Excursive questions

1.1 Sentences and speech acts

Let us start with the familiar distinction between sentences and speech acts. We presuppose the “generative” approach to syntax which has been presented and motivated in several works during the last few decades (Chomsky 1981, 1995, Haegeman 1991, Radford 2004, among others). According to this view, sentences are structured objects generated by rules of syntax and interpreted by rules of phonology and semantics. The phonological rules apply to the “phonetic form” (PF) of a sentence, while the semantic rules apply to its “logical form” (LF). In what follows, we will not be concerned with PF, and we will represent LF in a simplified manner which we believe suffices for the purpose at hand. Consider, for example, the exchange in (1). The LF of the sentences is given below them.

What about speech acts? These are events that transpire when sentences are used in discourse. In the aforementioned scenario, the acts performed by the utterance of the sentences in (1)-A and (1)-B are, intuitively, those described in (2a) and (2b), respectively.

It is, of course, true that when A utters her sentence, a lot more happens than what is described in (2a): it also happens that A breathes, the earth rotates, etc. These are not part of what we call the “speech act” performed by A. Speech acts, then, are not intended to be all that happens when sentences are uttered in discourse. They are meant to be just those events that constitute “the point” of using the sentence, the essential communicative purpose of the utterance. A more technical term is “illocutionary acts” (Austin 1962). There is a clear sense in which the point of A’s producing her sentence is to ask B whether John used the car yesterday. Similarly, B produces her sentence clearly in order to tell A that John did not use the car yesterday.

1.2 The performative hypothesis

Let us note two facts about speech acts. First, they can be described very accurately. Consider the descriptions in (2), for example. The verbs used to name the speech acts in question, i.e., ask and tell, seem to do their job perfectly. Let us call these verbs “illocutionary verbs.” This state of affairs is not obvious. We perform many acts in our everyday life which cannot be named so precisely by a single word. That we have lexical materials for such an exact characterization of speech acts should in fact be considered a puzzle about natural language (Searle 1969). The second fact to note about speech acts is that they seem to relate to the form of the sentence (Katz and Postal 1964, Schiffer 1972, Strawson 1964). Intuitively, the syntactic profile of A’s sentence, which is an “interrogative,” relates to the fact that it is used to pose question and that of B’s sentence, which is a “declarative,” to the fact that it is used to make a statement. The terms “interrogative” and “declarative,” while intended to be labels of purely formal types, are clearly mnemonic of the use to which sentences of these types are put. In this case, it is to ask (interrogate) and to tell (declare).

The facts just mentioned suggest the possibility that the linguistic system is designed to represent what we have been calling “speech acts.” This hypothesis has in fact been made, and it goes by the name of “Performative Hypothesis” (Levinson 1983). What the performative hypothesis (PH) says, essentially, is that what is expressed by the LF of the sentence also contains information about the speaker, the hearer, and the illocution of the utterance (Chomsky 1981, Frege 1879, Krifka 1995, 2001, 2019, 2020, Lakoff 1970, Miyagawa 2022, Ross 1970, Sadock 1974, Sauerland and Yatsushiro 2017, Stenius 1967, Trinh 2022, Trinh and Truckenbrodt 2018, Whitehead and Russell 1910, Wiltschko 2021). Thus, when A asks B whether p, the literal meaning of A’s sentence is not “whether p,” but actually “A asks B whether p.” To illustrate, let us revisit our scenario again. Under the perspective of PH, the LFs would not be as in (1), but would be as in (3).

As we can see, the LFs of the sentences now mirror the descriptions in (2). What is added, under PH, is the “performative prefixes,” i.e., the part about the speaker, the hearer, and the action performed.

Note, importantly, that performative prefixes are always silent. We never hear them. However, unembedded sentences are always parsed with them. We can immediately see that linguistic communication under PH becomes quite different from how it was traditionally conceived. There is now, technically, only one speech “act,” namely the act of conveying the message which is identified with the literal meaning of the sentence. For example, when A utters the interrogative in (3)-A, she conveys the message that she, A, is asking her hearer, B, whether John used the car yesterday. This message is expressed completely by the LF of (3)-A. And when B utters her sentence, she is conveying to A the message that she, B, is telling her hearer, A, that John used the car yesterday, and this message is expressed completely by the LF of (3)-B. A consequence of PH is that “speech act theory” would now be factored into the theory of grammar, which investigates the semantics and syntax of such items as ask and tell, and the theory of discourse, which investigate how people employ and react to sentences containing them.^[1]

What would constitute empirical arguments for PH? Since the performative prefix is, by hypothesis, silent, such arguments would have to be either semantic or syntactic. Semantic evidence would be cases where we find that the interpretation of the sentence cannot be adequately described without reference to the speaker, the hearer, and the illocutionary verb. Syntactic evidence would be cases where dependencies such as those pertaining to argument structure, variable binding, and island hood span across the position which is semantically identified with the performative prefix.

In this article, we argue that PH is correct. In Section 2, we discuss types of questions which occur frequently in discourse but which, to the best of our knowledge, have not received attention in the literature. We show how they constitute semantic and syntactic evidence for PH. In Section 3, we propose an analysis of such questions. Our analysis relies crucially on the assumption that speech acts are represented in the grammar.

2.1 Semantic evidence

Consider the following exchange between A and B (Schmitt 2021). We will use subscripts to number a discourse participants’ utterances. Thus, A₂ is the second utterance made by A, etc.

Let us focus on B₁. It is, obviously, a question, which is answered by A₂. What is the meaning of B₁? Clearly, it cannot be the question in (6).

That (6) is not the meaning of B₁ is evidenced by the fact that A₂, which answers B₁, clearly does not mean (7).

It turns out that the interpretation of B₁, and consequently of A₂ which answers it, can be straightforwardly described when we make reference to the illocutionary acts which are intuitively associated with these utterances. Thus, (8a) and (8b) seem to deliver just the right meaning of B₁ and A₂, respectively.

Thus, B₁ is most accurately described as a question about a question, or more precisely, about a question act. It represents a sort of excursus in the conversation, a detour that is taken before the conversation gets back on track. We will hence call such questions “excursive questions” (EQs). The question about which an EQ inquires will be called the “antecedent question” (AQ). In the scenario in (5), A₁ is the AQ, while B₁ is the EQ. Note that AQ in this case is a polar question, but AQ can be a wh-question as well, as (9) illustrates.

Again, we see that we cannot avoid mentioning the illocutionary act in specifying the interpretation of the EQ and its answer. The utterance where in (9) can intuitively be described as asking “which place x i is such that A is asking B when people eat dinner at the place x i .” Similarly, the answer to the EQ can intuitively be described as saying “A is asking B when people eat dinner in Spain.” EQs, therefore, present semantic evidence for PH. Let us now turn to some syntactic evidence.

2.2 Syntactic evidence

3.1 Deriving the LF of EQs

The starting point of our analysis is the claim that HP is correct: sentences may contain performative prefixes. Moreover, we will allow the grammar to append such prefixes recursively. This means a structure such as (25) is, in principle, well-formed.

The sentences in (1), repeated below in (26), can now be ascribed to the following LFs.^[7]

Let us go into more detail about how LFs such as that in (26)-B are generated and interpreted. First, we assume that wh-phrases are underlying of the form which NP, where NP is time in the case of when, thing in the case of what, person in the case of who, etc. Second, we assume that determiners, which include which and the, come with indices (Hackl 2019). The index on the determiner is interpreted as an appositive of the complement NP.

The interpretation of the EQ in (26) would then be (28).

As we can see, this interpretation amounts to a description of the speech act event that transpires when B utters the EQ. Under the perspective of PH, this is precisely the meaning of the EQ (as uttered by B).

But how is this LF constructed? Here is our crucial claim about the derivation of EQ.

This means EQs are derived as any sentence is, except that the input must be the structure of AQ. What are the “syntactic rules”? One of them is, of course, merge. This is the operation which maps two expressions α and β to the expression containing α and β as immediate constituents. We will assume that merge is subject to only one constraint: one of its arguments must be unembedded. Thus, we allow for movement and adjunction to non-matrix nodes. Let us now see how the LF in (26)-B is derived. By hypothesis, we start with the AQ, which is (30).

We then merge α with at which _i time, which has been constructed independently in the derivational workspace. This operation extends α to α′, resulting in (31).

Next, we merge which _i time with γ , extending it to δ , as shown in (32). As γ contains which _i time as one of its subconstituents, this is a case of wh-movement, resulting in which _i time occupying two different positions in the same tree. We indicate the lower position, i.e., the “trace,” with angled brackets.

What follows is an incremental merge of ask, A, and B to yield ε .

The final step involves a rule which is not merged. This is Fox’s 2003 rule of Trace Conversion (TC) which applies to lower copies of wh-movement and changes which _i to the _i . Note that this means lower copies must be interpreted as definites (Fox 2003, Sauerland 2004).

Applying TC to (33) results in (35), which is our target LF.^[8]

A question which needs to be addressed at this point is whether the step from (31) to (32) violates any locality constraint. This step involves movement of which _i time from inside α ′ to γ .

As the movement crosses the wh-expression whether, the objection might be raised that it violates the Wh-Island Constraint (Ross 1967). However, note that this constraint ended up being derivable from Chomsky’s theory of bounding nodes (Chomsky 1981, 1986a). The two propositions we need from this theory are the following two.

Standard cases of Wh-Island Constraint violations turn out to be cases in which a step of movement crosses two bounding nodes. An example is given in (38a). The theory also explains the Complex NP Island Constraint, as illustrated in (38b).^[9] The relevant bounding nodes are framed and the offending movement indicated with dashed instead of solid arrows.

Now, let us say that performative prefixes are not tensed. Specifically, they extend the CP domain, but do not introduce any new TP. Then, there will be no problem with the movement in (36). For clarity, we reproduce (36) in (39), with the Greek letters replaced by category labels.

As we can see, the movement crosses a single bounding node.

3.2 Explaining the observations

It is clear from the discussion above that the LFs which we derive for EQs deliver the intuitively correct interpretation of these expressions. It should also be clear how we would explain the connectivity effects (Section 2.2.1), the reconstruction effects (Section 2.2.2), and the A ¯ -dependency effects (Section 2.2.3).^[10] These effects all reduce to properties of wh-questions, which is what we take EQs to be. We leave it to the reader to verify this.

There remain the definiteness effects discussed in Section 2.2.4. Recall the relevant contrast, reproduced below in (40) and (41).

It turns out that the contrast follows straightforwardly from (29); i.e., the claim that EQ is derived from AQ by syntactic rules. Let us first consider how the EQ in (40) is derived. The starting point, by hypothesis, is the AQ, which is (42).

Next, we merge which _i book to α, extending it to β, as shown in (43).

Note, at this point, that we have not performed any movement operation. The step from (42) to (43) is an external merge, not an internal merge. Nevertheless, the output is the same as that which would result from movement and TC: we end up with which _i book in the operator position and the _i book at the base position. The subsequent steps are those of merging A, B, and ask to yield (44), which delivers the intuitively correct meaning.

Now, let us turn to the deviant EQ in (41). It is intended to mean the same as the non-deviant EQ in (40). Thus, we want to derive for it the same LF as that in (44). Can we? The answer is no. Given the claim in 3.1, the derivation would have to start from the AQ in (41), which is (45).

Subsequent steps of the merge will then give us (46).

The problem with this structure is that there is nothing in it which could be interpreted as the trace of which _i book. Traces must be definite, and a _i book is not definite. Thus, which _i book would be a vacuous quantifier, which is something not tolerated in natural language grammar (Chomsky 1986b). Can we change a _i book into the _i book? The answer is no. There is simply no syntactic rule which can bring about this change. Note that TC would not do the job, because TC applies to traces (lower copies), and a _i book is not a trace.^[11]

This account of the definiteness effect extends to cases of implicit argument also. Consider (47).

The deviance of the EQ in (47) would be explained if we make the standard assumption that the implicit argument of eat is an indefinite (“something”), but not a definite (Chomsky 1981).