2 th - century theories of perception A similar problem is supposed to arise in the other modalities, but although this is true of audition, it is not clear that the same applies to the rest of the senses (Burge 2010) In the case of seeing, the question of how we derive a rich view of objects from what is projected on the retina is answered in different ways Structuralism points to associations based on past experience Gestalt psychologists hypothesize a number of principles of organization, focusing mostly on how the principles regulate conscious appearances Constructivism, not unlike Gestalt psychology, centers on some principles of organization and views them as part of an unconscious process that not only organizes, but also adds information to what is contained in sensory stimulation (Helmholtz 1867).13 According to constructivism, the perceptual process supplements the initial stimulation by performing a type of inference In vision, the retinal ‘image’ is processed by using hidden ‘assumptions’ to reach perceptual ‘conclusions’ about the environment A given discontinuity in light intensity, for example, may be taken to be an edge based on the assumption that discontinuities of a certain kind are typically edges in our world The perceptual conclusions are ‘guesses’ as to what is the most likely cause of the stimulation In this view, the world appears as it does because we construct it to appear as it does In constructivism, the internal states of perceptual processing that encode information about the world are typically called ‘representations’ Representations are not themselves objects to the experiencing subject Generally, in cognitive science, representations are subconscious states that carry information about the world, where carrying information is spelled out in different ways (Dretske 1981) The conscious subject is not aware of representations, and she has little or no control over them A representation, in this context, is a theoretical notion introduced to explanatory work It is not clear how representations, understood in this way, relate to the intentional entities postulated by representationalism (see section 1.3.) Because the inferential view conceives of perception as mediated by representational states, it also conceives of perception as indirect, a term reminiscent of indirect realism The sense in which perception is indirect in the inferential framework, however, is different from the sense employed in sense-data theory (see section 1.1) According to the latter, sense-data are the immediate objects of awareness, and everything else is perceived by or in virtue of perceiving sensedata This is not the case in inferential accounts of perception Percepts are built out of sensory states that carry information, but the sensory states are not perceived or given to the individual In this respect, constructivism also differs from structuralism, according to which the atoms of sensation are introspectively available Another important difference with both the sense-data tradition and the basic elements of structuralism is that sensory states in constructivism need not be modality specific Although inferential views typically take vision to be their paradigm example, and some proponents think of the senses as distinct (Fodor 1983), some constructivists have been open to the idea that much of sensation and of perception are multimodal This is true of recent Bayesian accounts to which I return later.14 115