Aggressively non-D-linked construction and ellipsis: A Direct Interpretation approach

Abstract

The so-called aggressively non-D-linked construction (ANDC) involving wh-the-hell phrases like what the hell is of empirical and theoretical interest due to its complex morphosyntactic and semantic/pragmatic properties. This paper focuses on the construction in general as well as in ellipsis phenomena. We first explore its grammatical properties on the basis of attested corpus data and show that the construction can occur more widely in elliptical constructions than suggested by previous literature. We then suggest that the licensing conditions of the ANDC in ellipsis are not solely syntax-based but due to tight interactions among a variety of grammatical components such as morphosyntax, semantics, and discourse/pragmatics. We also argue that the authentic uses of the construction favor a Direct Interpretation (DI) approach that can account for its uses in a variety of environments.

1. Introduction

It is widely accepted that wh-expressions can be classified into two types: D-linked (discourse-linked) and non-D-linked ones (Pesetsky 1987, 2000: 16):

The key difference of these two types, as pointed out by Pesetsky (1987) and subsequent work, comes from a discourse structure. The expression which book in (1a) implies the existence of a set of contextually determined entities from which the speaker asks for a choice, whereas what in (1b) carries no such implication. That is, in (1a) with the D-linked wh-phrase which book, there is a set of books determined in the discourse and it is questioned to select one from the members of this set that Kim read. However, in (1b) with the non-D-linked wh-phrase what book, there is no discourse-provided set referring to the entities Kim read.

In addition to these two types, there is another related wh-type that combines with an emotive expression like the hell, the heck, on earth, and the Dickens. This phrase is taken to be an ‘aggressively non-D-linked’ wh-phrase since it is non-D-linked and further expresses a strong negative feeling, as illustrated by the following examples (Pesetsky 1987; Ginzburg & Sag 2000; Den Dikken & Giannakidou 2002; Huang & Ochi 2004):

The contrast here can be attributed to the difference in the D-linking property of what and which. ² The interrogatives in (2) are information-asking questions, but can accompany a negative inference such that Kim should not buy anything for (2a), or the speaker’s negative attitude (surprise, frustration, annoyance, etc.) toward the sentence in question.

Literature has noted that the wh-the-hell phrase displays unusual properties with respect to ellipsis. The noted observation has been that the wh-the-hell phrase is disallowed in Sluicing, but can occur in the so-called Swiping (Merchant 2001: 111–112, 2002; Den Dikken & Giannakidou 2002; Sprouse 2006; Hartman & Ai 2009). Consider the following data:

Examples in (4a, b) are typical Sluicing examples, but, as observed, cannot have the emotive phrase the hell following the wh-expression. However, as in the Swiping example in (4c), this ungrammaticality can be saved by having a preposition after the emotive phrase. These three elliptical constructions have been often argued to involve movement as well as clausal ellipsis while attributing the illegitimate presence of the wh-the-hell phrase in ellipsis to a phonological constraint such that the rightmost expression needs to be given stress (Merchant 2001, 2002; Hartman & Ai 2009; Güneş & Lipták 2021).

However, a corpus search yields a significant number of wh-the-hell phrases in Sluicing environments: ³

Such attested examples, in which Sluicing after the wh-the-hell phrase occurs in matrix and embedded clauses, indicate that we cannot simply rule out the uses of the wh-the-hell phrase in elliptical environments. If the wh-the-hell phrase is licensed in Sluicing and other related ellipsis phenomena, questions then arise: when and how the construction can be used, and what licenses the wh-the-hell phrase in ellipsis environments. This paper attempts to answer these questions.

We organize the paper as follows. In Section 2, we first review some key grammatical properties of the ANDC (aggressively non-D-linked construction) noted in literature. Section 3 discusses the findings of our corpus investigation to understand its authentic uses in real life. This section also suggests that attested data do not countenance some of the observations made in previous literature. In Section 4, we then offer a construction-based Direct Interpretation (DI) analysis of the construction that introduces no hidden syntactic structures in the putative ellipsis site. In Section 5, we summarize our main findings and conclude the paper.

2. Some key properties

The wh-the-hell phrase has several unique morphosyntactic properties. First, the combination of the wh-expression with an emotive phrase is an inseparable syntactic unit, as illustrated in the following examples (Ginzburg & Sag 2000: 229; Merchant 2002; Huang & Ochi 2004): ⁴

As the data tell us, the emotive phrase like the hell forms a tight syntactic unit with the preceding wh-expression. This syntactic cohesion is further evidenced from attested examples like the following:

In these examples, the contracted auxiliary or the possessive marker ’s hosts the wh-the-hell phrase. Given the fact that the former combines with a subject and the latter with an NP, the attested examples here suggest that the wh-the-hell phrase is a single constituent as a whole.

A key defining property of the wh-the-hell phrase is that unlike normal wh-phrases it cannot occur in situ (Pesetsky 1987; Ginzburg & Sag 2000: 229–230; Den Dikken & Giannakidou 2002; Huang & Ochi 2004). Consider the following examples:

Examples as in (9) show that a normal wh-phrase what alone can occur in situ, but its wh-the-hell phrase counterpart what the hell needs to be ‘fronted’ to the sentence initial position. The requirement for fronting the wh-the-hell phrase to the sentence initial position also holds when it is in a lower clause, as demonstrated in (10). ⁵

As discussed earlier, a salient property of the ANDC concerns the discourse information. The wh-the-hell phrase in general has no referent available in the previous discourse. This discourse requirement disallows it from combining with which (N):

Different from the which (N) phrase, wh-expressions like what, who, when, and how many (N) do not require a determined set of individuals in discourse (Pesetsky 1987; Den Dikken & Giannakidou 2002; Huang & Ochi 2004). ⁶ For instance, consider the examples below:

In unmarked situations, the interlocutors of these interrogatives do not need to share a particular set of individuals in the discourse to make a felicitous answer. That is, no previous discourse is necessary with respect to the referent of a what- or who-phrase, as opposed to a which-phrase. This is why it is rather infelicitous to utter *What do you like most/more? while it is fine to say sentences like Which one do you like most/more?

This discourse property of having no salient set in the discourse also seems to lead to a difference in the following examples (Den Dikken & Giannakidou 2002; Huang & Ochi 2004):

The badness of (14b), in contrast to (14c), can be attributed to the contradiction that the wh-the-hell phrase inherently involves no knowledge of the referent for the non-D-linked wh-expression who, but the speaker says that she knows who that person is. Observing such a contrast, Den Dikken & Giannakidou (2002) suggest the parallelism between the wh-the-hell phrase and NPIs (negative polarity items):

The contrast here indicates that the wh-the-hell phrase appears only in nonveridical contexts, like NPI licensing items like not. This fact is also related to the non-D-linking constraint. Both the nonveridicality and wh-the-hell phrase contexts do not express certainty about, or commitment to, the truth of a sentence. That is, the wh-the-hell phrase implies that the referent of the wh-expression is unavailable to the speaker.

Another prominent discourse property of the ANDC, as briefly noted above, is that the construction with the wh-the-hell phrase expresses the speaker’s negative attitude toward the possible value of the wh-the-hell phrase, as seen from the following data:

The examples in (17), just like those in (16), can be information-asking in that they ask for a value of the variable introduced by the wh-phrase. However, those in (17) at the same time convey the speaker’s negative attitude (frustration, anger, or surprise) toward the proposition evoked by the question.

In addition, the wh-the-hell phrase gives rise to only a wide scope reading with respect to a quantifier unlike normal wh-phrases (Den Dikken & Giannakidou 2002):

In the example in (18a) with a normal wh-phrase what, either a wide or narrow scope reading of what is available with respect to the universal quantifier everyone. On the other hand, in the example in (18b) with its wh-the-hell phrase counterpart, only a wide scope reading of what the hell is available with respect to the universal quantifier. Similar to this scope restriction, no non-local reading is available to the wh-the-hell phrase (Ochi 2004, 2015):

In the example in (19a), the normal wh-phrase why can be related to either the event of saying in the matrix clause or the event of Kim’s being mad in the embedded clause. In other words, it is ambiguous in that the normal wh-phrase why allows both a local reading and a non-local reading. However, in the example in (19b), its corresponding wh-the-hell phrase can only be related to the event of saying in the matrix clause, not the event of Kim’s being mad in the embedded clause, disallowing a non-local reading.

As discussed so far, the ANDC introduced by a wh-the-hell phrase shows several intriguing morphosyntactic and semantic/pragmatic properties, which distinguish the construction from other related constructions. In the next section, we discuss our corpus findings for its uses in real life situations.

3. Corpus findings and discussion

3.1 Corpus used and search methods

In order to investigate the authentic uses and grammatical properties of the ANDC, we performed a corpus investigation using COCA (Corpus of Contemporary American English). To extract ANDC examples from COCA, we first used simple string searches with some regular expressions as given in (20):

Such string searches gave us a total of 17,249 tokens and we then manually excluded irrelevant examples like those in (21):

In (21a), which on Earth has a literal meaning with the relative pronoun use of which rather than functioning as a wh-the-hell phrase. In (21b), the devil is not used as an emotive expression in a wh-the-hell phrase but it functions as the subject of the verb looks. After manually filtering out such irrelevant examples from the extracted data, we have identified a total of 15,651 tokens with the wh-the-hell phrase, for which we have performed a quantitative and qualitative investigation, as discussed in what follows.

3.2 General distributions of the ANDC examples

As for the data extracted from COCA, we first looked into the registers of the identified tokens. Figure 1 shows the uses of the construction by registers.

Figure 1 Frequencies of the ANDC by registers in COCA.

As seen from Figure 1, the ANDC is mainly used in informal, colloquial contexts such as fiction and spoken registers and it is less preferably used in formal contexts like academic register. This seems to support the discourse uses of the construction to express the speaker’s negative attitude toward the proposition associated with the interrogative.

We also identified that the predominant wh-expression used in the wh-the-hell phrase is what, followed by how, why, who, and where. Figure 2 shows the distribution of the identified ANDC examples from COCA by wh-words.

Figure 2 Frequencies of the ANDC by wh-words in COCA.

As seen in Figure 2, the frequencies of how, why, who, and where are quite similar. In the meantime, the frequency of when and that of whom in the construction are quite low, compared to the other wh-expressions. The corpus yielded no token of the wh-the-hell phrase with which, supporting the traditional dichotomy between D-linked and non-D-linked wh-expressions. The following include some examples of the construction we identified from the corpus:

The corpus data also include tokens where the wh-the-hell phrase is used as part of a complex phrase:

Moreover, the identified tokens have examples where the emotive expression in the wh-the-hell phrase includes a pre-modifier as in (24), which is unnoticed by previous literature:

In the examples in (24) modifiers like bloody and doggone are used in the wh-the-hell phrase to emphasize the negative connotation inherent in the construction.

In terms of syntactic combination, the data show us that the wh-the-hell phrase can combine with either a finite or a nonfinite dependent. In most cases, the wh-the-hell phrase combines with a finite sentence (13,126 instances), but it can also combine with a nonfinite sentence or XP:

In (25), the wh-the-hell phrase combines with a nonfinite sentence. One may take the examples in (25a–d) to involve the absence of a finite auxiliary verb, but an example like (25e) indicates that the wh-the-hell phrase can combine with an infinitival VP, as does a regular wh-phrase.

We have also checked the variations of the ANDC in matrix and embedded environments. The ANDC can occur in both matrix and embedded environments, but it is dominantly used in the former. A total of 13,390 instances of the ANDC (85.6%) occur in matrix environments while only 2,261 instances (14.4%) occur in embedded environments, some of which are given in (26):

As shown in (26a–d), the ANDC can occur as the complement clause of a verb, adjective, noun, and preposition. It can even occur as a sentential subject as in (26e).

Related to this, Den Dikken & Giannakidou (2002), as mentioned earlier, noted that the wh-the-hell phrase is only used in nonveridical contexts. However, our corpus search yields examples of the wh-the-hell phrase in veridical contexts, as illustrated in (27):

Note, at this juncture, that predicates like know, recognize, and find out do not license an NPI, as seen from the following:

Within the assumption that NPIs occur only in nonveridical environments (Giannakidou 2002), such examples imply either the extended uses of the wh-the-hell phrase or a need to revise the claim that the wh-the-hell phrase occurs only in NPI environments.

3.3 Uses in elliptical environments

As noted in the beginning, the wh-the-hell phrase displays several unexpected properties in elliptical constructions including Sluicing. We have identified a total of 2,290 tokens of the wh-the-hell phrase (about 15% of total 15,651 tokens) in the elliptical environments. The tokens are distributed over the three main types, whose frequencies are given in Figure 3.

Figure 3 Frequencies of the wh-the-hell phrase in ellipsis.

As represented in Figure 3, the most predominant elliptical environment for the wh-the-hell phrase is Sluicing. The frequency of the wh-the-hell phrase in Stripping and Swiping is significantly low, but is consistently observed.

3.3.1 Sluicing

One unexpected finding from the corpus data, as in Figure 3, is a significant number of tokens with the wh-the-hell phrase in matrix and embedded Sluicing: ⁷

Of these Sluicing tokens, 2,091 (99%) occur in matrix environments and only 21 (1%) occur in embedded environments. ⁸

Much of the prior literature has noted that Sluicing introduced by the wh-the-hell phrase is unacceptable in matrix environments (Ginzburg & Sag 2000; Merchant 2002). Consider the following example provided by Ginzburg & Sag (2000: 314):

The example indicates that the wh-the-hell phrase cannot refer to an overt antecedent or correlate in the previous discourse, reflecting the non-D-linking property of the wh-the-hell phrase. This contrasts with examples like the following:

The key difference from (31) is that this example has no overt correlate or antecedent introduced by the discourse. The wh-the-hell phrase is used as a nonsentential utterance referring to the situation in question.

This discourse constraint, as we discussed earlier in (4), also holds in the embedded environment. Observe similar examples below (Den Dikken & Giannakidou 2002; Merchant 2002; Sprouse 2006; Almeida & Yoshida 2007; Hartman & Ai 2009):

Examples like (33) belong to the so-called merger type of Sluicing in that the antecedent clause includes an overt correlate someone linked to the wh-remnant who in the second clause. ⁹ The ungrammaticality of these merger examples is rather expected when considering the non-D-linking nature of the wh-the-hell phrase. Just like the wh-the-hell phrase in (31), the wh-the-hell phrase here is linked to an overt correlate introduced by the previous discourse.

Observing these discourse factors, we classified the identified Sluicing examples into three different groups, depending on the correlate/antecedent type, as in (34):

The example in (34a) is taken to be pragmatically controlled (or exophoric) in the sense that the context with no linguistic antecedent can provide a key to the intended meaning of who on earth? Hearing the doorbell ring at 10 p.m., the speaker utters the nonsentential utterance (NSU), who on earth? There can be more than one putative source sentence for this NSU, such as Who on earth is visiting me at this time? and Who on earth is ringing the bell at this time? The sprouting example in (34b) is a case where the correlate is implicit but the previous antecedent clause provides a basis for the intended meaning of the remnant, as in How on earth is being a vegetarian a positive one? Lastly, the type involved in (34c) is referred to as ‘pseudo-merger’ in that it is the same as merger in terms of having an overt antecedent clause and an overt correlate but their correlate types and functions are different. For instance, in (34c) the antecedent clause is a wh-question Where did you put her? and the correlate is the wh-expression where, not a simple indefinite expression like somewhere, as in He put her somewhere but I don’t know where. The pseudo-merger example in (34c) is specifically used to emphasize the previously uttered antecedent/correlate. Since it is different from merger in these respects, it is termed ‘pseudo-merger’ here. The frequencies of these three types are given in Figure 4.

Figure 4 Frequencies of the wh-the-hell phrase in Sluicing by the correlate/antecedent types.

As given in Figure 4, the pragmatically controlled type is the most dominant one used in Sluicing with the wh-the-hell phrase, possibly due to the main discourse functions of the construction. That is, the key function of the wh-the-hell phrase is to add the speaker’s negative attitude toward the situation evoked from the wh-question. For instance, (34a) could ask himself who is the one knocking the doorbell, but at the same time expresses the speaker’s negative attitude such that no one should knock the doorbell at that time. Further, since the wh-the-hell phrase is non-D-linked, the discourse does not need to provide a discourse referent of the wh-expression. This non-D-linking property seems to result in the high frequency of the pragmatically controlled type followed by the sprouting type, but no instances of the true merger type that has an overt correlate in the discourse.

3.3.2 Stripping and Swiping

Stripping or bare argument ellipsis (BAE) is an ellipsis that elides everything from a clause except one constituent. Since the wh-the-hell phrase must involve a wh-expression, we investigated Stripping with the wh-the-hell phrase where the wh-the-hell phrase has one remaining constituent. First, consider the following Stripping data with normal wh-phrases from Ortega-Santos, Yoshida & Nakao (2014: 58):

Why-Stripping as in (35a) and wh-Stripping as in (35b) differ in that the former is only introduced by why while the latter is by a wh-expression other than why. ¹⁰ From these, we would expect examples like Why the hell about syntax? and Who the hell about phonology? Among the identified examples from COCA, 130 tokens involve these kinds of Stripping as shown in Figure 3. We classified these Stripping tokens with the wh-the-hell phrase on the basis of wh-words and the categories of the stripped remnant and Table 1 shows their distribution.

Table 1 Frequencies of Stripping with the wh-the-hell phrase based on wh-words and the categories of the stripped remnant.

Of these 130 Stripping tokens with the wh-the-hell phrase, 117 belong to why-Stripping while the remaining 13 tokens involve what (12 tokens) and how (1 token). The categories of the stripped remnant with the wh-the-hell phrase vary, including NP, VP, AdvP, PP, and the negation marker not: ¹¹

In terms of the correlate/antecedent type, Stripping with the wh-the-hell phrase also has three different types: pragmatically controlled (48 tokens), sprouting (80 tokens), and pseudo-merger (2 tokens). The following demonstrate these three types:

As noted in Figure 3, another type of ellipsis we have identified with the wh-the-hell phrase is Swiping. The following are a few from the 48 Swiping examples with the wh-the-hell phrase:

In these Swiping examples, the wh-the-hell phrase combines with a preposition. The wh-word and preposition combination patterns in the data are quite limited: of the total 48 tokens, 46 have the combination of what and for, and the remaining two are that of what and about as in (39b) and that of what and with as in (39c). In terms of the correlate/antecedent type in these Swiping examples, we could identify three tokens of the pragmatically controlled type and 45 tokens of the sprouting type, as given in (40):

Some key observations from the corpus data include the following. First, the ANDC is dominantly used in informal contexts such as fiction and spoken registers. Second, its uses are more diverse than observed in previous literature. Third, most notably, contrary to the previous observations, the wh-the-hell phrase can be used in a variety of elliptical constructions. When the wh-the-hell phrase is used to introduce the elliptical constructions, it occurs more frequently in sprouting or pragmatically controlled contexts than in merger contexts, which seems to be related to the non-anaphoric (non-D-linking) properties of the wh-the-hell phrase. In what follows, we try to offer a non-derivational analysis for the construction in general as well as in elliptical environments.

4. A construction-based analysis

4.1 Licensing the wh-the-hell phrase

Let us first discuss the formation of wh-the-hell phrases. In licensing wh-the-hell phrases, as noted earlier, the grammar needs to allow only a limited set of emotive expressions like the hell, the heck, and on earth to modify a wh-expression.

The possible emotive expressions that can occur in the wh-the-hell phase are quite idiosyncratic, as demonstrated in (42):

As shown here, the emotive expression disallows the indefinite article a/an and must be definite as in (42a). The emotive noun cannot be plural as in (42b), but can host an internal modifier (e.g. the bloody hell), as discussed earlier in (24) and illustrated here in (42c).

Another basic property of the wh-the-hell phrase is that it is only acceptable in contexts with question-orientation (Huddleston 1993; Ginzburg & Sag 2000: 9, 230): ¹²

The wh-word here has no interrogative use: the wh-word in (43a) is a relative pronoun and the one in (43b) is an exclamative pronoun. This requirement can also differentiate between the two examples below:

The example in (44a) is grammatical since the embedded clause introduced by the wh-the-hell phrase is selected for by a verb with question-orientation wonder; however, the example in (44b) is ungrammatical as it is selected for by a verb with answer-orientation recall.

To capture such unique combinatorial properties of the wh-the-hell phrase, we first accept Ginzburg & Sag’s (2000) suggestion that wh-words have nonempty specifications for the feature wh, as in the following feature structure specifications:

The lexical specifications here ensure that the interrogative wh-word who bears a nonempty wh feature value which is a parameter referring to a person with an index value x. The emotive expression then modifies the preceding wh-word with a nonempty wh value. As noted earlier, the emotive phrase cannot modify the wh-phrase with a nonempty wh value:

Further, we have seen that only a limited set of wh-words and emotive expressions can participate in the combination. Considering these peculiarities, we suggest that English has an idiosyncratic construction, as given in the following:

This construction rule licenses the combination of a non-D-linked wh-word with an emotive expression which is also predetermined in the grammar of English. The syntactic cohesion of the resulting expression, as noted earlier and further suggested by a reviewer, behaves like a lexical expression (marked with [light +]) with respect to syntactic distributions (see Section 4.3 also). ¹³ The specification of the feature light is to reflect that the construction is a light, quasi-lexical constituent. Within this system, words as well as the combination of two words are thus [light +], while phrases are typically [light –]. However, the combination of a wh-word with the emotive phrase results in a [light +] expression. ¹⁴ Note also that the emotive expression is a functor that combines with a wh-word via the feature sel in accordance with the Head-Functor Construction, which is independently motivated for treating specifiers and modifiers in a uniform manner (see, among others, Van Eynde 2007; Kim & Sells 2011; Kay & Sag 2012). ¹⁵ This construction rule would then project a structure like the following:

As represented in the structure, the NP emotive expression the hell is a functor and selects the interrogative wh-word what. This eventually results in the formation of a well-formed wh-the-hell construct bearing the feature [light +] so that it can behave like a lexical expression. As we will see in what follows, this light feature allows Swiping to be possible only with a [light +] expression (wh-word and wh-the-hell phrase) and a limited set of prepositions.

In the construction, the wh-word also needs to have a nonempty parameter value for the feature wh, which bars the emotive phrase from combining with a noninterrogative wh-word as in (43) and as in *the student who the hell we met last night and *I ate what the hell Kim ate. The requirement for the nonempty wh value could also block examples like the following:

The complementizer whether can introduce an interrogative clause, but inherently lacks a parameter value of the feature wh, as suggested by Ginzburg & Sag (2000: 214). ¹⁶

The present analysis can also make other predictions, accounting for the grammaticality and ungrammaticality of the following examples:

The wh-the-hell phrase can be an NP constituent and occur as the specifier of the possessive marker, as in (51). The ungrammaticality of the examples in (52) is expected as well, because the head wh-word and the emotive modifier phrase are in discontinuous positions so that the emotive phrase cannot select the head wh-word.

4.2 Licensing the construction

With the constructional formation of the well-formed wh-the-hell phrase, let us now consider how the grammar licenses its occurrences in syntactic environments. The defining property of the wh-the-hell phrase in English, as we have noted, is that it cannot stay in situ, whose key data we repeat here:

The wh-the-hell phrase is illicit when staying in situ, as in (53a, d). Considering that normal wh-phrases in English can appear in situ as in (54), this positional requirement is rather a constructional one. ¹⁷

Another motivation for a construction-based approach, as we have noted earlier, comes from its independent discourse constraint. As observed by Den Dikken & Giannakidou (2002) and others, the construction involving the wh-the-hell phrase occurs in NPI environments that trigger a widening effect. For instance, in (53b), the domain of who the heck is an open set including all the possible individuals that satisfy ‘Sandy visited x’. In addition to this widening effect, we have seen that the construction evokes a negative inference in (53b) such that Sandy should visit nobody or such that the speaker does not have any knowledge of the individual ‘x’ that Sandy visited. This kind of negative inference does not come from any individual expression in the sentence in (53b), but arises only when the wh-the-hell phrase occurs in a specific construction.

As a way to address these syntactic and pragmatic peculiarities, we suggest that English employs the following independent construction involving a wh-the-hell phrase: ¹⁸

The construction has two daughters: a wh-the-hell phrase and a head sentence which has this wh-the-hell as its gap (gap) value. ¹⁹ In addition, its constructional constraint also includes the contextual background information such that the speaker has a negative attitude toward the situation (s ₀) in question. Since this information refers to a discourse structure, it can be identified even when this head is elided as in How the hell?. The constructional rule thus allows a non-D-linked wh-the-hell phrase to combine with an incomplete S, which yields a head-filler unbounded construction:

In (56), the wh-the-hell phrase serves as a filler and combines with a sentence with a gap whose grammatical function corresponds to the direct object of the verb call or need. ²⁰ The examples in (57) even show a long distance dependency between the wh-the-hell phrase and the putative gap in the embedded clause. This becomes clear when considering a simplified structure of (56b):

The head S has an NP gap (gap) which is linked to the filler, the wh-the-hell phrase. The gap value can be in a long distance relation with the filler, as in (57).

The construction has a contextual background that evokes a pragmatic inference conveying the speaker’s negative attitude toward the situation (denoted by the head S) in question. ²¹ For instance, the examples in (56) allow us to infer that the addressee should not call anyone and Kim should not need anything. Even in information-seeking examples like (59), discussed by Güneş & Lipták (2021), there is a negative inference:

The speaker B wonders about a value for the wh-expression (‘I wonder who he has seen’), but at the same time has a negative attitude (unexpected surprise) or rhetorical question such as ‘he shouldn’t have seen anyone’.

Another advantage of this construction-based account comes from examples like the following (Sprouse 2006: 350): ²²

The contrast here tells us that we cannot simply disallow double wh-the-hell phrases in a sentence. The present system blocks examples like (60a) because the second wh-the-hell phrase is not licensed by the construction rule in (55). However, the rule licenses both wh-the-hell phrases in (60b) as the non-D-linked nonhead daughter. ²³

As discussed earlier, the construction typically occurs in nonveridical situations, but our corpus investigation yields examples that at first glance seem to be veridical:

If we take the wh-the-hell phrase as an NPI as claimed by Den Dikken & Giannakidou (2002), such examples would not be expected since the wh-the-hell phrase is in the embedded clause selected for by verbs like know, figure out, and find out. To make such sentences acceptable, there needs to be a licensor like not or a question operator as suggested by Den Dikken & Giannakidou (2002), but there exists none here. Instead, the present system can attribute the possibility of such examples to a discourse factor of the construction. What we can observe here is that the veridical predicate is further embedded by a construction like want to, need to, and try to. These contexts imply that the speaker seeks a value for the wh-expression (which is a nonveridical environment) and expresses his or her negative attitude toward the situation in question.

4.3 ANDC in ellipsis

Let us now discuss the distribution of wh-the-hell phrases in ellipsis. In accounting for ellipsis in general, there have been two main strands: movement and PF-deletion and Direct Interpretation (DI) approaches. The movement and PF-deletion approach basically assumes that fragments are canonical utterances of the type S (see, among others, Ross 1969; Merchant 2001, 2002, 2004; Weir 2014; Yoshida, Nakao & Ortega-Santos 2015). Within this kind of movement and PF-deletion view, an ellipsis site has internally structured material through derivation and PF-deletion renders some of it unpronounced under some kind of identity and the meaning composition is dependent upon the derivational source. For instance, according to the movement and PF-deletion approach, the Sluicing example in (62a) would be derived from (62b) (Merchant 2001):

The wh-expression what is moved to [Spec, CP] motivated by a focus assignment, and then the remaining clause is deleted. As seen earlier in (4), however, we cannot simply apply such a process as in (63b) because it would allow ungrammatical examples like (63a).

We cannot syntactically bar the clausal ellipsis after the wh-the-hell phrase as we have seen from the attested, possible Sluicing data in matrix as well as embedded environments. The existence of such attested examples also casts doubt on the assumption that the ungrammaticality of such examples is due to the lack of a phonological accent on the emotive expression, as suggested by Sprouse (2006) and Güneş & Lipták (2021).

Meanwhile, the Direct Interpretation (DI) approach for ellipsis, which we adopt in this paper, directly generates ellipsis with no clausal source and that allows the resolution of the elided part by structured discourse (Ginzburg & Sag 2000; Culicover & Jackendoff 2005; Sag & Nykiel 2011; Nykiel 2013; J.-B. Kim 2015; Jacobson 2016; Kim & Abeillé 2019; Kim & Nykiel 2020; J. Kim 2021; Nykiel & Kim 2022). Within the DI approach, there is no syntactic structure at the ellipsis site and the fragment is the sole daughter of an S-node, directly generated from a construction rule like the following (Ginzburg & Sag 2000):

All the attested NSUs with the wh-the-hell phrase belong to this Head-Fragment Construction. For instance, consider the following attested example:

This naturally occurring fragment would have a simple structure like the following:

Here, the interrogative wh-word who combines with the emotive expression the hell in accordance with the wh-the-hell Construction, forming an NP first, and then this NP is projected into an NSU (nonsentential utterance) S on its own as a type of the Head-Fragment Construction. This S can serve at the same time as an instance of the ANDC whose head S is unexpressed but supplied by the discourse (e.g. Who the hell is calling me?). ²⁴

To be more precise, as the resolution of this kind of fragment into a propositional meaning, the DI approach relies on the discourse structure, rather than on the putative clausal source. The resolution of the NSU is achieved by discourse-based machinery. That is, the interpretation of a fragment depends on the notion of ‘question-under-discussion’ (qud) in the dialogue. Dialogues are described via a Dialogue Game Board (dgb) where the contextual parameters are anchored and where there is a record of who said what to whom, and what/who they were referring to (see Ginzburg 2012). dgb monitors which questions are under discussion, what answers have been provided by whom, etc. The conversational events are tracked by various conversational ‘moves’ that have specific preconditions and effects. The main claim is that NSUs, corresponding to salient utterances, are resolved to the contextual parameters of the dgb. Since the value of qud is constantly being updated as the dialogue progresses, the relevant context offers the basis for the interpretation of fragments. In this system, dgb is part of the contextual information and has at least the attributes sal-utt (salient-utterance) and max-qud (maximal-question-under-discussion), given in (67):

The feature max-qud, representing the question currently under discussion, takes as its value questions. Meanwhile, the feature sal-utt, taking as its value syntactic as well as semantic information, represents the utterance which receives the widest scope within max-qud.

To see how this discourse-based system works, consider the following sprouting example:

Uttering the declarative sentence They survived can also introduce a qud, activating the appropriate dgb information, as given in (69):

As represented here, the declarative sentence can introduce a qud questioning how (the manner x) they survived. ²⁵ The fragment question How the hell? is basically asking a value for the variable x. The Head-Fragment Construction allows any phrase matching the focal or salient utterance (sal-utt) to be projected into a sentential expression S. The remnant wh-the-hell phrase matches the sal-utt, which is the manner they survived in the context (Ginzburg & Sag 2000), as shown in the following:

As shown here, the NSU is a stand-alone clause, following the Head-Fragment Construction. This NSU matching the sal-utt value refers to the qud introduced by the preceding declarative assertion sentence They survived:

The evoked qud is that the speaker asserts that they survived and she asks herself the manner for this, in particular, with the focus on the adverb wh-word how.

This discourse-based approach implies that the grammar would license the ANDC in matrix Sluicing as in (29) and embedded Sluicing as in (30). Consider similar examples below:

The present analysis assumes that any wh-the-hell phrase in Sluicing can be projected into an S so long as an appropriate context can be retrieved. This allows us to account for the cases where the wh-the-hell phrase has no linguistic antecedent clause at all but its antecedent is just pragmatically controlled. For instance, the wh-the-hell phrase in (34a), repeated in (73), can have several different types of max-qud as given in (74):

This discourse-based analysis thus could avoid pitfalls that any analysis resorting to syntactic identity between the antecedent clause and the unpronounced material encounters.

The present analysis can also be extended to Swiping with the wh-the-hell phrase. Note first that across Germanic languages, Swiping is for the most part possible with ‘simplex’ wh-words and not with which, as illustrated in (75) (Chomsky 1995; Uriagereka 1995; Merchant 2002):

The Swiping example in (75a) is well-formed since it is introduced by a simplex wh-word who, while the one in (75b) is ill-formed since it is introduced by which. The example in (75c) with a minimal pair shows a clear contrast between simplex wh-words and which in terms of the possibility to license Swiping. Our corpus search also yields Swiping examples with the wh-the-hell phrase involving a simplex wh-word as in (76) but no examples with which, supporting the observation made in previous literature:

Swiping could be dealt with by movement and deletion operations, as suggested by Merchant (2002), Hartman & Ai (2009), and Radford & Iwasaki (2015). For instance, Merchant (2002) introduces operations such as pied-piping wh-movement followed by PF-deletion of the remaining clausal material and then an additional head-movement of a wh-word to a preposition. For instance, the Swiping example Mary was talking, but I don’t know who to would be generated by the following processes:

However, the application of such complex syntactic operations must be quite restrictive since a limited set of wh-words and prepositions can participate in Swiping, as illustrated by the following:

As also noted by Merchant (2002) and Culicover & Jackendoff (2005), prepositions such as about, after, as, at, by, for, from, in, near(?), of, on, till, to, and with are possible in Swiping, but not those like above, before, between, despite, during, into, and so forth. The corpus investigation of Kim & Kim (2020) also shows the idiomatic combinations of wh-words and prepositions in Swiping, as shown in Table 2.

Table 2 Frequencies of wh-expression and preposition combinations in Swiping in COCA (from Kim & Kim 2020: 498).

The limit of Swiping with a restricted set of wh-expression and preposition combinations suggests that it is more plausible to assume that English speakers acquire the possible forms of Swiping directly, without reconstructing a derivation from a regular sentential underlying structure, as suggested by Culicover & Jackendoff (2005). This eventually supports the postulation of the Swiping Construction in the grammar of English.

Adopting Ginzburg & Sag (2000), Kim & Kim (2020) define the Swiping Construction as a subtype of slu-int-cl (sluice-interrogative-clause), which is in turn a subtype of hd-frag-cxt, as represented in the following:

The construction specifies that the combination of a wh-expression and a preposition can be projected into a sentential utterance with a special form-function mapping relation in English. The construction is a subtype of Sluicing since it occurs only in Sluicing environments, as can be observed from the contrast between I got a date. Who with? and *Who with did you get a date? The constructional constraint in (79) also indicates that the preposition functions as the sal-utt in the discourse and belongs to the type of strandable (str). This allows us to block Swiping Construction examples with nonstrandable prepositions as in (78) (cf. *What were they complaining during?) and *I know they fell out, but I don’t know what because of (cf. *What did they fall out because of?). Furthermore, in Swiping Construction, only the preposition can bear stress (e.g. Mary is going to the prom, but I’m not sure who WITH/*WHO with) (Merchant 2002; Hartman & Ai 2009; Radford & Iwasaki 2015). The construction constraint thus ensures that in Swiping Construction, only a limited set of wh-expressions (lexical as well as phrasal) can combine with a restricted set of prepositions bearing focus.

For instance, producing the antecedent, I got a date, would evoke a qud asking whom the speaker got a date with in the dgb as shown below:

The uninstantiated PP argument is linked to the NSU Who with? and this NSU is asking a value for this variable (x). According to Kim & Kim (2020), then, the Swiping Construction example Who with? in this context would have the following structure:

The sal-utt information associated with the uninstantiated argument is introduced by context, entering into the qud. The wh-word who combines with the following preposition with as a well-formed instance of the Head-Swiping Construction. The resulting PP is then projected into an S on its own with the desired interpretation as an instance of the Head-Fragment Construction.

With such a justification that English independently employs the Swiping Construction, let us consider how the present analysis can account for Swiping with the wh-the-hell phrase, making use of one attested example:

Adopting the analysis set forth by Kim & Kim (2020), we assume that English introduces the Head-Swiping Construction which licenses the combination of a wh-expression with a preposition in order. Since the present analysis takes a simple wh-word and the wh-the-hell phrase to be identical in bearing the feature [light +], we could expect the wh-the-hell phrase can also occur in the Head-Swiping Construction, combining with a limited set of prepositions.

Now consider the example in (82) again. Uttering the first sentence would activate the uninstantiated second argument in the dgb.

The NSU What the hell about? is asking a value for this variable (x). The present system then would license a structure like the following:

The sal-utt information linked to the unrealized or uninstantiated argument is introduced by context, entering into the qud. The wh-word what first combines with the emotive expression the hell as a well-formed instance of the Wh-the-hell Construction. The construction then combines with the following preposition as a legitimate instance of the Head-Swiping Construction. This resulting expression, also functioning as a wh-the-hell phrase, is projected into an S as an instance of the Head-Fragment Construction. Since Swiping is a subtype of Sluicing, it cannot combine with a head S as in *[What the hell about] [is he grinning]?

This direction also offers a possible account for the behavior of the ANDC in Sluicing, which we noted in the beginning. The claimed contrast in the literature has been that unlike normal wh-phrases, the wh-the-hell phrase only permits Swiping, but not pied-piping Sluicing. Consider similar data again:

The ungrammaticality of pied-piping Sluicing examples in (85b) could be attributed to a phonological constraint, as suggested by Sprouse (2006) and Güneş & Lipták (2021). They argue that these ellipsis phenomena must end with an accent-bearing material but the emotive expression the hell cannot have an accent. This then accounts for examples like (84)–(85), but as noted earlier, corpus data contain a great deal of matrix and embedded Sluicing examples with the wh-the-hell phrase. The present analysis would license the wh-the-hell phrase in Sluicing environments, but could block examples like (85b) by placing an additional prosodic licensing condition on the construction, as do Güneş & Lipták (2021).

5. Conclusion

The ANDC is an independent, idiomatic construction whose syntax as well as semantics/pragmatics overrides a one-to-one form-function relation. In this paper, we first reviewed some key properties of the construction that previous literature has noted. We then explored the real life uses of the construction, making use of authentic corpus data from COCA. The corpus data revealed a variety of interesting facts about it in terms of preferred registers, distribution by wh-words, grammatical functions, matrix/embedded environments, and more diverse uses than previous observations, including their occurrences in elliptical constructions.

To capture the peculiar morphosyntactic and semantic/pragmatic properties of the wh-the-hell phrase, we first postulated two key constructions: Wh-the-hell Construction and ANDC. The former guarantees the formation of idiosyncratic wh-the-hell phrases, and the latter controls its syntactic distribution. In addition, we noted that licensing the wh-the-hell phrase in elliptical constructions such as Sluicing and Swiping depends on the tight interplay of several different grammatical levels such as syntax, semantics, and discourse. In doing so, we discussed that corpus data pose several non-trivial theoretical and empirical problems for the movement and PF-deletion approach that requires a derivational source sentence. We then showed that the Direct Interpretation (DI) approach, making use of enriched discourse information as well as syntactic and semantic information, can account for much wider uses of the construction in question in a systematic way.