I've had a hard time finding answers to this question on Ixquick.

When I was young, transformational grammarians believed that sentences were derived from "kernel sentences," which were uniclausal, single-verb clauses with no modifiers. http://grammar.about.com/od/il/g/Kernel-Sentence-term.htm." For instance, "Mary bought and fed a red horse" would be derived form "Mary bought the horse." "Mary fed a horse." "The horse was red."

From what I've read, most modern linguists don't believe that sentences are represented in the brain as kernel sentences.

What considerations prompted generative linguists to abandon the notion of kernel sentences?

  • It seems they decided to concentrate on reverse engineering from surface structure, leaving ultimate models of deep structure for later generations. These kernel clauses with two links (S-V-O), reusing any existing items, make a very parsimonious model. Can you get that efficiency and precision out of a hidden-layer connectionist model?
    – amI
    Nov 15, 2018 at 10:46

2 Answers 2


I can't give any real answer of the kind you want, but I can tell you what I know about about this topic being an undergraduate 40 years after the event, hoping that it will be of any help, even though I am certainly wrong to some degree.

As far as I can see the idea originated with Zellig Harris, who is now taken to be one of few American strucuralists who was concerned with syntax in a serious manner. Harris even dealt with structures bigger than sentences, and since substitution tests were central to structuralism, he had the idea of establishing equivalence relations for sentences by checking which sentences could replace each other in a longer text, which for instance connected corresponding active/passive sentences. It seems that he continued from there by assuming that one member of such a relations is basic and the others derived, and consequently invented transformations to do that derivation.

In Chomsky's later transformational grammar, the notion of kernel sentece seems to have just been carried over without much deep consideration. I can't remember that he discussed the matter any more than saying that it allows for a more simple grammar. Later work by Katz and Fodor and Katz and Postal then introduced some form of deep structure, which led to the model in Aspects, which repurposed transformations to not deal with inter-sentence equivalences, but instead to map between deep-structures and surface-structures. As far as I know this was when kernel sentences were given up.

My best guess is that that notion was, at least in generative grammar, seen as mostly accidental and a matter of tradition from the start, and that there was no interest in defending it against what was intuitively seen as a more interesting model in Aspects.


Harris' work was very empirically based, and kernel sentences with their transformations were his tool for analysis. His student, Noam Chomsky, never worked empirically, yet introduced the notion of a generative grammar, which produces all the grammatical sentences of a language without producing any of the ungrammatical sentences. The type of transformation you use as an example does not operate on a single tree. It combines three trees ("Mary bought the horse." "Mary fed a horse." and "The horse was red."). Combining separate trees opens up too many possibilities for producing ungrammatical sentences, which is a failure to comply with the requirements of a generative grammar. Remember Chomsky was not working empirically. His move was to theorize something akin to a computer program that would spit out all and only the grammatical sentences of a language. After Syntactic Structures, Chomsky moved to eliminate transformations that combined trees, which were referred to as generalized transformations, in order to apply constraints against some of the ungrammatical outputs generated. Chomsky had an additional tool that Zellig Harris did not have, phrase structure rules. The distributional benefit, from a generative perspective, of limiting operations to single trees only had the effect of reducing everything to a single unified "deep structure". The creation of a unified deep structure allowed for a singular level in which meaning was initially thought to be brought in. This got a lot of linguists interested in meaning. No one at that time, therefore, had any motivation for going back to Zellig Harris' manner of combining multiple trees. However, you will find a return to combining separate structures more recently in what are referred to as unification-based grammars, like Construction Grammar.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.