The constituent unit is defined in Wikipedia as a word or a group of words that functions as a single unit within a hierarchical structure. When phrase structure trees are produced, each node in the tree marks a constituent. Thus, each individual word is a constituent in phrase structure grammars just by virtue of being a word.

There is a major problem in this area. The tests for identifying constituents employed in the same phrase structure grammars contradict this starting assumption concerning individual words as constituents. Most individual words fail the majority of tests for identifying constituents that are employed in syntax and grammar textbooks (e.g. topicalization, clefting, pseudoclefting, answer fragments, proform substitution, etc.). A bit of time spent reading the Wikipedia article on the constituent unit makes this point clear. That article discusses the constituent structure of the sentence Drunks could put off the customers. Of the six words present in the sentence, only one of them, i.e. the subject nominal Drunks, passes most of the tests. The other five fail most of the tests. The next data set and those further below are taken from the Wikipedia article; they illustrate that the other five individual words do not behave as constituents with respect to the tests. The first examples are of topicalization:

enter image description here

The same five words fail the clefting test:

enter image description here

They also fail the answer fragment test:

enter image description here

They also fail the proform substitution test:

enter image description here

Similar results obtain for the majority of tests for constituents discussed in the article (15 of them altogether). Only one of the tests, i.e. coordination, regularly suggests that individual words are constituents. The point, then, is that many tests for constituents actually suggest that most individual words are in fact NOT constituents.

The next questions derive from this insight. Isn't this a problem for phrase structure grammars, and if it is not a problem, why is it not a problem? Should the term tests for constituents be renamed to something else, perhaps to tests for phrases, since many (but certainly not all!) phrases pass most of the tests? I of course have a concrete opinion about all this, but I am respectfully requesting that the phrase structure people here share their views and understanding.

  • Comments are not for extended discussion; this conversation has been moved to chat.
    – prash
    Commented Jun 15, 2020 at 13:03
  • Reaction to your edit: You can’t conclude ¬A⊃¬B from A⊃B, only the contrapositive. This is actually an interesting question, are there negative constituency tests?
    – Atamiri
    Commented Jun 20, 2020 at 19:33
  • @Atamiri, Please note the wording, that is, the use of the verb "suggest". Yes, it is an interesting question and an important one. It goes to the heart of dependency vs. phrase structure distinction. Commented Jun 21, 2020 at 2:23
  • Yes, I see the verb “suggest”, but tests for constituency are material implications, they only suggest something if true. If a test comes out as false it doesn’t suggest anything. In particular, if one out of ten tests can be applied, it follows that the tested phrase IS a constituent.
    – Atamiri
    Commented Jun 21, 2020 at 8:13
  • 1
    It doesn't look to me as though you're going to get any further answers and I can't see the people I think you have in mind being motivated by a bounty. I hope you'll answer your own question anyway. I have not really grasped what is supposed to turn on this issue and suspect you are reifying constituents too much - I think it's more a question of what role constituency can / should play in a theory and how well different conceptions of it fit that role. At the moment you have no job description for constituency but are still asking for an appraisal.
    – rchivers
    Commented Jun 24, 2020 at 12:17

1 Answer 1


I only wanted to write a comment but I’d like to include a tree so here’s a more detailed answer. First of all, constituents are defined by context-free rules, not by tests. The tests can be used to devise the rules but they’re defeasible implications. Even if they were material implications, if one is true, the converse needn’t be. But this is elementary.

Consider the Eastern Armenian sentence Ես Երևան եմ գնում “I go to Yerevan.” Its syntax tree is

enter image description here

This is completely uncontroversial, there’s no alternative sensible way of parsing the sentence. The point here is that the enclitic auxiliary is a pragmatic marker and moving the NP Yerevan around the sentence drags the auxiliary with it (since it needs to remain focus marked). So Yerevan is the categorial head of the auxiliary, but at the same time the auxiliary is the dependency head of Yerevan — by virtue of being a cohead together with the main verb but this isn’t germane here. The point is, it doesn’t make sense to apply constituency tests to single words (in the sentence at hand the auxiliary could be dispensed with if the verb were in future tense) and the obvious (but apparently still worth pointing out) point is that a true implication doesn’t make its converse true.

I’ve digressed a bit — the short answer is that individual words (more precisely their preterminals) are constituents by definition.

Edit: A similar example would be “Robert goes to Yerevan” with the subject in focus, i.e., with the auxiliary farther away from the main verb. The auxiliary is the verb “to be” which can also serve as a copula.

enter image description here

  • For this answer to start to be coherent, the example from Armeninan would need a three-level gloss (romanization, element-for-element translation with morphological information, and English meaning). As it is, it is only comprehensible to those who know Armenian. Linguistic reasoning that is based on impenetrable examples requires faith, which has no place in science. Commented Jun 14, 2020 at 2:27
  • 1
    @TimOsborne Can't you see the tree with the romanised transcription?
    – Atamiri
    Commented Jun 14, 2020 at 8:03
  • 1
    Yes, I saw those. But I see you claiming that em receives the gloss I. What is that? Inflection? Does that mean that em is agreeing with a lingusitically absent subject pronoun? Or is yes the subject pronoun? What does NP with subscript p mean? Des the p stand for particle? It if does, then em is actually not an auxiliary verb.The failure to provide the standard three-level gloss means that the reader is unable to interpret your example in these areas. Commented Jun 14, 2020 at 11:29
  • @TimOsborne Yes, the auxiliary must agree with the subject, dropped or not. But the point here is the distinction between what linguists call “categorial” and “functional” heads, which you consistently overlook. The tree alone is a clear illustration of what’s going an, as there’s an apparent dependency cycle, and a particularly clear counterexample to (one aspect of) your trees.
    – Atamiri
    Commented Jun 14, 2020 at 12:41
  • So yes is the subject pronoun, right? Just to be sure, the form of the clitic changes if the subject changes. If the sentence were, say, 'You go to Yerevan', the clitic em would have a different form. Is that right? If possible, you could produce a similar example from a language closer to home? Can you illustrate the point with an example form English or German or some other related language? English also has clitic auxiliaries. It'd be good to see your point illustrated with an English example. Commented Jun 15, 2020 at 2:56

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.