Linguistics: An Introduction to Language and Communication (2017 7 ed). p. 196 Middle.

enter image description here

Why does [that]complementizer hinder questioning of constituents (in 80b), when [that]complementizer can be included in propositional attitude that-clauses (in 80a)?

I saw this on Reddit, and someone commented

You're asking why one particular syntactic rule has one particular exception. There are thousands of such rules, and many more exceptions; nobody knows the reasons for any of them. What kind of answer are you looking for? Oh, and I suppose I have to ask whether you're familiar with Ross Constraints, because this is one of them.

  • 1
    I'm not seeing an exception - there is no wh- movement in a. It's not a case of question vs statement - you can say e.g. Mary believed that WHO would marry John?!.
    – rchivers
    Apr 20, 2020 at 12:29
  • this is an excellent question because it does not have a simple (or even clear) answer. you can see in tsutsu's answer the kind of sophistication that an adequate answer requires. in the present state of syntactic theory, the particular answer you adopt depends on independent theoretical commitments, so no answer below is necessarily definitive (though some may be preferable to others) May 21, 2020 at 18:22

2 Answers 2


Consider the sentences in (1) and (2):

(1) I asked [CP which boxer [IP John could defeat t ]] ?
(2) *[CP Why did [IP you ask [CP which boxer [IP John could defeat t t ]]]] ?

I assume you're familiar with cyclic-movement of wh-constituents, and I want to generate the subjects as specifiers of IPs on purpose, their generation in Spec,VP won't affect my discussion. The sentence in (1) is grammatical. The object which boxer is extracted (moved) from an embedded position into an intermediate Spec,CP position leaving a trace behind. However, (2) is ungrammatical for an obvious reason related to the fact that there are two movements involved (why and which boxer). So, either why or which boxer has to move first into the embedded position Spec,CP. Let us say that which boxer moves first to the embedded Spec,CP and leaves a trace behind. When the turn comes for why to move, it finds that an already wh-constituent is filling the embedded Spec,CP position, as a result it will have to move straight away to the matrix Spec,CP as (2) shows. Ross has noticed that this movement is constrained, hence the term Ross Constraints. This movement violates an island-constraint on movement (wh-movement in our context). Therefore, wh-constituents are islands or barriers to movement. A wh-constituent does not allow another wh-constituent to cross over it. Remember that we have to generalize that this position (i.e., Spec,CP) though empty, must host the first movement of wh-constituent (Spec,CP acts as a rest station for this movement). The wh-island above is similar to the complex NP in the following sentence:

(3) *[CP Who did [IP Mary make [NP the claim [CP that [IP John defeated t ]]]] ?

In (3), the modified NP is marked in bold. Here, there's only one instance of wh-movement, but still the sentence is ungrammatical. The similarity between (2) and (3) lies again in the notion island. Ross identifies this type of NPs as islands along with wh-constituents. Again wh-movement cannot cross NPs of this sort (he calls them COMPLEX NPs, hence the notion Complex NP Constraint). Keep in mind that each language varies according to how many islands an extracted wh-element can cross.

With this logic, let us consider the following sentences:

(4) Whom did Mary believe that John will marry?
(5) *Who did Mary believe that will marry John?

(4) is a wh-extraction of an object (object of the embedded verb marry {John will marry whom), whilst (5) is a wh-extraction of a subject (subject of the embedded clause {whowill marry John}). Now, consider their representations:

(4’) [CP Whom did[IP Mary believe[CP t that [IP John will marry t ]]]]?
(5’) *[ CP Who did [ IP Mary believe [CP t that [IP t will marry John]]]]?

In (4’), the object whom is extracted from the foot of the tree leaving a trace behind. I assume you're familiar with the notion government. V governs the object and assigns it both an internal theta-role and an accusative case. The movement from this position into the matrix Spec,CP is not allowed (Ross, constraint). Remember that movements must rest in Spec,CP first (cyclic movement). So, it should land in the embedded Spec,CP first, then move right away to the matrix Spec,CP. This is exactly what it does in (4’) shown with the intermediate trace in the embedded Spec,CP. Let’s turn to (5’) now. This sentence is a problem. Notice that it is similar to your example in (80b) above.

In (5’), it is a subject which is extracted from Spec,IP leaving a trace there. There is a Spec,CP position available in this case, it has to move to move there first then later to the matrix Spec,CP (hence no violation of Ross constraint). Still, the sentence in (5’) is ungrammatical. I will deliberately escape another notion called ‘subjacency’ (Chomsky, 1973) which reformulated Ross constraints, and move straight away to what (5’) violates. Before we do so, notice that in terms of representation there is no difference between (4’) and (5’), they both have Spec,CP as a landing site.

Chomsky (1986) observed that the sentence in (5’) and (80b) violate the following principle:

Empty Category Principle (ECP):
Traces must be properly governed.
A properly governs B iff A theta-marks and govern B or A antecedent-governs B.
A theta-governs B iff A governs B and A theta-marks B.
A antecedent-govern B iff A governs B and A is coindexed with B.


A governs B iff there is no Z such that:
Z is a potential governor for B; Z c-commands B; Z does not c-command A.

Now, let us turn to the grammatical (4’) and ungrammatical (5’):

(4’) [CP Whom did[IP Mary believe[CP t that [IP John will marry t ]]]]?
(5’) *[ CP Who did [ IP Mary believe [CP t that [IP t will marry John]]]]?

In (4’), the trace of whom in the foot is properly governed, because it’s theta marked and governed by a V marry. However, the trace of who in the Spec,IP is a problem. It is indeed governed by I (which assigns it the nominative case), but it is not governed by V (although we assume that who moved from Spec,VP; still, it is governed by a V’ not a V; X’ projections are not governors). Now, what remains is antecedent-government. The subject who moves from Spec,IP to Spec,CP as Ross Constraint (and Chomsky’s Subjacency) dictate. But look carefully that the trace in Spec,IP cannot be antecedent governed by who in the intermediate Spec,CP (before moving to the matrix Spec,CP). It cannot be antecedent governed because the complementizer that intervenes. that is a head and hence a potential governor. So what this that does is it intervenes between a proper antecedent-government between Spec,CP and Spec,IP. It’s in the middle.

Therefore, (5’) and (80b) violate ECP.

So, that is omitted because it acts as a barrier between the intermediate trace in Spec,CP and the trace in Spec,IP, which again triggers ECP. When it's omitted, no such barrier intervenes and no ECP is triggered= Grammaticality.

  • Your answer is long and complex and therefore difficult to follow. One has to take many theoretical assumptions for granted for it to work (e.g. X-bar structures, cyclic movement, ECP, Minimality, etc.). Furthermore, it does not address the additional data from reduced relative clauses. Or does it? Apr 21, 2020 at 12:36
  • 1
    I answered his question using the theoretical framework he referred to. He mentioned 'Ross Constraints', this is Generative Syntax. If you think it's complex, I mentioned that I cut the discussion in the middle (neglected Subjacency). It was too long, because I wanted to clarify derivationally why is it ungrammatical. You said it doesn't address Reduced Relative Clause? His question above is WHY (80b) IS UNGRAMMATICAL. This is what I accounted for.
    – Tsutsu
    Apr 21, 2020 at 12:56
  • Your explanation is GB syntax, circa mid 1980s. The question does not mention GB syntax. Ross's works were much earlier. That is all beside the point, though. What good is an explanation that almost no one can understand? Apr 21, 2020 at 14:28
  • Should I delete it?
    – Tsutsu
    Apr 21, 2020 at 14:36
  • No, I don't think so. The sort of explanation you present definitely exists, so there is some value in your account. For me personally, though, your account belongs in the history of syntactic theorizing rather than with current efforts to find good explanations. I can recommend to any student of syntax to view all that GB (and MP) theorizing from decades ago from a distance, with a critical eye. Apr 21, 2020 at 14:52

I think there is a straightforward explanation for the observed acceptability contrast. The word that functions in various ways: it can be a pronoun (e.g. I heard that, too), a determiner (e.g. We've seen that movie), or a subordinator (e.g. They said that they are interested). If we assume that that is necessarily interpreted as a pronoun when it (immediately) precedes a finite verb (whereby the modal verb will is considered finite), then the ungrammaticality of (80b) is predictable. The word that is interpreted as the subject argument of will marry, and John is of course the object argument of will marry. Hence the transitive predicate will marry is satiated, both of its arguments being immediately present in its clause. This situation renders the sentence-initial Who an orphan, for it cannot find a home.

The plausibility of this explanation relies on a key assumption about the nature of sentence structure and the manner in which it is produced and processed. The subject-verb combination (verb = finite verb) is the core of sentence structure. Speakers and listeners of a language produce and process sentence structures in terms of the central dependency that exists connecting the subject argument to the finite verb, the finite verb being the hierarchically dominate verb in the clause. Once subject and finite verb have been identified, then the processing of the rest of the sentence structure is easy.

The statements in the previous paragraph are based on a further assumption, namely that sentence structure is produced and processed online, that is, in time from earlier to later. The desire is hence to identify the subject argument as soon as possible in order to process the rest the sentence more smoothly. Thus, the first candidate for subject status is necessarily construed as the subject. This situation forces the interpretation of that in (80b) as the subject pronoun.

There is another phenomenon that is closely related to the main observation in the question and that indirectly supports the sort of explanation being put forward here. This further phenomenon occurs in reduced relative clauses, e.g.

(1a) the boy I know __

(1b) the boy you talked to __

(1c) the boy you told __ about the virus

(1d) *the boy __ told us about the virus (cf. the boy that told us about the virus)

Reduced relative clauses are possible in the event that the relativized position is an object. If the relativized position is a subject, the relative pronoun or subordinator that cannot be omitted. Given the reasoning developed above, example (1d) fails because the boy is necessarily interpreted as the subject argument of the predicate told, which renders the whole thing a matrix clause (contrary to what the gap indicates).

Finally, I should comment that I have not encountered the explanation put forward here anywhere in the linguistics literature on syntax. I have, though, encountered the basic observation in the question in the literature, although I cannot remember where.

  • That would make imperatives non-finite, I think - is that orthodox?
    – rchivers
    Apr 21, 2020 at 18:07
  • @rchivers I do not understand. Modal verbs are frozen; their forms do not change, so viewing them as finite is somewhat problematic. They are, however, the hierarchically dominate verbs in their clauses; they appear left-most. In this regard, they are very much finite-like. In the same way, the imperative forms of verbs can be viewed as finite-like because those forms are also hierarchically dominant, although they typically appear without any auxiliaries. Apr 22, 2020 at 2:48
  • Sorry, the misunderstanding was on my side.
    – rchivers
    Apr 22, 2020 at 8:58
  • this does not work for at least three reasons. first, when "that" is a complementizer, it is always unstressed. when it is a demonstrative (pronoun or determiner) it is stressed. this makes it difficult to imagine that the garden-path-like effect you suggest is happening. second, the effect is not limited to the complementizer "that." it occurs with other complementizers too. for example with "if." you can say "who did you ask if I like" (maybe marginally) but not "who did you ask if likes me." third, the same effect occurs in other languages with unambiguous complementizers, eg Nupe. May 21, 2020 at 16:57

Your Answer

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