The Question Operator

Linguistics 322

Intermediate Syntax

Contents: question as modified proposition | questions in main clause | Q as strong | raising | dummy verb insertion | Inseparability | Feature Bundle Copying

There are two kinds of interrogative sentences. The first type are affirmation questions wherein the speaker is trying to get the addressee to assert whether or not a proposition underlying the interrogative sentence is true or false. The addressee may respond with a declarative sentence, a negative declarative sentence (he is asserting that the proposition is false), or he may respond with the answer that he does not know the truth value of the proposition. The second type are questions in which the speaker is seeking information about a given proposition. The addressee may respond with the information, refuse to give the information, or claim that the does not know the information., or he does not wish to divulge the information.

Questions are marked in English syntax by moving the first auxiliary verb in the clause to the beginning of the clause, or, more specifically, the features of the auxiliary verb are copied upwards. The auxiliary verb fills the C node, which marks mood:

(1).     Mary may go to the movies.

(2).     May Mary go to the movies?

In the minimalist type of grammar we are following here, no movement (transformation ) occurs unless motivated. The head of the interrogative M-proposition is an operator--a question operator ([+Q]). The question operator triggers movement. In English the question operator is a strong operator which means that the movement is obligatory. We will explain this in more detail below.

  A question is a modified lexical proposition. Let's look at the following two sentences:

(3).     John smokes stogies.
(4)      Does John smoke stogies?

The lexical E- proposition here is

(5).     SMOKE <patient: STOGIE> <agent: JOHN> .

Both (3) and (4) are modified by tense--the present tense:

(6).     [TENSE [-Past]] < SMOKE <patient: STOGIE> <agent: JOHN> >

(6) is a T-proposition; i.e., it is a semantic proposition. The difference between them is that (3) is a declaration and (4) is a question. In English as in most languages declarations are formally unmarked whereas questions are marked in one of several ways. We could represent the underlying structure of (3) and (4) as (6) and (7), respectively, where [+Q] is a feature of MOOD (mood). "Q" here is a symbol meaning [+Irreal, +Interr]. It represents the question modifier: Like other features that are not represented by a Grammatical word, it requires the host T. The operators with a null output are not included for clarity:

(7).    [MOOD [-Irreal [+Q]]] <[TP [TENSE [-Past]] [VP [V1 SMOKE <patient: STOGIE>] <agent: JOHN> ]]]].


But before we do so let us look at the properties of questions at the surface in more detail.

In main clauses questions are formed by placing a tensed auxiliary verb at the beginning of the sentence. In transformational theory, it is assumed that the tensed auxiliary verb moves there from its underlying or D-structure position. First, Mood is realized by convention as 'C' in the syntax. 'C' stands for complementizer. This name was chose long before anybody thought that complementizers were markers of mood. There is no grammatical word representing [+Q] in English.

There is a problem with the conventional hypothesis that nodes govern nodes. In (8) above, C governs Prom, and Prom governs T. The main problem that occurs is that we have an intervening governor: Prom governs T; hence C cannot govern T. There are at least two around this dilemma. The one we will choose here is to rethink government. A node is nothing more than a bundle of features. Do nodes govern? What is the exact form of government? Perhaps features governs. They govern in order to build a link for agreement and targeting purposes. Let us redefine government, from governent of nodes to government of features. The same basic concept of governent still holds, just government is a property of features:

(9)    A feature X 1governs governs another feature X2 where X is a common feature occuring in two (or more) different positions, iff:

X1 c-commands X2,

X1 is a governor.

There is no intervening governor X3, such that X3 governs X2, and X1 governs X2.

A link now links features rather than nodes.

Now, in (8) [+T] in C governs [+T], the head of TP. Currently, the feature [External Argument] governs the same in VP. There is no intervening goverenor:



All questioned main clauses must be marked for tense. In the discussion on tense and aspect, we showed that T (tense) is lowered to the main verb, because T needs a verbal host:

(11)     Rule:The features of T are copied to V (which it governs)

Motivation: T needs a host.

(12)     Rule: The features of [+Q] are copied to T.

(13)     Property: T functions as host for Q

Motivation: [+Q] in root sentences needs a host.

However, that cannot be the case for questions. Note in (4) that does is a tensed auxiliary verb and the main verb smoke remains in situ; it does not move to the beginning of the sentence:

(14).      *Smokes John stogies?

(15)     Parametric Rule: The features of T+Q are not copied to V

The features of T are not copied to the main verb. How can we explain this? Suppose that when the features of Q (and of T) are copied, they form an outer constituent or feature block:



Let's replace T in (16) with (17):



Recall that a sentence will crash if T does not have a verbal host:

(18).     *John -s smoke stogies.

Now suppose that the features of T cannot "escape" out of Q, and that since Q has a host, there is no need to copy the features of Q to V. This can be attributed to the Least Effort Principle: don't do it if you don't have to. If the features of Q and hence T cannot be copied to the verb, then how is T saved which still needs a host? It is saved by the Last Resort: the insertion of a dummy verb. The dummy verb in this case will be {BE}:


The Raising of {BE}

Now comes the difficult part. The auxiliary verb must precede the subject as we saw in (2). Before we account for this, let's copies the features of the appropriate argument to the subject position.

Case must be assigned to both NP arguments of V.

(20)     Rule: [Acc] is assigned to the direct complement of V.

Motivation: Case Theory (in part).

(21)     Principle: *NP if NP has no Case, and NP is phonetic.

The accusative Case is assigned to the direct complement of smoke, and the features of the NP John are copied to the inserted subject NP which is adjoined to TP (or to Spec-T see x-bar theory):



Raising takes place in two ways. The first is in the syntax. If the feature is strong, it raises back to its original position in the syntax. If the feature is weak, it raises in the interpretive component. The interpretive component is required for listeners who must interpret the meaning of the sentence. Tense is a weak operator in English. The verb containing T only does not raise in the syntax. If Q is adjoined to T, then Q becomes strong and it raises (in normal questions). When an affix raises, its host raises with it. Otherwise, the affix would end up without a host. For example, in the following sentence:

(23)     *-Ed John do sing a song?

The host for the suffix remains behind. The result is hopelessly ungrammatical. In the approach we are taking here, the features [+Past, +Q] are copied upwards to C and is spelled out as *-ed. "-ed" is possible only when it has a host.

Note that T alone is weak and it does not raise in the syntax. Normally, T+Q is strong and it raises in the syntax. This phenomenon of raising to empty categories in the syntax is indeed a curious one. It seems to be motivated in part by a syntactic need to overtly mark questions. This motivation is very arguable and we are not claiming to be the case. There are two constructions where Q remains weak when adjoined to T and the two together are lowered. (WH+Q lowering, ). In disbelief constructions, Q is lowered, and the intonation contour marks both Q and disbelief together. In some languages the marker of a question can occasionally be the question intonation contour. Unmarked questions in Russian are one such example.

(22)     Rule: Q is raised back to C when Q+T is adjoined to an auxiliary verb.

(23).    Property: Q is an operator.

Q is said to be strong when it moves back to C. Why it becomes strong is not clear. The reason for stipulating strong Q is found in WH-questions where the WH-word is the subject of sentence (WH+Q lowering,). The tree structure representing (19) is following Case assignment and raising to subject is:



What principle motivates strong operators to raise? There are two possible answers, of which only is presumed to be the correct answer. Chomsky has noted that traces must be c-commanded by their antecedents. The subject NP always c-commands its trace:

(25)     John swims [trace].

(26)     Bears eat berries [trace].

However, when an operator is lowered, the empty position it leaves behind does not c-command its antecedent (the moved node):

(27)     [Empty] John swim+s [trace].

Here, [empty] = [trace] c-commands its antecedent '+s'. This bothered Chomsky for awhile. Finally he (or Pollack?) decided that if the lowered node raises back to original position, the resulting trace is the raising is valid, and the empty original node is now filled again. Hence, the problem disappears. This motivates raising then. However, raising occurs either in the syntax or in Logical Form (the same component as the propositional form). Chomsky soon came to the conclusion that the effect of traces occurs in Logical Form, not in the syntax.

The second approach states that in Logical Form, all moved items must be moved back to their original place for proper interpretation. That is, for example, a question must take scope (c-command) over its argument the T-proposition. By raising Q back to C, Q then takes scope over its argument. This approach seems slightly more desirable, but a thorough comparison of the two approaches must be made.

The enigma that awaits us is why do some operators raise in the syntax and others wait and do it in Logical Form (this is called Procrastinate) by Chomsky. This phenomenon is without doubt a parameter.

Let us return to the structure in Figure (24). T is marked as [+Bound], but it has no host. Recall that T cannot lower becomes of the strong Q. The host requirement cannot be overridden. There must be a host. Is there a solution, or will the structure fail, or is the theory wrong?

Fortunately, there is a solution. English has in its repertoire of strategies a set of dummy verbs. The dummy verb do is inserted to function as a host for T when it cannot lower to the main verb:

(28)     Rule: Insert dummy verb do as a last resort.

(29)     Property: Do functions as a host for T when the complement of T is VP whose head is lexical.

At the final stage we return to the lexicon to fill in the final phonological form of each node or mark it empty. Let's start with the dummy verb. The host for T+Q must be phonetic. Rule (28) is applied selecting the morpheme {DO}. This morpheme is linked to the dummy verb as well the lexical entry for the main verb do.

(30).     [+T, +Q] John smoke stogies. -->

(31).    Do+[-Past, +Q]  John smoke stogies.

The lexical entry for the morpheme {DO} contains information that the stem do is spelled out as 'do-' /d+caret/, and following agreement the inflectional ending for the third person singular is regular:. {+Q] determines the interrogative intonation contour

(32).     Do+es John smoke stogies?:

JOHN is spelled out s John. The nodes containing the omega features are linked to an antecedent (the blue arrow), also called the head of a link and consequently of a chain. The tail ed of a link is the original position before movement. In (33) John is head of the link (drawn with a blue line) and the omega features occur at the tail of the link. In links of this sort, the head receives phonological form, and the tail is nearly always empty. When empty, it is called a trace. In days of linguistic yore, it was thought that the phonological item moved. Nowadays it is believe that the bundle of features is copied to the new position (the head of the link), and then spelled out:


The feature [+Strong] in [+Q, +Strong] means that the form to which Q is attached must raise back in the syntax to its original position. This is motivated by the need for interpretation of the sentence containing Q. This a property of lowered forms. It has been observed that traces, an empty form remaining when a node is raised, must be c-commanded by the raised form. If a weak feature is lowered, and spelled out in the lowered position, the feature does not c-command the empty form of C[+Q] as shown in (13).

Finally, this structure is sent to the phonological component where phonological rules spell out the final phonetic form.

Next, note the difficulty of representing the head of CP. This node should be C, but T is moved to that node. At first glance, you might expect there to be a prohibition of a dual node of this sort. The prohibition that we do find is against a node representing two lexical items simultaneously. The problem is with labelling. Node labels are artifacts of linguists, not of the grammar. Grammars do not contain such labels. Linguists have invented them for convenience. It is easy to read N, and realize that this N is a short hand for [+N, -V], two features. C is a label for mood: [MOOD [+Irreal [+Q]]]. T is label for tense. Both mood and tense are features in the node which we have labelled as T, though we could have labelled it T/C or C/T. Whatever labels are selected should be used consistently.

Consider the interrogative of sentences which contain a modal verb:

(34)     John could write good papers.

(35)     Could John write good papers.

(36)     *-d John can write good papers.

It first glance it appears that when a feature raises, its host raises, too. If the feature (or an affix) raises ('-d' in (36)), it the host doesn't raise as well, the feature will end up without a host, which is not permitted. We adopt the the following rule of separability here, which is derived from the Unattached Affix principle:

(37)      Principle: Inseparability

In the syntax once X has been adjoined to some node Y, X cannot be separated from Y.

This principle, then, explains that when C raises back to C, it is T that must raise, since C is adjoined to T.

However, if lexical items are assigned phonetic form which includes affix development as a final step, then this above issue disappears, except to say that if one feature in a feature bundle is copied to another position, all features in that bundle are copied to that position. We can now replace (37) with the following principle which is similar in its scope:

(38)     Principle: Feature Bundle Copying

In a feature bundle if one feature is copied to a new structural position, all features in the bundle are copied to the position.

(30)     Definition: Feature Bundle

A feature bundle is the set of features that are assigned to a terminal grammatical category.

(31)     Definition: Terminal Category

A terminal category is one which dominates no node at some specific point in a derivation in a grammar.

The intent of (31) is to ensure that the features in phrasal nodes which are later expanded into adjunction structures are all copied. When T +Q is raised, the principle ensures that all the features in T+Q are copied, which may be spelled out as a stem plus an inflectional suffix as in the case of di+d. T+Q is a feature bundle, and it is a terminal category before the second lexical pass.

Contents: question as modified proposition | questions in main clause | Q as strong | raising | dummy verb insertion |

To go to the negation operator Click here.

To go to agreement Click here.

To return to the 322 course outline Click here.

This page last updated 21 FE 02