There are several facets of'binding' (definition 1) that excite philosophers. A classical aspect has to do with the persistence of the identity of things whose constituents turn-over with time, such as the self (see the 'Ship of Theseus' problem in "persistence). Neuroscience and the philosophy of mind presently focus on a distinct type of the 'binding problem', which refers to the ability of the brain to bind, within a fraction of a second, the features of a complex "stimulus into a coherent, meaningful percept (definition 2). Interest in this type of problem has a long history (e.g. Hume 1739). Neuroscience has dragged it into the laboratory, although for many scientists it still retains an excessively 'soft' connotation (also see below). Consider vision: different types and combinations of visual attributes are processed in the brain in multiple streams (Knierim and Van Essen 1992). How do they recombine to yield a coherent visual percept (Treisman 1993; Singer and Gray 1995; Shadlen and Movshon 1999)? This is the 'Humpty Dumpt/ problem: 'Humpty Dumpty sat on a wall/ Humpty Dumpty had a great fall/All the King's horses and all the King's men/ Couldn't put Humpty together again' (Carroll 1872). In the brain Humpty is put together again. Or at least so we sense. How sad is it not to: 'On an incredibly clear day/. I saw ./That Great Mystery the false poets speak of.. ./That there are hills, valleys and plains/That there are trees, flowers and grass/There are rivers and stones/But there is no whole to which all this belongs/ That a true and real ensemble/ Is a disease of our own ideas.' (Pessoa 1914).1
Binding is related to the coherency of all kinds of internal representations (definition 3), not necessarily in the context of sensory perception. Hence it surfaces, either implicitly or explicitly, in discussions of "memory (Squire et al. 1984; Teyler and DiScenna 1986; Damasio 1989; Hommel 1998; Dudai and Morris 2000). These discussions usually refer to "declarative memory, but sometimes generalize to simple stimulus-response representations. Clearly, although the mainstream interest in the 'binding problem' is still in the context of perception, whatever will be gained there will contribute to the understanding of memory as well.
The binding problem binds several subproblems. Here is a selection:
1. Parsing: How are the relevant elements selected among other elements in the perceptual or mental space (Treisman 1999)? And how much of this selection is constrained by "a priori rules?
2. Encoding: How is the binding marked, maintained, and read by other systems in the brain (ibid.; "cell assembly)?
3. Mapping: How are the elements, once bound, kept in the correct structured relations (ibid.; "map)?
4. Flexibility: How are the bound elements reused in binding without lingering interference of the previous binding(s)?
Each of these questions could be tackled at multiple "levels, from that of the computational theory, via the "algorithms that implement the computations, down to the biological hardware that implements the algorithms. Discussion of binding in cellular neurobiology is still rather uncommon. The main focus is on the higher levels of neuronal circuits, brain systems, and cognition. At these levels, it is methodologically convenient to distinguish two types of approaches: top-down or cognitive, and bottom-up or neurobiological.2 The "classic top-down approach is that of the Gestalt School (Gestalt, from German for 'shape'; Koffka 1935; Hochberg 1998; "insight). This school of psychology, founded in Germany in the early twentieth century, has promoted the view that the nature of perceptual parts is determined by the whole, and that enquiry into the mind should consider global organization and proceed top-down. Unfortunately not much top-down analysis of the brain was possible during the formative years of the Gestalt. In more recent cognitive psychology, an influential "model is that of 'feature integration' (Treisman and Gelade 1980; Treisman 1993). This model considers "attention as the binding agent. It proposes that simple perceptual features are registered in parallel across the visual field, in a number of specialized subsystems. Focused attention scans serially, within milliseconds, through a 'master-map' of locations, accessing the features present there at that point in time. The features are integrated, or 'glued', by the attentional 'beam' ("metaphor; for critical discussions of 'feature integration', see M. Green 1991;Van der Heijden 1995; Treisman 1995).
The neurobiological approach attempts to identify computations and algorithms relevant to binding in the brain and their physiological implementation. It focuses on the "cortex and on thalamocortical interconnections; in discussions of the role of binding in memory, attention is also devoted to the "hippocam-pal formation and to its role in coherency of internal representations. Two major types of solutions come up in neurobiological models of binding. The first type of solution is that binding is based on a place code ("map), and is performed by hierarchical combination of coding units, which converge anatomically on a master location ("homunculus; Barlow 1972; also discussions in Singer and Gray 1995; Grossberg et al. 1997; Bartels and Zeki 1998). The second type of solution proposes that binding is based on a temporal code (Eckhorn et al. 1988; Hardcastle 1994; von der Malsburg 1995; Engel et al. 1997). The basic idea in this case is that feature-detecting neurons are bound into coherent representations of objects if they fire in synchrony. Neurons in the cortex have been indeed observed to engage in recurrent bursts at frequencies of 30-70 Hz, and this has specifically been proposed as a candidate mechanism of binding. It also fits psychophysical data, which suggest 20-30 ms as the time scale of a 'cognitive beat' ("capacity, "percept). At this stage, the temporal synchrony hypothesis is still mostly phenomenological. It is not yet clear whether the oscillations represent a causal mechanism, a phenomenon, or an epiphenomenon ("criterion). To understand what's going on, one would wish to identify the semantics of the representational code(s), the source of the oscillations (i.e. intrinsic, emergent ensemble properties, top-down induction or executive control), and the hardware components (e.g. "coincidence detector).
So is 'binding' as defined above a problem, or a pseudoproblem? The same question applies to other "enigmas of the brain. What distinguishes 'binding' from some other unresolved brain processes and mechanisms, and occasionally endows it with a mystic flavour, is probably its association with major philosophical aspects (or some would say 'spin-offs') of the neurosciences. These include the mind-body problem and "consciousness (e.g. Crick and Koch 1990). Many scientists hesitate to touch these issues, others do it rather enthusiastically. Crick (1994) remarks on the 'binding problem' that 'it is not completely certain that this is a real problem or the brain gets around it by some unknown trick'. Sure the brain does its trick, and the problem is hence only ours to solve. 'That wonder is the effect of ignorance has been often observed' (Johnson 1751). It is therefore likely that with time, the 'binding' will stay but the 'problem' dissipate.
Selected associations: Algorithm, Attention, Cell assembly, Coincidence detection, Percept
'Pessoa's lines seem to echo a neurological disorder, Balint's syndrome, in which the ability to perceive the visual field as a whole is disturbed due to bilateral damage to the occipitoparietal region (Halligan and Marshall 1996).
2Bottom-up analysis of perceptual binding commonly attempts to account for cognitive phenomena by circuit and multicircuit properties. In the process, it could still employ top-down analysis of synaptic properties.
Was this article helpful?