Introduction

That causality has a central role in cognition, whether human and nonhuman, is not controversial. What might be controversial is whether ‘cause’ means exactly the same thing in human and nonhuman cognition. Or, in other, more philosophical, words, would attribution of a common causal belief — e.g., “The fact that it rains will cause Mother not to take me for walk” — to my dog, Tolkien, and to my 11-year-old daughter, Abigaël, make sense? On a superficial view, we might say that both exemplify, mutatis mutandis, the same behavior, Abigaël is curled in an armchair reading a book, Tolkien has sneaked into another armchair and both are casting melancholy eyes at the rain beating on the windows. The question, at a deeper level, is whether there is more to Abigaël’s causal belief than a mere association between rain and no walk and whether, if there is, it might be legitimate to attribute that additional feature to Tolkien’s causal belief too. For instance, Abigaël may have a mentalist explanation to the effect that I believe that rain makes one wet and that I don’t like to get wet which is why I choose to stay indoors when it rains. This explanation, presumably, is not something that it would make sense to attribute to Tolkien. Can we say exactly in what the difference between the causal belief as attributed to Abigaël and as attributed to Tolkien lies? It seems to me that the difference lies in the fact that Abigaël has an explanation for the association whereas Tolkien is left with the association, naked as it were. What is more, Tolkien is not, and would not be, interested by an explanation, while Abigaël would not, indeed should not, be satisfied with the naked association. Though it may be adventurous to see the difference between human and nonhuman causal cognition as lying in the existence in the first and absence in the second of explanation, this is the claim I want to make. I even want to go slightly farther and say that association is mostly between perceptible entities, while explanation, more often than not goes beyond the observable (as is the case with Abigaël’s explanation for why I won’t take her for a walk when it rains). As Hume (1975, 74) famously noted, “All events seem entirely loose and separate. One event follows another; but we never can observe any tie between them. They seem conjoined, but never connected”. Hume deduced from that basic observation to the perceptibility of the association and the non-perceptibility of the causal link the inexistence of the second, but I will not be concerned with that metaphysical claim here.

Are humans associative animals?

I claimed above that causal cognition in humans is not or is not merely associative. This claim can be (and has been) cashed in different ways. To begin with, Premack (1995) distinguished between arbitrary causal knowledge (hereafter ACK), resulting from associative learning — dependent on contiguity and repetition —, and natural causal knowledge (hereafter NCK), strongly domain specific and a priori — not dependent on contiguity and repetition. Another way of putting it might be to say that ACK is based on induction, while NCK can be used as the basis for deduction. Typical NCK in humans is relative to folk psychology, folk physics and folk biology. It is difficult if not impossible to attribute it to nonhuman animals. To the extent that it is not based on associative learning, it does obviously justify my claim. NCK is however not what I want to discuss here. I’ll concentrate on ACK. Regarding ACK, there are two possibilities:

• It is based solely on associative learning in both human and nonhuman animals;
• Though it is based on associative learning in both human and nonhuman animals, associative learning is not sufficient for ACK in human animals, though it is sufficient in nonhuman animals.

This raises a further question, which has to do with why association is not sufficient for ACK in humans. After all, if association is adaptive for nonhuman animals, why should it not be enough for humans? Another obvious question concerns what explanation exactly is.

My (tentative) answer will be that the response to both questions goes somehow through the fact that humans are the only linguistic species. The rest of this paper will be devoted to a short review of those experimental works which purport to show that association is not the whole story of human ACK and to some, admittedly speculative, hypotheses on the role of language in the difference between human and nonhuman ACK.

To begin with a pivotal point of contemporary philosophical literature on causality (now being investigated experimentally — see Roese 1994, Roese & Olson 1996, 2003, Pennington & Roese 2003), there is a strong link between counterfactuals and causal reasoning. A very common philosophical view is that saying “C caused E” is tantamount to saying two counterfactuals: “If C had occurred, then E would have occurred” and “If C had not occurred, then E would not have occurred”. However, though it is plausible that counterfactual reasoning is uniquely human, it is not clear that the link between causal and counterfactual reasoning would support a more than associative view of human causal cognition. This is because counterfactual reasoning commonly bears on the associated facts rather than on explanation. This does not mean that counterfactuals have nothing to do with explanation: they often represent the idea of a necessary link, which may be the first point of departure between human and nonhuman animals. However, they do not, in and of themselves, constitute an explanation. So we come back to the first question: what is an explanation?

This may be the right place to introduce a distinction advocated by Waldmann (2000, 2001), between predictive and diagnostic learning: while predictive learning goes from cause to effect, diagnostic  learning goes from effect to cause. It should be fairly clear that explanation, whatever other feature it might possess, is diagnostic in its direction. However, Waldmann goes farther than this simple distinction, saying that a sheer associationist model (e.g., Rescorla & Wagner 1972) cannot account for the whole of inductive causal — both predictive and diagnostic — learning,  because it is basically indifferent to causal asymmetry, being concerned not with causes and effects, but with cues and outcomes, where cues and outcomes can be either causes or effects. Waldman goes on to show that depending on whether the task is a predictive or a diagnostic task, some well-known effects of associative learning (e.g., blocking and overshadowing) do not operate identically. In other words, causal directionality plays a role in causal learning, contrary to what associationist models would predict. In a series of papers, Waldmann and colleagues (Waldmann & Hagmayer 1998, Waldmann & Martignon 1999, Hagmayer and Waldmann 2004) have gone farther, defending a more abstract view of causal cognition, based on a Bayesian network model, arguing that human learners rely on abstract causal categories (e.g., multiple causes, multiple effects, causal chain) to learn causal relations.

This, then, is the first intimation of the fact that ACK does not only rely on association. However, the Bayesian model proposed by Waldmann and his colleagues still strongly relies on covariation of causes and effects. A further question is whether covariation really is the central factor in human causal learning. Dennis & Hahn (2001) examined order effect in causal relations judgment, through presenting subjects with the same covariation data but in different sequences. There was a strong primacy effect, suggesting that covariation is not the whole story behind causal learning and that a belief-updating process may be operative. An indirect support for this view could be found in Lovibond’s (2003) experimental work (using a simple fear conditioning paradigm in humans) who has shown that associative models might profitably be reinterpreted as modeling inferential and propositional learning, given that fear conditioning in humans operates indifferently from physical stimuli or from linguistic instruction. Ahn and colleagues (1995) were also responsible for another body of work relevant to the problem, questioning the impact of covariation relative to that of mechanism information in causal attribution. Ahn and colleagues devised a series of tasks in which subjects, to give a causal explanation, could ask either covariational (who-, what-questions) or mechanism (how-questions). They found a strong preference for information on causal mechanisms rather than on covariation in all of their tasks. They went on to point out that explanation is more readily understood as mechanism-based (i.e., general laws) than as covariation-based, noting that mechanism-based explanations are truly explicative in that they are generative, allowing one to make predictions about new situations in an abstract way. A view that indirectly agrees is that of Eagleman & Holcombe (2002), where the authors discuss Haggard et al. (2002) paper reporting on subjective judgments of the timing of events when the subject reports on the result of one of his/her own action versus an isolated similar event. There was a general reduction in the reports relative to the actual delay for the result of intentionally produced action. Eagleman & Holcombe  explain this surprising result via the idea that “events known to be causally related are more likely to be close in time and space than unrelated events” (p. 325). This supports the view that you can deduce association (and, hence, temporal contiguity) from causal mechanism.

To sum up this section, then, both human and non-human animals do rely on association for ACK, but that’s not the end of the story for humans, who also use abstract causal schemata and rely on general explanations, rather than staying with simple association. Finally, we are now in a position to give an (informal) definition of explanation: an explanation invokes a general mechanism that accounts for the correlation of a given effect with a given cause.

Is language the answer to why humans are not simply associative animals?

The first question to ask may be if indeed humans are the only animals not satisfied with association. Granted, my dog Tolkien does not go farther than association, but might not more cognitively complex animals, e.g., chimpanzees, actually go farther than association and use, as humans do, though maybe in a more restricted fashion, abstract causal models, as well as look for general explanations possibly based on invisible mechanisms? And if they don’t, while we certainly do, what can explain the discrepancy between us and them?

The question of whether chimpanzees look for explanation was explored by Povinelli and Dunphy-Lelii (2001) in two ingenious experiments comparing preschool children and chimpanzees (9;4 to 10;3 year in the first experiment). The task in both cases was simply to stand several blocks on platforms covered with an irregular mat in which some holes with a regular surface had been made. In both experiments one sham block was proposed: in the first experiment, this block couldn’t be made to stand because its ends had been beveled; in the second experiment, regular and sham blocks were visually identical, being L-shaped, though weights were placed in either the long or short size, making possible or impossible to stand the block on its long axis. The results were interesting: in the second experiment, where the difference between sham and regular blocks was not visible, 61% of the children investigated the sham block to try and find out why it couldn’t be stood up in the wanted position, while chimpanzees didn’t. This has led to the unobservability hypothesis (see Vonk & Povinelli in press), according to which “one of the important ways in which humans differ from other species is that our minds form and reason about concepts that refer to unobservable entities or processes” (p. 5). Vonk and Povinelli go on to “suspect that the underlying ‘abstractive depth’ that makes reasoning about unobservables possible co-evolved with natural language” (idem).

This hypothesis seems to fall foul of a series of experiments made by Varley and her team (see Siegal et al. 2001, Varley & Siegal 2000, Varley et al. 2001) which show that agrammatic aphasics can nevertheless still solve reasoning, causal and theory of mind tasks (and ToM tasks, by definition, involve unobservable entities). Thus, operational language is not mandatory to succeed at such tasks. On the face of it, this seems to contradict Vonk & Povinelli’s hypothesis about a link between the ability to conceptualize unobservables and language. However, Varley (1998) has herself observed that her patients had normal linguistic abilities until mit-adulthood, leading her to the conclusion that her “results have nothing to say about the role of language in the development of thinking. It may well be that language is necessary to configure central cognition for certain types of cognitive activity” (p. 145). More troubling might be the fact that pre-linguistic children are supposed to engage in sophisticated reasoning abilities as evidenced in NCK. Even this, however, should be nuanced: false belief test does not seem to be passed before language has set in (see Reboul 2004 for a review) and it is possible that babies’ performances at folk physics habituation/dishabituation tests could be explained through more basic abilities than has been supposed, as proposed by Povinelli (2000). Let us in fact suppose that, as claimed by some researchers, NCK develops through time (which, by the way, does not contradict innateness factors). In this case, the apparent contradiction between the unobservability hypothesis and its link with language and prelinguistic or aphasic abstract thought disappears.

What is still mysterious, however, is how and in what way language is linked to the conceptualization for unobservables, that Vonk and Povinelli see as specific to humans by contrast to nonhuman animals. To try and clarify that link, let us go back to what is usually said about the evolution of language. The current opinion is that it evolved for communicative purposes. Apart from the fact that Chomsky thinks that it didn’t evolve but just emerged, he has violently attacked the communicative account in a number of papers, of which I will only quote the most recent (Chomsky 2005). At the beginning of this paper, Chomsky quotes a number of eminent biologists (Jacob, Monod, Luria) to the effect that communication would not have produced a great selective pressure to produce language. As both a philosopher and a linguist, I entirely concur with this view, being firmly convinced of the cognitive import of language. However, it is interesting to ask what exactly is meant by language. A current and popular model of language evolution was given by Jackendoff (1994) who saw it as involving a series of steps or stages: From animal communication — protolanguage — to Chomskyan universal grammar (UG). Animal communication differs from proto language in being finite in number of items and in being unable of displacement (the ability to refer to absent or non-existent objects). Protolanguage has a non-finite lexicon and allows for two-words inference but has no function items (if, that, the, where, etc.) or morpho-syntax which distinguishes it from UG. According to most linguists (including Jackendoff), the big evolutionary step is from protolanguage to UG. Chomsky’s view is interestingly different: UG — now reduced to very few operations — emerged as a function of complexity, being triggered by the necessity of linking isolated but numerous concepts in a generative (and potentially infinite) way, without any evolutive — in the major adaptive sense — process being involved (it should be noted that this Chomskyan hypothesis receives support through the mathematical models being developed by Nowak and colleagues: see Nowak 2001). If this is right, the major step was the augmentation of the number of available concepts, which, pace Anderson 2004, Maynard Smith & Szathmary 1999, may have been the decisive evolutionary step. In other words and supposing that the protolanguage hypothesis does make sense, the major step would have been going from the closed systems characteristic of nonhuman animal communication to the open systems characteristic of human cognition with their lexical and conceptual open-endedness. It has often been pointed out that displacement does not exist in nonhuman communication system and it can be argued that it is not clearly and uncontroversially present in the so-called talking apes (see Anderson 2004 for a discussion). This, I think, is what takes you from a closed to a truly open-ended system and this, one should emphasize, is what allows one to develop concepts for unobservables, of which it should be noted that they are strongly implicated in NCK, and, to close the loop, the explanations behind ACK frequently make use of NCK.

Conclusion

I’ve tried to show that causal cognition, though partly common in human and nonhuman animals through the associative basis of ACK, can nevertheless not be reduced to a simple associative process in humans, due to the fact that it involves a need for explanation which is not to be found in nonhuman animals. This major difference between human and nonhuman causal cognition has been explained by the unobservability hypothesis. I’ve tried in the last part of the paper to sketch an account of how and why the human ability to conceptualize unobservables is intimately linked with the human capacity for language.

References

Ahn, W-K., Kalish, C.W., Medin, D.L. & Gelman, S.A. (1995), “The role of covariation versus mechanism information in causal attribution”, in Cognition 54, 299-352.

Anderson, S.R. (2004), Doctor Dolittle’s delusion: animals and the uniqueness of human language, New Haven/London, Yale University Press.

Chomsky, N. (2005), “Three factors in language design”, in Linguistic Inquiry 36/1, 1-22.

Dennis, M.J. & Ahn, W-K. (2001), “Primacy in causal strength judgments: the effect of initial evidence for generative versus inhibitory relationships”, in Memory & Cognition 29/1, 152-164.

Eagleman, D.M. & Holcombe, A.O. (2002), “Causality and the perception of time”, in TICS 6/8, 323-325.

Haggard, P., Clark, S. & Kalogeras, J. (2002), “Voluntary action and conscious awareness”, in Nature Neuroscience 5/4, 382-385.

Hagmayer, Y. & Waldmann, M.R. (2004), “Seeing the unobservable — inferring the probability and impact of hidden causes”, in Proceedings of the 26^th animal conference of the Cognitive Science Society, Mahwah, NJ, Erlbaum.

Hume, D. (1975), Enquiries concerning human understanding and concerning the principles of morals, Oxford, Oxford University.

Jackendoff, R. (1994), Patterns in the mind: language and human nature, New York, Basic Books.

Lovibond, P.F. (2003), “Causal beliefs and conditioned responses: retrospective revaluation induced by experience and by instruction”, in Journal of experimental psychology: Learning, memory & Cognition 29/1, 97-106.

Maynard Smith, J. & Szathmary, E. (1999), The origins of life: from the birth of life to the origins of language, Oxford, Oxford University Press.

Nowak, M.A. (2001), “Evolution of universal grammar”, in Science 291, 114-118.

Pennington, G.L. & Roese, N.J. (2003), “Regulatory focus and temporal distance”, in Journal of experimental social psychology 39, 563-576.

Povinelli, D. (2000), Folk Physics for apes, Oxford, Oxford University Press.

Povinelli, D.J. & Dunphy-Lelii, S. (2001), “Do chimpanzees seek explanations? Preliminary comparative investigations”, in Canadian journal of experimental psychology 52/2, 93-101.

Premack, D. (1995), “Cause/induced motion: intention/spontaneous motion”, in Changeux, J.P. & Chavaillon, J. (eds), The Origins of the human brain, Oxford, Clarendon.

Reboul, A. (2004), “Evolution of language from theory of mind or coevolution of language from theory of mind?”, Webconference Issues in the coevolution of language and theory of mind, available at URL: http://www.interdisciplines.org/coevolution/papers/1.

Rescorla, R.A. & Wagner, A.R. (1972), “A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and non-reinforcement”, in Black, A.H. & Prokasy, W.F. (eds), Classical conditioning II. Current research and theory, New York, Appleton-Century-Crofts.

Roese, N.J. & Olson, J.M. (1996), “Counterfactuals, causal attributions, and the hindsight bias: a conceptual integration”, in Journal of Experimental Social Psychology 32, 197-227.

Roese, N.J. & Olson, J.M. (2003), “Counterfactual thinking”, in Nadel, L., Chalmers, D., Culicover, P., French, B. & Goldstone, R. (eds): Encyclopedia of cognitive science, New York, Macmillan.

Roese, N.J. (1994), “The functional basis of counterfactual thinking”, in Journal of personality and social psychology 66/5, 805-818.

Siegal, M., Varley, R.A. & Want, S. (2001), “Mind over grammar: reasoning in aphasia and development”, in TICS 5, 296-301.

Varley, R.A. & Siegal, M. (2000), “Evidence for cognition without grammar from causal reasoning and ‘theory of mind’ in an agrammatic aphasic patient”, in Current biology 10/12, 723-726.

Varley, R.A. (1998), “Aphasic language, aphasic thought: propositional thought in an apropositional aphasic”, in Carruthers, P. & Boucher, J. (eds), Language and thought: interdisciplinary themes, Cambridge, Cambridge University Press.

Varley, R.A., Siegal, M. & Want, S. (2001), “Severe impairment in grammar does not preclude theory of mind”, in Neurocase 7, 489-493.

Vonk, J. & Povinelli, D. (in press), “Similarity and difference in the conceptual systems of primates: the unobservability hypothesis”, in Zentall, T. & Wasserman, E. (eds), Comparative cognition, available at URL: http://www.cognitiveevolutiongroup.org/site100-01/1001369/docs/preliminary_similarity.pdf.

Waldmann, M.R. (2000), “Competition among causes but not effects in predictive and diagnostic learning”, in Journal of experimental psychology: Learning, memory and cognition 26/1, 53-76.

Waldmann, M.R. (2001), “Predictive versus diagnostic causal learning: evidence from an overshadowing paradigm”, in Psychonomic Bulletin & Review 8, 600-608.

Waldmann, M.R. & Hagmayer, Y. (1998), “How categories shape causality”, in Hahn, M. & Stenoss, S.C. (eds), Proceedings of the 21rst annual conference of the Cognitive Science Society, Mahwah, NJ, Erlbaum.

Waldmann, M.R. & Martignon, L. (1999), “A Bayesian network model of causal learning”, in Gernsbasher, M.A. & Derry, S.J. (eds), Proceeding of the 20^th annual conference of the Cognitive Science Society, Mahwah, NJ, Erlbaum.