As noted by Quilty-Dunn et al., the language-of-thought hypothesis (LoTH) has fallen out of favor in philosophy. But why? I deeply appreciate the authors’ careful and comprehensive review of contemporary empirical and mathematical work that supports the LoTH. I especially welcome their clarification of its core commitments, which enables us to see the LoTH at work in areas where its presence may not be apparent. But I feel that they are too concessive to LoT's critics. All of the considerations originally adduced in favor of the model still stand, particularly those that appeal to facts about person-level thought.
Understanding the propositional attitudes – believing, wanting, supposing, and so on – is one of the central goals of the philosophy-of-mind. What should constrain theorizing about them? Here are two surpassingly important data:
(1) Propositional attitudes – particularly beliefs and desires – can combine to produce actions in ways that conform to what Aristotle called the “practical syllogism.”
(2) A perfectly rational thinker can hold incompatible thoughts without realizing it.
As Jerry Fodor (Reference Fodor1978) pointed out long ago, the hypothesis that propositional attitudes are functional states involving physically realized, syntactically structured representations offers smooth explanations of both these data. On the contrary, two leading anti-LoT theories cannot.
(1) That the mental states of believing and desiring something can cause movements in the body is explained simply by their being realized physically – one doesn't need the LoT to do that. But not just any materialist theory can also explain why mental states can rationalize the movements they cause. If I want to snowshoe, and believe that I will be able to snowshoe if I go outside now, I will go outside now. That this belief–desire complex makes rational a certain course of action is explained by the salutary formal relation among the structured representations that underlie the attitudes: If I want to X, and believe that if I do Y then I can accomplish X, then (ceteris paribus) I should do Y. The architecture of LoT, which guarantees the authors’ properties 1, 2, 4, and 5, enables causal relations to track rational relations. (That the substituends of X and Y in [e.g.] the practical syllogism represent propositions is presupposed by the logic, although we'd need examples of different kinds of inference to secure property 3.)
(2) Lois Lane believes that Superman can fly, and she believes that Clark Kent cannot. But Superman is Clark Kent, so Lois's beliefs conflict. But Lois is no dope; why can't she see the problem? According to the LoTH, it's because she has two lexically distinct representations for the same individual. Because her mental processes are sensitive only to the form of the (physically realized) representations, they have no compunction about placing semantically incompatible sentences into her belief box.
How can these data be explained without appeal to an LoT? I'll briefly discuss two views that are officially agnostic about the existence of language-like representations, but that hold that such representations need play no role in accounts of human mentality and behavior. They both fall short with respect to (1) and (2).
Dispositionalism, defended recently by Eric Schwitzgebel (Reference Schwitzgebel2002), holds that believing (and desiring, etc.) is primarily a matter of being disposed to behave in certain characteristic patterns. A well-known problem for this view is specifying the pattern specific to a given belief. But there is a deeper problem. Beliefs do not have proprietary behavioral consequences – because they are inferentially promiscuous, they are willing to combine with any desire whatsoever to generate action. A dispositionalist can accommodate this point by saying that a signature behavioral profile is only determined holistically, by taking into account of all of an agent's beliefs and desires. But what does “taking into account” mean? The dispositionalist cannot rely on the logic of the practical syllogism to say what difference it would make to my behavior, given (say) my belief that it's snowing, whether I'm up for some wintry recreation or want to stay someplace warm.
Neither can the dispositionalist explain why the same set of motor movements is predictable when it's described one way, but not another. It's rational for me to, as I think of it, go outside. But going outside might, unbeknownst to me, involve stepping on a slippery surface. (I thought the walkway had been sanded.) My belief and desire rationalized my going outside, but they would not have rationalized my stepping on a slippery surface. The LoT explains why and how we do things “under a description,” as philosophers like to say (Antony, Reference Antony1987).
Another view of belief, championed recently by Frances Egan (Reference Egan2014) is deflationism. Egan argues that the LoTH (a version of what she calls “Hyper Representationalism”) founders on the failure of efforts to give a naturalistic account of the representation relation. Characterizations of mental processes in terms of the manipulation of representations, she argues, should be viewed as merely useful “glosses,” not as serious posits of a mature cognitive science. Although I have to concede that we still don't know how to reduce intentionality to nonintentional conditions, I fault Egan for failing to recognize the role that the syntactic properties of LoT representations play in psychological dynamics. Whatever “Superman” means, its lexical distinctness from “Clark Kent” is sufficient to explain why Lois behaves differently when she deploys the first representation rather than the second.
To conclude: Why is the LoTH so unpopular? I suspect that it's because of a residual allegiance to behaviorism, with its commitment to empiricism (hence the enthusiasm for pattern-extraction models of thought, like deep neural networks [DNNs]), and its rejection of mentality as a genuine domain in nature. The person-level data stand on their own, but many thanks to the authors for demonstrating the utility, fecundity, and ubiquity of the LoTH in so many areas of contemporary psychology.
You have
Access
As noted by Quilty-Dunn et al., the language-of-thought hypothesis (LoTH) has fallen out of favor in philosophy. But why? I deeply appreciate the authors’ careful and comprehensive review of contemporary empirical and mathematical work that supports the LoTH. I especially welcome their clarification of its core commitments, which enables us to see the LoTH at work in areas where its presence may not be apparent. But I feel that they are too concessive to LoT's critics. All of the considerations originally adduced in favor of the model still stand, particularly those that appeal to facts about person-level thought.
Understanding the propositional attitudes – believing, wanting, supposing, and so on – is one of the central goals of the philosophy-of-mind. What should constrain theorizing about them? Here are two surpassingly important data:
(1) Propositional attitudes – particularly beliefs and desires – can combine to produce actions in ways that conform to what Aristotle called the “practical syllogism.”
(2) A perfectly rational thinker can hold incompatible thoughts without realizing it.
As Jerry Fodor (Reference Fodor1978) pointed out long ago, the hypothesis that propositional attitudes are functional states involving physically realized, syntactically structured representations offers smooth explanations of both these data. On the contrary, two leading anti-LoT theories cannot.
(1) That the mental states of believing and desiring something can cause movements in the body is explained simply by their being realized physically – one doesn't need the LoT to do that. But not just any materialist theory can also explain why mental states can rationalize the movements they cause. If I want to snowshoe, and believe that I will be able to snowshoe if I go outside now, I will go outside now. That this belief–desire complex makes rational a certain course of action is explained by the salutary formal relation among the structured representations that underlie the attitudes: If I want to X, and believe that if I do Y then I can accomplish X, then (ceteris paribus) I should do Y. The architecture of LoT, which guarantees the authors’ properties 1, 2, 4, and 5, enables causal relations to track rational relations. (That the substituends of X and Y in [e.g.] the practical syllogism represent propositions is presupposed by the logic, although we'd need examples of different kinds of inference to secure property 3.)
(2) Lois Lane believes that Superman can fly, and she believes that Clark Kent cannot. But Superman is Clark Kent, so Lois's beliefs conflict. But Lois is no dope; why can't she see the problem? According to the LoTH, it's because she has two lexically distinct representations for the same individual. Because her mental processes are sensitive only to the form of the (physically realized) representations, they have no compunction about placing semantically incompatible sentences into her belief box.
How can these data be explained without appeal to an LoT? I'll briefly discuss two views that are officially agnostic about the existence of language-like representations, but that hold that such representations need play no role in accounts of human mentality and behavior. They both fall short with respect to (1) and (2).
Dispositionalism, defended recently by Eric Schwitzgebel (Reference Schwitzgebel2002), holds that believing (and desiring, etc.) is primarily a matter of being disposed to behave in certain characteristic patterns. A well-known problem for this view is specifying the pattern specific to a given belief. But there is a deeper problem. Beliefs do not have proprietary behavioral consequences – because they are inferentially promiscuous, they are willing to combine with any desire whatsoever to generate action. A dispositionalist can accommodate this point by saying that a signature behavioral profile is only determined holistically, by taking into account of all of an agent's beliefs and desires. But what does “taking into account” mean? The dispositionalist cannot rely on the logic of the practical syllogism to say what difference it would make to my behavior, given (say) my belief that it's snowing, whether I'm up for some wintry recreation or want to stay someplace warm.
Neither can the dispositionalist explain why the same set of motor movements is predictable when it's described one way, but not another. It's rational for me to, as I think of it, go outside. But going outside might, unbeknownst to me, involve stepping on a slippery surface. (I thought the walkway had been sanded.) My belief and desire rationalized my going outside, but they would not have rationalized my stepping on a slippery surface. The LoT explains why and how we do things “under a description,” as philosophers like to say (Antony, Reference Antony1987).
Another view of belief, championed recently by Frances Egan (Reference Egan2014) is deflationism. Egan argues that the LoTH (a version of what she calls “Hyper Representationalism”) founders on the failure of efforts to give a naturalistic account of the representation relation. Characterizations of mental processes in terms of the manipulation of representations, she argues, should be viewed as merely useful “glosses,” not as serious posits of a mature cognitive science. Although I have to concede that we still don't know how to reduce intentionality to nonintentional conditions, I fault Egan for failing to recognize the role that the syntactic properties of LoT representations play in psychological dynamics. Whatever “Superman” means, its lexical distinctness from “Clark Kent” is sufficient to explain why Lois behaves differently when she deploys the first representation rather than the second.
To conclude: Why is the LoTH so unpopular? I suspect that it's because of a residual allegiance to behaviorism, with its commitment to empiricism (hence the enthusiasm for pattern-extraction models of thought, like deep neural networks [DNNs]), and its rejection of mentality as a genuine domain in nature. The person-level data stand on their own, but many thanks to the authors for demonstrating the utility, fecundity, and ubiquity of the LoTH in so many areas of contemporary psychology.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.