Thomas Johnston

(1) In the first part of this paper, I review Chomsky's meandering journey from the formalism/mentalism of Syntactic Structures, through several methodological positions, to the minimalist theory of his latest work. Infected with mentalism from first to last, each and every position vitiates Chomsky's repeated claims that his theories will provide useful guidance to later theories in such fields as cognitive psychology and cognitive neuroscience. With the guidance of his insights, he claims, psychologists and neuroscientists will be able to avoid costly dead-end lines of research. (2) This never happened. As I have shown, this never could have happened. (See Johnston, 2018). What has happened, instead, is that current neurolinguistic research (with the arguable exception of the now-dated Lemma Model of Willem Levelt) proceeds without reference to Chomsky. It also wholeheartedly rejects the mentalism of the associated Language of Thought theory of Jerry Fodor. (See Johnston, 2018). (3) I make this argument in the first part of this paper. I would also like to point out that most of my argument was developed in 1972, when I was a graduate student. I know of no other sustained criticisms of Chomsky at that time, and certainly none along the lines I had developed back then. (4) In the second part of this paper, I present my own account of the methodology of science. When I was a graduate student, philosophy of science was dominated by an attempt to describe a methodology common to all the specific sciences, i.e. Hempel’s deductive-nomological model. These days, Hempel’s emphasis on the methodological unity of science has been rejected by such “dis-unity” philosophers of science as Ian Hacking, Patrick Suppes and Nancy Cartwright (see Cat, 2021). (5) I view this change as the swing of a pendulum or, to change the metaphor, a journey from one end point of a continuum to another. As the level of abstraction at which one tries to describe scientific method is raised, the descriptions become increasingly general. Whether or not unity-of-science theories become so general as to be vacuous, is ultimately a subjective judgment. And so I expect that philosophers will eventually become tired of increasingly specific “close to the workbench” descriptions of how scientists work, and begin to turn back to methodological “big pictures”, finding in them powerful abstractions rather than empty irrelevancies. (6) In the second part of this paper, I present my own account of the methodology of science, which I would situate somewhere between the “unity” and “dis-unity” accounts. However, I am not a scientist. My own views about scientific method have three origins: (6a) my work as a graduate student from 1966 until I passed my comprehensive exams in 1973 (at a different university); (6b) reading every issue of Scientific American from 1972 until nearly 2000, (at which point I continued to read it only sporadically, since I concluded that, around that time, it had evolved from a serious science magazine to a popular science magazine); and (6c) my three-year immersion in the cognitive neuroscience of language after I retired, based on repeated study of and note-taking for (Banesh & Compton, 2018), (Kemmerer, 2015), several other books and, finally, numerous articles not hidden behind a paywall. So, as always with my writings: caveat emptor.

(1) This is Part 2 of the semantic theory I call TM. In Part 1, I developed TM as a theory in the analytic philosophy of language, in lexical semantics, and in the sociology of relating occasions of statement production and comprehension to formal and informal lexicographic conclusions about statements and lexical items – roughly, as showing how synchronic semantics is a sociological derivative of diachronic, person-relative acts of linguistic behavior. I included descriptions of new cognitive psychology experimental paradigms which would allow us to precisely measure the two constituents of semantics – meaning and reference – both at the level of individual speech acts and at the level of societal convergences, i.e. at both the token and type levels. (2) In the Introduction, I recapitulate the arguments of Part 1. The Introduction also develops some analytic philosophical and lexical semantics themes not discussed in Part 1. (3) After the Introduction, I present neural TM (nTM) as a theory of the neural mechanisms and processes which give rise to these person/occasion-relative acts of linguistic behavior. I develop nTM at three levels, the first two of which describe linguistic/semantic functions independently of their cortical locations. At the first level, I describe individual word-to-word and word-to-object connections. At the second level, I describe the corresponding structuralist networks of which they are the individual components. At this level, I introduce some key linguistic concepts of TM – its graded meaning, reference, and generalization sets, and the types of statements which express various levels of word-to-word and word-to-object relationships among lexical items which, because of the constraints they impose on the use of those lexical items in statements we produce and comprehend, are concepts. This constitutes the second structural level of nTM. (4) At the third level, I associate the non-localized structures of the previous levels with cortically located neural structures and with the fasciculi that connect them. I distinguish neural areas in which primary (phonetic) and secondary (orthographic) lexicons are stored in long-term memory. I also describe the embodied concepts which co-exist in the anterior temporal lobes with the images they lexicalize. These concepts are often said to name physical objects and their features, although what they in fact name are kinds of physical objects and features. I describe how conceptual constraints and referential constraints interact to channel our intentions to say how things are into statements which are semantically well-formed, and which consequently successfully communicate information. (5) Following this presentation of nTM, I examine five prominent neural semantic theories. I point out what is wrong with each of them as far as their explanations of semantics are concerned, and I also indicate how nTM can replace the “semantic cores” of those theories. (6) The two basic mistakes made by neuroscience semantic theories, as I will explain, are (i) that all but one of them regard semantics as a matter of the association of words with perceptual images, and of generalizations from those associations; and (ii) that they all rely on an unspecified set of neural structures which purportedly encode the meaning of concepts in abstraction from their phonological and orthographic forms. nTM maintains, in contrast, that there are no abstract neural representations of semantic content. Neural constraints on our linguistic behavior, especially on our ascriptive and co-ascriptive use of words, express the semantic constraints on those words which make them concepts. That is the semantic content of words. (7) I next consider several results from neuroscience experimental data which have been given one interpretation by one or another of the standard neurosemantic theories, but to which nTM gives a different interpretation. I include several predictions which I have found neither confirmed nor disconfirmed in the experimental neuroscience literature. (8) After a concluding section in which I summarize the major changes to neurosemantic theory introduced by TM, and the analytic philosophy of language and lexical semantics contexts within which TM is situated, there follows an appendix in which I discuss neural net AI, and make some recommendations for implementing nTM in silicon.

Lexical semantics has a problem. As Allesandro Lenci put it, the problem is that it cannot distinguish semantic from non-semantic relationships within its data. (2008, 2014). The data it relies on are patterns of co-occurrence of lexemes within linguistic corpora. But patterns of co-occurrence can reflect either our knowledge of what the world is like or our knowledge of what words mean -- matters of fact or matters of meaning. In this essay, I develop a semantic theory which draws this distinction in a way which makes it discernible in lexical semantics and cognitive science research. In doing so, this theory unifies truth-functional and structuralist approaches to semantics, and provides an integrated explanation of meaning and reference. I base this semantic theory on linguistic dispositions to pair words with words, and to pair words with things, both based on learned patterns of association. These dispositions manifest themselves in verbal behavior, and (in Part 2) I propose a new neurophysical model to account for these dispositions which underlie both the intensional and extensional patterns in that behavior.