers, for time to influence anything. Temporal causation would seem to be entirely superfluous. Again, time is simply an odd feature of our world. How then do we explain how animals may come to have mental representation that track such an odd aspect of our world? As we’ll see in subsequent chapters, I do not believe that the peculiarity of time poses any genuine problems for providing a theory of content that makes sense of our attributions of temporal content. In fact, in what follows I argue that the resources of information theoretic accounts of content (e.g. Dretske (1981)) are sufficient. However, a number of researchers have doubts that standard stories can be given for temporal representation. In particular, if we look at the theories that are proposed by psychologists working on the temporal reasoning capacities of humans, we find that they often endorse theories that make use of some rather sophisticated machinery. Perhaps the most salient of these approaches has its origins in the work of Whorf (1944) and the later work of Lakoff and Johnson (1980). According to the spatio-cultural approach to explaining temporal representation our ability to reason about time involves the scaffolding of temporal representations on an existing system for the mental representation of space.19 Time, they argue, is spatialized. The details of the accounts vary from author to author but the general evidence for the view remains the same across writers. Perhaps the most salient of the data points that they take to support their claim is the widespread use of terms with a primarily spatial meaning to speak about temporal relations.                                                      19 Instances of this sort of approach can be found in Lakoff and Johnson (1980), Lera Boroditsky (2000, 2001, 2011; Boroditsky & Ramscar 2002), Daniel Casasanto and colleagues (Casasanto 2008; Casasanto & Boroditsky 2008; Casasanto et al 2010; Merritt et al 2010), and Prinz (Boroditsky & Prinz 2008). 30  We speak of long or short times. We speak of the past being behind us and your future being ahead of you. Also, the specific ways in which time is spatialized will vary from culture to culture. For instance, native speakers of languages that are written from left-to-right will typically associate the past with the left and the future to the right. While native speakers of language that are written from right-to-left will typically associate the past with the right and the future to the left (Boroditsky 2011).20 Now given the widespread use of spatial terms for speaking about time and the associations that can be found between spatial and temporal properties, there does seem to be something correct about the spatio-cultural approach to explaining temporal reasoning. However, it’s unclear how this sort of approach actually solves the problem that Boroditsky described in the quote above (this fact is worth noticing as Boroditsky herself is one of the major figures behind the spatio-cultural approach to temporal reasoning).  The original question that we were tackling in this section was how mental representations of time come to have their temporal content. The story that we are considering here would explain this by saying that temporal representations involve distinct uses of the underlying machinery used in spatial representation. Presumably, since these approaches take space to be more concrete, it should be the case that the underlying representational system has the ability to represent space. That much should be granted. However, the question then becomes how do we explain how a system used for representing space comes to represent time. The question of how the temporal representations we attribute in the explanation of some capacity come to have their temporal content simply remains unanswered. Furthermore, if we require a sophisticated explanation for how these temporal representations acquire their content, then what should we say of the vast number of other temporal representations that cannot be explained via these sophisticated means? Surely prior to the acquisition of language and prior to a                                                      20 Additional evidence for the view comes from the existence of SNARC effects. See chapter 5 for a discussion of this evidence. 31  significant amount of cultural influence, children are capable of navigating their environment, and this likely involves some temporal representation. Therefore, it’s likely that not all temporal representations should be given such a complicated explanation. Another of these sophisticated explanations for how humans come to acquire their temporal reasoning capacities bases temporal reasoning on an understanding of causation (Hoerl, 2009; Hoerl & McCormack, 2011, 2016). The theoretical ancestor for this view comes from a failed metaphysical project of Hans Reichenbach’s (1956, 1957). Reichenbach attempted to give a metaphysical analysis of time in terms of causation. Time, or the temporal structure of our world, he argued is not fundamental but is based in the causal order of the world. It’s widely accepted in the literature that this analysis fails (see Hoerl and McCormack (2011) for a discussion), however, several researchers have taken this analysis as a plausible one for understanding the development of temporal reasoning capacities in children. There is some evidence to show that infants have difficulties in reasoning about time, especially for novel temporal sequences as opposed to highly familiar scripted encounters. However, this ability to reason about time seems to be come online as more and more sophisticated capacities for causal reasoning emerge. The proposal is then that the concepts of EARLIER THAN and LATER THAN emerge from the concepts of CAUSE and EFFECT.  Whether or not this causal explanation of the origins of temporal concepts succeeds, the same explanatory burden remains for this theory as it did for the spatio-cultural approach. That is, the question was one of how the temporal representations we attribute in the explanation of some capacity come to have their temporal content. Saying that these temporal concepts are born out of representations of causal relations does not tell us that story. We still lack the explanation that we require. 21                                                      21 This isn’t to say that these sophisticated accounts couldn’t provide the needed explanation. Rather, it’s just to point out that the important question of how we come to acquire representations with a particular type of content remains unanswered. 32   The point to take from all of this is that even granting that time itself is a somewhat peculiar aspect of our environment, the difficulty of explaining how we come to represent time will not be exhaustively solved by proposing complex and sophisticated explanations of those temporal capacities. Furthermore, we simply have no a priori reasons for assuming that there will be a single explanation for how we represent time as opposed to there being a multiplicity of ways in which animals across the animal kingdom come to represent time. And, as I will argue throughout the thesis, by looking at the specific neural machinery involved in the production of time sensitive behavior we will get a better sense as to what theories of content are viable for our explanatory purposes, and in doing so we will see how the temporal coordination problem is overcome.  1.5. Moving forward  The interest in the mental representation of time has been around for an incredibly long time and throughout this history the relation between the temporal structure of the mind and the temporal content of mental states has always been up for debate. As Locke put it, It is evidence to anyone who will but observe what passes in his own mind, that there is a train of ideas which constantly succeed one another in his understanding, as long as he is awake. Reflection on these appearances of several ideas one after another in our minds, is that which furnishes us with the idea of succession… (Locke, 1689 Chapter XIV, 3) In order to explain how we come to have ideas, or in modern parlance concepts, of time we simply needed to appeal to the actual temporal structure of our experiences themselves. It was the very temporal sequence of non-temporal mental states that would furnish us with mental states that were about time. 33   This story of Locke’s quickly met resistance. Both Reid (1855) and Kant (1781/1998) argued that a mere succession of experiences could not amount to an experience of succession. That is, the mere temporal structure of experience could not account for the content of experience. Instead, it was argued that something further is needed to explain how mental states come to be directed at or about time.  Since the discussion of temporal representation in the early modern period quite a lot has occurred. In particular, perhaps the most widespread and influential discussions of the mental representation of time in philosophy came in the form of William James’ (1890) and Edmund Husserl’s (1917/2008) discussions of temporal experience. Both authors proposed accounts of how we could come to experience the temporally dynamic world around us, and both authors have been interpreted in a number of distinct ways.22   Unfortunately, much of the philosophical discussion of the mental representation of time remained focused on the narrowly defined experience of time. That is, philosophers have focused on how we come to consciously experience the temporal structure of the world around us and the flow of time’s passage. Furthermore, while there has been an impressive explosion of empirical work on temporal representation coming from computer science, cognitive psychology, neuroscience, zoology, linguistics, and anthropology, there has been only limited engagement between that empirical literature and philosophical discussions of temporal representation. Perhaps the reason for why this additional relevant literature hasn’t been appealed to by philosophers is due to the fact that philosophers have concerned themselves primarily with the conscious experience of time, and this other work doesn’t directly bear on the conscious experience of time.                                                      22 For interesting historical work on these authors see (Andersen & Grush, 2009) and (Andersen, 2014). As we go through the thesis, one might notice that I do not speak about the views of either James or Husserl in any detail. There is good reason for this. Both historical figures produced views that have been interpreted in a great variety of different ways. As a result, in order to fruitfully discuss either view, either as a topic in and of themselves or in the service of some further aim, would require a significant amount of historical analysis. And that is something that would simply take us too far from the main themes of this thesis.  34   In what’s to come we’ll be looking at temporal representation in a way that goes beyond the mere conscious experience of time. By beginning with the temporal coordination problem more generally we are able to consider the various ways that animals come to keep track of and coordinate their behaviors with the temporal structure of the world around them. Once we consider temporal experience as being just one aspect of this more general temporal coordination problem the wealth of empirical work from various fields instantly becomes relevant to our research. In discussing this point, we’ll see that positing internal representations of time is often required, especially when it comes to discussing how we experience time, and I will make strides towards explaining how it is that those internal representations of time come to have their temporal contents. Ultimately, we’ll not only understand something about how the mind comes to be directed at or about time, but also, we will make some headway in seeing that the mind can be given a naturalistic understanding through seeing how the mental representation of time can itself be naturalized.   35  Chapter 2: Empirical Timing Models and the Explicit / Implicit Distinction  As described in the introduction, all animals face a form of the temporal coordination problem in that all animals need to coordinate their actions with the temporal structure of the world around them. Rather than explain how any aspect of the temporal coordination problem is solved, the goal of this chapter is to lay out some of the general ways in which the problem could be solved by giving a taxonomy of the explanatory approaches found in the empirical and philosophical literature.  As we’ll see, the major axis according to which the taxonomy in this chapter will be structured concerns the degree to which these explanations appeal to explicit representations of time. At one extreme, the temporal coordination problem is overcome through the use of dedicated mechanisms for the explicit representation of time. Towards the other extreme are positions that simply do away with any representational states whatsoever. Between these extremes are a number of different positions that preserve some role for representational states and even some role for representations of time in explaining temporal coordination.  In order to get a handle on these different explanatory approaches it will be useful to distinguish between two ways that a system might be said to represent some content, that is, to distinguish between explicit and implicit representation. The chapter will begin by tackling this task. In section 1, I will go over a number of different means by which people have attempted to draw the explicit / implicit representation distinction and argue that many of them, while tracking interesting psychological divides, do not track genuine differences in how some content is represented. In section 2, I do the work of articulating a sense of the explicit / implicit distinction that does in fact draw a line between two ways of representing some content. With that distinction in hand, I’ll turn to the main task of this chapter and provide a taxonomy of the various explanatory strategies for understanding how animals overcome the 36  temporal coordination problem. Since, as we’ll see throughout the thesis, the temporal coordination problem that any animal faces actually breaks down into a number of distinct coordination problems, there is unlikely to be any single explanatory strategy that will explain the entirety of the coordination problem. Therefore, instead of arguing that any of these strategies is correct, the goal here is to note the variation in explanatory approaches and to simply lay them out so that we may appeal to them later on when needed.  2.1. The explicit / implicit distinction in cognitive science  In this section, we’ll look at how some in the cognitive sciences have attempted to draw the implicit / explicit representation distinction and argue that none of these attempts track genuine differences in how representations encode their content. In the next section, we’ll turn to providing an account of the implicit / explicit distinction that does track a difference in the manner of encoding some content. Before we lay out an account of the explicit / implicit distinction that tracks a genuine difference in ways of representing some content, it will be helpful to survey some of the ways that people have attempted to draw the explicit / implicit distinction in the cognitive sciences. In each case, the distinction drawn, while tracking some distinction amongst representational states, fails to distinguish between two ways of encoding some information in a representation. Instead, these existing approaches mark the distinction not in how the representation encodes its content, but how the representations fit within a larger functional system.  How then has the explicit / implicit distinction been made? Perhaps the most common way of marking the distinction is in terms of the conscious availability of some representational content (Schacter et al, 1993). According to this way of marking the boundary, only information that is available to 37  conscious report is explicitly represented whereas implicitly represented information is not so available. What this distinction rests on, however, is not a distinction in how representations encode their content, but rather in how content encoded by a representation is used. Just a quick survey of different theories of consciousness shows us that many theories of consciousness distinguish between conscious and unconscious mental states (or processes) not in terms of how they encode information, but rather, in terms of the functional role played by particular representations.23 For instance, on the popular Global Workspace Model of Consciousness (Baars 2005; Dehaene et al 1998) what distinguishes conscious from unconscious representations is whether the content encoded in the representation is being made available to a wide range of consumer systems as a result of entering into the global workspace. Notice that on this sort of model the very same representation can be conscious at one moment and unconscious at another, and as a result, the same representation will be at times explicit and at other times implicit.24  To further illustrate how this way of carving the distinction fails to track ways of encoding content in a representation, notice that we can render a representation unconscious to the point of not even being possibly available to consumer systems by manipulating aspects of the cognitive architecture that lie beyond the confines of the individual representation. Cognitive science is full of cases like this. One particular case comes from the use of transcranial magnetic stimulation (TMS) (Lamme 2001; 2006). In these cases, stimuli that would normally be consciously perceived are rendered unconscious by the application of a precisely timed TMS pulse that disrupts the feedback mechanisms that would place the representation into a dynamic feedback loop with higher-level cortical areas. Again, the particular representation in question that may have become conscious is left untouched. What is changed is how                                                      23 Some theories, such as Tye’s PANIC theory (Tye 1995; 2000), also require that conscious representations possess certain types of content. In Tye’s case, conscious representations must possess abstract non-conceptual intentional content. However, in addition to the type of content conscious states possess they must also be poised to impact downstream systems, and in this way his theory relies on the functional difference I am describing here. 24 Besides the global workspace, we find functional characterizations of the conscious / unconscious distinction in Lamme (2001; 2006), Prinz (2000; 2012), Dennett (1992), Dretske (1995), Rosenthal (2002), Koch (2004).  38  other parts of the brain are functioning. Since the representation is left the same, then the difference in consciousness cannot mark a difference in how the (untouched) representation encodes its content.  An alternative, but somewhat related, attempt to mark the distinction between implicit and explicit representations is given in terms of the amount of processing25 needed to utilize information (Kirsh, 2006). According to this way of marking the distinction between implicit and explicit representations the difference is a graded one. Explicit representations are those whose contents are readily made use of with minimal effort and with minimal demands on limited processing resources. Implicit representations are, on the other hand, ones that require significant processing resources in order to access their content. In other words, explicitly represented content is capable of driving other behaviors with minimal additional processing while in order for implicitly represented content to drive further activity requires a significant amount of intervening processing. While there is something to this way of marking the distinction, there is nevertheless a significant shortcoming to using the amount of processing needed to extract the content as a means of dividing explicit from implicit representations.  Take the sentence ‘the police officer wore a hat’. Armed with a decent understanding of English the content of this sentence is readily extractable. It tells us something about what a particular person had on their head. In fact, the content of this sentence, on standard linguistic theories, is simply a result of a compositional process whereby the words contribute meaning to the sentence in a manner determined by the grammatical structure of the sentence. Now, take the sentence ‘police police police police police’ (from (Kirsh 2006)). On a first glance this sentence seems to be just a meaningless stringing together of the same word five times over. However, this sentence is in fact a well-formed sentence of English, whose content is determined through a straightforward compositional process, but as anyone coming across this sentence would notice, the content of the sentence requires a significant amount of mental                                                      25 This notion of amount of processing is left rather vague in these accounts. Typically, however, this amount of processing can be equated with the difficulty for the subject to extract the relevant information. 39  effort to extract. A grammatically correct interpretation of the sentence is policemen who are policed by policemen also police policemen.   The way that these two sentences carry their content is the same. They both possess their content as a result of the compositional semantics that dictates how the word level meanings are brought together to form the sentence level meaning. Yet they differ in how much effort it takes to utilize their content. If, as we’re assuming, the explicit / implicit distinction is one about the way in which content is encoded by a representation, then the amount of information processing resources required for the extraction of that content cannot be what defines the distinction, since at most, the amount of information processing resources required to extract some content tells us about how some representation enters into the rest of the cognitive economy.  One final approach from the literature for marking the implicit / explicit representation distinction is given in terms of whether some content is hardwired into the operations of a system, i.e. implicitly represented, or whether it is explicitly represented through some variable aspect of the system.26 As a matter of biological fact we have been endowed with a number of different neurological mechanisms that allow us to interact with the world in certain ways. Some of these mechanisms, it has been argued, have been shaped in just the right way that they seem to provide us with some rather sophisticated capacities. For instance, in the literature on object perception27 a class of theories (e.g. Carey 2009; Leslie et al 1998) appeal to the operation of the object tracking system in order to explain how infants and adults come to form perceptual representations of enduring physical objects. At the core of these theories is a perceptual system that is designed by evolution to track stimuli that adhere to certain principles, such as continuity of form, spatio-temporal continuity of trajectory, etc. Only stimuli that adhere to these principles will be tracked by the object tracking system. The result is that even very young infants will appear to have an                                                      26 See (Bernal, 2005) for a discussion of this sort of approach to the explicit / implicit distinction with regards to the purported object-specific principles employed in infant object cognition. 27 Similar conclusions could be drawn by discussing the literature on biological motion perception. 40  understanding of objects as being spatiotemporally coherent enduring entities. Yet, on many of the interpretations of this system, the object tracking system does not explicitly represent any of these principles, rather it only ever produces representations that pick out or refer to the objects they track (Bernal 2005; Pylyshyn, 1989, 2001, 2007). The system is hardwired to appear as though it is adhering to some rules, and in this way, some have described the object tracking system, and its operations, as providing the infant with implicit knowledge of objecthood. Nowhere is this knowledge explicitly represented, but it is embodied in the operations of the system.  In this way, we have something that seems to be a genuine difference in how a system encodes certain information. Implicitly represented contents are those that are possessed by a system due to the way that it is hardwired to operate. While explicitly represented contents are those that are carried by malleable or variable properties of the system. The purpose of this section isn’t to dictate how someone should use the terms ‘explicit representation’ or ‘implicit representation’ as each of the various proposals covered so far seem to latch onto psychological distinctions that might have some explanatory use. Nevertheless, the worry with this particular means of drawing the implicit / explicit distinction is that it seems to trivialize the notion of representation. The only compulsion to claim that the principles of objecthood are represented at all is that the behavior of infants and adults seems to be best characterized as though it were following certain rules. However, this way of characterizing things applies to every lawful phenomenon in the world. A plant, as it grows towards the light, could be interpreted as having the implicit knowledge that light is good. Yet it seems as though this sort of interpretation isn’t needed to understand how the system is actually operating. We can just describe the plant as having certain biological mechanisms that respond to light in very mechanical ways. Similarly, in the case of object perception, we needn’t ascribe any representational component to the operation of the object tracking system, since we only need to say that it responds in a very mechanical way to retinal stimulation. Instead, as we’ll see in the next section, the notion of representation is typically tied closely to the states of a system, how these states track the world around it, and how ascribing semantic content to these states 41  helps us understand how the representational system is put to use by consumer systems of the representation.  2.2. Minimal semantic interpretations and the implicit / explicit distinction  The approaches to marking the implicit / explicit distinction either trivialize the notion of mental representation (as in the case of taking hardwired operations as implicitly representing some principles) or simply fail to mark a distinction between forms of representation. In this section, I’ll articulate a way of understanding the distinction that is not only found in the empirical literature but that also tracks a genuine distinction between ways that a representation can be said to carry some content.  Let’s begin by drawing an analogy with linguistic communication where we can draw a distinction between the literal linguistic meaning of an utterance and what is implied by the utterance.28 Take the following sentence: 1. Sally and Johnny are the only students in the course. The literal meaning of the sentence is given by what we can call the minimal semantic interpretation (MSI) of the language. Given the expressive demands of a symbol system, the MSI of the system is the simplest assignment of semantic contents to the symbols of the system that given rules for the combination of these symbols provides for the full expressive power of the entire symbol system. The meaning of (1) is given by the assignment of semantic contents to the individual words that compose the sentence and the rules by which these symbols are combined. There needn’t be any additional semantic interpretation of the sentence as a whole, apart from the interpretation of the individual symbols and a                                                      28 The distinction here is very similar to the Gricean distinction between what is said and what is communicated (H. Grice, 1975), however, we should resist the urge to identity the two distinction. According to Grice, what is communicated by a sentence are the implications of the sentence that rely on the acceptance of some auxiliary assumptions. If one were to reject those auxiliary assumptions, then the implications would not follow. However, the implications that I will be discussing here needn’t rely on any auxiliary assumptions for their acceptance. 42  specification of how those symbols combine (that is, (1) is non-idiomatic). Take the literal meaning of (1), that Sally and Johnny are students in the course and no other students are in the course, as the explicit content of the sentence. Notice, that the sentence as a whole only has a meaning, or evaluable semantic content, in a context given that the phrase 'the course' is only given an interpretation relative to a context (whether that is the context of utterance or interpretation doesn't matter). Due to the context dependency of how we interpret many symbols, the notion of a MSI needs to be modified to allow for this context dependency. The MSI of a system is, given a context, the simplest assignment of semantic content to the basic constituents of a symbol system that in conjunction with rules of combination provide for the full expressive power of that system. Given the MSI of the language, in this case English, we can then describe two senses in which (1) has certain implications that a competent speaker of English would be warranted in inferring given (1). Some of these implications, such as (2) and (3), are guaranteed in a context free manner: 2. There are two students in the course. 3. If x is a student in the course and x is not Johnny, then x is Sally. Both (2) and (3) can be directly derived from (1) regardless of the context in which (1) is used.29 All that is needed to make these sorts of inferences is that the inference maker possess the appropriate concepts required to articulate the implications and that the inference maker grasp the principles that ground these inferences. If, in addition to the logical inferences that lead to (2) and (3) we allow for some analytic inferences, then the class of context free implications of (1) can be greatly expanded to include things, such as:                                                      29 In fact, all logical truths can be derived, and are therefore, implications of any sentence of English. 43  4. There are at least two individuals capable of learning in existence.30 If we accept that 'x is a student' analytically implies that 'x is capable of learning and exists’, then we can infer (4) from (1) independently of any factual information about the world that extends beyond semantic or conceptual relations between concepts. Therefore, if we accept that analyticities exist, then inferences like (4) from (1) will be guaranteed by the MSI of English and (1). Of course, however, the existence of analyticities is a highly contentious matter in analytic philosophy and their existence is doubted by many (Fodor & Lepore 1991, 1992; Quine 1951). So, perhaps inferences like (4) do in fact require factual knowledge of the world. Regardless, the discussion here can continue regardless of whether analyticities exist, as the important distinction is between the literal content of a representation and that which can be inferred from the representation. This leads us to the next category of implications from (1). These implications are secured by bringing to bear other knowledge of the situation represented by the original sentence. In other words, these inferences are only derived in a context dependent manner that relies on information not specified by the MSI of the language and the logical structure of the language. Imagine that the context in which (1) appears is such that courses at that institution rarely ever have less than 10 students unless the course meets very early Saturday mornings. Given this additional information, we can infer from (1) the following: 5. Sally and Johnny are likely morning people. 6. The course most likely meets on Saturday mornings. 7. Other people are not interested in attending this course.                                                      30 If the inference from x is a student in a course to x is capable of learning and exists isn’t one that strikes you as being an analytic statement, then substitute the sentence in (4) for whatever strikes you as more plausibly analytic. If nothing strikes you as a plausible analytic statement, then the discussion can continue on the assumption that there are no, or only very limited, analytic statements. 44  The MSI of the English does not by itself give us reasons for inferring (5) – (7). What unites (2) – (4) and (5) – (7) into one coherent group is that in every case the inference requires that we first interpret the literal meaning of (1) to derive any of (2) – (7). More specifically, we can say (1) implicitly represents X, just in case X is a context-free implication of (1). We can alternatively say (1) implicitly represent X given C, just in case X is a context dependent implication of (1). This distinction is important, since the context dependent implications of (1) are only warranted given some further facts that are not provided by the MSI. Given the notion of implicit / explicit representation given here, we need to distinguish this from an outwardly similar appearing phenomena. Imagine a situation in which you and I need to coordinate our behavior, but we wish to do so in a way that goes unnoticed by onlookers. We might establish a code word convention where we use (1) to indicate that one should turn off their phone. When you use (1), I then know that you are telling me to turn off my phone. One might be tempted to say that in this case (1) implicitly represents that I should turn off my phone. However, this would be mistaken. The use of (1) as a code word fails to depend on the MSI of (1) necessary to understand the language in general. Instead, by adopting the convention that (1) be used to tell someone to turn off their phone, we no longer are using the original MSI. Instead, we have provided a new MSI.31 The actual phonetic string associated with (1) is now subject to two distinct interpretations. In this case, then, a usage of (1) will have two explicit contents, each associated with a distinct MSI. With this usage of the implicit / explicit distinction in hand, we can understand how it can be applied to mental representations. While linguistic expressions consist in discrete units that can in some sense be interpreted however the language users like, as indicated by the use of (1) to mean that someone should turn off their phone, mental representations do not possess either characteristic. In providing an                                                      31 In effect, the use of (1) in this coded context could be replaced with something like a wink of an eye or a random noise as long as we have previously agreed to use whatever “expression” in order to convey that one should turn off their phone. 45  analysis of the representational content of some mental mechanism we don't have clearly defined symbols that can then be assigned some semantic value. Instead, we must avail ourselves of states of physical systems. These states are the units that we can then assign semantic values to. However, in assigning semantic values to the system, we are constrained by our scientific practices to make the functioning of the system intelligible and by the causal efficacy of the states of the system in making the representation’s contents available to downstream processes.  On the causal efficacy of the relevant states, it is crucial for a system to be a genuine representation that its contents be available to some consumer system. In the linguistic case, the actual physical marks on the paper, or the waves in the air, must be such that they are capable of causally impacting other systems that allow for the extraction of the encoded content. Some piece of paper may be my sister’s favorite piece of paper, and we could plausibly try and ascribe some semantic content to the paper in virtue of having this property. But that would be a mistake. The property of being my sister’s favorite piece of paper simply isn’t something that changes the paper’s causal powers thereby making some content extractable by consumer systems. In the case of mental representations realized by neural systems, whatever property of the system that we appeal to as being the semantically significant properties (i.e. the properties that encoded or bear semantic contents) must be capable of causally influencing downstream neural processes. So, while a neural system might change its mass through various metabolic processes, we could not take these states of the system as being semantically significant since downstream neural processes are not sensitive to these states of the system. That is, the mass of a neural system simply isn’t the sort of state that is causally efficacious within the system (see chapter 3 for a more detailed discussion of this point). Turning to the intelligibility constraint, let’s take an extreme example. If we want to provide a semantic characterization of the functioning of the vestibular system in humans, we would be misguided to interpret the state of the vestibular system as having cities in France as their semantic values. While an 46  interpretation of this sort may be made formally consistent, in that there is a one-to-one mapping between cities in France and states of the vestibular system, it would not make human behavior intelligible since cities in France seemingly have no significance for the balancing behavior of humans. Instead, if the vestibular system is to be given a semantic characterization in order to make some feature of human behavior intelligible, then we need to assign semantic values to that system that have this intelligible making role. This notion of intelligibility can be understood in terms of the difference that accurate or inaccurate representation has in the production of behavior. Cities in France simply aren't significant for a person's ability to balance themselves, and it goes without saying that accurately representing the state of cities in France is unlikely to be of any significance for human balancing. In other words, if we assign the states of the vestibular system a semantic interpretation that maps the states of the system onto cities in France, the fact that the system is accurately or inaccurately representing some state of affairs will have no explanatory value in our understanding of the cause of successful balancing. However, the direction of the gravitational pull is significant for human balancing in that if the system is not accurately keeping tabs of the current direction of acceleration, then the system will not be able to properly orient itself in its environment. So if we have two consistent interpretation of the representational role of the vestibular system, according to one where the system represents cities in France and the other that represents gravitational pull, then it seems clear that we should adopt the semantic interpretation that assigns direction of gravitational pull to the states of the vestibular system. None of this should strike anyone as controversial.  Taking the example of the human vestibular system, we have a system with various states – e.g. influence on cilia by fluid in the semi-circular canals. Let us also assume that this system can be provided 47  a MSI32 in that each state of the semi-circular canals represents the acceleration of the head in a particular direction. Furthermore, this assignment of semantic values to the states of the semi-circular canals makes the balancing behaviors of humans intelligible. We can then say that the states of the semi-circular canal explicitly represent accelerations of the head. Let us take a particular state of the semi-circular canals in the following way: 8. Head accelerates at X m/s2 in direction Y. From (8) it can be inferred that: 9. Head accelerates at speed < X+1 in direction Y. (9) can then be said to be implicitly represented by N, where N is a state of the semi-circular canal. Given some further contextual factors, say, that if one's head accelerates at a speed greater than X-1, then some damage is occurring to the organism, then it can be inferred that: 10. The organism is incurring damage. Both (9) and (10) are implicitly represented by (8), however, (10) is only implicitly represented given some further contextual facts. In what follows, the distinction between contextual and non-contextual implicit representation will be of some importance, but for now just treat them as both forms of implicit representation. This way of marking the implicit / explicit representation distinction explicitly appeals to a distinction in the way in which information can be encoded by a system. Take one further example, the implicit representation of numerosity by elements in working memory (e.g. Carey (2009)). It has been argued that the rudimentary numerical abilities of infants can be accounted for by the implicit                                                      32 While the original statement of an MSI appeals to a combinatorial syntax for the system being interpreted, a non-combinatorial system like the vestibular system can still be given a MSI since the class of rules for the combinatorial syntax would simply be empty. 48  representation of numerosity by working memory representations. Infants, it is argued, are able to open mental files, or explicitly working memory representations, that demonstratively refer to a limited number of elements (between 2 – 5 elements). For instance, provided with a display of four puppies the infant might form a working memory representation of the following form: 11. That1, That2, That3, That4. Where each 'ThatX' explicitly represents a particular element. However, the entire representation, consisting of the four demonstrative representations, can be said to implicitly represent that there are four elements because an implication of (11) is: 12. There are four elements. Number needn't be explicitly represented, in that number never needs to appear as the semantic content that is attributed to a mental representation, rather the implicit content is something that could be extracted or derived given the MSI of the initial representation, given that (11) along with some auxiliary assumptions would entail (12). Only later may the numerical content be explicitly represented. However, for the explanation of behavior, the MSI of the working memory representations of mental files needn't attribute to any single representation numerical semantic content. Notice that for both explicit and implicit representations the information is only made available for further processing provided that the further processes are causally sensitive to the states of the system that encode the information (see Kulvicki (2005) for a discussion of the notion of extractability). Since the implicitly encoded information is carried in virtue of the explicitly encoded information the availability of both sorts of information depends on the availability of the explicitly represented information. Furthermore, in order for any downstream process to act on the encoded information it must be the case that the downstream processes not only are causally sensitive to the encoding properties but 49  that there is a mechanism that treats those encoding properties of the representation in a way that respects their semantic properties. This point is one that was emphasized by Fodor in the defence of methodological solipsism (1980). A representational system in psychological explanation is first and foremost a mechanistic system, the functioning of which is governed by the causal relations amongst the parts of the system (Bechtel 2005, 2011; Craver 2007; Machamer et al 2000). While the interpretation of a representational system might be governed by non-local factors, as in the case of externalist theories, the operation of that system is local and mechanistic. As a result, whatever local properties are being mapped onto the semantic contents of the system must be such that the operations of the system can be explained by appealing to those local properties, as those local properties are the properties that enter into the mechanistic explanations that explain the transitions of the states of the particular system. In this way, the difference between implicit and explicit representation cannot be cashed out in terms of, nor does it even fall along the same lines as, needing further downstream processes for the extraction of information. In order for either explicitly represented information or implicitly represented information to be extracted the representation must be related to downstream systems that are appropriately causally sensitive to the encoding properties of the representation and whose causal operations respect the semantic properties of the encoded information.  2.3. Explicit and dedicated timing mechanisms – internal clocks With the distinction between implicit and explicit representation in hand, we can now turn our attention towards providing a taxonomy of the various general approaches to solving the temporal coordination problem. 50  The two sorts of timing mechanisms described here fall within the explicit timing category; these mechanisms involve states that constitute explicit representations of time (or temporal features). The important difference between these two sorts of models is in how they represent time and what temporal features they explicitly represent. It is a difference in the mechanisms involved, or in other terms, it is a difference in the structure of the representational vehicle, that distinguishes these timing models, and, as we'll see, this difference in the structure of the vehicles results in a semantic difference in what these models are capable of representing.  Period timers  Consider the way in which a standard 12-hour analog clock keeps track of time.  The underlying mechanism behind the functioning of the clock is an oscillatory mechanism, where the oscillator is a mechanism that exhibits periodic behavior with a fixed period.  When one cycle ends, the underlying mechanics of the oscillator guarantee that the system will enter into a subsequent cycle with the same period as the previous cycle.33  Given our time keeping practices, we take the hands of the clock, as they move around the clock, to be explicit representations of the time of day.  Our time keeping practices assign times as the semantic values of the states of the clock – this assignment constitutes the MSI of clocks.  Keep in mind that 12-hour analog clocks are human artifacts. They are technological achievements brought about by the conventional practices of modern cultures. The appeal of period timing mechanisms when employed in cognitive science is that the very same sort of mechanism, an                                                      33 The idea of the “same period” has to be taken somewhat loosely. No clock is perfect, so unless we idealize the functioning of the system, we should talk about periods that average a certain period. For the most part, nothing hangs on this qualification, but there will be places where a stochastic or probabilistic understanding of oscillator periods will become useful – i.e. when explaining variance in timing behaviors by appealing to stochastic processes as a source of noise. See (Wearden, 2001) for a discussion of how variation in the periods of oscillatory mechanisms can be of use in explaining variations in timing behaviors. 51  oscillatory mechanism, can be plausibly appealed to in order to explain psychological phenomena. However, in order to understand the analogy between the cultural artifacts that are clocks and psychological period timers, not only is it important to notice the underlying mechanics of these clocks but we also need to know what the semantic contents of clocks are. At a rough pass, we know that we use clocks to tell us what time of day it is at the moment of reading the clock, but this leaves two possible interpretations of what the semantic contents of clock states are. Let us take a standard 12-hour clock that reads 5:30 as an example. The first way to characterize the content of this system is as follows. 13. It is currently 5:30. 14. It is 5.5 hours since the cycle began. The first reading of the clock has the clock picking out a specific moment in time and saying of that moment that it has the property of being 5:30. The second reading of the clock, however, does not only mention a single moment in time, the present moment, but it states a relation – a temporal interval – that holds between two moments in time. For the time being we can remain neutral as to which of these readings is how we ought to interpret period timers, or clocks more generally.34 We can even remain neutral as to whether there is a single semantic interpretation available for period timers or whether there may be cases in which period timers are used to express a semantic content in line with (1) while other cases may involve a semantic content in line with (2). Finally, we can even remain neutral for now as to what these temporal properties are that are picked out by clocks. For our purposes in this chapter we are merely laying out the mechanisms that are at the heart of these different models. Despite their non-conventional nature, oscillator mechanisms realized in neural systems have many of the same features as clocks. Let's consider a concrete case to help our discussion. Circadian                                                      34 In fact, for the time being we can be neutral as to whether or not these two options are genuinely distinct options. One might cash out the predicate “is currently 5:30” in terms of its being 5.5 hours after the beginning of the cycle. If that is how we understand the property, then (1) and (2) are merely notational variants of one another. However, if we think we can give an analysis of the predicate in (1) that respects the surface appearance of a non-relational property, then (1) and (2) will turn out to have distinct metaphysical commitments. 52  behaviors in mammals are thought to be realized by a neural system that centers on the suprachiasmatic nucleus (SCN).35 Individual neurons in the SCN exhibit oscillatory behavior in their spontaneous firing rates that follow a roughly sinusoidal pattern with an approximately 24-hour period. If we take the firing rate of SCN neurons to be the states of the system that carry information concerning the time of day, then we run into a problem in that considering the firing rates of individual SCN cells introduces a representational ambiguity in the system. For any given firing rate of an SCN neuron, there will be two times of day associated with that firing rate. As a result, the system could not by itself indicate one specific time of day. The situation is similar to seeing a 12-hour analog clock in isolation. The hands of the clock will point to particular places on the clock face, but without any further disambiguating information, you will not be in a position to know whether the clock indicated an AM or PM time and will be unable to provide a precise time as the semantic content (as opposed to a disjunctive content that specifies either AM or PM). There is further information that is needed to properly assign semantic values to the state of clock beyond what is given by merely looking at the clock face. While this situation isn't problematic for clocks, since we often have disambiguating information that allows us to appropriately interpret the clock, the same can't be necessarily said for neural mechanisms.36 Perhaps we needn't appeal to any time keeping devices that have less ambiguity than clocks, however, we have the theoretical machinery to grasp how networks of oscillator mechanism can override this ambiguity. The most influential of these disambiguation techniques is seen in the Church-Broadbent Model (Church & Broadbent, 1990) that instead of employing a single oscillator employs a set of oscillators. Their original model (1990) involved 11 distinct oscillators with different oscillatory periods – the least                                                      35 This system is discussed at length in chapter 3. 36 Perhaps some disambiguating information can be acquired through other sensory means, but for reasons discussed in chapter 3 this way of disambiguating oscillator information will not work for circadian oscillators. The circadian system is realized by an oscillator system that unambiguously indicated times of day.  53  being 0.2 sec and progressively doubling up to 204.8 sec. Temporal information is not encoded by any single oscillator, instead the unit of semantic significance, the unit that is assigned a semantic value, is the state of the entire oscillator array represented as a vector of oscillator phases where each oscillator has only two possible states, + or -.37 Using this set up, we can represent the state of the oscillator network that indicates that 15 seconds has passed as “+1 +1 -1 +1 -1 -1 +1 -1 -1 -1 -1” and 40 seconds as “-1 -1 -1 +1 -1 -1 +1 +1 -1 -1 -1”. Given individual oscillators with a limited number of states, a system of these oscillators can successfully, and unambiguously, represent a number of temporal durations.38 Someone could object here, however, in that the state of the oscillator will still ambiguously pick out (at least) two distinct times. The objection goes that no matter how many distinct oscillators we employ if we consider the operation of the oscillators over a sufficiently long period of time, then the states of the oscillator will ambiguously pick out at least two times. Suppose, that the oscillators cycle through a complete series of states every 24 hours. If we interpret the states of the oscillator over a 24-hour period, then the oscillator states will unambiguously pick out a specific time within that 24-hour window. However, if we were to instead interpret the states of the oscillator over a 48-hour period, then the states of the oscillator will ambiguously pick out two distinct times (say, 10pm today or 10p tomorrow). In order to avoid this ambiguity, we would need some means of restricting the time interval over which the cycles of the oscillator are being interpreted.  Perhaps we could overcome this sort of ambiguity by appealing to external means of demarcating the intervals according to which the oscillator states are being interpreted (e.g. day / night cycles). However, an entirely different approach would be to simply embrace the ambiguity. All that would be                                                      37 One could provide a semantic interpretation of each individual oscillator, however, if each oscillator were interpreted separately, then there would be a much greater level of ambiguity in the content of the system. 38 See (Wearden, 2001) for an overview of this literature. The values and numbers of oscillators in the Church-Broadbent model were arbitrary, but also see (Wearden & Doherty, 1995) for how the choice of number of oscillators and their possible phases are important for the representational abilities of the system.  54  required is that the ambiguity be one that does not cause problems in the use of the system. If for instance the oscillator were used to tell the time of day, and the sort of ambiguity present in the system was one that disjunctively picked out a specific time of day on distinct days, then no behavioral consequences would follow from the particular sort of ambiguity. A single oscillator that ambiguously (or disjunctively) picks out a time every 12 hours would likely have problematic consequences for behavior if the oscillator was supposed to coordinate the activity of the organism with the specific time of day. However, if the ambiguity resulting in a disjunctive content that picked out times separated by 24 hours, then this would be unlikely to lead to any behavioral consequences, as the same time of day would be picked out. Given these techniques for getting around the ambiguity introduced by appealing to a single oscillatory mechanism, we can assume that if period timers are involved, and there is no system independent means to disambiguate the information they carry (i.e. other sensory information such as ambient light levels), then we can assume that they are employing some technique, like the one found in the Church-Broadbent model, to avoid ambiguity within a particular temporal interval. One standout limitation of period timers is that on either of the two possible semantic interpretations, period timers pick out specific times of day by either picking out that time directly (as in option 1) or by picking out that time by noting the temporal interval between some initial moment and the current time. These are the two possible ways of understanding what is explicitly represented by these timers. However, what is often important for animal behaviour is the ability to track the temporal interval between two arbitrarily chosen events.39 If the animal is capable of remembering when two events occur, via noting the state of the oscillator system when those events occur, i.e. noting that a occurred at T1 and b occurred at T2, where 'a' and 'b' are events, and 'T1' and 'T2' are times of day picked out by states of the oscillator, then the combination of these representations will implicitly represent the temporal interval                                                      39 Montemayor (2013) describes these limitations to period timers, although, his notion of explicit representation is the functional notion of needing further processing for the extraction of information and not the semantic notion employed here. 55  between a and b. It is only in conjunction with some memory store that we even get implicit representations of intervals because a memory store is required for the two time of day representations that bound the specific interval, and then this information is only extractable given the appropriate cognitive machinery to extract the interval information.40 Nevertheless, if the period timer is combined with a memory system that records states of the oscillator system and the events that occurred simultaneously with those oscillator states, and the organism is equipped with the rudimentary mathematical ability to calculate interval lengths, then we have a powerful system that allows us to register both specific times of day and interval lengths.  Interval timers The guiding analogy for understanding period timers was the 12-hour analog clock. The guiding analogy for interval timers is the hourglass. Hourglasses measure intervals between two events via the accumulation of some medium that accrues at a fixed rate and that indicates how much time has passed since the hourglass was first “activated”. If you want to measure how long it takes you to complete a puzzle, you can turn over the hourglass when you begin the puzzle, and then by noting how much sand, let's say, has accumulated in the lower bulb of the hourglass you can determine how much time has passed. We can notice two striking differences between hourglasses and 12 hour clocks. First, the mechanisms that are involved, and grant the devices their time keeping abilities, are significantly different. The 12-hour clock keeps track of time by some underlying oscillatory mechanism, whereas the hourglass functions not through any oscillatory mechanism but instead by the accumulation of some                                                      40 For this reason, when period timers are employed to measure intervals, they are embedded within larger networks that invariably include a memory store (see Wearden (2001)). 56  medium.41 Second, hourglasses do not explicitly represent the time of day, rather, hourglasses explicitly represent the temporal interval between two events, a start event (marker event) and end event (target event).42 As a result, to accurately use an hourglass you must be able to recall the earlier event that demarcated the beginning of the timed interval and notice the later event that serves as the end of the time interval. These lessons, comparing hourglasses with 12 hour clocks, can be extended to period timers and interval timers directly. Interval timers are modeled as accumulation processes that mark the temporal interval between a start event (marker event) and an end event (target event). As a result, interval timers require that there be a memory store that can recall the pair of events that demarcate the timed interval. Furthermore, what bears the informational content of the interval timer is the quantity of the accumulated medium – most likely in the case of neurally realized interval timers, the accumulated medium is an aggregation of neuron spikes in some memory store. The function, then, of interval timers is to measure the temporal interval between two events. We can take this description to be giving what Marr (1982) calls the computational level description.43 But what we need to really understand how interval timers differ from period timers is the algorithmic level. The standard model of an interval timer is given by the scalar expectancy theory (Gibbon 1977; Gibbon et al 1984) or more broadly pacemaker-accumulator timing models (Treisman 1963).                                                      41 Wearden (2001) puts the difference in terms of period timers utilizing qualitative encodings of time in which different times are represented by different states of a system with a constant number of elements, versus, interval timers that employ a quantitative encoding of time in which “longer times are represented by more of something” (Wearden 2001). 42 One could of course use an hourglass to tell the time of day by starting the hourglass at the beginning of the day, however, the readout from the hourglass doesn't by itself tell you this. In this case time of day is represented implicitly via the explicit representation of a temporal interval. 43 Marr (1982) distinguished three levels of description that were needed to properly characterize and study the visual system – the computational, the algorithmic, and the implementational. While Marr was specifically concerned with vision, the same explanatory schema can usefully be appealed to throughout the study of the mind and brain. 57  On these approaches the interval timer consists of a pacemaker mechanism and an accumulator mechanism. The pacemaker is a mechanism that produces some medium (some ticks of the clock) at some fixed rate that can be accumulated. In the case of an hourglass, the pacemaker is the upper bulb and the accumulator is the lower bulb with the sand being the accumulated medium. The pacemaker can be modeled as producing ticks at a regular rate (Treisman 1963) or may produce ticks at random but with some average rate provided a long enough interval (Gibbon, 1977).44 Unlike traditional hourglasses, however, interval timers are thought to involve an additional component – a gate or switch that controls the flow of the accumulated medium between the pacemaker and the accumulator. This whole process is then rounded out by a memory store that can record the marker event that opens the gate between the pacemaker and accumulator and the target event that closes the gate. Given that the pacemaker produces ticks at a fixed rate (at least when averaged over a sufficiently long interval), the quantity of ticks recorded by the accumulator carries information about how much time has passed since the gate was opened. As a result, we can interpret the system as providing an explicit representation of the temporal interval between two events.45 Since the point of this chapter is to present these timing models independently of any particular timing task, we would do well to note some of the flaws in these models. These models seem well suited for prospective timing judgments in which the animal knows ahead of time that it should judge how long a specific interval is between two events. To see the reason why the model is useful for these sorts of tasks consider the scenario where I ask you to see how long I can hold my breath. I give you strict instructions that when I start to hold my breath, the timing should commence. I then fidget around for a bit exhale completely and then inhale deeply holding my breath. You can then take the event of my inhaling deeply as the marker event that starts the timer, and then the timer ends when I gasp for breath.                                                      44 One can even construct an interval timer in which the pacemaker is an oscillatory mechanism and the accumulator records how many cycles the oscillator goes through. 45 Notice that the interval timer proper doesn’t represent the events that open and close the gate. Instead, the interval timer merely represents that some interval of some specific length has passed. 58  However, consider the other scenario in which I never give you instructions. The holding of my breath would have been just one among many events that you may have noticed occurring, but without any specific reason for doing so you wouldn't have started a timer based on noticing my having started to hold my breath. As a result, if given a retrospective timing task, in which the animal doesn't know what they should be timing, or even that they are timing anything, the interval timing mechanisms face a problem. We need a reason to open the gate between the ticker and the accumulator. We obviously can't open an interval timer for every event we perceive, but also, it's obvious that many of the timing tasks that are relevant for our behaviour do not involve explicitly prospective timing judgements.46  2.4. No dedicating timing mechanism approaches In order to understand the motivation behind the three approaches to explaining temporal coordination that we're about to discuss we need to take a step back and recall some very general points about scientific theory choice that most people seem to accept.  When confronted with some phenomena that we want to explain, if we have two theories that equally account for the observed phenomena, then, all else being equal, we should accept the simpler of the two theories. So far so good. The tricky business comes in when we try and understand what's meant by a 'simpler theory'. We can distinguish between (at least) two forms of simplicity. The first, ontological simplicity. The second, law simplicity.  A theory is said to enjoy ontological simplicity over another theory if the first theory posits fewer kinds of things in order to explain the phenomena. Famously, David Lewis (1986) argued that his theory of possible worlds realism, in which there exists an enormous number of possible worlds (perhaps even uncountably many) that are just as real as our actual world, was ontologically simpler than many of the                                                      46 Gallistel (1996) gives this line of reasoning that interval timers by themselves are unable to account for a number of timing tasks. 59  actualist accounts of possibility that tried to construct ersatz possible worlds out of abstract entities like properties or propositions.47 Why was Lewis' possible world realism ontologically simpler? Well, even though the theory required that there exist many worlds, which in a sense seems extravagant, the theory didn't multiply the kinds of things out in the world. The non-actual possible worlds are just like our world – they are of the same kind. Along these lines, a psychological theory would enjoy ontological simplicity over another theory if that theory posits fewer kinds of psychological entities. For instance, a theory that posits fewer kinds of representational mechanisms would be ontologically simpler than a theory that posited many different representational mechanisms.  We can say a theory enjoys law simplicity if the laws that explain how the phenomena come about are simpler. This is version of simplicity is to a certain degree more difficult to characterize. Simplicity of laws might be understood in the number of laws required to explain the phenomena. If, for instance, a distinct law is required for every state of a system in order to explain how that system evolves over time, then that theory would be lawfully complex in comparison to a theory that had a very small number of laws that could cover a wide range of possible states of the system.48  Now, when it comes to providing an explanation for the temporal coordination problems, one motivation behind these explanatory approaches is that the explicit timing models are costly in terms of their ontological simplicity. On the explicit and dedicated timer models in addition to the standard perceptual and cognitive machinery that is often posited in order to explain other aspects of animal (and human) psychology, we must also posit an additional type of mental mechanism – a mental timer. And                                                      47 Lewis calls this form of ontological simplicity qualitative simplicity in contrast to quantitative simplicity. Qualitative simplicity is characterized with regards to the number of kinds of entities posited by a theory, whereas quantitative simplicity is characterized with regards to the number of entities posited. 48 Now, we run into a difficulty here, given that we have a simple procedure for creating a theory with fewer laws given a theory that consists of many laws. Take the set of laws from the complex theory, now simply conjoin them in a long conjunction. The conjunction is logically equivalent with the set of individual laws, yet the revised theory consists of a single law. Obviously this won't do as a simpler theory. What is needed is some characterization of simple laws – laws that are not merely the result of conjoining individual laws. See Sober (2006) for a discussion of these issues. 60  not necessarily just one timer, but possibly many. If we could explain the temporal coordination tasks in some way that didn't appeal to dedicated timing mechanisms, then it would seem as though we would have on our hands an ontologically simpler model. We can characterize the two approaches to explaining temporal coordination in relation to the explicit timing models from above by looking at how sparse these No Timer Models are in comparison to the explicit models above.  Intrinsic timing models  Of the no-dedicated timing models we’ll discuss here, the intrinsic timing models are the only ones that still explicitly represent time. According to these approaches, many of the systems involved in the representation of non-temporal aspects of the world (e.g. the color of a flash of light, the pitch of a tone, etc) have dual-contents. They represent the non-temporal features of the world as well as some of the temporal features of that very same aspect of the world. For instance, the very mechanism involved in the representation of a particular tone will also have states that allow it to carry information about the duration of that tone.  According to these models, the ability to keep track of time is ubiquitous in the brain as the ability to encode the temporal features of the world are intrinsic features of neural systems. In this way, there is no single timing mechanism, but rather, a number of timing mechanisms that are hyper-specific in that their timing capacities are tied to very specific and localized neural systems. The particular features of neural systems that intrinsic timing models appeal to differ from theory to theory. State-dependent timing models (discussed in detail in chapter 5) appeal to particular time-dependent spatial patterns of activation in neural systems to encode time (Buonomano 2000; Buonomano & Karmarkar 2002; Finnerty et al 2015; Ivry & Schlerf 2008; Karmarkar & Buonomano 2007). Efficiency coding models (Eagleman & Pariyadath 2009) appeal to the efficiency of neural signaling as a means of encoding duration. Finally, 61  ramping activation models (Lebedev et al 2008) appeal to the time-dependent ramping of neural activity to encode temporal content.  In all cases, whatever intrinsic property the models exploit, the appeal of these models is the ontological austerity of how they describe timing behavior. Dedicating timing models have to posit systems that are in addition to what is required to explain how animals come to represent other non-temporal aspects of their environment, yet intrinsic timing models attempt to explain these timing capacities by appealing to properties of neural systems that we have independent reasons for accepting. Still these intrinsic models all suffer from similar problems. First, since their operations are tied to the particular neural systems involved in representing specific non-temporal aspects of the world, they have difficulty in explaining certain phenomena like cross-modal transfer (see discussion in Ivry & Schlerf (2008)). Specifically, it is well documented that subjects who are trained to discriminate durations with auditory stimuli show a benefit in their ability to discriminate durations in vision. However, the reverse is not true in that duration discrimination training in vision does not improve duration discrimination in audition (Alais & Cass, 2010; Bratzke et al, 2012).49 If, as according to intrinsic timing models, the operations of the individual timing mechanisms are independent of each other, then there simply is no reason to suppose that training in one modality should enhance discrimination in another modality. Furthermore, there is no reason to suppose that there should be any asymmetries between how training benefits transfer across modalities.50 As a result, there seems to be no a priori, or theoretical,                                                      49 Although see Lapid et al (2009) for cases in which cross-modal transfer from audition to vision do not occur. The data in this study, however, is up for interpretation. Other studies have found that that there is a transfer of perceptual learning benefit from audition to vision (Alais & Cass, 2010; Bratzke et al., 2012), however, not only did the Lapid et al study fail to find this transfer, but they were also unable to replicate the within modality perceptual learning benefit that is also commonly found. This indicates that perhaps the failure to find a cross-modal transfer benefit was due to a more general problem with the methodology in their study. 50 If we supposed that the training benefits were the result of a post-perceptual decision process, perhaps something akin to what occurs in signal detection theory, then we might expect there to be cross-modal training benefits as a single (type of) decision process might underpin the perceptual decision in the various modalities, and the training benefit afforded by the enhancement of this decision process might cover all of the various intrinsic time keeping mechanisms. However, we then run into a problem of explaining why there are asymmetries in how readily training benefits transfer across specific modalities. Of course one could likely come up with a story that makes sense of this 62  reason to suppose that training that modifies how one mechanism operates or is interpreted should also influence how other distinct mechanisms operate. Also, intrinsic timing models are plausible for representing only very short durations, as the intrinsic properties they appeal to often are not stable enough to underpin the perception of temporal properties beyond several seconds (at an absolute maximum) (Ivry & Schlerf 2008).  Cued Synchronization Accounts  To get a sense of how cued synchronization models explain how animals overcome the temporal coordination problem, let’s begin with the picture of the mind as given by models that appeal to explicit representations of time. The mind contains a number of sensory / perceptual mechanisms that produce representation of the world. Once these perceptual mechanisms produce some initial representations of the world, then the content encoded in these perceptual representations gets taken up by cognitive and motor processes in order to accomplish a variety of different tasks. Along the way, the content or information provided by perception and manipulated by cognition is integrated with dedicated mechanisms that explicitly represent time and space, thereby providing the information needed to properly orient the organism within its environment. This story should seem old hat, because in a sense, it is very old hat. This just is the traditional representationalist picture of how the mind works and synchronizes behavior with the environment. Cued synchronization accounts agree with almost everything in the old hat picture of the mind except for the need for the explicit representations of time. According to these approaches, all that is needed to account for the mind is all of the non-time-specific machinery of the old hat picture. With the                                                      all, however, the main reason being cited in this argument is that the intrinsic timing models by themselves do not help us understand this aspect of our temporal perception. 63  ability to represent the what and where of things and events in their environment, minds can solve the temporal coordination problem. But how could that be? How could one appropriately coordinate their behavior with their environment if they had no way of explicitly representing time? According to these approaches, we needn't form internal representations of time because the events in the world we interact with themselves exhibit temporal regularities51 – there's no need to represent time because the temporal structure of the environment scaffolds behavior in the appropriate ways needed for the coordination of behavior. For example, one means of explaining the migratory behavior of birds, like the Pied Flycatcher (Ficedula hypoleuca), a small European migratory bird, is to appeal to some internal clock mechanism that lets the bird initiate its migration from Sub-Saharan Africa where it winters to Northern Europe where it breeds. Perhaps the clock is capable of telling the length of the day and that allows the bird to know when it should migrate. Or perhaps the bird possesses a longer-period timing mechanism that allows it to know approximately what time of year it is. In either case, the bird’s migratory behaviors seem to have a regular temporal pattern to them, and some internal clocklike mechanism could explain that pattern. However, as it turns out, there really isn’t any need to posit an internal clock of this sort, since much of the temporal regularity in the bird’s migratory behaviors are due to non-temporal cues that the bird is exposed to. Particularly, it appears that the bird’s migration is due to noticing the temperature changes starting in its wintering grounds in Africa and its stopover location in the Iberian Peninsula (Both et al 2005).52 Since the changes in temperature at these locations have a temporal pattern to them, the bird by cueing into these temperature changes exhibits behaviors that themselves have a temporal pattern to them. The bird is                                                      51 In some cases, the temporal regularities appealed to might be internal endogenous regularities of the organism, but the question would still remain as to what extent this sort of internal cuing could orient the organism properly without appealing to either a representation of time or the external cuing. 52 In fact, the explanation in terms of sensitivity to temperature better explains odd irregularities in the timing of the bird’s migration, as the temperature in the Iberian Peninsula has changed less drastically than the temperature in the ultimate destination of the migration. See discussion in Both et al (2005). 64  successful in coordinating its actions with the temporal structure of its world without any temporal representations.  Now, once we have some animal behaviors that can be explained as being scaffolded from the temporal structure of events in the world, we can explain a number of other animal behaviors that use the initial animal behavior (the one scaffolded from some temporal pattern in the world) as a further scaffold. For instance, if birds begin to migrate due to certain changes in temperatures, then predatory behavior is likely to have a temporal pattern that builds off of the temporal patterns of their prey’s behavior. This building of temporal patterns on top of temporal patterns can continue. Also, given the sheer number of temporally regular patterns in the environment, primarily the light dark cycle, it’s likely that quite a bit of the temporal regularity in the animal world can be explained in this way.53 On these approaches, the animal can still be said to be receiving information about the temporal features of the environment. If, some event, like sunrise, is being represented, and sunrise always occurs at a particular time of day, then by having an internal state that is sensitive to the appearance of the sunrise, then that internal state will implicitly represent the time of day given the contextual facts about the environment. Given that certain events correspond with particular times of the day, the representation of some event's occurrence can be used to infer that it is a particular time of day, given an understanding of the temporal relations. However, and this is an important aspect of the no-timer models, in order to act in a temporally coordinated manner, the organism needn't go through the inferential process to extract the temporal information from the representation of events. Rather, merely by representing the occurrence of an event, and knowing what to do when that event occurs (whenever that may be) the animal will exhibit temporally coordinated behavior. All the animal needs to internalize is a series of hypothetical imperatives, or conditional rules, that describe sequences of behavior and cuing events, such as, <if sunrise, then descend from tree>. Nevertheless, there is a legitimate sense in which the representation of                                                      53 A variation of this cued synchronization account can be found in Killeen & Gregor (1988). 65  events can implicitly represent the temporal features of the environment, in that if the animal possesses the appropriate cognitive machinery, then the animal could make the inference and explicitly represent the temporal information. The cued synchronization accounts have the virtue that if they successfully explain how animal behavior becomes temporally coordinated with its environment, then it will have done so for a small cost. Nothing has to be added to the ontology of the mind. The ability to represent the what and where, an ability that is well studied in all areas of cognitive science, is enough to explain how from the theoretician's perspective animal behavior's seem to track the temporal features of their environments. And as a matter of fact, we can find many temporal regularities in the events in nature that an animal might be sensitive to.  Cued synchronization accounts seem to have some plausibility as it's really easy to find examples where we report the timing of our own behaviors as being the product of external cues. To use a case from my own childhood, as a child, living in the tristate area during the mid-nineties I can remember knowing when dinner was going to be ready without having to look in the kitchen or at a clock. Instead, I had internalized the following rule <when Jeopardy ends, then help set the table>. Looking upon my 10-year old self someone might have been taken aback by the temporal regularities that my behaviors showed, “He helps set the table most every night at 7:30PM”. One way of explaining this is to posit something like an explicit representation of the time of day that is guiding my behavior – perhaps I look at a clock. However, the cued synchronization accounts have an economical, and most importantly, an accurate, explanation of my behavior. My behavior was cued by a certain TV-show and exhibited a temporal regularity not because I was explicitly representing the temporal features of my environment, but because the environment itself possessed a temporal regularity. While in the best cases we can easily point to some cueing stimulus that regulates temporal behavior, the vast majority of cases seem to lack obvious candidates for cueing stimuli. In extreme 66  experimental cases, some timing behaviors persist in conditions in which all possible cueing stimuli have been removed (this line of reasoning is discussed in detail in chapter 3). Furthermore, some behaviors require the ability to keep track of the very contingent temporal structure of events in the immediate environment. With contingent events, perhaps trying to coordinate your behavior with the specific duration of a specific event, there will simply be no pattern in the environment to notice and as a result there can be no learning of any conditional rules that could explain how one coordinates their behaviors on the basis of some event in the world. Finally, in many cases, the cues for behaviors are the very temporal properties the representation of which we are trying to understand. For instance, Russell Church and Warren Meck (Church & Meck 1984; Meck & Church 1983) showed that rats could be trained to expect food at a particular location given either an auditory or visual cue of a particular duration. It is the duration itself that is the cue, and as a result, the animal requires some means of gathering information about the temporal structure of the stimuli. Cued synchronization simply can’t help in this setting since the non-temporal cues fail to provide the required information to the rat. While cued synchronization accounts seem capable of explaining some behaviors, there is a significant difficulty in understanding how this approach can generalize to explain a wider range of temporal properties.  Non-representational Accounts While cued synchronization accounts depart from the orthodoxy in the cognitive science of temporal coordination, they constitute only a minor departure from the traditional representational theory of temporal coordination in their denial of any explicit temporal representations. Whatever picture of the mind that emerges from the cued synchronization accounts, we can be fairly safe in claiming that it doesn't constitute a radical departure from the mainstream. However, more extreme approaches are possible. We can attempt to explain how temporal coordination of behavior occurs in the complete 67  absence of any temporal representations (local non-representationalism) – either explicit or implicit – or even more radically, in the absence of any representations whatsoever (global non-representationalism).  Both the local and global forms of non-representationalism are motivated (in large part) by parsimony reasons. Local non-representationalists claim that intentional explanations of certain psychological phenomena are explanatorily superfluous (Hutto & Myin 2013). Instead, particular temporal coordination tasks are explained in terms of particular organism–environment interactions and the dynamic features of the interaction. If any representational interpretation is possible for the particular system in question, then it happens to be the case that the representational interpretation fails to add to any explanatory goal.  Global non-representationalists motivate their position with many of the same concerns that motivate local forms of non-representationalism. One common approach by globalist non-representationalists is to lay out some conditions that must be satisfied in order for some system to count as representational, and then claim that the mind / brain fails to satisfy those conditions. What criteria are chosen to characterize representationalism varies from anti-representationalist to anti-representationalist and their individual preferences as to what is the most plausible candidate for representationalism.54 This project, of showing that nothing in the mind / brain satisfies the conditions needed to count as a genuinely representational system, is given in conjunction with a positive sketch of how explanations should proceed in the cognitive sciences in the absence of intentional states. This leads us to the second motivation, global non-representationalists argue that there are non-representational ways of describing the functioning of the mind which render any representationalist account of how the mind works as                                                      54 Chemero (2000) directs his argument against a representationalism styled after Millikan (1989). Hutto and Myin (2013) direct their arguments against an information-theoretic account of representationalism styled after Dretske (1981, 1995). In both cases, the presumption is that the arguments weighed against these versions of representationalism scale to argue against all forms of representationalism as the authors take their target opponents as providing the strongest representationalist case. 68  explanatorily unnecessary. Furthermore, it's claimed that the non-representationalist explanations provide better more specific explanations / prediction of behavior. The general feature of all of the non-representationalist approaches is that we shouldn't think of the mind as consisting of systems that manipulates representational states in the way that a computer manipulates computational states. Computation, it is said, occurs in an “abstract” time (van Gelder & Port 1995). The unfolding of the computations has no intrinsic connection with the temporal dynamics of the computation. However, the alternative approach endorsed by the non-representationalists, and especially those coming from the dynamical systems approach to cognition (Chemero  2000; van Gelder & Port 1995) understand the mind to be a dynamic system that is continuously evolving over time, and it is this being in time that is crucial to how minds arise from dynamical systems.55 Temporal coordination then arises not because there is anything like an explicit representation of time, instead, temporal coordination arises due to the dynamical unfolding of the organism and its environment.56 Consider the case in which a batter is receiving the pitch in a baseball game. The pitcher releases the ball, and according to the standard computational story, information picked up by the visual system is used to create a representation of the trajectory of the ball and the rate at which the ball is moving. A motor plan is then constructed instructing the motor systems to swing the bat at a particular moment and at a particular location (or not swing at all). The appropriate command is then sent to the motor systems and the batter swings. The dynamical approach instead would understand the batter as being in a dynamic causal relationship with the oncoming pitch. The light reflected off of the baseball and received by the batter starts a cascade of dynamical interactions between different aspects of the cognitive system – a                                                      55 It’s important to notice, however, that not all dynamicists do away with the notion of representation. Some, for instance (Bechtel, 1998), view the dynamical account as providing a means of implementing a representational system in much the way that earlier connectionists viewed connectionist architectures as implementing a classical computational system. 56 Again, this is only the case for those dynamicists that take the dynamical systems approach to cognition as an alternative to representational or computational models. For those that see dynamical accounts as explaining the implementation of a representational system, the adherence to a dynamical model of cognition is not in conflict with the representational explanation of temporal coordination. 69  feedback loop is produced between motor and perceptual systems. This interaction unfolds over time, and without the need for an explicit representation of when to swing. The successful timing of the swing is a result of the evolution of the dynamical system over time – if the process takes too long, the swing will be late, if the process is too quick, then the swing will be too early. This is an admittedly sketchy explanation of how temporal coordination occurs in a dynamical system approach, one which surely doesn’t do justice to the complexities and sophistication of their explanations, but the goal of this chapter is only to give an overview of the sorts of explanations that are produced by different approaches to temporal coordination problems. To give more details to the picture would require detailing the dynamics of each component cognitive process and detailing the way in which these different components interact. For any given timing task, we can assume that the details of the non-representationalist approach will differ depending on what systems are in causal interaction at a given moment. It's important to note that the coordination between the contingencies of the environment and animal behavior is brought about by the causal connections that hold between the stimuli that impinge upon the sensory systems and the dynamical systems internal to the organism. In this sense, if the non-representationalist approach to timing tasks is successful, then we will have an extremely economic theory of the mind, but the problems arise in cases where it seems that the environment does not provide sufficient stimulation to properly constrain and scaffold behavior. This same problem arose for the cued-synchronization accounts. So, if there aren’t even enough stimuli to allow cued-synchronization accounts to get off the ground, then the non-representationalist versions will also suffer.  70  2.5. Summing up  While this hasn’t been intended as an exhaustive survey of the different explanatory approaches available for explaining how animal behavior comes to be coordinated with the temporal structure of the environment, it nevertheless gives us the background to go forward and investigate the various ways in which animals overcome the temporal coordination problem. In providing this taxonomy of positions, the focus on the distinction between ways of representing time (i.e. the explicit / implicit distinction) provides us not only with the basis for forming the taxonomy but for having a better understanding of how mental representations should be understood. In particular, the aspects of mental representation discussed here – how it makes animal behavior intelligible and the requirement that semantically significant states of the system be ones that can exert causal influence on downstream systems – will be put to work in later chapters. Something to emphasize throughout the project of this thesis is to not take these models to be in all out competition. No single approach is correct tout court. Instead, some approaches may be successful in explaining certain psychological tasks whereas other approaches may be successful in others. The only competition that arises between the various approaches one might take is when multiple approaches attempt to explain a single phenomenon. Yet, as we’ll see in chapter 5, even cases in which a single phenomenon seems to be under investigation, and thereby engendering a conflict between different models, it may turn out that what initially appears to be a single unified explanandum turns out to fragment into distinct phenomena to be explained independently.    71  Chapter 3: The Sense of Time  Most of cognitive science and philosophy of mind understands the architecture of the mind as having three main divisions. Sensory systems acquire information from the environment, cognitive systems extract and manipulate information from the sensory systems to allow for flexible thought and stimulus independent thinking, and motor systems translate the information carried by sensory and cognitive systems into motor signals that move creatures. With this general architectural picture in place we can ask about any mental process or system where in the cognitive architecture that process sits – is it a sensory, cognitive, motor, or some hybrid process? This paper will begin by asking where in the cognitive architecture we can find the representation of time.  Our initial starting point is the following: across the animal kingdom, and within any given individual creature, there are likely to be a number of different means by which the temporal structure of the world is represented. It’s known that animal behaviors are sensitive to the temporal structure of their world over an incredibly wide time scale – over 10 orders of magnitude (Buhusi & Meck 2005). Humans, aided by cultural, mathematical, and technological achievements, can synchronize their behaviors with even greater time scales of incredible length and with incredibly high precision. In addition to the very different time scales that can be mentally represented, the temporal properties themselves that are mentally represented can be distinguished both semantically and in terms of the mechanisms that are involved in their representation. Semantically, representations of durations (e.g. being 5 seconds long) and locations in temporal sequences (e.g. being earlier than or later than) have a predicative role. They are applied to objects or events in much the same way that we apply predicates like being green or being smaller than. On the other hand, representations of particular moments in time (e.g. now or March 15, 1992) have a referential role. These representations refer to moments in time in a way that allows us to predicate properties of those times. Logically, or semantically, the representation of moments in time operates in the same way as other referential representations such as here or Newark, NJ. Mechanistically, 72  we also know that various aspects of our ability to mentally represent time come apart from one another since they are underpinned by distinct neural mechanisms.57  We know that there are a variety of different means by which we mentally represent time. What we do not know, at the moment, is just how varied this underlying machinery in fact is. Given that there is such a variation in how we mentally represent time, there simply is no guarantee, prima facie or not, for thinking that all of the various components involved in temporal representation will fall within one or another category of the overall cognitive architecture. So, the question that we should ask is not where in the cognitive architecture does the mental representation of time belong but rather where in the cognitive architecture does a specific capacity for the mental representation of time belong.  Besides the question of in which division of the cognitive architecture some mental capacity belongs another broad question about cognitive architecture concerns the domain specificity or generality of any given mental process or system. Are the psychological capacities we know from introspection or through the observation of behavior the product of mechanisms or processes that are specific and dedicated to the capacities in question or do they follow from a more general capacity? Are there dedicated motor abilities for walking? Or do the movements involved in walking simply come from a more general motor ability to move one’s legs? Similarly, is the sensory capacity for visual form detection due to a task specific mechanism or does the capacity arise from a more general visual capacity, say for the detection of luminance contrast?  In this chapter we are going to ask two questions. First, in which broad category (or categories) of the cognitive architecture will we find the particular capacities for the mental representation of time? Second, within any one of those broad architectural categories, is the capacity due to a task specific mechanism or not? In this paper, I will argue that at least some of our capacity for the mental                                                      57 See Buhusi and Meck (2005) and Hinton and Meck (1997) for an overview of how the mechanisms involved in various aspects of the mental representation of time come apart. However, for more detailed discussions see the arguments for the fragmentary model of temporal perception given in chapter 4 and 5. 73  representation of time is due to a genuine sense of time. That is, there is a task specific, and dedicated, sensory mechanism for the mental representation of time. The chapter will go as follows: In section 1, I will provide an account of how to distinguish the senses from the rest of the cognitive architecture in terms of their role in an information processing architecture. In section 2, I will turn to arguments in the philosophical literature that attempt to show that there is not (or simply cannot be) a genuine sense of time. In section 3, I show how none of the arguments against the sense of time presented in section 2 succeed in establishing their intended conclusion. Finally, in section 4, I argue that when we turn to the literature on the circadian systems of mammals with the characterization of sensory systems given in section 1 that we find good reasons for supposing that mammals, as well as many other animals, have a genuine sense of time.  3.1. The senses in an information processing architecture  Folk wisdom, and philosophical tradition stretching back at least as far as Aristotle’s De Anima (2004), has it that there are five and only five senses – touch, taste, vision, smell, and hearing. However, like many aspects of our folk understanding of psychology, the notion of a sensory system has come to play an important role in our scientific theorizing about the mind. Within those scientific practices that study, or appeal to, the sensory systems, it is common to find attributions of senses that go beyond the classic five Aristotelean senses. Given the wide range of uses of ‘sense’, in both folk and scientific contexts, it’s not surprising that some philosophers have attempted to argue that the uses of ‘sense’ in scientific contexts to describe non-Aristotelean senses in a way manages to change the subject (Nudds 2004). The senses, they argue, are in some way analytically restricted to the five classic Aristotelean senses.  74   Now, it may strike some as though the disagreement as to whether there are only 5 or more senses is something of a verbal mistake. Defenders of only applying the term ‘sense’ to the Aristotelean senses simply have a more restrictive notion of ‘sense’ than those scientists that apply the term more liberally. All that is required, it might seem, to avoid this conflict is to simply specify how to apply our terms, and there are coherent accounts of the senses that apply to only the Aristotelean senses and coherent accounts that apply to the more liberal usage of ‘sense’. In this section, I will argue that the debate between the uses of ‘sense’ is not a merely verbal dispute and that liberal uses of ‘sense’ are not merely changing the subject. There is importantly a common core that lies at the heart of both the more restricted and liberal uses of ‘sense’ and it is by acknowledging this common core that we gain an understanding of how the senses function in our understanding, both scientific and folk, of the mind. In both folk and scientific cases, the notion of a sense is used to describe those aspects of the cognitive architecture that provide an immediate and ongoing tracking of the world and function as the initial stage in the information gathering process of the mind. The disagreement, however, simply arises in how deep into the cognitive architecture the independent sensory systems are supposed to extend and how these sensory systems are individuated from one another.  In order to see this common core, we need to distinguish between two questions that are often confused in the literature.58 First, what distinguishes the senses from the rest of the cognitive architecture? In order to answer this question, we need to understand what is distinctive of the sensory systems as a psychological kind. An answer to this question will tell us in virtue of what audition and vision are of the same psychological kind while both being different in kind from motor systems. Second, what distinguishes one sensory system from another? In order to answer this question, we need to know something other than what is needed to answer the first. An answer to this question will tell us in virtue of what audition and vision, while being of the same broad psychological kind, are nevertheless distinct                                                      58 Matthen (2015) makes a distinction between these two questions in his overview of the philosophical literature on the senses. 75  types of sensory systems. As we shall see, part of the reluctance by some philosophers to admit of non-Aristotelean sensory systems is the result of confusing these two questions. Both folk and scientific understandings of sensory systems give roughly the same answer to the first question – the senses, as opposed to other aspects of the cognitive architecture, are involved in explaining how organisms gather information about the world. Where folk and scientific uses of ‘sense’ differ is in how they understand the individuation question. They appeal to distinct means of individuating the system. Yet, I will not argue that any one way of individuating the sensory systems is the correct way. Rather, there are a number of distinct ways of individuating the sensory systems that are responsive to our particular explanatory aims, and these means of individuating the sensory systems are closely tied to how to fully cash out what distinguishes the sensory systems from the rest of the cognitive architecture. As a first step towards understanding these questions, let’s turn to some examples of how the senses are attributed in scientific and folk explanations.  One of the overarching goals of the cognitive sciences is to explain behavior. Crucial to this explanatory task is the explanation of how individual creatures are able to coordinate their behaviors with the contingencies of their environments. It is in the service of explaining this coordination that sensory systems are attributed to organisms as a means of explaining the initial and ongoing contact that the mind makes with the world.  Let’s first look at research on the feeding habits of elasmobranch fish (these include sharks, skate, and rays).59 These fish have what once seemed a puzzling ability to feed on creatures that are hidden in the sand on the ocean floor. What puzzled researchers about this feeding ability is how these fish were able to detect their prey. None of the classic Aristotelean senses seemed to be capable of providing the fish with the information required to locate their prey. However, through a series of nicely controlled behavioral and anatomical studies, it was determined that the information needed to find their prey was                                                      59 See Kalmijn (1982) for a detailed discussion of these capacities. 76  not provided by any of the Aristotelean senses, but was instead the product of a distinct system that could detect the electrical fields produced by the prey. As a result, researchers began to attribute to these fish an electroreception sense. That is, they began to attribute a distinct sense that would explain how a certain sort of environmental information could be gathered by these organisms in order to drive subsequent behavior.  A somewhat different example comes from research on human proprioception (Proske & Gandevia 2012). Even in the absence of other possible sources of sensory input humans possess the ability to know the relative position of their own body parts. Even in the absence of any light, you can know whether your hand is right in front of your face simply by having a sensation of where your body parts are. As it turns out, this fact, that we can know where our body parts are, cannot be explained on the basis of information from the classic Aristotelean senses, but instead, requires an appeal to a novel source of sensory information based in a system that monitors the location of the body. Researchers working on this topic have called this novel sensory system the proprioceptive system.   Another, more controversial example, is found in the literature on the human vomeronasal system.60 It is known that a number of animals possess a distinct sensory system for the detection of the hormones of conspecifics called the vomeronasal system (Halpern 1987). Originally, while it was thought that human fetuses had a vomeronasal system (or at least the remnants of one), it was widely accepted that adults lacked a functioning vomeronasal system. It was thought that whatever system the fetus might have had was lost during development. However, more recent research has revealed that adult humans may in fact possess an intact and functioning vomeronasal system. Careful anatomical and behavioral studies (Taylor 1994) have shown that not only is there the anatomical machinery that would be expected for a vomeronasal system, but that activation of this system has behavioral consequences. Importantly,                                                      60 A longer discussion of the vomeronasal system and how its existence impacts philosophical discussions concerning the role of phenomenology in the individuation of sensory systems can be found in (Keeley, 2002). 77  this research shows that adult humans may possess a novel, non-Aristotelean, system for gathering information that is not made apparent through introspection. The discovery of this system was a thoroughly empirical one as there seems to be no notable phenomenal aspect associated with the operation of this system. Researchers in discussing this system have begun to call it the vomeronasal sense.   And finally, let us consider a case from science fiction. It’s a common device in fiction to describe certain people as having a sixth sense. A sensory system that goes beyond the classic five Aristotelean senses and often is used to describe how someone is capable of having some mysterious and supernatural epistemic ability. In the movie The Sixth Sense, for instance, the main child actor has the ability to see dead people. In other movies, the term sixth sense gets used to describe people who can know when a disaster is going to happen or who have an uncanny ability to tell when someone is lying or bluffing at a hand of poker. The details about what is detected don’t really matter, but what is important is the context in which a sixth sense is posited. In all of these cases, the additional sense is posited when people try to explain how some individual has the ability to know or detect something about the world that does not seem to be explainable by simply appealing to the Aristotelean senses. Nothing about neuroanatomy is involved in these cases. Also, nothing introspective seems to be involved either, as the attributions are often made to a third person. Rather, the basis for the attribution of a novel sense is due to explanatory pressures of explaining human behavior.  It is here where we find the common-core to both the folk and scientific uses of the notion of a sensory system. In all cases, the senses are attributed in situations where we need to explain how the individual organism comes to carry and act upon information about some aspect of the world. If we find that the organism is acting on some contingent aspect of its environment, and we cannot explain how the organism has information about these contingencies by appealing to the senses we already attribute to the organism, then we posit a novel sensory system. However, the senses cannot be distinguished from the 78  rest of the cognitive architecture by merely claiming that they acquire information about the environment, since other, non-sensory systems do this. The situation is a little more delicate.  The sensory systems do not merely carry information about the environment, they directly track the state of the environment as it changes. They provide an ongoing epistemic contact with the environment. The accurate deployment of a concept in a thought will also carry information about the environment. When I have the structured conceptual thought BEAR THERE NOW, the deployment of these concepts carries information about the state of the world. However, there is a way in which these conceptual resources do not track the world, but are deployed when other systems that do track the world provide the right sort of triggering information. For instance, I might deploy a thought of this sort when I see a large brown fuzzy creature standing on the trail in front of me. Furthermore, none of the information given to me by vision might be in some way identical to the information given to me by my thought BEAR THERE NOW. That is, the information given to me by my thoughts, that there is a bear there, needn’t be identical to or reducible to the information given to me by vision, that there is a large brown fuzzy thing there. So, while the deployment of BEAR THERE NOW might carry information that is not found in any other mental representation, it is nevertheless the case that the deployment of these concepts are under the control of information provided by systems that directly track the state of the world. Concepts, and other mental representations, might carry information that is not found elsewhere in the cognitive system, but they are nevertheless triggered or come to be deployed so as to accurately track the world, through information that is given through purely sensory means.61 Sensory systems, are in a manner of speaking, on the frontlines of the information processing system. The initial representations                                                      61 Some application of concepts might come to track the world through non-sensory triggering conditions. For instance, we may succeed in inferring the existence of something in the world based on purely rational grounds. In this case, we have a situation in which the deployment of a concept accurately tracks the world, but not on the basis of sensory information. 79  deployed in the sensory systems are triggered through means that do not rely on other representational states for their deployment.  A useful point of contrast here is with the so-called number sense (Dehaene 1997). I say ‘so-called’ because according to the understanding of sensory systems as systems that directly track the environment, the number sense is not in fact a sense. Dehaene takes the number sense to consist in a variety of systems that allow for rudimentary mathematical thought, such as the ability to quickly tell the approximate number of items in a collection. However, what distinguishes the “number sense” from the genuine senses is that the numerical representations that it produces are subservient to, or triggered by, the information carried by representations in more general sensory systems. When I quickly judge that one plate of cookies has more cookies than another plate (or that a plate has more cookies than the number of cookies I can feel inside of a black box), I deploy some of the systems included in the number sense, but it is on the basis of information gathered by vision that I make my mathematical judgment. The operation of the “number sense” requires the operation of the other sensory systems. A genuine sense would not require information gathered from other systems in this way.  A complication, arises, however. We know that there are a number of cross-modal interactions between sensory systems at even very early levels of processing (Eckert et al 2008; Falchier et al 2002). For instance, auditory information can influence how the visual system segments the visual scene into discrete objects over time (Shams et al 2000). This takes us into the very next problem in giving a general account of the senses. Sensory systems serve on the frontline of the information processing architecture of the mind, but there is a question about how deep into the cognitive architecture the sensory systems extend.  Consider a simplified account of the relation between sensation, cognition, and action (figure #1), in which we have the initial transducer systems that feed into a complicated system of perceptual processes that extract information from the activity of the transducers. This perceptual information is then 80  transferred to cognition for further processing and cross-modal integration, and then it is delivered to the motor systems to drive behavior.  Figure 1: The simplified feedforward architecture. Transducer activity feeds into several layers of sensory / perceptual processing, which feed into central cognition, which ultimately feeds into the motor systems. On the account of the sensory systems that we have been developing so far, with the senses as the initial stage in the information processing architecture, the distinction between sensory systems and the rest of the cognitive architecture is simple to draw. We simply draw a vertical line prior to the stage at which sensory information becomes integrated in central cognition. The individual senses remain relatively isolated and stimulus driven as we move from transducers to central cognition.  Unfortunately, though, the architecture pictured in figure 1 simply doesn’t exist. We have two reasons for thinking this. First, the operation of even “early” visual areas seems to be sensitive to the re-81  entrant processing found in “later” visual areas.62 Evidence for re-entrant processing has been found in both behavioral studies (e.g. where higher-level task goals influence early sensory processing (Watanabe et al 1998)) and anatomical studies (e.g. neural projections from frontal areas to earlier sensory areas are common (Clavagnier et al 2004)). Second, we know that not only do “later” areas influence “earlier” areas, but that there are a number of cross-modal influences in which we have lateral influences from “early” areas of one system on the “early” areas of other systems. For instance, hearing two clicks will influence how the early visual system processes a flash of light. In the absence of the two auditory clicks, the visual system will parse the flash of light as a single flash. However, in the presence of the two auditory clicks, the visual system will parse the flash of light as two distinct flashes (Shams et al 2000). Furthermore, there are neuroanatomical studies that show cross-modal projections between early sensory areas (Eckert et al 2008; Falchier et al 2002). With all of this informational cross talk and back chatter the resulting picture of the architecture is less like that in figure 1 and more like that in figure 2.                                                      62 “Early” and “late” need scare quotes here since once you admit of re-entrant processing, then the sense in which sensory processing can be divided into early and late starts to fall apart as you trace the flow of information in the system. 82   Figure 2: The complex interactive architecture. The information streams leading from the transducers are not segregated throughout the subsequent sensory / perceptual processing. There is extensive horizontal and feedback influences.  Whatever picture of the cognitive architecture we might adopt has to be one that allows for systems to influence each other at very early processing stages. So, if we are to draw the sensory / cognition boundary at a stage of processing that occurs prior to integration, then we seem to be forced to draw the boundary between sensory systems and the rest of the cognitive architecture at an incredibly early stage of processing – perhaps as far as the sensory transducers. This might be an acceptable place to draw the line between sensory systems and the rest of the cognitive architecture, but it seems difficult to maintain in light of how the notion of a sensory system actually gets used. Just look at any textbook on perceptual psychology and they will have chapters devoted to individual sensory systems and in their discussion of these systems they will include stages of sub-cortical and cortical processing.63 The distinction between the sensory systems and cognition seems to be drawn at some intermediary point                                                      63 Really almost any psychology textbook will do. 83  between the transduction organs, which are clearly sensory, and some cognitive or associative areas that are (by most lights64) non-sensory. However, as I will now argue, where we draw the line between sensory systems and cognition is not a fixed one, but is importantly tied up with drawing a line for the individuation of the specific sensory systems. The two questions, while conceptually distinct, come together in a way that each plays a role in determining the other.  How then are sensory systems individuated? What distinguishes one sense from another and what makes my visual system the same type of sensory system as yours or a dog’s? Being able to answer this last question is crucial to many scientific projects. It’s assumed by developmental cognitive scientists that the visual system (or any sensory system) they study in the developing neonate be the same sort of sensory system, namely a visual system, as the mature visual system whose development they wish to explain (despite the fact that there might be radical differences between the visual systems of the neonate and the mature individual). It is in fact the development of the visual system that they study. It’s vital to psychology as a whole that the visual system in one person be the same in type as the visual system in other people, even with some abnormalities, in order to make any sort of generalization about sensory systems across the human population65. Finally, it’s vital to many ethologists that they be able to make generalizations or claims about sensory system types across the species boundary. For instance, many claims about the human visual system come from studies of the visual system of macaques and cats. Even more so are researchers that study the development of very general sensory capacities, for instance those who study things as general as the development of color vision (Braddick & Atkinson 2011; Brown                                                      64 Some people think that everything the brain does is sensory. These concept empiricists come in many guises, but for all intents and purposes they all agree that even amodal association areas should actually be interpreted as multisensory association areas which importantly do not count depart from processing purely sensory information. Examples of this sort of view are found in (Barsalou, 1999; Langacker, 1986; Prinz, 2002). For a critical discussion of the attempt to view these association areas as being multisensory as opposed to amodal (and therefore non-sensory) see Weiskopf (2007). 65 Being of the same type does not require that the visual systems of the distinct individuals be exactly the same or identical. Variation is allowed. However, there must be something in common that makes the visual systems of distinct individual’s visual systems and not something else. It’s this something in common that we’re after. 84  1990), whose research again requires that sensory systems be typed in a way that individual token sensory systems (the actual sensory systems realized in individual brains) can all be type-identical. If we are to give an account of how sensory systems are individuated (at least in our scientific explanations of the mind), then we must respect these taxonomic boundaries, since these boundaries are crucial to the explanatory tasks for which the senses are posited.66  Perhaps one of the most influential recent philosophical pieces on how to individuate the senses is Brian Keeley’s Making Sense of the Senses: Individuating Modalities in Humans and Other Animals (2002). In his paper he does not distinguish between the question of what distinguishes the senses from the rest of the cognitive architecture and the question of how to individuate the senses from one another, but his paper still provides us with a helpful starting point for discussing the question of individuation. Keeley proposes four criteria that can be used for individuating the senses. They are: Physics: each sense responds to a distinct aspect of the environment as determined by physics. Neurobiology: each senses is constituted by a distinct neurobiological mechanism. Dedication: the sense must have the function of detecting the magnitudes specified by physics. Behavior: the information acquired must be used for guiding behavior. If we look at these criteria we find that only the first three, physics, neurobiology, and dedication, can be used to individuate the senses. The fourth, the behavior condition, actually plays no role in individuating the senses. Keeley, in stating this condition, is claiming that whatever system we are considering as a potential sensory system must be capable of driving behavior. Since all sensory systems are posited in                                                      66 Of course a philosopher could jump in here and insist that scientists are individuating the senses in a way that is simply incoherent. Scientists, they might say, are simply getting things wrong and they should stop. However, as we’ll see there is nothing more incoherent in how scientists individuate the senses as do some philosophers. That is, there is nothing incoherent here, and unless there are outright contradictions or insurmountable theoretical puzzles that emerge from some use of language, then we should refrain from taking a normative approach and policing language use. 85  order to explain some sort of behavior, this condition does not distinguish between sensory types.67 It is really the job of the other three criteria to individuate the senses. Let’s take the simple (or at least highly studied!) examples of vision and audition and see how these criteria are supposed to function.  Vision and audition respond to distinct types of energy in the environment.68 Vision responds to electromagnetic radiation within a certain range. Audition responds to mechanical energy. However, it is unclear whether the physics criterion is enough to individuate the senses since both vision and audition seem to be responsive to the same energy types as other sensory systems. The visual system can also be activated by mechanical energy (use your finger to push on the side of your eye). Also, the visual system, while it is responsive to electromagnetic energy, it is not the only sensory system that is responsive to this feature of the world. Thermoception capacities in animals like pit vipers (Buning 1983; Gracheva et al 2010) also are responsive to electromagnetic energy, although of a distinct range. If we are simply individuating sensory types in virtue of the energy types, as defined by physics, that they respond to, then our visual systems would be of the same type as the thermoreceptive system of pit vipers. Perhaps one could insist that they are of the same type, although any insistence of this sort obviously seems strained. Turning to audition, similar worries arise, and in this case, they arise within humans. The mechanical waves detected by audition can also be felt (if they are strong enough), yet it is clear that we should not admit that audition and touch are in fact one sense. That would flout any of the distinctions that appeals to sensory systems require.                                                      67 The behavior condition plays a specific role for Keeley. He’s concerned with the role of phenomenology in understanding the mind and focuses on theories of sensory individuation that give a central role to phenomenology. However, as we’ll see later in this chapter, Keeley gives reasons why phenomenology cannot be used to individuate the senses, when they are used in scientific contexts, because while we attribute sensory systems to a range of animals we simply have no idea what their phenomenology is like. So, instead of using phenomenology to understand the senses we must understand them in terms of the behavioral effects they have on the organism. It is here that the behavior condition becomes crucial. Sensory systems are not just systems that are responsive to the environment, but they are ones that can drive animal behavior. In humans, this sometimes comes along with some distinctive phenomenology, as in the case of the Aristotelean senses, but we needn’t tie the sensory systems to this phenomenology. Rather, it is their ability to drive behavior, along with the other criteria, that makes something a genuine sense. 68 The arguments in this paragraph and the next are closely related to those given by Macpherson (2011).  86   So, the physics criterion by itself won’t work. But perhaps we can augment it by adding in the neurobiology criterion. Audition and touch might both respond to the same types of energy. And so do vision and thermoception. However, in both cases we have very distinct neurobiological systems at play. Take vision and thermoception. Both involve radically different sorts of transduction organs. The visual system of all animals have transducer mechanisms that employ photosensitive pigments whereas the heat sensors of the pit viper actually involve temperature changes in the receptor membrane that directly influences ion channels leading to changes in neural activity (Bullock & Cowles 1952; Gracheva et al 2010). Similar considerations apply to the audition / touch case. Both involve significantly different sensory organs. And yet, we again find reasons for thinking that neurobiology isn’t enough. Take vision. The retina, the end organ of the visual system, is not a homogenous mass of photoreceptors. In fact, the retina of most humans contains 4 standard photoreceptor types – rods, S-cones, M-cones, and L-cones (Solomon & Lennie 2007), as well as photosensitive retinal ganglion cells that possess their own photosensitive pigment, melanopsin. But, not all human visual systems contain this array of photoreceptor types. Dichromats lack one of the cone types. Monochromats lack two of the cone types. Anomalous trichromats, while possessing three distinct and functional cone types, have photoreceptors that are tuned to different wavelengths than the cones of the typical human. Even more odd are cases exist in which the female offspring of a male anomalous trichromat will sometimes have 4 functional cone types (Jordan et al 2010; Jordan & Mollon 1993). All of these cases point to sometimes radical differences in the visual systems of humans. The neurobiology is different in all of these cases. If we consider non-human visual systems, then things get even more complex. Animals across the animal kingdom possess a wide variety of different photoreceptor types, mantis shrimp have up to 16! (Chiao et al 2000; Cronin et al 2001; Marshall & Oberwinkler 1999), and these animals all possess different systems that process the initial sensory information. And yet, in all of these cases, we still often describe these various sensory systems as all being of the same type. That is, despite their neurobiological differences we still consider them visual systems. 87   But, what binds the visual system in any single animal as being a single system? Why not consider the distinct photoreceptor types as giving rise to distinct sensory systems? Perhaps we could conceptualize the human visual system as containing 4 distinct sensory systems (one for each photoreceptor type), or 2 sensory systems (one for the rods and another for the cones), or perhaps we could distinguish sensory systems according to their projections to the LGN (so, there could be three systems, the magnocellular, parvocellular, and koniocellular sensory system). Perhaps one approach to unifying the visual system into a single sensory system would appeal to the similarities between the photoreceptors. One salient similarity is simply that they all employ opsin molecules to convert light into neural activity. In this regard there is a neurobiological fact that would group all of the distinct types of photoreceptors as being of a single kind. But, opsin molecules are also found in the pineal gland, so the presence of opsin molecules is not sufficient either for making the visual system distinct from other parts of the cognitive architecture (Peirson et al 2009; Velarde et al 2005). Furthermore, we could equally emphasize the differences in these biological system. They all employ opsin molecules that have peak sensitivities to distinct wavelengths of light. They also enter into different post-receptor processes (e.g. the M and L cones enter into the magnocellular and parvocellular systems, while the S cones enter into the koniocellular system).69 Maybe another neurobiological similarity can be found that appropriately individuates sensory systems such that we have a sense of vision that is of the same type as the sense of vision possessed by a dog. We can actually stay neutral with regards to this point.70  If we add the dedication criterion things start to change. We can nicely describe the photoreceptors in the retina as having a common function. They are all dedicated to detecting luminance and color, or dedicated to detecting the visual features of the world, or dedicated to the detection of distal                                                      69 There might be some reason to doubt this initial parsing of the human visual system into distinct pathways of this sort as there may be reason to think that there is a mixing of cone type information in the distinct retinal-LGN pathways in infancy. See Dobkins & Anderson (2002). 70 My own two cents on this are that I believe that quite a bit can be made from this neurobiological criterion for sensory individuation.  88  objects and their properties. In this way, all of the photoreceptors can be grouped within a single sensory system. Now, audition also has the function of detecting the distal world, and yet, we needn’t group it together with vision since they differ with regards to the other two criteria.71  We seem to having something that works here. We can individuate the senses by employing a host of different criteria, each of which is insufficient for the task when taken individually, but jointly succeed in distinguishing the sense. However, the success of these criteria for individuation depend on where we draw the boundary between sensory systems and the rest of the cognitive architecture. If we draw the sensory system / cognition border immediately after the transducers, then we run into problems with all three criteria. The physics criterion becomes difficult to maintain since, just taking vision as an example, all of the photoreceptor types while responsive to electromagnetic energy are responsive to distinct ranges of the EM spectrum. Similarly, the neurobiology criterion becomes difficult to maintain as the distinct photoreceptors are all distinct. And finally, the dedication criterion becomes particularly difficult to maintain as it’s unclear what single dedicated function each of the different photoreceptors possess. Saying that they are all for seeing the world won’t work, since that presupposes the very thing that we are trying to understand. We would need a non-question begging understanding of what it is to see the world and such an understanding could not employ the fact that seeing is done by vision. An evolutionary account of function also won’t work as the distinct photoreceptors emerged various points in our evolutionary history in response to evolving ecological needs (Jacobs 2009). Similarly, if we adopt a historical approach to determining function (or for determining some neurobiological criterion), then we run into problems in that different evolutionary paths that stem from a common ancestor were driven by radically difference ecological situations (e.g. the octopus and humans) (Gehring 2014). If we draw the                                                      71 How we cash out the function of the sensory systems is actually a very tricky issue to figure out. Akins (K. Akins, 2014) nicely argues that in many cases we cannot understand the sense as picking out distal features of the world. Instead, she argues that the sensory systems track narcissistic properties that importantly are not found out in the mind independent world described by physics but are instead only found in relation to the organism doing the sensing. This is a thorny issue, especially for an account of the senses like Keeley’s, but as we’ll see we needn’t decide this matter for what it going to come in this chapter.  89  sensory system / cognition boundary at the periphery, then we seem to have to individuate the senses at an incredibly fine-grain which would do damage to divisions that we wanted to respect.  But, of course, nobody would draw the sensory system / cognition boundary so close to the periphery. We need to look a little deeper into the information processing architecture (with regards to figure 2, we need to move the vertical division separating sensory systems from cognition farther to the right). Yet when we do that we run into further problems.   Once we begin to draw the boundary between sensory systems and cognition further into the processing stream we get further difficulties in individuating the senses. We want to avoid the problem of carving the senses too finely, but at this point we now have to deal with the ubiquitous cross-modal interactions that happen between cortical processes. A purely neurobiological account may be able to work if we choose to provide a coarse-grained account of neural anatomy. Audition might be just that sensory system that extends from the ears through the auditory cortex. Similarly, vision could just be that sensory system that extends from the eyes through the visual cortex. Now, in order to do this there better be a non-question begging way of individuating the auditory and visual cortices that do not appeal to a prior definition of the sensory systems. Perhaps they are the regions that receive input only from particular transducers, but it’s unlikely that that criterion could hold, since we know that cross-modal interactions influence processing in early sensory areas. Perhaps there is some other anatomical marker that could cordon off the appropriate cortical areas, but the worry here is whatever marker we adopt would likely also separate sub-regions of the traditionally defined auditory and visual cortices. Considering just the visual system, we find a number of distinct anatomical structures in various areas of the visual system (e.g. the striation in V1 and V2, the blobs and interblobs of early visual areas, various retinotopic maps vs tonotopic maps, etc), however, none of these anatomical structures are capable of carving off the entirety of the visual system. If we took these anatomical structures as means of distinguishing sensory types, then the visual system would fail to be a single sense. If we take a more 90  general type of anatomical structure, for instance the presence of projections from the thalamus, then we end up grouping together disparate sensory systems as the presence of these projections is common throughout the various sensory cortical areas. The result is either that we lump together supposedly disparate sensory systems, or we again carve them too finely.  The problem of jointly using many criteria to individuate the senses is that at any level that we carve off the sensory systems from cognition, at least one of these criteria seems like it will carve the senses too finely. It will distinguish between types of sensory system, when we want to think of those sensory systems as being of a single type. Adding further criteria will simply be of no help in those situations since they can only help to carve things even more finely.   Keeley’s aren’t the only criteria in town, though. There is another approach to individuating the senses that has a long and venerable tradition within philosophy according to which the senses are individuated according to some introspective criteria. A number of views fall under this heading, but they all share in common the idea that what is distinctive of each individual sensory system is some property of experience that is available to us introspectively. According to some, the senses are individuated by their distinctive contents (Aristotle 2004; Dretske 1995; O’Dea 2011). According to this content criterion, audition and vision are distinct sensory systems since they represent different features of the environment and this representational difference is something that we can be aware of through introspection. According to others, the introspectible property that individuates the senses is the phenomenal character that is associated with the individual senses. According to this phenomenology criterion, it may be the case that vision and audition might both represent the same sort of property in the world, they may both, for instance, represent the spatial location of an object, but they differ in that what it is like to represent the spatial location in audition is different than what it is like to represent the spatial location in vision.72                                                      72 Both sorts of introspective approaches to individuating the senses have met resistance. Against the content approach see (H. P. Grice, 1962; Lopes, 2000). Against the phenomenal approach see (Keeley 2002; Nudds 2004, 2011). 91  Note how different these introspective approaches are to Keeley’s. Keeley was attempting to give a third-person account of the sensory systems, while these approaches are clearly first-person accounts73. This difference will be important in what immediately follows.  One main motivation for the introspective accounts is that they seem to do well with the Aristotelean senses. It’s common across cultures that people acknowledge the existence of the Aristotelean senses. This widespread acknowledgement happens even though there are varying states of scientific knowledge about the mind. In fact, even wildly misinformed accounts of the mind / brain, such as Aristotle’s view that the brain cooled the blood, do not seem to cause problems with acknowledging the existence of the Aristotelean senses. How else, then, might all these various people in various epistemic situations be coming to make the same distinctions? Well, the front runner of an answer seems to simply be that it’s clear in our experience of the world.   While perhaps this approach is helpful for discussions of how folk psychology might individuate the senses, it undermines a large portion of the scientific usage of the senses.  As Keeley argues, if the notion of the senses is to play a role in our scientific understanding of the mind, then we need to divorce our account of the senses from phenomenology, since we readily attribute senses to animals while we have no access to their phenomenology.  Second, we attribute senses to humans as well, like the vomeronasal system, that have no phenomenology.  If the senses were individuated according to their introspective qualities, then we have an argument for why these non-phenomenal senses would fail to be genuine senses.  The argument is as follows74: 1. X is a sense with non-phenomenal sense. [assumption]                                                      73 There’s another way that the difference between these approaches could be characterized. The introspective accounts are giving metaphysical accounts of sensory systems that also builds in an epistemic criterion according to how we determine whether some experience is given to us by one or another sensory system. The approach by Keeley doesn’t necessarily build in any epistemic claims about how we should go about deciding whether two systems are of the same sensory type or not. 74 The formalization here makes the argument much clearer than could be done with prose. 92  2. If something is a sense, then it must be a sense of a particular type. [senses as types] 3. If a sense is of a particular type, then it must have a distinct introspective qualities particular to that sense-type. (introspective individuation) 4. X has no distinctive introspective qualities. [from 1] 5. X is not a sense of a particular type. [from 4, 3] 6. X is not a sense.  [from 5, 2] If the individuation of the senses is based on some introspective property, then no non-phenomenal system could count as a genuine sense, and as a result, large swaths of the empirical literature concerning the senses would be misguided. A way of understanding this might be to say that if we individuate the senses according to their introspectible qualities, then we must draw the boundary between the senses and the rest of the cognitive architecture late enough in the processing so that the sensory systems will include whatever processes are involved in conscious experience. But we simply don’t know where to draw this boundary in humans, and we have even less of a handle on how to draw this line when it comes to non-human animals. On top of that we simply have no idea how to determine whether there is some special introspectible quality that would individuate sensory system in non-human animals, since we do not have access to their conscious states. We seem to be running into an impasse. The folk notion of sensory systems, that philosophers have typically analyzed, seems to rely on some introspectible quality and therefore draw the boundary between sensory systems and cognition rather late. However, this approach to individuating the senses simply doesn’t apply well to non-human cases, and as a result undermines a swath of cognitive science that discusses the sensory systems of non-human animals. Perhaps this is where the changing of the subject objection arises against the third-person approaches to individuating the senses. However, if there is a core notion of sensory system that applies to both folk and scientific account of the senses, then we 93  can see this disagreement not as one of changing the subject, but there simply being multiple acceptable means of individuating the senses. The argument for there being a common core notion of sensory system builds on the above characterization of senses as the initial stage in an information processing architecture – the senses for both the folk and scientists are those systems that gather information and directly track the world. Furthermore, in both folk and scientific cases, we as a matter of fact can individuate sensory systems without appeal to any introspectible quality, and therefore, there must be some other sort of feature that we can latch onto. Therefore, the above argument for the non-existence of non-phenomenal sensory systems fail (that is, premise three of the argument turns out to be false). As mentioned above, it appears as though in folk contexts the senses are individuated according to their introspectible qualities. Simply reflect upon your experience and the divisions between the senses should be apparent. However, the folk notion of the senses is permissive enough to allow for the proper typing of sensory states as being of a certain sense despite there being no phenomenology associated with those states. A brief foray into recent history makes this clear enough. In the 1950’s, in the midst of Cold War paranoia, the idea of subliminal perception was capturing the public attention. The fear was that the broadcasting of “foreign ideologies” through television, radio, and film could influence behavior without the public becoming aware of what was happening (Acland 2012). The fear was so strong that the National Association of Radio and Television Broadcasters asked its members to refrain from using subliminal messages in their broadcasts. Despite the fact that these subliminal messages were by definition something beyond our conscious experience and therefore had no associated phenomenology, people were still capable of conceptually distinguishing and being fearful of auditory vs visual subliminal messaging. The folk could do this. To put it differently, the folk could distinguish between subliminal perception of different sensory types. But for this to be possible, the typing of the subliminal experiences could not rely on the existence of any phenomenal or otherwise introspectible quality. The folk must have 94  a means for typing or individuating senses that appeals to a third-person available feature. This isn’t to say that in some cases the senses are individuated by the folk in terms of some introspectible quality, but rather, it’s simply to say that this can’t be the only means by which the individuation occurs.  Premise three of the argument for the non-existence of non-phenomenal senses fails. There must be some other means that even folk senses can be individuated. What that is, we can remain neutral.75 However, importantly the core notion of the sensory system remains constant. The senses play a role in directly tracking and gathering information about the world. It is here where we can finally put the point. The two questions that began this section, what makes something a sense as opposed to some other psychological kind, and what makes one sense a different type than another sense, can be answered in a variety of ways. Furthermore, the two questions are not entirely independent. Given where we draw the boundary between sensory systems and cognition and motor systems will influence what sort of accounts of individuating the senses are plausible. That is, given a particular boundary between sensory systems and the rest of the cognitive architecture, certain accounts of individuating the senses will provide us with a particular taxonomy of the senses. However, if we adopt a different sense / cognition boundary, then that same account of individuating the senses will result in a different taxonomy of the senses. In some cases, we may which to push the sensory boundary towards the periphery, and employ some third person criterion for individuating the senses, resulting in a likely very fine-grained taxonomy of sensory systems. In other cases, we may push the sensory boundary deeper into the architecture and adopt a first-person criterion, thereby resulting in a taxonomy that approaches (or is identical to) our folk notions. Furthermore, a range of intermediary positions are available. If you want to study how the visual system has developed in mammals, then you may draw the                                                      75 A plausible approach would be to employ some rudimentary (neuro)biological criterion. The folk can distinguish outward sensory organs and perhaps they use this to individuate sensory systems. Another approach is that of Nudds (Nudds, 2004, 2011) where the individual sensory systems are individuated by the epistemic activities involved in using those senses. Vision, according to Nudds, is for seeing. Audition is for hearing. Touch is for feeling or touching. Olfaction is for smelling. And taste is for tasting. Or perhaps there is an appeal to something like dedication or the content criterion.  95  sensory boundary at some intermediary position and draw upon third person criterion. The options are many. And yet, we have no reason to suppose that any of these combinations gets it right.  The brain is real and the operations in the brain are also real. However, how we produce a taxonomy of the brain and its operations are up to the goals of the researchers. Importantly, this isn’t to adopt some form of radical anti-realism about the mind. Rather, it’s admitting that there are real patterns in the world and we can group these together in a variety of different ways that track real clusters of behaviors and phenomena in the world. What dictates our general category of sensory system is a common-core commitment to the idea that the sensory systems are directly tracking and gathering information about the world. It is in this way that scientific discussions of non-Aristotelean senses do not fall prey of merely changing the subject. They maintain the important core, but are concerned with slightly different explanatory aims, in much the same way that the folk in different contexts will appeal to non-Aristotelean senses and non-introspective individuation criteria in the explanation of behavior.  3.2. Against the sense of time  Now that we have an understanding of the information gathering role of the senses in our theorizing about the mind we are in a position to see the arguments that some philosophers have put forward arguing that there simply is no (or cannot be) a genuine sense of time. It’s somewhat interesting that it is actually fairly common to find people, both philosophers and scientists, claiming that there is no sense of time, however, it is much less common to find those claims backed up with any significant argument. So, the two arguments that I’ll present here are ones that are not only explicitly found in the literature, but they are also arguments that seem to be behind some of the un-argued for claims that there is no sense of time.76                                                      76 These two arguments aren’t the only arguments that could be found in the literature. One could imagine someone running an argument against the existence of a sense of time on the basis of some general characterization of the 96   The Non-Causality Argument  The following quotes by the psychologist Lera Boroditsky and the philosopher Mohan Matthen give the general idea behind the non-causality argument.77 All of our experience of the world is physical, accomplished through sensory perception and motor action. And yet our internal mental lives go far beyond those things observable through physical experience: we invent sophisticated notions of number and time […] So how is it possible that physical organisms who collect photons through their eyes, respond to physical pressure in their ears, and bend their knees and flex their toes in just the right amount to defy gravity are able to invent and reason about the unperceivable and abstract? (Boroditsky 2011, p. 333) Should [the systems responsible for the representation of time] be regarded as transducers for a sense of time? That is, do periods of time cause them to emit a pulse that carries information about these periods of time? Both sides of the question can be argued. A negative answer might be a reason to exclude the sense of time. (Matthen 2015, p. 573)  As I described earlier, an essential characteristic of the senses, which distinguishes them from other representational systems, is that the senses track the ongoing changes of the particular aspects of the world that they detect. In other words, the senses provide us with a real-time information link with the                                                      sensory systems. Perhaps, by appealing to a particular account of sensory individuation, like Keeley’s physics criterion that can be spelled out in terms of there being a causal connection to the world that individuates senses. However, as we saw there isn’t any single means of individuating the senses, so any sort of argument of this sort will likely fail. The two sorts of arguments to be presented here both accept the general characterization of the senses as information gathering mechanisms that was argued for in the previous section. 77 Similar arguments can also be found in Gallistel (1996) and Coull (2011). 97  environment. It is in virtue of the epistemic connection with the world given to us through the senses that we can then go and correctly deploy many of our other representational and conceptual systems. If we look at the classic Aristotelean senses, we find that this information link with the environment is established via a causal connection with the aspect of the environment that each sense detects and the sensory systems themselves. For example, it is through the causal interaction of photons impacting the retina that the visual system is able to acquire information about the changing environment. Similarly, for many non-Aristotelean senses, such as proprioception or electroreception, it is through the causal interaction between the to-be-detected aspects of the world and the sensory systems themselves that the appropriate information link is established.  However, it is commonly claimed in the literature on the metaphysics of time that time itself is not causally efficacious, as a result, there simply could not be a sense of time since no sensory system could ever stand in the right causal relation with time.78 Putting the argument explicitly, we get the following: 1. The senses must keep track of the immediate environment in real time. [assumption] 2. The only way of keeping track of the contingencies of the immediate environment is to be in causal contact with those contingencies in the environment. [the causality constraint]                                                      78 See Benovsky (2012) for a discussion of the debate in the metaphysics literature on whether time can be causally efficacious. Benovsky argues that time in fact must be causally efficacious to explain the half-life decay of a number of elements. However, see (Lewis, 1973; Maudlin, 2002; Newton-Smith, 1980) for arguments that in cases where it appears as though time is in fact causally efficacious there is in fact a more parsimonious reading of the causal relations that does not require that time is actually causally effecting anything. To give just one example, someone might leave the potato salad out on a hot day and the salad spoils. We might say, the cause of the salad spoiling is the duration of time that it was in the hot sun. But, Newton-Smith (1980) for instance, argues that we can interpret the spoiling of the salad as not being an effect of the time in the sun, but as a chain of molecular events that bring about the spoiling (e.g. the bacteria in the salad feed on the ingredients of the salad, and thereby multiply, etc). This sort of story, while occurring in time, does not appeal to time, or any temporal properties themselves, as causes of any of the subsequent states. Furthermore, many have taken Shoemaker’s (1969) thought experiment concerning frozen worlds as showing that the mere passage of time isn’t causally efficacious. The details of all of these arguments trade not only on how to provide understand time but also importantly how to understand causation. Making sense of these issues would be a project all to itself. 98  3. If there is a sense of time [i.e. a sense that keeps track of the temporal structure of the events in the world], then we must be in causal contact with time. [from 1 & 2] 4. Time is not itself causal. [the non-causality assumption] 5. Therefore, there is no sense of time. [from 3 & 4] Now the purpose of this chapter is to understand something about the senses and not about the metaphysics of time. So, we can simply take the claim that time is not itself causally efficacious as a fair assumption. Furthermore, if we were to abandon this claim about the causal inefficacy of time, then the arguments considered here would simply fail out of the gate. But, by giving them their space, and to evaluate them for what they say about sensory systems and not about the metaphysical nature of time, we actually can gain insight into a very peculiar aspect of how we mentally represent time. So, in the subsequent section when I consider how this argument fails, I will focus on the causality constraint and not the non-causality assumption.  The Integration Argument  Kant famously argued that intuitions of space and time serve as pre-conditions for the possibility of any experience of an objective world whatsoever. The central idea being that it is by embedding experiences, or the deliverances of the individual senses, into a spatial and temporal framework that one is able to have experience of and make sense of the outer world. In this picture, the representations of space and time are in an important sense prior to the individual senses, since it is only by embedding experiences within a spatio-temporal framework that we can have experience of an objective world, and as a result, the representation of space and time are not proper parts of any of the individual senses.  While it’s difficult to evaluate Kant’s own arguments for the priority of temporal representations, Matthen (2014) has recently argued that while Kant’s arguments likely fail, a Kantian-inspired argument can be 99  run where the representation of time is not sensory since it provides a framework for the interpretation of the individual senses.  The starting point for Matthen’s argument is to notice that, as a matter of fact, representations of time serve as a “common measure” for the deliverances of the individual senses. The information acquired by the individual senses is ultimately organized within a unified temporal framework to create a unified representation of the external world.79 For instance, when I see and hear a thunder storm, the flashes of lightning that I see and the crashes of thunder that I hear are placed within a single temporal order, so that at times they are simultaneous, but at other times the lightning precedes the thunder.  Importantly for Matthen, not only does the representation of time serve as a common measure for the various sensory systems, but the way in which temporal properties are attributed to events in the world gives us further reasons for thinking that the representation of time is at one remove from the sensory systems. Adopting the account from Phillips (2010), Matthen argues that temporal properties are attributed to events in the world through a metaexperience. The individual sensory systems, in tracking the world, bring it about that the creature has certain experiences that are specific to those individual senses. Furthermore, these experiences themselves are temporally structured in that the experiences themselves occur at particular moments in time, have durations, and stand in temporal relations to other experiences and events in the world. The attribution of temporal properties to the events detected by the individual senses occurs by some system that notes the temporal relations that hold between the experiences of worldly events (e.g. earlier than, later than, and simultaneous with), and then these temporal relations are exported and attributed to the events out in the world originally detected by the individual senses. It is through these metaexperiences that operate over the individual senses that we are able to coherently integrate the various deliverances of the individual sensory systems into a single unified temporal order. Since the temporal exportation process operates over the individual senses it is not                                                      79 This sort of integration is the focus of chapter 5.  100  itself a sensory system nor is it a part of any of the senses. In this way, Matthen argues that due to the role of temporal representation in structuring and ordering the deliverances of the individual sensory systems we have reasons for denying that the representation of time is due to a genuine sense of time.  We can put the argument more explicitly as follows: 1. The deliverances of the individual senses are organized / integrated within a temporal framework of earlier than, later than, and simultaneous with relations. [time as organizing framework] 2. The represented order of events is due to a mechanism that tracks the temporal relations between the experiences of those events, and then the temporal relations that hold between those experiences are exported and attributed to the experienced events. [temporal exportation] 3. Therefore, the attribution of temporal properties is through a mechanism that takes experiences as its inputs. [from 1 & 2] 4. The senses do not take as their inputs other sensory systems or representational states of the organism. [established in the previous section] 5. Therefore, there is no sense of time. [from 4] In what follows, we’ll see that the argument as given by Matthen fails to establish its intended conclusion since as it stands it simply isn’t valid. Even if all of the premises are true (although, see chapter 4 for an argument that premise 2 is false), the conclusion still would not follow as there may be other mechanisms for temporal representation that do not fit the story given here. But, with these two arguments on the table, let’s now turn to see why they both fail.  3.3. Towards a sense of time  In this section I will argue that both the non-causality argument and the integration argument fail to establish that there is no sense of time. Once we notice how these arguments fail, we will be in a 101  position to determine whether the existing psychological / neurological mechanisms possessed by animals provide them with a genuine sense of time. Against the non-causality argument  As we’ve been discussing, the senses are distinguished from the rest of the cognitive architecture by their information gathering role. They directly track and gather information about the ongoing contingencies of the environment. As some have described them, the sensory systems are servile to the environment (Akins, 1996) in that the states of the sensory systems are importantly tied to the ongoing and changing state of the environment. When we open our eyes, the sensory system cannot help but begin to respond to the environment. The same goes for touch and the various other sensory systems. None of this is to say that the organism itself doesn’t contribute to the process of perception. But, it is nevertheless the case that at the very periphery of the sensory system the senses are servile to the environment.   It’s for these reasons that a causal connection between the environment and the senses seems to be necessary. How else might a sensory system keep track of the changing contingencies of the environment if not through a causal connection? There’s a sense that if there were no causal connection, then any correlation between the states of the sensory system and states of the world would be mysterious. Again, if we look at the classic Aristotelean senses we find that the capacity that the senses have to keep track of the environment as it changes and as we move within the environment is due to the causal sensitivities of the sensory systems. If time simply can’t be a cause, then we can’t be in the appropriate causal connection. Notice how strong this argument is. It’s not simply that as a matter of fact some creature lacks a sense of time. Rather, the conclusion drawn from the non-causality argument is that there simply could not be by necessity a sense of time.  However, as Dretske emphasized in his Knowledge and the Flow of Information (1981), for there to be an informational link between a representational system and some state of the world (or in Dretske’s terms, between a signal and its source) a causal link need not be present. For a system to carry 102  information about some state of the world simply requires that there be some nomic relation between the state of a representational system and the state of the world that meets the following condition: A system, s, in state R carries the information that t is F (where t is a particular instantiating the property F), just in case, t’s being F is guaranteed by S’s being in state R (given appropriate background channel conditions80 and the laws of nature).81  In other (more epistemically loaded) words, the fact that s is in state R allows one to know that t is F. Often what guarantees this nomic relation is a causal connection. However, if the appropriate information link can be established in the absence of such a causal connection, then we would still have the appropriate information link.  Dretske gave an example that tried to show that an informational link between two states of the world, say s and t, needn’t depend on there being a causal influence from s to t (or vice versa). His example is the following: Consider a pair of TVs, TV1 and TV2, located in isolated rooms and connected to a single closed-circuit television network. Both TVs are capable of only showing the signal that is being sent from a single central broadcast (see figure 3). Now, despite the fact that TV1 and TV2 exert absolutely no causal influence on one another, there isn’t even any physical means by which they might do so, there is still an information link between them. We know that TV1’s being in state 1 carries information about the state of TV2 since the two TVs are being controlled by a single source.                                                      80 These channel conditions are the physical states of the world that when in place mediate the nomic relation between s and t. 81 This definition could also be given in terms of conditional probabilities. A system, s, carries information about t if the states of s eliminate some of the possible states of t. Dretske in giving his analysis of semantic or representational content in terms of information required that the conditional probability of t’s being F given that s is R be 1 since he needed to rule out cases in which s’s being R is an ambiguous signal that could indicate either s’s being R or some other state of affairs entirely such as q’s being F. 103   Figure 3: Common-cause information channel. Notice that there is not causal connection between TV1 and TV2 yet they nevertheless carry information about one another.  Now, Dretske’s example shows that there needn’t be a causal connection between the information carrying system and the state of the world that it carries information about but it does so by appealing to a common cause. However, in what follows, we’ll see that this common cause condition doesn’t even seem necessary.  Since a causal connection isn't necessary for an information channel to be established, one can understand the non-causality argument to be a sort of abduction argument. Since the senses put us into real time informational contact with the environment, and causal connections seem to be the only means of establishing these connections, then it seems as though there can be no sense of time, since a causal connection is not available. However, if one can show that we possess psychological mechanisms that establish the appropriate informational link with time, then we will have done part of the work required to 104  avoid the conclusion of this argument. What remains to be shown is that the information link established is an automatic one in the sense that the relevant feature of the environment is in control of the information gathered by the senses. Importantly, in what follows I will show that this additional constraint is also satisfied, even though time does not exert causal control over the sense of time, it is nevertheless true that the sense of time is nomically related to time, and is in that sense not under the control of the individual organism that possesses the sense.  Against the integration argument  As Matthen argued, it appears as though at least some of our access to the temporal features of mind-external events is indirect. We export the temporal relations in which our experiences stand and attribute them to the events those experiences are of. As a result, the attribution of temporal properties is through a process that operates over the individual senses, and not a genuine sense in and of itself. Whether or not Matthen is correct in his account of how the temporal relations of earlier than, later than, and simultaneous with are attributed to the deliverances of the individual senses can actually be sidestepped here. We can avoid the conclusion to his argument in either of two ways while still granting him all of his premises. (In fact, in the next chapter we’ll see that the exportation mechanism doesn’t explain how we perceive temporal relations.)  The first way of denying his conclusion is to argue that the mechanism he proposes may not be the only mechanism available for the attribution of temporal ordering relations. The temporal exportation mechanism he describes could simply be one of several mechanisms that attributes temporal relations to the world.  The resulting situation would be similar to the situation in which we are asked to detect the texture of an object.  Merely pointing to the fact that someone can come to infer the expected feel of an object based on how it looks, doesn't suffice to deny that we have a genuine sensory access to texture 105  given by touch.  We often have multiple routes according to which we might detect properties in the environment, yet this fact doesn't prohibit the possession of a sense for any of these features.  The second approach is to argue that perhaps the mechanism Matthen discusses is the only mechanism for the attribution of temporal ordering relations (although for this argument it needn't be, but we can grant this and still run this line of argument), but there may be mechanisms that attribute other temporal features to the external world that do not employ any sort of temporal exportation. Furthermore, of these other mechanisms there may be a mechanism that can be properly described as a genuine sense of time. Both of these approaches would amount to a denial of the exhaustiveness claim Matthen needs, but it is the second approach that I will follow in the coming section. There exists a mechanism, distinct from the mechanism discussed by Matthen, that has the features of a genuine sense and is used to detect the approximate time of day.  It is by noting that we have multiple means of representing time that we open up the possibility for understanding which of these representational abilities are genuinely sensory or not.  3.4. Circadian rhythms and internal clocks  Let’s begin with a very general point about animal behavior that we’ve already mentioned. In order for animals to be able to successfully navigate their environments they need to be able to coordinate their behaviors, at a variety of time scales, with the temporal structure of the events in their environment. Some behaviors need to be coordinated with the temporal properties of events, regardless of when they occur. For instance, a baseball batter has to be able to coordinate their swing with the spatio-temporal trajectory of an incoming pitch regardless of when that pitch is occurring (i.e. regardless of what specific point in time or time of day the pitch is occurring). Some behaviors, however, need to be coordinated with the temporal properties of events that occur at particular times (e.g. foraging behavior often must be timed to specific times of day when food sources are available). In this section we’ll look at coordination of the 106  second sort. Specifically, we’ll be looking at relevant literature on the role of the circadian system in the coordination of circadian behaviors.  Throughout the animal kingdom we find that many animals exhibit patterns of behavior and patterns of internal activity that have roughly 24-hour periods.82 These behaviors include things like sleep-wake cycles, eating patterns, hormone regulation, body temperature regulation (Moore 1997), and even patterns pertaining to how effectively new memories can be formed (Ruby et al 2008). While the existence of a number of daily patterns with nearly the same 24-hour period may seem to imply that there is some sort of mechanism (perhaps even a single mechanism) that represents the approximate time of day and allows for the regulation of these pattern that inference could fail.  One could deny that animals need to represent the time of day in order to coordinate their behaviors with the temporal structure of the environment. Take for instance the feeding behavior of certain fish. As many fishermen might know, in certain places fish can be found in certain locations at certain times of day. Now, one might explain this predictability of fish behavior in terms of the fish having a robust internal representation of the time of day that they use to navigate to a particular location at a particular time. However, another explanation of the regular pattern in fish behavior offloads the explanation from an internal representation of time to a causal sensitivity to rhythmic patterns in the environment (in chapter 2 I called this the cued-synchronization model). One possible example can appeal to the rhythmic pattern of tides in particular locations. In some places, tides have a nearly 24-hour period, and fish might simply be cuing into the fact that at certain tides, food sources will be located in particular locations, and as a result the fish will congregate in particular locations at particular times of day. No representation of time is needed to explain the regular pattern of fish behavior. Instead, the fish are simply                                                      82 In fact, circadian rhythms seem to be ubiquitous in not just the animal kingdom but also the plant and bacteria kingdoms. Ultimately, I will argue that many animals have a sense of time that is centered on the circadian system, but I would likely stop short of attributing a sense of time to non-animal organisms. In those cases, the mechanisms that underpin the circadian rhythms of activity simply do not have the functional characteristics needed for being a sensory system. 107  causally sensitive to some aspect of the environment that itself exhibits a 24-hour period, and the circadian behavior of the fish is simply scaffolded off of this external rhythm. So, what seems to be a sophisticated situation in which a rich system of internal temporal representations were required actually can turn out to be a rather simple situation in which no temporal representation (and possibly, no representation whatsoever) is needed.  Furthermore, once we begin to notice that there are patterns in the environment that can be cueing and driving temporally regular behaviors in animals we open the door for a complex iterative process in which further complex behaviors are further scaffolded off of the initial environmental regularities. Suppose, again, that the temporal patterns of fish behavior can be explained by appealing to a causal sensitivity to environmental patterns. If that were the case, then we can likely expect to find further animal behaviors that build off of the fish behavior. Predators that prey upon fish will also exhibit behaviors that are coordinated with the temporal structure of the environment by merely tracking the location of the fish whose movements are themselves temporally structured. Once we notice how patterns can build upon patterns we can actually explain how a complex of temporally structured behaviors might simply be built on top of some environmental rhythm.   In fact, the prevalence of contextual cues in the environment led to many researchers denying that animals synchronized their behaviors with the environment by representing time itself. Certainly, this line of reasoning has something to it. In general, we choose scientific theories that posit the least amount of novel machinery. Since we already, for many other reasons, have to posit that animals have means of keeping track of the non-temporal aspects of their environment, if we can also explain their circadian behaviors by merely appealing to these other systems, then we would have a simpler theory than one that posits mechanisms specific to temporal representation. In other words, researchers tried to explain the temporal patterns of animal behaviors by positing mechanisms that keep track of the what and the where. In some cases, surely something like this happens. As a child, I would know that dinner was going to be 108  soon when Jeopardy began playing on TV, and more often than not, this would result in me, or someone in the family, setting the table. So, there would be a temporally regular behavior in the household, but we can explain the temporal regularity without appealing to any temporal representation. Rather, the regularity was due to a response to the regularity in the TV broadcast.  However, there are good reasons for thinking that many, if not most, circadian behaviors cannot be explained as a result of some temporal patterns of external cues. The main reason is simply that many of these circadian behaviors, even those that are learned, persist in the absence of any external cues. In this way, the experiments that revealed the temporal mechanisms that underpin mirror those that revealed the mechanisms that underpinned the electroreceptive capacities of elasmobranch fish. First, eliminate the influence of any of the existing sensory system, then, if the behaviors persist there must be some other means by which the animal is keeping track of its environment.  Consider the following sorts of studies reported in Gallistel’s The Organization of Learning (Gallistel, 1996). In one study (Holloway & Wansley 1973), rats were administered shocks (e.g. negative reinforcement) when they attempted to enter a chamber containing sugared milk at a particular time of day. The very same rats were also given positive reinforcement when they tried to enter the very same chamber but at a different time of day. So, the rats were receiving feedback that could only be interpreted as not conflicting if the rats were able to keep track of when they were receiving feedback. That is, if they were not keeping track of when they were receiving feedback, then the feedback they were receiving would be contradictory and the positive and negative reinforcement should cancel each other out (if not fully, then at least there would be a weakening of any reinforcement effect). However, the results were that the positive and negative reinforcement would only compete if they were receiving the feedback at the same time of day (on separate days, of course). The conditioning of the rat behavior was clearly sensitive to time of day. 109   What wasn’t clear from these conditioning experiments was whether the resulting circadian behaviors (i.e. the feeding pattern with approximately 24-hour rhythms) were still the product of some external cue. Further studies by Rosenwasser, Pelchat, and Adler (Rosenwasser et al 1984) showed that rats that were trained to anticipate food in certain locations at particular times of day would continue to show these aniticipatory behaviors even when they were put into free running conditions in which all contextual cues were removed. The circadian behaviors of the rats simply could not be explained by appealing to any external cues since there simply were no available cues in the rats’ environment.  However, Gallistel points out that some theorists attempted to maintain that some still undiscovered cues could explain the free running behavior of rats. However, even this sort of know not what external cue theory is ultimately untenable once you realize that the circadian rhythms of individual animals kept in free running conditions will begin to drift away from one another. Within a short amount of time, often simply within a day or two, animals that were originally showing synchronized behaviors began to show behaviors that slowly drifted away from synchrony. But, importantly, their behavior was still rhythmic (or had a nearly 24-hour period). Some animals would show circadian behavior that was operating on a contracted schedule (e.g. they would perform some action every 22 hours, as opposed to every 24 hours). Some animals, on the other hand, would show circadian behavior that was operating on an extended schedule (e.g. they would perform some action every 26 hours). Furthermore, once the period of behavior for an animal was determined, then the timing of other of that animal’s circadian behaviors could be predicted as falling on that animal specific period (Bolles & Moot 1973). This shifting away from each other could not explained if each animal was responding to the same unmentionable cue unless in addition to some unknown cue we posited a specific unknown cue for each animal. However, this is clearly starting to strain any desire for a parsimonious theory.83                                                       83 The astute reader will notice that in free-running conditions the circadian behaviors of animals began to drift, thus indicating some role for external cues in synchronizing behavior. We’ll discuss this influence of external cues on circadian behaviors slightly later. 110   Instead, the explanation that was given, and that is widely accepted today, is that the circadian behaviors of a wide-range of animals are in many cases best explained by the operation of an internal time keeping mechanism with a fixed period. That is, explanations of circadian behavior often appeal to the operation of an internal time keeping device while downplaying the influence of contextual cues. This still, however, left two options open for how to understand the operation of this internal time keeping mechanism. The first option describes the time keeping mechanism as an interval timing mechanism, while the second option describes the mechanism as a period timing mechanism.  According to the explanation based on an interval timing mechanism, circadian behaviors would be explained in the following way: A given event is used as a marker and a timer is started with the occurrence of that event. Subsequent behavior is then governed by noting the temporal interval between the marker event and some reading of the interval timing mechanism. For example, in time-place-learning experiments, a rat may learn when to find food at a particular location by measuring the temporal interval between some marker event and some feeding event.  According to the alternative period timing mechanism, circadian behaviors would be explained in the following way: No marker event is required. Rather, there is some oscillator system with a regular period and particular times of day are recorded as corresponding to particular states of the oscillator. Taking the time-place-learning example in hand again, the feeding behavior of the rat could be explained by directly appealing to the rat encoding the particular time of day at which food is found in particular locations in terms of states of the oscillator. In a sense, this means of keeping track of the timing of events involves a time-stamping process.  Importantly, the differences between an interval timer and a period timer are noticeable both in terms of the implementation of the timers and in the content of the timer.  In terms of implementation interval times typically involve some sort of accumulation process. Considering the classic pacemaker-accumulator mechanism of Treisman (1963), the pacemaker will produce neural spikes at a regular rate 111  and the accumulator system keeps track of these spikes in format that is neurally significant84 for downstream processes. On the other hand, period timers needn’t involve any accumulation process. Rather, they simply appeal to the rhythmic change of some neurally significant properties of a system that can be exploited to keep track of time. For instance, we cannot describe the accumulation or the rhythmic states of a time keeping mechanism as being purely temporal properties. The accumulation cannot merely be that the system has been turned on for 5 seconds. Similarly, the rhythmic properties of an oscillator cannot be that the clock is now at 5pm. Pure temporal properties simply are not the sorts of things that neural systems can exploit to guide their behavior, and any temporal information encoded by a timing mechanism must be such that it can be exploited by downstream systems.85 Otherwise, the positing of such mechanisms would do nothing for explaining animal behavior. Crucially, whatever properties of the timing mechanisms we appeal to as being the semantically significant properties must also be causally efficacious within the overall cognitive architecture.   Also, the two sorts of timers differ in their semantic contents. Interval timers invariably appeal to a marker event – their content is always of the form <e is I since m>, where e is the target event, m is the marker event, and I is the temporal interval between e and m. Period timers, on the other hand, do not reference any marker state. Instead, period timers predicate of particular moments some temporal property – their content is always of the form <tp is p>, where tp is a particular moment in time and is p predicates of tp a certain temporal property (we’ll say more about this later). If an interval timer model is to account for circadian behavior, then we need to give an account of what the relevant target and marker                                                      84 Recall, from chapter 2 it was argued that representational systems are constrained by the downstream accessibility of information. In the case of neural representations, the information encoded by a system must be stored in a manner that downstream neural systems can access that information. That is, the information must be encoded in a neurally significant format. 85 Surely, there are properties in the vicinity of these pure temporal properties that are neurally significant. Suppose that a neural system in a particular state, say state A, results in the system producing 50 spikes per second. As a result, being in neural state A for 2 second, will be causally efficacious since the system will have produced a total of 100 spikes over that interval.  112  events might be for any circadian task. If no such account can be given, then we ought to abandon the interval timing model for circadian behavior.  With an understanding of the typical implementation of these timing mechanisms and the differences in their semantic contents we notice that we have ways of adjudicating between these different models of circadian behavior.  Several considerations count against adopting an interval timing model at both the neuroscientific level and the behavioral / functional levels. First, the interval approach still requires some initial marker event according to which subsequent behaviours can be synchronized. In the studies discussed, all attempts were made to remove any contextual cues that could be used in order to regulate circadian behaviour, as a result, it seems as though there would be no marker events according to which one might initiate the interval timing mechanism. However, its possible that the animal's own behaviours could serve as marker events in order to synchronize circadian behaviours. Perhaps, the animal knows that 4 hours after it wakes it should look for food in location A and that 6 hours after looking for food in location A it should look for food in location B. And so on for the rest of the animal's behaviour.  However, there is a feature of the interval timing mechanisms that is not found in period timing mechanisms.  As the length of the interval the interval timer is measuring increases, the timer will increase in noise and become less accurate (Dehaene 2003; Gallistel & Gelman 1992; Malapani & Fairhurst 2002; Wearden 2001). In some cases, this sort of noise in the model is appreciated as we find a corresponding noisiness in timing behaviors. In particular, many interval timing studies have shown that animals throughout the animal kingdom discriminate durations in accordance with Weber’s Law, according to which the just noticeable difference between two stimuli is a constant ratio of the values of the two intensities (e.g. as the overall intensity or duration of a stimulus increases individuals will require a larger difference for another stimulus to be judged as different).  As a result, we should find that a degree of inaccuracy in the circadian behaviors that simply isn't found in the above studies (Ko et al 113  2003). The circadian behaviors are timed with equal accuracy regardless of when in the day they are occurring.  So, on the behavioral / functional level, we seem to not find the signatures associated with interval timing mechanisms. While the behavioral / functional data were telling against the interval timing model, the final nail in the coffin came from the direction of implementation. What really changed the way in which researchers thought about the circadian system was the neuroscientific findings concerning the neural systems that regulate the circadian behaviours. The first task in understanding the circadian system was the task of localization. It was discovered through a series of studies that many circadian behaviours in mammals are regulated by a specific neural region called the suprachiasmatic nucleus (SNC), a region of the hypothalamus, alongside a variety of more peripheral oscillatory systems whose rhythms are synchronized through the SCN.  It was shown in a number of studies that damage to the SCN results in disruptions to the circadian rhythms of animals.  For instance, one telling study involved the long-term monitoring of two chipmunk populations (DeCoursey et al 2000). The first population underwent surgery in which their SCN's were removed. The second, control, population underwent control operations in which the animal underwent much the same operation as the chipmunks in the first population underwent, however, they were left with intact SCNs. The chipmunks were then released back into the wild and monitored over the course of 18 months. After the end of the 18-month monitoring period, the control group had a significantly higher survival rate than the experimental group that had lost their functioning SCNs. It was determined that one contributing factor in the lowered survival rate of the experimental group was that the chipmunks with excised SCNs were much more active at night when predatory weasels were out on the hunt. As a result, the SCN excised chipmunks were noticed at a much higher rate by the predatory weasels and for that reason they were victims of weasel predation at a much higher rate than the control group of chipmunks. Due to the 114  loss of the SCN circadian behaviours were not properly synchronized with the environment, and as a result, the chipmunks in the experimental group suffered from a greatly reduced survival rate.  Another telling experiment involved the behaviour of hamsters kept in free-running conditions (Ralph et al 1990). Through selective breeding practices two populations of hamster were bred with distinct circadian cycles. One group of hamsters possessed the normal circadian system with a nearly 24-hour period, while the second, mutant, hamster group possessed a circadian system with a greatly reduced period of approximately 16 hours. Upon the ablation of the SCN in hamsters of either group circadian rhythms halted – sleep/wake cycles, daily activities, body temperature regulation, and hormone regulation all became erratic. However, circadian behaviours were recovered upon surgical transplantation of healthy SCN tissue. Perhaps most interesting however is that when mutant SCN tissue was transplanted in a hamster from the normal population circadian rhymicity was restored but the circadian patterns of the normal hamster displayed the 16 hour period of the mutant SCN. Similar findings were found when the mutant hamsters were given normal SCN transplants.86 The most natural explanation of these behaviours is that it is the SCN that plays a crucial role in determining and regulating circadian behaviours. One final example is useful to understand the role of the SCN.  It is known that mammalian body temperatures fluctuate over a circadian cycle.  In a study by (Megumi Maruyama 2007) it was shown that the SCN facilitates the formation of heat stress memories.  Two groups of rats were exposed to elevated temperatures for a fixed five-hour period of the day.  One group of rats had bilateral damage to their SCN whereas the control group had an intact SCN.  Both groups exhibited lowered core temperatures for an extended time of the day, as a means of combatting the heat stress.  However, only the control group of rats with intact SCNs exhibited lowered core temperatures that coincided with the time of day during                                                      86 A fascinating aspect of these studies was that upon receiving a transplant SCN, animals would show circadian behaviors but upon anatomical examination it was found that the neural connections between the SCN and the rest of the brain were not properly formed. The influence that the SCN had on behavior was seemingly driven through humoral factors released into the bloodstream. This seems to be a manner in which a neural system may convey information to other regions of the brain in a way that bypasses direct neural connections. 115  which the training heat stress stimuli were administered.  The rats with SCN damage exhibited lowered core temperature but this lowered core temperature was not specific to any time of day.  The SCN damaged rats were registering the heat stress but were unable to synchronize their body temperature with when the heat stress occurred. Now, merely localizing the SCN as the mechanism that regulates circadian behaviour doesn't adjudicate between the interval timing explanation and the period timing explanation. However, it is by looking at the manner in which the SCN functions that we differentiate between the two models. Through a series of experiments the mechanism by which circadian timing occurs was discovered. While the details are still being worked out (see Bechtel (2011b)), the general story is now well known, and a simplified version of the model can be given here. The general mechanics of the SCN is that of a molecular clock that is governed by the reciprocal interaction of the transcription and translation of multiple genes and proteins, but for our purposes we can simplify the issue and consider a system that consists of only a single gene / protein pair87 Consider the gene PER (short for period). During gene expression, PER is transcribed to produce the protein per, however, as the level of per in the cell increases, the expression of PER is inhibited, thereby reducing the production rate of per. As the protein levels decrease, the gene PER is once again readily expressed. It was then discovered that through some at present unknown mechanism, the level of per in the cell directly related to the spontaneous firing rate of the SCN neurons88. As a result, no accumulation process was found in the functioning of the SCN, instead, what was found was a cyclical behaviour in the SCN neurons. Furthermore, the SCN's behaviour is controlled by the endogenous behaviour of the SCN neurons, and not through some causal interaction                                                      87 Most current models of SCN functioning in mammals involve the reciprocal interaction of five or more gene / protein pairs (Bechtel, 2011b), however, the simplifications made in this chapter do not mischaracterize the general point. 88 While the exact mechanism by which per levels influence spontaneous firing rates is unknown, see (Vasalou & Henson 2010) for an account.  116  or registering of an external state, and as a result, there is no marker state according to which the functioning of the SCN depends. The spontaneous firing rate of the SCN is then used as a master regulator that regulates a variety of other circadian systems in the body. For instance, as the firing rate of the SCN increases, it will either activate or inhibit downstream neural populations that themselves will directly influence various conditions in the body that are behaviourally relevant. In some cases these systems will involve the production of hormones during certain times of day, such as the production of melatonin at night (Moore 1997). In these cases, the behaviour of the SCN is used to provide coarse synchronizations with the environment – sleep related hormones are released at night and this becomes behaviourally important because if the animal were not asleep at night, as in the chipmunk case described above, then the animal will incur a reduction in its ability to survive. In other cases, these downstream processes will involve the coordination of behaviour to much more specific times. For instance, it is known that certain birds navigate by calculating target paths by integrating the location of the sun and the current time of day (Cassone et al 2009; Cassone & Westneat 2012). This sort of navigation requires a much more specific representation of the time of day than merely day / night representations, furthermore, merely appealing to the location of the sun (i.e. contextual cues) is insufficient for the determination of a flight path. It should be noted that since the circadian system, like any clock, acts as an oscillator, going through a complete cycle approximately every 24 hours, that the system occasionally needs to be recalibrated or entrained.  This is particularly important when as the seasons change and the day / night ratio changes.  This recalibration mainly occurs through the input from a dedicated visual pathway that indicates the overall light level of the environment.  Unlike most visual processing, which begins with input from the cones and rods, circadian entrainment begins with photosensitive retinal ganglion cells.  These cells input directly to the circadian system and do not enter into the pathways that lead to the visual cortex.  Importantly, as is witnessed by animal behavior in free-running conditions, the light entrainment 117  from retinal ganglion cells only serves to recalibrate the system, as the circadian system continues to function even in the absence of light / dark information.  Much like the way that a clock that is recalibrated on occasion still tracks time, the circadian system, even though it is recalibrated by light / dark information, still tracks time of day, and not the light / dark cycle. The need for this occasional recalibration is what explains why in normal situations we might find a whole group of animals to have synchronized behaviors but when those animals are put into free-running conditions the timing of their behaviors begins to drift apart. This fact can often be used as a way of raising objections to my discussion so far.89 The objection states that perhaps what this system is representing is not the time of day, but rather, the location of the sun.  Let’s take this objection that the circadian system isn’t representing time per say, but is instead measuring the location of the sun and see how it fails. As we’ll see in chapter 5, there’s actually something very plausible to this suggestion yet that doesn’t undermine the claim that the circadian system represents the time of day. Many of our own cultural time keeping mechanisms, the analog clocks you see on the wall, for instance, were for many years calibrated to local noon. Clocks were synchronized to read 12:00 pm at the point when the sun was highest in the sky. So, in this way, we could have interpreted these clocks as representing the location of the sun, and not the time of day. However, this isn’t how we made use of the clocks. Given that the time of day and the location of the sun would often go hand in hand (less so, of course as we move away from the equator), the movement of the hands of the clock could carry information about both the time of day and the location of the sun. However, the information that we would access from these devices would be used to coordinate the timing of our actions.90                                                      89 These objections have been raised by audiences at the 2016 Eastern Meeting of the American Philosophical Association and the 2014 Society for Philosophy and Psychology Meeting. Also, thanks to my commentator at the Eastern APA, David Sackris, for raising this objection in detail. 90 Yes, we could also coordinate our time sensitive actions by following rules like, “when the sun is at X location, do action Y”. In this way, however, we would be using the location of the sun in much the same way that we use a clock. The trajectory of the sun itself would the periodic system whose states we use to tell us the time. Regardless, the system would still be representing the time of day. 118  Furthermore, even in the absence of sunlight, i.e. when the sun was simply gone, we could still use the clocks to coordinate our behaviors. Similarly, for animals. The circadian system can still be used, for instance in time-place-learning, when the sun is gone. Finally, while input from the retina is perhaps the strongest source of calibration for the SCN, it is not the only source. One significant source, and one that many people know from merely battling jetlag, is food (Schibler et al 2003; Stephan 2002). If food can also calibrate the system, and have similar results as the calibration through light levels, then the system really cannot be interpreted as representing the location of the sun. The very same considerations would force us to attribute a contrary interpretation in which the SCN is tracking the presence of food, but this conflicts with the SCN tracking the location of the sun. Rather, the more general interpretation, that the SCN is tracking the approximate time of day, makes all the other circadian behaviors intelligible. And as a result should be our preferred interpretation. Someone might object however that on purely empirical grounds, we shouldn't take the SCN to be the system that regulates all of these distinct circadian behaviours. This is due to some findings that certain circadian behaviours persist in animals that have had their SCN damaged through selective surgical procedures (P. A. Lewis et al., 2003). For instance, it has been shown that hamsters show conditioned place preferences to specific times of day even though they have had bilateral SCN damage (Marchant & Mistlberger 1997).  However, in these cases the question never shifts away from understanding the internal oscillator unpinning circadian behaviors.  Instead, the only shift in the question is a shift to what other oscillator systems might be in place, since the behavioral / functional markers of an oscillator system are still present – there are no available contextual cues and there are no signs of increased inaccuracy as length of time increases (ruling out a contextual account and an interval timing model respectively).  Further data shows that the SCN is still involved to some degree in the timing of these other oscillators since a damaged SCN that is still capable of sending signals, although misleading signals, to downstream processes seems to impair the functioning of these other oscillator systems.  The resulting view that seems to be coming out of the chronobiology literature is the SCN serves as a master 119  clock that is capable of synchronizing a variety of other mechanisms in the body and brain that display circadian behaviors.  The important point for this paper is that the way in which circadian behaviors are implemented is through oscillator systems that directly and automatically keep track of time.  Whether those oscillators are wholly located in the SCN or not is only of secondary concern.  What we have here is an argument against the argument from integration.  Recall that that argument had an implicit premise that the mechanism described there that takes the temporal relation between experiences and attributes that temporal relation to the objects that those experiences are of was the only mechanism that we possessed to track the temporal properties of the world.  However, as has been shown so far in this discussion, the mechanisms that subserve the circadian behaviours are not like that at all.  Instead, they function by representing the time of day through the functioning of an internal oscillator(s).  Nothing like temporal exportation happens here.  Instead, we seem to have system that tracks the time of day without extracting the time of day information from any other sense.  So while it may be true that the way in which we consciously attribute temporal properties to many objects and events at short time scales is through something like what Matthen (2014) and Phillips (2010) discuss, the point remains that there is an additional system that keeps track of longer periods of time involved in circadian behaviour (although, see chapter 4 for an argument against the exportation mechanism).  3.5. The semantics of (internal) clocks  In order to use the empirical literature on the SCN to show that the non-causality argument fails, we need to first say something about the semantics of clocks. Once we have a general story about the semantics of clocks on the table, then we’ll be in a position to give the meta-semantic account of how that content or information is acquired in a direct and automatic manner that does not depend on a causal relationship between time itself and the states of the clock.  120   Let’s begin with a general point. In order to get clear on the semantic content of any representation it’s important to look at the correctness conditions for that representation. For instance, the sentence “Matthew is a student” is correctly asserted, or is true, just in case Matthew, the referent of ‘Matthew’, has the property of being a student, the semantic value of ‘is a student’. The sentence is false, just in case, Matthew, or whatever object is referred to by the sentence, fails to have the property of being a student.91 Individual names or predicates do not have correctness conditions in and of themselves.92 To put the idea in Fregean terms, a predicate or name by itself does not express a complete thought, or proposition, that can be evaluated for truth. In order for a sentence, or a representation, to have correctness conditions it must be evaluable against the predication of a property to an object.  Given that we take clocks to be correct or incorrect it follows that clocks do more than merely refer to an object or pick out a property. Clocks must predicate of particular objects some sort of property, but what object clocks refer to and what property they attribute to those objects is less than obvious. In order to clarify this matter, we can look at the way clocks are used as a guide to their content, since whatever content we attribute to clocks must be in line with the way that we make semantic, or informational, use of clocks. Otherwise, if there weren’t some connection between the uses of clocks and their contents, then our uses of clocks would remain mysterious.  Let’s start with the following claim, called CLOCK, in order to understand the semantics of clocks: CLOCK: It is 5pm.                                                      91 The sense in which whatever referred to by the sentence is “an object” takes object in an incredibly broad sense. Anything that can be referred to or predicated of counts as an object. We shouldn’t restrict this semantic structure from applying to things that do not live up to the standards of genuinely material objects. 92 We can of course talk about correct or incorrect applications of individual names or predicates. For instance, I might incorrectly call some philosopher ‘Bruce’ in which case I would have incorrectly used the name. But in these cases the correctness of the application of a name, or predicate, is really short hand for something like that is Bruce. The correctness is still given at the level of full subject predicate wholes. 121  For CLOCK to be true, the following must be the case: the referent of ‘it’ must stand in the relation expressed by ‘is’ to whatever is expressed (or referred to) by ‘5pm’. Given what we’ve said so far in this section, this much should be obvious.  Let’s begin with understanding ‘it’. What does ‘it’ refer to in CLOCK? Something important about clocks is their indexical nature. A properly functioning clock does not tell you something about some arbitrary time. Clocks tell you something about what time it is now. To clarify this point, consider the following scenario involving Bonnie. Bonnie is a part in a complicated bank heist and she has explicit instructions to cut the power to the bank’s security system at 5pm sharp. Since the timing of this task is of the utmost importance, she brings with her two clocks as a redundancy just in case one of the clocks stops working. As she is sitting there in front of the electrical wires that feed power to the bank’s security system she goes to consult her clocks but accidently drops both of them. When she picks them up and looks at them, she realizes that one clock reads 4:45pm while the other clock reads 4:48pm. Both clocks can’t be accurate, but how are we supposed to characterize this difference? What makes one inaccurate?  The intuitive answer, and the answer that is crucial for how Bonnie is using the clocks, is that one of the clocks is inaccurate since it is misrepresenting what time it is now.93 While it’s a matter of debate in the metaphysics of time as to what particular moments in time amount to and whether there is anything special about the current moment in time, however, for our purposes here we can remain neutral towards these metaphysical debates. Whatever metaphysics of time we adopt, there will be something that we pick out with the phrase ‘this moment in time’, and it is to that sort of entity that clocks attribute particular temporal properties.94                                                       93 There are clear parallels between the semantics of clocks and the role of indexicals that John Perry isolates in his paper The problem of the essential indexical (1979). In both cases, the indexical representations are crucial for guiding behavior. 94 Whether particular moments in time are in some sense primitive or whether they are derived from other aspects of the world is a particularly pressing issue in the philosophy of time. For discussion see Meyer (2013). 122   To further make the case for the indexical nature of clocks, we can see that clocks possess a property that has often been restricted to demonstratives in the philosophy of language (Recanati 2012). Clocks are immune to error through misidentification. A sentence like ‘John is running’ can fail to be true on a particular use in two ways. Either John fails to be running or the sentence, as it is applied to some situation in the world, does not actually pick out John. I may misapply the sentence to someone who looks like John, but isn’t in fact John. However, some sentences are unable to fail to be true due to this second sort of situation. These are sentences that are thought to be immune to error through misidentification. Consider a demonstrative statement “that is running” that I utter without any presuppositions about the thing that I am referring to. I intend to demonstratively refer to whatever it is that I’ve selected. This particular demonstrative utterance may be wrong because whatever I demonstratively pick out fails to be running. Or I may be under an illusion and fail to pick anything out at all. However, this statement cannot be wrong as a result of not picking out the right object (in the same way as when I mistook someone as John). In this way, clocks also are immune through error through misidentification, since they cannot be describing the wrong time, since they always pick out now. However, unlike demonstratives, clock representations cannot even fail to pick out an entity. Whenever a clock is representing a time it is invariably referring successfully, since the clock’s operations will necessarily be occurring in time.   Having pinned down what ‘it’ refers to in CLOCK, we must now turn to understanding the overall logical structure of CLOCK and how this logical structure depends on how we interpret ‘is’. ‘Is’ famously has two distinct interpretations (Russell 1905). There is the ‘is’ of predication, as in ‘2 is prime’ in which we attribute a property to some object. There is also the ‘is’ of identity, as in ‘2 is 2’, in which we say that one object stands in the relation of being identical to an object (namely, itself). In cases in which the is of identity is being used, we can typically add the phrase ‘identical to’ after the ‘is’ in the sentence without any problem (e.g. ‘2 is 2’ can be read as ‘2 is identical to 2’). Correspondingly, there are two possible interpretations of CLOCK that depend on differing uses of ‘is’. 123  I think there are clear reasons for rejecting the identity reading of CLOCK. If we read CLOCK as expressing an identity statement, then we are making the claim that now = 5pm. Yet to understand the truth of this statement, we need to look at the referent of the indexical, and we get a particular instant in time, call that instant i1, so the claim states that i1= 5pm. However, suppose that a day goes by, and again we utter CLOCK, in that context of utterance we would understand clock to be saying of a particular instance of time, i2, that i2 = 5pm. Yet, transitivity of identity fails in these cases since i1 ≠ i2, since these are times on different days, and therefore are different times, yet we take each utterance of CLOCK to be true.  As a result, we can't coherently hold the identificational reading of CLOCK, instead the logical structure of CLOCK is one of predication. As a result, we need to find some property that is being picked out by '5pm'. This should strike most people as odd. What property is picked out by 'is 5pm'? It seems like 'is 5pm' shouldn't pick out any objective property of the world since our timing practices seem to be the product of merely contingent convention and as a result the division of any length of time into something like hours, minutes, and seconds, is just some sort of societal construction. If there really isn't any property out in the world that is expressed by 'is 5pm', then it would be impossible to understand clocks as being accurate or not (given that accuracy is typically understood in terms of something like correspondence with reality), yet we do in fact takes clocks to have satisfiable accuracy conditions, and therefore there must be some property out in the world that is expressed by 'is 5pm'. In order to make sense of all of this we need to notice the way in which our timing practices provide a framework according to which clocks can be understood as making accurate or inaccurate claims about the objective world.  The divisions (e.g. the temporal properties) that are picked out by our timing practices are just those that are imposed on temporal reality by our timing practices themselves. Let us consider a spatial analogy, and then we will make connections between the spatial case and the temporal case. Consider a long road along which we are moving at a fixed rate and along this road 124  there are rest stops at regular intervals. We might want to know what the distance is between any location on the road and the next rest stop. Imagine that in order to determine how far between two rest stops we are we decide to extend a semi-transparent overlay with a regular 24 color pattern over the length of the road in which the entire 24 color spectrum is repeated through every interval between rest stops. We can then describe our location between any two rest stops in the following way “here is blue” or “here is orange”. When we say 'here is orange' we are saying that our current location has the property of being in a particular position in a sequence of locations. The predicate 'is orange' has as its extension all of those objective spatial locations that fall under the orange overlay. Accordingly, these descriptions provide us with a characterization of our objective location relative to any two rest stops, even though the subdivision of the road was imposed by something of our choosing. Once that convention is in place, we can then speak accurately or inaccurately about our location along the road.  Notice importantly, that in employing this colored overlay on the road as a means of dividing the length of the road, we do not end up adopting some anti-realism about space.  The spatial locations are all independent of our way of carving up the intervals between rest stops.  The only thing that we are imposing is a means of classifying these locations such that we have a means of describing the objective spatial locations in terms of the class of locations they fall under given our means of carving up the road. We can do the same with the time of day. Consider each passing day, or sunrise / sunset cycle, as being analogous to the regularly spaced rest stops along the road. We then simply establish some representational system that allows for that interval to be subdivided into discrete units. The degree to which we subdivide this interval is simply a product of the specificity we need in order to characterize our time keeping practices. But there is no circularity here in trying to account for our time keeping abilities in terms of arbitrary subdivision that are then explained by our time keeping practices. The time keeping practices go hand in hand with how we subdivide the temporal interval. The two things come together since you cannot have one without the other. 125  When it comes to how we analyse the semantics of 24-hour clocks, we essentially lay out a “transparent” overlay along the temporal dimension dividing up the objective stretch of time into hours, minutes, seconds, etc. We then use this subdivision of the world to predicate of the present moment. When it comes to internal clocks, however, someone might object that as theorists we are anthropomorphizing circadian behaviour in that we describe the content of the circadian system in hours, minutes and seconds. In general there is a risk in the cognitive sciences that as conceptually sophisticated adult humans we may be systematically mischaracterizing the informational content of not only sub-personal states but of non-adult human animals in general. There is some reason to worry about this concern, in general, however, the concern is misplaced here. We may reformulate the content of the circadian system using our language as theorists without mischaracterizing the content of the circadian system.  Our theoretical language may involve different predicates as the circadian system, yet in attempting to describe the circadian system we won't cause any problems unless we start attributing to the system a greater amount of sophistication than is present.  For instance, perhaps we do not want to attribute to the circadian system the same level of precision in time keeping as we would to a watch with a second hand.  However, we can constrain the content we attribute by limiting our attribution of content to the minimum precision required to explain the circadian behaviour and the functioning of the circadian mechanism.    3.6. An information-theoretic account  As mentioned earlier, a system s carries information about some state of the world, t’s being F (i.e. some object being some way), provided that t’s being F is guaranteed by s’s being in state R and as a matter of fact s is in state R. Since we have an understanding of the semantics of clocks, we need a theory of content that is capable of explaining how it is that the neurological mechanisms of the circadian system come to bear their semantic content. While theories of content aren’t often explicitly discussed by 126  cognitive scientists, there is a common approach in the empirical sciences that lends itself (at least initially) to a resemblance theory of content (see chapter 4 for a more detailed discussion of resemblance theories). In particular, many researchers in trying to explain the temporal content of some psychological state will point to the temporal properties of the neurological state.95 For instance, in order to explain the temporal content of a perceptual state, say hearing a loud tone as preceding a quiet tone, theorists might simply point to the fact that the neural state representing the loud tone temporally precedes the neural state that represents the quiet tone.  However plausible some sort of resemblance theory might be for the representation of duration or temporal ordering the situation is rather different when it comes to representations of the time of day (although, see the discussion in Chapter 4 for why even in those cases resemblance is implausible). Resemblance theories are inheritance theories. A representation will represent some property X only if the representation itself possesses that property X (think for example how a paint swatch at the hardware store will represent the color of paint in a paint can by itself being that color). But what would the inheritance story be for the circadian system? As discussed earlier, the content of the circadian oscillator will be something along the lines of <it is now tp> where ‘tp‘ picks out a temporal property that applies to a particular time of day. So, if an inheritance theory is correct, then the representation will have this content only if the representational vehicle itself possesses the property tp. But notice, that this will trivially be the case, as the representation itself is located in time, it will necessarily have the property of being at the particular moment in time that it is supposed to be misrepresenting. In this way, someone could argue that every aspect of the world carries this content as every current state of affairs in existence will exist currently.                                                      95 For a critical discussion of this approach in the cognitive sciences see Dennett & Kinsbourne (1992) and more recently Grush (2006). For a discussion of this approach as it appears in the philosophical literature see Phillips (Phillips, 2014a, 2014b) and Lee (2014a, 2014b) as well as the extended discussion of the topic in chapter 4. 127  Several problems arise when trying to use an inheritance theory for the circadian clock. First, consider cases of misrepresentation (i.e. where the clock fails to accurately keep track of time and produces maladaptive behaviors). Notice that a paint swatch might misrepresent the color of paint in a can by there being a misalignment between the actual paint in the can and the color represented by the paint swatch. The reason is that there is nothing that by necessity guarantees the color of the paint swatch (i.e. the property of the representational vehicle) will match the color of the paint in the can. However, in the case of the time of day there is a necessary connection. The time of day applies to everything that’s occurring at that moment (and in a particular location). So, there can’t be the misalignment between the content of the representation and what is supposed to be represented. Therefore, no possibility of misrepresentation. Second, it’s unclear how the time of day at which a neural state occurs could itself be neurally significant. If neural systems could just access these time of day properties, then there would be no need for a dedicated clock. But as is evidenced by the behavioral effects of damaging the SCN, there is a need for a dedicated clock, so it’s unlikely that the neural system can actually access time of day properties directly. A distinct theory of content is needed, yet there is something importantly right about the resemblance theorist’s attempt to explain the content of the circadian system.  An important aspect of how clocks come to have their contents is that clocks not only represent time, but to use a term from Brian Smith (1988), clocks participate in time. Vital to a clock's ability to keep track of time is that the mechanics of clocks are themselves governed by processes that occur in time, and that the occurrence of these processes in time is regular. A grandfather clock is able to keep track of time because of the regular way in which the movement of a pendulum progresses through time. Similarly, it is the progression of the mechanics of the circadian system that allow the circadian clock to keep track of time.  Contrast both of these cases with an analog clock whose hands are stuck at 5:30. This clock is never correct – not even twice per day – it simply fails to represent the time. The participatory aspect of clocks is vital to their keeping track of time, yet, the fact that clocks represent time is not due to any sort of resemblance 128  between the clock and the temporal features of the environment as shown by cases of misrepresentation. The importance of the participatory nature of clocks is that it is their participation in time that allows them to carry information about time. Perhaps one could employ a causal theory of content (Fodor 1987; Prinz 2004; Stampe 1977), however, given that it is a controversial idea in the metaphysics of time to think that particular moments in time are causally efficacious it seems that a causal theory of content can be ruled out a priori. I propose to understand the content of the circadian system in information-theoretic terms. Recall the earlier characterization:  A system s in state R carries the information that t is F (where t is an object instantiating F), just in case, t's being F is guaranteed by S's being in state r (given appropriate background channel conditions and the laws of nature).  All that is required for an information relation to hold is that there is a relation between pairs of property instantiations. In this case, the relation holds between states of the circadian system and states of the world, as described in the previous section. When the circadian system is in state S1, the organism is in a position to know that the current time of day is T1, because there is a nomic relation that holds that whenever the time of day is T1, the circadian system will be in state S1. As time progresses from T1 to T2, the circadian system will itself progress from state S1 to state S2, thereby putting the organism in the position to know that the time of day is T2. The endogenous activity of the circadian system guarantees the appropriate nomic relation between the states of the circadian system and the time of day, in the very same way in that the physical behavior of a pendulum clock guarantees that when the clock's hand moves, the movement of the hand indicates the current time. 129  What we have so far is an account of how the circadian system successfully represents time, but the true test is whether the circadian system is capable of misrepresentation. This brings up a general problem for information theoretic accounts of content. Pure information theories seem unable to explain misrepresentation since if the state of the world that typically corresponds to a state of the representational system fails to occur when the representational system is in the particular corresponding state, then the representational system will simply fail to carry any information. On a strict account of information, there is no such thing as misinformation (Dretske 1981). Either the system carries accurate information or it doesn't carry any information at all. In these cases we can augment a pure information theoretic account by distinguishing between indication and representation (Dretske 1991). A system indicates that some state of affairs is the case, just in case the system carries information about that state of affairs. In other words, indication is a factive relation that holds between an information bearing system and that which it carries information about. However, genuine representation occurs when a state that typically carries information acquires the function of carrying that information and can thereby be said to represent what it typically indicates. A system will acquire the function of representing something when the system typically carries information about some state of the world, and then the representation is in some sense selected for and maintained in order to signal the state of the world.  A clock has the function of telling the time, because typically, when everything is functioning properly, the state of the clock carries information about the time and we use the clock on the assumption that it is successfully carrying this information. Therefore, when the background channel conditions that underpin the information channel are damaged, and the system misfires and fails to genuinely carry information, we can still treat the system as though it were successfully carrying that information.96  Similarly, the circadian clock has the function of representing the time of day, because it has the ability to carry this information, and the adaptive value of the circadian clock resulting in the circadian clock being selected for as a means of                                                      96 Misrepresentation, then given this analysis, can be understood in terms of a failure of channel conditions. For instance, the speedometer on your car my misrepresent the speed at which you are travelling due to some damage in the channel that mediates the location of the needle on your dashboard and the movement of your wheels. 130  serving this function, and downstream systems operate on the assumption that the circadian clock is successfully carrying this information.  As a result, even when the clock fails to carry information about the environment, the clock is still used by the organism as though it were carrying information.  This is all because the clock's internal states typically signal the current time of day, and in cases of misrepresentation, the internal state of the clock is understood as signalling the time of day even though the clock fails to carry any information. What we have now is an account of how the circadian system comes to have its particular contents. In addition, we have the answer to the original worry posed by the non-causality argument. Recall, the non-causality argument against the sense of time made the following claims: the senses perform their information gathering function in virtue of a causal relation between the senses and that which they are senses of. Time cannot be a cause. Therefore, there cannot be a sense of time. However, we are now in a position to see how this argument fails. The circadian system is a system that is capable of directly establishing a real-time information link with a particular aspect of the environment. But there is something interesting to notice here. The reason why there can be a sense of time is that the circadian system participates in the very thing that it represents. As time progresses in the environment, the internal operations of the circadian system themselves develop and unfold in time. It is this sort of parallelism that guarantees the real-time information link. This is something that is plausibly unique to time and temporal representation. If we look at any of the other senses, there is no similar parallelism. Time, in this sense, is a unique aspect of our environment. However, this strangeness does not create any roadblocks in our understanding of how animals come to represent time. Instead, it is this very feature of time that makes it possible for there to be a sense of time.  131  3.7. Conclusion  As we typically understand the mind, in both scientific and folk contexts, the senses play a crucial role in explaining our epistemic access to the world around us. They are the first step in how we come to acquire information about the contingent goings on of our environment. However, we find that there is quite a bit of disagreement over how to even distinguish the sensory systems from the rest of the cognitive architecture and also how to distinguish one sensory system from another. In many ways, the Aristotelean senses are often treated by some philosophers as being special in that only they deserve to be categorized as genuine sensory systems and that scientific appeals to non-Aristotelean senses are in some sense changing the subject. But, as we’ve seen, there is in fact a common core to the notion of sensory system that extends through both the folk and scientific uses of the term. The senses are the information uptake systems. Disagreement arises not over this common core idea but over the place to draw the line between cognition and sensory systems and how to individuate one sensory system from another. I have argued, however, that there simply is no single place where to draw these divisions. For various explanatory needs we may need to distinguish the senses from the rest of the cognitive architecture at a certain place as opposed to another place. Similarly, for various explanatory needs we may need to individuate the senses according to different criteria. These two questions are related, and how we determine one will constrain how we answer the other, but importantly both questions are amenable to fluid answers.  Once we have this understanding of the sensory systems in hand, we can then turn to the question of whether there is a sense of time that parallels our other sensory systems. We found that many researchers have tried to deny that there is in fact a sense of time, but argument for this position are far less commonly found. But we were able to isolate two arguments that are explicitly found in the literature. Both arguments focus on the point that there is no system for temporal representation that picks up temporal information in the direct manner that is typical of the sensory systems. In my argument, I 132  have shown that this simply isn’t the case. The systems that coordinate animal activity with the approximate time of day, the circadian systems, pick up information about the temporal structure of the world in a way that cannot be explained by appealing to the informational capacities of the other sensory systems.   In arguing for this claim we actual came across a rather interesting point that has to do with temporal representation. While the other sensory systems require a causal connection between the aspects of the environment they track and the states of the sensory system itself, the circadian system doesn’t require any sort of causal connection of this sort to maintain its information gathering role. It is because the circadian system’s operations occur in time at a regular rate that they are able to keep track of and represent the approximate time of day in a format that is accessible to down stream neural processes. So, unlike Dretske’s example of two TV’s that can carry information about one another without having any causal influence on one another, the case of does not even require that there be a continuous causal path (i.e. TV1 to broadcaster to TV2) between the information carrying system and that aspect of the world that it carries information about. Clocks, as Brian Smith pointed out, not only represent time but they also participate in time. As a result, time may stand alone as the only sensory system that does not require a causal connection to its content.  Ultimately, however, the goal of this chapter is to point out one way in which animals come to coordinate their behavior with their environments. In this case, the coordination occurs at the relatively long time scale of daily patterns and is subserved by a dedicated system that gathers information about the approximate time of day that can then be used to coordinate a number of behavior. Importantly, though, this system only accounts for a very small aspect of how we can come to represent the temporal structure of our world, but by having looked at this system, we get a first glimpse at the ways that the representation of time is similar to and also differs from the representation of the non-temporal aspects of our world.  133  Chapter 4: The Temporal Structure of Experience: Against the Atomism / Extensionalism Debate  In the previous chapter we looked at the role of the circadian system in coordinating activity with the temporal structure of the environment. Specifically, it was argued that the circadian system counts as a genuine sensory system that provides a wide range of organisms with information about the approximate time of day. However, the extent to which the circadian system is implicated in coordinating animal behavior with the temporal world is rather restricted as it is not capable of producing the fine grained temporal discriminations that are apparent in our experience of the world and that are needed for the coordination of fine-grained motor activity.  In this chapter, we’ll turn to the ways in which humans97 gather information about time at the much shorter time scale of milliseconds to seconds. It is here where we find the temporal capacities that pervade our experience of the world. From perceiving the slight variations in phonemic structure, music, and movement to being able to navigate a busy sidewalk or catch a ball to feeling the painful sting of an awkward pause in conversation or a mis-delivered joke, all of these activities and experiences require that we be able to keep track of an notice the temporal structure of the events around us at very short time scales through multiple modalities.  As we’ll see, there is no single system or mechanism that provides us with the ability to perceive or experience the temporal features of the world at this time scale. In fact, ‘temporal experience’ does not even pick out a single phenomenon or capacity. Rather, ‘temporal experience’ admits of a number of polysemous interpretations that are held together by the fact that the various uses or interpretations of the term apply to phenomena whereby we pick out, track, or represent the temporal structure of the world around us. Instead of the unitary approach, where temporal experience is a single phenomenon, I will                                                      97 The restriction is to humans at this point since we’ll largely be talking about temporal experience. 134  argue for the fragmentary model of temporal perception according to which our capacity to perceive time is the product of a great many different mechanisms that represent distinct aspects of the temporal structure of our world.  While the argument for the fragmentary model of temporal perception rests on largely empirical considerations I will show how this model of temporal perception has direct implications for philosophical accounts of temporal perception. In particular, as was briefly mentioned in the previous chapter, there is tendency within philosophy and the cognitive sciences to explain the temporal content of perception by appealing to the temporal structure of the perceptual process itself. For instance, you may ask someone to explain in virtue of what their experience of a thunderstorm presents the flash of lightning as being prior to the crash of the thunder. A perfectly common reply would simply be to appeal to the fact that first they had an experience of the flash of lightning and then they had an experience of the crash of thunder. In this way, the temporal structure of experience has a role in explaining how experience has its particular temporal contents through a commitment to the mirroring constraint.  Mirroring: the temporal contents of an experience mirror the temporal structure of that experience itself.  Now, quite a bit of the philosophical literature on temporal experience is centered around debates about the truth of the mirroring constraint. Extensionalists claim that the mirroring constraint, in a sense to be further described below, must be true. Atomists, on the other hand, deny the mirroring constraint and claim that the temporal contents of experience rarely (or never!) mirror the temporal structure of experience itself. One thing, however, that both views have in common is that they agree that the 135  mirroring constraint is something that is either true or false across the board.98 That is, there should be a univocal answer to whether the mirroring constraint is satisfied or not for temporal experience in general.  The main conclusion of the fragmentary model for the debate over the mirroring constraint is simply that we should not expect a univocal answer to whether the mirroring constraint is true or not. Temporal experience is not a unified phenomenon and therefore we shouldn’t expect a unified response. Second, and perhaps more importantly, I argue that once we look at the distinct aspects that make up the fragmentary model of temporal experience we’ll see that in fact there isn’t a single answer to whether or not the mirroring constraint is satisfied. Rather, distinct aspects of our temporal experience are subserved by mechanisms that employ radically different strategies for representing the temporal structure of our world. Some of these employ representational strategies that respect the mirroring constraint while others do not.  Once we see that there isn’t a single answer to the question of whether or not temporal experience satisfies the mirroring constraint the debate over mirroring begins to lose much of its philosophical appeal. Instead, I argue, that the genuine puzzlement about temporal experience that originally motivated discussions of the mirroring constraint is largely left untouched. Instead, in order to properly understand how temporal experience is possible and how it is that we come to perceive the fine-grained temporal structure of our world we