Commentary on Stuart Shanker and Barbara King 'The Emergence of a New Paradigm in Ape Language Research' to appear in Behavioral and Brain Sciences.

Information processing and dynamical systems approaches are complementary

David Spurrett

University of Natal
South Africa
+27 (31) 260 2309


Shanker and King trumpet the adoption of a 'new paradigm' in communication studies, exemplified by Ape Language Research. While cautiously sympathetic, I maintain that their argument relies on a false dichotomy between 'information' and 'dynamical systems' theory, and that the resulting confusion prevents them from recognizing the main chance their line of thinking suggests.


Speaking very generally, information theory is primarily concerned with structure (understood via concepts like compressibility, probability, entropy, etc.), dynamical systems theory (DST) with change (making use of differential equations, trajectories through state-spaces, etc.). For the purposes of science, the two, thus understood, are pragmatically complementary, and probably conceptually inextricable (Collier 1999). The behavioral sciences in particular are simultaneously about some distinctive types of structure, and the processes (themselves structured through time) by which these structures are produced and maintained. Not only is it not required that we chose between the two, Shanker and King’s conception of them as being alternatives is probably incoherent.

As a mathematical apparatus, information theory makes no claims at all about the nature of cognition. In particular it does not demand an episodic turn-taking model of communication between cognitive systems, require that such communication make use of conventionalized encryption, require a theory of cognitive processing dependent on passive ‘internal’ representations, or entail anything at all about whether non-humans are capable of language. Confusing information theory with these commitments means relinquishing crucial tools for rejecting them as requirements for a model of cognition (for example that they demand the execution of unlikely encoding and decoding tasks), and for developing the details of a distinctive alternative, insofar as these involve specifying what acts of discrimination are to take place, under what time constraints, in what ways different streams of information processing relate to one another, and how they manifest in behavior.

Conversely DST, itself a mathematical apparatus, can in principle be used to describe any distribution of cognitive labor, including highly internalist approaches involving just the types of representation and communication Shanker and King reject. This fact is obscured by Shanker and King’s tendencies to conflate dynamical with ‘distributed’ as though dynamical systems approaches automatically involve control parameters in the wider environment, and to limit much of what they say about DST to metaphorical claims about ‘dance’, and impressionistic references to co-regulation, canalization, and so forth.

Shanker and King, then, draw a coherent-seeming contrast, by means of two expedients. First, saddling information theory with a range of non-essential commitments amounting to a restrictive and intrinsically implausible conception of mind, and second, providing almost no non-metaphorical detail about DST. Acceptance of the resulting false dichotomy makes all references to information, encoding, and representation seem suspect, thereby leading Shanker and King to miss their main chance.

I take it, for present purposes without argument, that some of our most empirically and theoretically powerful treatments of language proceed on the assumption that language is in some sense symbolic and systematic. Many take the fact that language presents this ‘digital’ aspect as justification for thinking that it is fundamentally digital. Even those wary of viewing language as digital (due to worries about neural implementation, adaptive explanation, developmental plausibility, etc.) nonetheless regard the digital aspects of language as setting an important explanatory target for cognitive science. What motivates this caution is recognition that even if any given theory of language turns out incorrect, we’d be unwise to relinquish the data, or the powerful and economical ways of representing it made possible by treating language as in some sense digital. This type of conservatism is entirely proper, and furthermore provides a key constraint on any ‘paradigm change’ likely to affect the course of scientific research. By ignoring or undervaluing it, Shanker and King demand too destructive a revolution.

Shanker and King’s positive case involves drawing our attention to a variety of reasons (some from well-established research programmes) for thinking that non-digital aspects of interactive, situated behavior, are crucially important for language learning, and for the functioning of language. I agree, and also grant that taking these aspects of behavior seriously (as increasing numbers of researchers do) could lead to significant changes in how language is understood. The crucial question that Shanker and King’s destructive radicalism prevent them from taking seriously, though, is how do the digital and non-digital aspects of language relate to one another?

Emotions, which are sometimes co-regulated, can be seen (Ross and Dumouchel MS) as strategic signals encoding preference intensities (typically more difficult to infer than orderings) in ways that, unlike standard commitment devices, do not require explicit construction prior to strategic interaction. By having preference intensities thus (even if roughly) represented, otherwise intractable strategic problems can be negotiated. I speculate that ‘dance-like’ aspects of interactive behavior could support the digital aspects of language, in effect by constraining the computational searches required for handling symbolic and systematic tasks.

To evaluate this hypothesis would require exploring the relationships between different streams of information processing (some overlapping or interacting), different types of control system (some enabling co-regulation), each with particular processing capacities, time-budgets, varyingly distinctive histories and so forth. It is interesting to know that ‘cross-modal’ matching (sect. 1, para. 4) takes place, for example when dynamical properties of an infant’s gesture are copied in its mother’s responding vocalization. But saying we should eschew talk of information in favor of viewing such activities as part of a ‘dance’ is unhelpful. As behavioral scientists we want to know how such matching is possible, and what difference it makes to learning and using language. This demands asking about the (information) processing capacities and properties of different organisms and their cognitive sub-systems, and how these relate to the (dynamical) control of behavior. If research in this direction is to be pursued, though, we need to eradicate the reflex to go on the offensive whenever we hear reference to ‘signals’, ‘codes’ or ‘representations’, let alone ‘symbols’ or ‘grammar’.


Collier, J. (1999) Causation is the Transfer of Information. In: Causation, Natural Laws and Explanation, ed. Howard Sankey, Kluwer: 279-331.

Ross, D. and Dumouchel, P. (MS) Emotions as Strategic Signals.