How confusing spoken language with written language undermines linguistic thought

Robert Port, Linguistics and Cognitive Science, Indiana University

For over a hundred years, almost all scientific studies of language (i.e., in linguistics, speech and hearing science, experimental psychology of language, etc.) have assumed that every language is defined by a fixed set of phonemes, the discrete sound types used for spelling words in memory, plus a set of words (or morphemes), short chunks of speech with fairly distinct meanings, that are composed into sentences, which are themselves word sequences that supposedly represent a ``complete thought’’ (and contain both a subject and a predicate).  Chomsky took these traditional ideas about discrete symbol types nested within each other (Phonemes within Words within Phrases within Sentences), into patterns that closely resemble the conventional structure of the written form of most languages, and invented a formal mathematics of symbol strings called generative grammar.  His contributions to formalizing this idea made him famous since these ideas were directly applicable to the development of programming languages for computers.   But all formal grammars (from binary machine code to Matlab and Word) are human artifacts (just like arithmetic, calculus and chess) and rely on idealizations like perfectly discrete symbols that are manipulated outside of real time by a discrete-time clock.   But human speech lives in continuous time and consists of conventional motor and acoustic patterns that only in very rough terms resemble formal symbols.   Without idealized symbols (all of which are apparently modelled historically on orthographic letters), formal grammars cannot be constructed.  In fact, none of the basic unit types for spoken language (e.g., phonemes and words) have ever been given definitions that are empirically satisfactory.   It has never been possible to specify exactly what the set of psychological phonemes is for any language (and thus what the phonological spelling is for any word) or what the list of words is (unless one takes the traditional conventions of orthography as decisive), nor why it is that most spontaneous speech fails to satisfy the constraints of sentence definitions offered by school teachers or linguists.   The implication of all this for those who seek to understand early stages of the evolution of language is that much of what linguists have offered to address language evolution, in particular, the notion that human languages are, in fact, instances of formal systems of any kind cannot possibly be true.  Of course, the use of formal systems to merely model human language is reasonable and may offer insights.  However, human language cannot actually BE a formal system so the problem of when and how humans began to have formal systems built in to their brains simply does not arise.  Formal systems, from chess to integer arithmetic to logic to computer programs to generative grammars are always, without exception, deliberate human artifacts.   Therefore, ideas such as postulating an abrupt genetic change that creates neural mechanisms capable of processing formal symbol strings (e.g., Chomsky, Bickerton, etc.) can be safely ignored.  But how could the idea that a language is a formal system be so persuasive that the basics of this idea have remained powerful for over a century despite many obvious forms of evidence that the assumption is false?  I believe it is because the (edited) written orthography (of English, at least) very roughly approximates the product of a formal system, e.g., it uses a discrete alphabet of fixed size, a relatively fixed inventory of conventional words and word sequences divided into units separated by a [period+space+capital letter].   Those of us with education began pir training for our language in terms of formal tokens like letters, words and sentences when we were extremely young.  Breaking free of that terminology is extremely difficult despite the failure of formal models to account for human speech.