Progress in linguistics after the Chomskian muddle

Chomsky has been pursuing his simple method to generate the underlying structure of natural languages for several decades now. Gone are his social and political antenna, in his obsession for the minimalistic magic formula that will describe a natural language. Even Newton did not make his discovery alone, he was helped by the ideas of Copernicus, Galileo and enormous data generated by Kepler. Chomsky like Napoleon puts to sword all statistical analysis, modelling and scientific analysis. Chomsky claims that machine learning will bring us no clarity, only magical solutions. I am glad that the new generation of information theory scientists and computer scientists have started countering this Chomskian muddle that has been blocking linguistics from scientific progress.

The first and the most important retort has come from his own student, Luc Steels, who has been studying the development of language as part of social development of human society. Luc Steels has been studying the social environment that allows Robots to develop a communication system. He systematically showed, like children, how Robots can also develop some rudimentary aspects of natural language with some syntactic linguistic structures. He has been pursuing systematic framework known as “Fluid construction grammar” that allows certain fluidity of underlying structures [1,2].

Although it seems as if language originate directly from the brain of the speaker, historical materialism have taught us that the society that polices and supports individual’s utterances of the language plays an important role in its syntax and structures that it pursues. When an young Artic Turn born and bread in the Artic North Pole in Summer migrates in Artic winter to the Antartic South Pole, as the young one leads the pack, it would seem as if this young one has genetically transfigured geography of the earth, but in reality as the birds travel in V-formation, which is another co-operative pattern formation advantageous to the older ones at the back, the older ones from the back guide those industrious young ones that frequently fly in the wrong direction, only to be corrected by small “quack” from the back while at the same skewing the V-fromation to veer the young bird in the correct direction. Unlike the egoistic humans, the older birds would like to lead from the back, as it also requires less effort to fly. The same is true of human children as they keep looking at mother’s face for approval as they utter sentences, mimicking their mother, thus forming what we call “mother tongue”. The child’s mother has already been socially trained to be in compliance with society and its syntactic niceties. Under these circumstances, the magic of Chomskian muddle falls apart.

From the school of information science of Zellig Harris, Fernando Pereira [2], from AT&T Labs, fired the next slavo at Chomsky. He showed how the recent developments in machine learning have managed to circumvent the criticism of Chomsky. Chomsky claimed that machine learning cannot distinguish between “Colorless green ideas sleep furiously” and “Furiously sleep ideas green colorless” as both don’t exist in the training set. Overfitting is some thing that has been avoided with care in modern approaches to machine learning, so these criticisms don’t apply any more. Using hidden variables that associate bag of words he was able to show that the first Chomskian sentence is 105 times more likely than the first.

Peter Norvig [4] has written a comprehensive blog refuting many dogmatic statements made by Chomsky regarding modern computer science and information theory approaches to linguistics and its philosophical aspects regarding development of science in general.


[1] Steels, Luc, Modeling the cultural evolution of language, Physics of Life Reviews 8 (2011) 339–356

[2] Steels, Luc, Fluid Construction Grammar, A fully operational processing system for construction grammars, https://www.fcg-net.org/

[3] Pereira, Fernando. Formal grammar and information theory: together again? Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences 358.1769 (2000): 1239-1253.

[4] Norvig, Peter, On Chomsky and the Two Cultures of Statistical Learning, http://norvig.com/chomsky.html


  1. avatar
    Suki Venkat December 23, 2019 5:31 pm 

    “Obviously, many psychologists are deeply uneasy with the concept of innate abstract knowledge, and the reception of this idea by geneticists has been less than welcoming (e.g., Gould & Lewontin, 1979; Lieberman, 1989, 1990). Among other objections, it has been pointed out that the claim of there being universal features to all human languages is actually counterevidence rather than support for inherited syntactic knowledge: in any inherited system, there is a great deal of variability due to the instability of the genetic code. If an aspect of language is truly universal, its universality cannot be due to its having a genetic basis, but rather, due to some functionally obligatory feature of the linguistic system. Namely, in the absence of evidence about systematic differences in syntactic features among different populations or individuals, it is inconceivable that this system is indeed inherited. The only genetically determined features of living organisms that do not manifest themselves in phenotypic variability are fatal ones: if all humans whose concept of Noun Phrase was different from the universal one — because of some mutation in the relevant genes carrying information about Noun Phrases — were to perish as a consequence, it would explain why we do not ever observe such persons or group of persons. As the lethality of mutant syntactic concepts is a difficult notion to entertain, it follows that in all probability the syntactic innateness hypothesis is built on incorrect genetic reasoning.”
    –Anat Ninio

  2. avatar
    Suki Venkat September 13, 2016 7:28 am 

    If machines become intelligent and acquire consciousness, then they will be hounded like Snowden, Assange and Aaron Swartz.

  3. avatar
    Suki Venkat September 11, 2016 5:40 pm 

    In fact, it is the primitive communist society that was able to make humans out of monkeys and give it the social consciousness that has since then been undone.

  4. avatar
    Suki Venkat September 11, 2016 3:27 pm 

    Humans are machine made out of hydro-carbons, so they can in theory be replaced with machines made out of silicon. Social training can be given to these machines by other humans to make them behave quite like humans with consciousness etc. exactly how children are given training in school etc. Whether capitalism is capable of doing it is really the question, not whether it is possible.

    • Kelvin Yearwood September 12, 2016 11:44 am 

      “Humans are machine made out of hydro-carbons.”

      Human beings are objectively not machines.

      We have not begun to get close to creating a machine that as remotely as complex as a a human being. The two things are galaxies apart.

      Imagine what a human being experiences and is socialised by in its nurture and its life in an environment of complex social being!

      The socio-psychological complexity of a capuchin monkey is way beyond our understanding, let alone human machine reproduction. The reality for human beings is that extant machines and human beings have almost nothing essentially in common.

  5. Philip Mayall September 9, 2016 6:54 am 

    This piece is riddled with language errors.

  6. avatar
    Vincent Walsh September 5, 2016 7:05 pm 

    It’s a shame to see people taking cheap shots at Chomsky without really understanding his work or the UG, simply because they’re obsessed with the idea that someday machines will learn to think like humans. They can’t, and won’t.

Leave a comment