In Griffiths and De Vries 2013 (G&dV), we offer an argument in favor of treating appositive relative clauses (ARCs) as syntactically integrated into their hosts, an argument that revolves around the distribution of ARCs in clausal ellipsis environments. In a reply, Ott (2016) counters this specific argument, rejects the more general integration analysis adopted in G&dV on conceptual grounds, and contends that an orphanage analysis of ARCs provides a more parsimonious explanation for the data introduced there. In this rejoinder, we demonstrate that, while Ott presents some relevant data and provides welcome discussion, his specific counterargument does not withstand scrutiny. We also defend the integration approach to ARCs on conceptual and empirical grounds and examine the orphanage analysis of ARCs, arguing that such an approach has conceptual and empirical inadequacies that no integration approach exhibits.
A question that fuels sustained debate in the generative literature on parenthesis is whether or not parentheticals are syntactically connected to their host clauses. All three conceivable stances on this matter—that (a) all parentheticals are syntactically integrated, (b) all parentheticals are syntactic orphans, and (c) some parentheticals are integrated but others are orphans—have been defended.1 One class of parentheticals that has received much recent attention with regard to this issue is appositive relative clauses (ARCs).
In support of a more general integrational view, in Griffiths and De Vries 2013 (henceforth G&dV) we offer an additional specific argument in favor of treating ARCs as syntactically integrated into their hosts. This argument revolves around the distribution of ARCs in clausal ellipsis environments. In a reply to G&dV, Ott (2016) counters this specific argument, rejects the more general integration analysis adopted in G&dV on conceptual grounds, and contends that an orphanage analysis of ARCs provides a more parsimonious explanation for the data introduced there.
In this rejoinder to Ott’s (2016) reply, we demonstrate that, while Ott presents some relevant data and provides welcome discussion, his specific counterargument to G&dV does not withstand scrutiny. More importantly, we also defend the integration approach to ARCs (and to parenthesis more generally) on conceptual and empirical grounds. In particular, we wish to stress that, if one given implementation of the integration analysis exhibits shortcomings, this does not logically imply that parentheticals must be orphans. Indeed, we will demonstrate that recent research on the “syntacticization” of speech acts engenders an integration analysis that does not exhibit the conceptual deficiency that Ott attributes to the G&dV analysis. Finally, we examine the orphanage analysis of ARCs and argue that such an approach has conceptual and empirical inadequacies that no integration approach exhibits.
2 The Narrow Issue
In this section, we address the narrow issue at hand. We demonstrate that Ott’s (2016) counterarguments to the G&dV analysis of ARCs are not warranted. We outline the analysis in section 2.1 and then address Ott’s critique in sections 2.2 and 2.3. In section 3, we return to the broader issue concerning the conceptual and empirical viability of integration and orphanage analyses of parenthesis more generally.
2.1 The Original Argument
The relevant data with which G&dV and Ott’s reply are concerned involve dialogues as in (1), where (1a–c) are alternative answers to A’s question by B. (For previous discussion of comparable data, see also Arnold and Borsley 2008.)
(1) A: Who stole Mary’s car?
B: John, who’s a notorious thief. (who=John)
B: John, which is awful. (which=John stole Mary’s car)
B: *John, which is blue. (which=Mary’s car)
In G&dV, we claim that, when analyzed from an integrationist perspective, the pattern of acceptability observed in (1) provides additional evidence for the idea that ARCs are syntactically connected to their hosts. For current purposes, the details of our integration analysis are irrelevant and it therefore suffices to state that we assume that, all things being equal, ARCs are merged in some way within their anchors’ maximal projection (2).2 (We return to discuss the mechanics of the analysis in section 3.2.)
(2) . . . [XP anchor [ARC. . .]]. . .
According to Merchant (2004), who claims that ellipsis is nonpronunciation of syntactic material, fragment answers are Ā-moved remnants of constituent deletion. Adopting this proposal, in G&dV we claim that utterances such as (1a–b) are acceptable because constituent deletion successfully applies (3a–b) and that utterances such as (1c) are unacceptable because it does not (3c); that is, the string to be deleted (indicated by strikethrough) does not match the actually licensed ellipsis site (enclosed in angled brackets).
2.2 ARCs, Ellipsis, and Extraposition
Ott’s first argument against the G&dV analysis revolves around extraposition. He argues that the analysis is flawed because the ARC in (1c) could occupy an extraposed position above the ellipsis site (4).3 If (4) is a possible representation of (1c), then the G&dV analysis incorrectly predicts that (1c) should be acceptable, as no nonconstituent deletion occurs in (4).
Because ARCs are indeed attested in extraposed positions in nonelliptical environments,4 Ott’s reasoning initially seems compelling. However, we show that on closer inspection it is not.
For Ott’s counterargument to go through, it must be demonstrated that the ARC is not only extraposed, but in fact extraposed to a position outside of a possible ellipsis site. In examples such as (5B), this is evidently not the case. Here, ellipsis targets everything below the CP level (including the auxiliaries), and the string to be deleted is not a constituent; therefore, the structure cannot be derived.
However, there are also cases that require a bit more discussion. We argue that even in a worst-case scenario (involving verb phrase deletion; see below), a potentially extraposed ARC cannot be outside an ellipsis site that contains its anchor. In doing so, we must rely on independent knowledge about ellipsis and extraposition, since we cannot directly see what is going on below the surface in the relevant examples.
Importantly, ARCs and restrictive relative clauses (RRCs) in English and other languages show the same limitations on their linear distribution (see, e.g., Arnold 2007).5 In both cases, the relative clause must occupy a position either linearly adjacent to the anchor/head noun or at the right periphery of the clause that immediately contains the anchor/head noun. In other words, both ARCs and RRCs appear able to occupy extraposed positions, provided that the Right Roof Constraint (Ross 1967:185) is satisfied (compare (6) and (7)).
(6) Appositive relative clauses
[That Jo met Adami yesterday], [whoi’s a dissident], is scandalous.
*[That Jo met Adami yesterday] is scandalous, [whoi’s a dissident].
*[That Jo met Adami yesterday] is, [whoi’s a dissident], scandalous.
(7) Restrictive relative clauses
[That Jo met someonei yesterday] [whoi’s a dissident] is scandalous.
*[That Jo met someonei yesterday] is scandalous [whoi’s a dissident].
*[That Jo met someonei yesterday] is [whoi’s a dissident] scandalous.
Given such distributional similarities, let us see if restrictive relatives can occupy extraposed positions outside of ellipsis sites. Baltin (2006:241) demonstrates that this is not the case. In (8a), we represent the unabbreviated example (which is acceptable) for ease of comparison. Small capitals represent pitch accents. In (8b), the combination of verb phrase ellipsis—indicated by strikethrough—and RRC extraposition is blocked.
Taking the correspondence between RRCs and ARCs seriously,6 we are not surprised that the ARC counterpart to (8b) in (9a) is also unacceptable, as is the question-answer sequence in (9b), where the answer displays verb phrase ellipsis.
In our analysis, this is because an illicit instance of nonconstituent deletion occurs. This is evident if the ARC resides in its regular position directly attached to the anchor (10a), but even extraposition to the vP level (10b) does not solve the problem: the ARC is still within the licensed ellipsis site (the structural complement of did).
Regarding (10b), notice that extraposed modifiers of objects attach within the verbal domain, as is well-known (see, e.g., Baltin 1981, Büring and Hartmann 1997, De Vries 2002).7 By contrast, extraposed modifiers of subjects may attach within the tense domain (Culicover and Rochemont 1990), as is evidenced by, among other things, their ability to escape verb phrase ellipsis; see (11), for instance.
Although not [many people]i would ride with Ad whoi knew just him, somek would
ride with Ad
Furthermore, a comparison of the examples in (12) demonstrates that the attachment site of extraposed subject relative clauses is contained within the site of clausal ellipsis (i.e., within the tense domain). The response in (12b), when uttered with pair-list intonation, shows that a “multiple-fragment” response to (12A) is available.8 The response in (12c) demonstrates that, although a multiple-fragment response is potentially available, unacceptability arises because the subject relative clause does not extrapose to a position above the ellipsis site. This means that such relative clauses must be caught within the clausal ellipse. (Note that the relative clause is necessarily interpreted as restrictive in this fragment answer context.)
Again taking the correspondence between RRCs and ARCs seriously, the unacceptability of (12c) supports the conclusion that (1c), which is repeated in (13B), is unacceptable because an illicit instance of nonconstituent deletion occurs (i.e., where what is to be deleted does not match the licensed ellipsis domain), regardless of whether the ARC is extraposed (14a) or not (14b).
A: Who stole Mary’s car?
B: *John, which is blue.
To summarize, there are strong reasons to reject Ott’s counterargument. Reasonably assuming that ARCs and RRCs extrapose to the same positions in the same contexts, one may conclude from the observation that extraposed RRCs do not escape ellipsis that ARCs do not escape ellipsis either. This is supported by independent work on ellipsis and extraposition. Roughly, verb phrase ellipsis targets vP (see Merchant 2013:80), while clausal ellipsis targets TP (Merchant 2001) or perhaps higher (Thoms 2010). When we combine this with the generalization that an extraposed element attaches to the lowest node that contains its regular surface position (next to the anchor) and the intervening material (see footnote 7 and the references cited in the main text), it follows that ARCs do not occupy a position above the ellipsis site.9 Consequently, the possibility of extraposition does not straightforwardly invalidate the G&dV explanation of the pattern in (1), contrary to Ott’s suggestion.
2.3 The Prominence of the Anchor
In the dialogue in (1), which is repeated in (15), the nominal anchor for who in (15a) is fully pronounced, the clausal anchor for which in (15b) is partially pronounced, and the nominal anchor for which in (15c) is entirely unpronounced. In the early stages of our analysis in G&dV, we form the generalization in (16) from this observation.
(15) A: Who stole Mary’s car?
B: John, who’s a notorious thief.
B: John, which is awful.
B: *John, which is blue.
(16) An ARC can only surface next to an anchor that is at least partially overt.
Ott argues that the generalization in (16) can be subsumed under the generalization in (17). He claims that, once (16) is thus subsumed, our appeal to nonconstituent deletion to capture the unacceptability of utterances such as (15c) is unnecessary.
Appositive relative pronouns anaphorically resume the most prominent featurally compatible antecedent in the preceding discourse.
(modified from Ott 2016:584; to be rejected)
Here, preceding discourse refers to all linguistic material that precedes the ARC, including that portion of the host clause that precedes the ARC. By prominence, Ott means information-structural prominence: the anchor of an ARC must be relatively new/focused/nontopical. In what follows, we show that this is incorrect.
When an ARC does not occupy an extraposed position, it turns out that the actual antecedent is always the linearly closest one—in fact, the adjacent phrase. 10 Manipulating the information-structural configuration of the host utterance does not override this preference. Even when a linearly more distant potential antecedent β displays contrastive (18) or presentational (19) focus while the closest antecedent α does not, α is still the actual antecedent.
A: I heard that the vizier presented a suitor to the princess.
B: No, [the SULTAN]i presented [a suitor]k, whok/*i I’m told is an old man, to her.
A: Who presented a suitor to the princess?
B: [The SULTAN]i presented [a suitor]k, whok/*i I’m told is an old man, to her.
Manipulating the information-structural configuration of the host utterance can influence the choice of antecedent only if an ARC occupies an extraposed position. In (20B), who can be associated with the sultan, a relatively distant but contrastively focused antecedent. In (21B), who can also, but not exclusively, be associated with the sultan, which is presentationally focused.11 In both cases, a suitor, which is discourse-old, remains a possible antecedent.12
A: I heard that the vizier presented a suitor to the princess.
B: No, [the SULTAN]i presented [a suitor]k to her, whoi/k I’m told is an old man.
A: Who presented a suitor to the princess?
B: [The SULTAN]i presented [a suitor]k to her, whoi/k I’m told is an old man.
However, if both candidate anchors are contrastively focused, then who can only be interpreted as coindexed with the linearly closest anchor.
A: I heard that the vizier presented a merchant to the princess.
B: No, [the SULTAN]i presented [a SUITOR]k to her, whok/*i I’m told is an old man.
An interesting case is the following. In (23), the appositive relative pronoun which can be interpreted as coindexed with the immediately preceding clause or with the noun phrase Jane’s autobiography (in which case the ARC is extraposed).
A: That Mary’s autobiography was stolen is worrying.
B: No, [that [JANE’s autobiography]k was stolen]i, whichi/k is awful (by the way), is worrying.
From the observations in (18)–(23) and footnotes 10 and 12, we conclude that the information-structural status of the host clause constituents is irrelevant for determining the antecedent of an ARC. On almost all occasions, the decisive factor is linear proximity. The only partial exception is utterances in which the ARC is extraposed and the host clause contains contrastive or presentational foci and the linearly closest phrase is less prominent. In such utterances, the antecedent is either still the linearly closest one or the closest possible one that bears focus.
(24) In nonextraposed ARCs, appositive relative pronouns anaphorically resume the linearly adjacent (necessarily compatible) anchor in the preceding discourse.
The two generalizations are obviously incompatible: (16) is concerned with whether the anchor is pronounced, while (24) is concerned with the anchor’s linear position. For extraposed ARCs, the situation is slightly more complicated, but, as we have shown, even in those cases linear proximity plays a role, so the conclusion still stands (in fact, the particulars related to extraposition can be ignored for current purposes because extraposed ARCs never escape ellipsis, as was demonstrated in section 2.2).13
Thus, we have shown that information structure plays a negligible role in determining which compatible antecedents ARCs take as their anchors. Consequently, Ott’s suggestion that ARCs must take “antitopical” antecedents (so that nonprominent anchors, including unpronounced ones, are excluded) does not withstand scrutiny.14
A weaker position might still be defended, however. One might argue that the G&dV analysis is superfluous because the generalization in (16) is actually an irreducible rule of grammar or discourse structure that has no deeper explanation. This is unattractive from a Minimalist point of view. But what is more, the generalization in (16) appears to be empirically inadequate, as we make clear in G&dV. (Indeed, its inadequacy was the impetus for the more syntactic analysis we provided there.) Specifically, it fails to capture the observation that (25B) is unacceptable even though part of the anchor a book about Henry V is overt.
A: A book about WHICH English king did you read?
B: *Henry V, which was published in hardback.
Ott claims that sentences such as (25B) are unacceptable because the appositive relative pronoun in (25B) cannot corefer with a partially overt anchor (i.e., a book about Henry V) when a fully overt anchor is also available (i.e., Henry V), semantic plausibility aside. This explanation is untenable, however, because it leads to wrong predictions. In (26B), a fully overt compatible anchor (i.e., a shark) is available. Still, the utterance is ambiguous, as there is also an elliptical clausal antecedent for which. Ott’s suggestion predicts that the appositive relative pronoun cannot corefer with this partially overt clausal anchor a shark bit John, contrary to observation.
To summarize: Ott offers two arguments that specifically target the G&dV analysis of the data in (1). The first is that the analysis fails to account for the possibility that ARCs may occupy extraposed positions (a possibility that purportedly invalidates our analysis), while the second is that the analysis is superfluous, as a simpler explanation is readily available. We have shown that neither of these criticisms is valid. We demonstrated in section 2.2 that extraposition has no bearing on the G&dV analysis, and in section 2.3 that no simpler explanation of the relevant data is available.
3 The Broader Issue
The current debate about how best to capture the distribution of ARCs in elliptical environments touches on the broader issue of how parenthesis should be analyzed more generally: are parentheticals syntactically integrated into their hosts, or are they orphans? In this section, we address this broader issue, returning to additional comments provided by Ott (2016) when relevant.
3.1 Integration vs. Orphanage within the Minimalist Framework
In order to place this section’s discussion on clear conceptual foundations, we begin by explicating precisely what integration and orphanage mean within the Minimalist framework of grammar (Chomsky 1995 et seq.). We do this because no previous theory of orphanage—each of which varies dramatically from the next (e.g., compare Espinal 1991, Haegeman 1991, Burton-Roberts 1999, 2006, and Peterson 1999; see also footnote 1 for references)—is couched in Minimalist terms, and because Ott (2016) neither aligns himself with an outstanding theory of orphanage nor outlines his own.
For Chomsky (1995), linguistic expressions are pairs of instructions (i.e., algorithms) that can be executed by the articulatory-perceptual (AP) and conceptual-intentional (CI) performance systems. These algorithms are the products of two factors: (a) interactions that occur between the ordered computational operations that constitute the “Y-model” of grammar—that is, grammatical operations; and (b) executability constraints imposed by the interface between the derivational procedure described by the Y-model and the AP/CI systems—that is, grammatical constraints.
The products of executing linguistic expressions are utterances. Minimalism stipulates that linguistic expressions cannot contain subroutines that make reference to the products of executing other linguistic expressions. In other words, the grammaticality of an utterance expressed at any particular point in conversational time cannot be contingent upon the properties displayed by an utterance expressed at a preceding point in conversational time. As a result, additional cognitive algorithms that may pertain to language production and/or comprehension that make reference to (subroutines of) two or more linguistic expressions must be treated as extraneous to the Y-model of grammar and its interface with the AP/CI systems. For brevity, we will refer to these extraneous algorithms as discursive instructions and the cognitive operations whose interaction creates these algorithms as discursive operations.
To see a concrete illustration of this division of labor across classes of algorithms, consider the second sentence in (27). Here, the phonological string /lu:si: blʌʃt/ and its rough interpretation blushed′(Lucy′) arise from executing a linguistic expression, while the successful conveyance (or comprehension) of this utterance as referring to the consequence of Bill’s kissing Susie arises from executing discourse instructions, which take conversational context and world knowledge as their input.
(27) Bill kissed Susie. Lucy blushed.
The “integration vs. orphanage” debate is therefore a disagreement about how many linguistic expressions must be executed to generate the entire utterance in (28).
(28) Jill, who is my neighbor, is now my boss.
Orphanage claims that two linguistic expressions must be executed to generate (28): one linguistic expression generates the host clause Jill is now my boss, while another generates who is my neighbor. On this approach, the interpolation of the ARC into the host clause is achieved by executing a discourse instruction. Conversely, integration analyses maintain that only one linguistic expression needs to be executed to generate the entire utterance in (28). On this approach, the linear position of the ARC is determined by grammatical operations (i.e., the Y-model operations Merge and Linearization).
3.2 Accounting for Opacity Effects: Orphanage vs. Integration
Opacity effects provide the principal empirical motivation for adopting an orphanage analysis of ARCs. As is well-documented (e.g., Jackendoff 1977:176, Safir 1986, McCawley 1998, Potts 2005, De Vries 2007), nonlocal c-command dependencies cannot be established across the boundary between an ARC and its host clause, which causes ARCs to appear “opaque.”
The orphanage explanation of opacity effects runs as follows. Syntactic dependencies rely on c-command, which arises from (repeated applications of) syntactic Merge. As a grammatical operation but not a discourse operation, Merge is instrumental in the formation of linguistic expressions, but plays no role in forming discourse instructions. Put another way, Merge recursively concatenates atoms of utterances (i.e., lexical items and collocations thereof) but not atoms of discourse (i.e., utterances). Because the host clause and the ARC in (28) are distinct utterances (as they are generated from distinct linguistic expressions), no element of the host clause and no element of the ARC have merged with each other. Therefore, no element of the host clause c-commands (any element of) the ARC. This precludes nonlocal dependencies and thus creates the opacity effects observed in (29).
Because it introduces no new theoretical machinery, orphanage explains why opacity effects arise in a more parsimonious manner than (almost) all previous integration analyses, including the integration analysis adopted in G&dV. The G&dV analysis invokes Parenthetical Merge (Par-Merge), a syntactic concatenation operation whose output does not “dominate” its input (in a technical sense) and that is triggered only when one of its inputs is the functional head Par (De Vries 2012); in effect, this creates a new c-command domain, since c-command is defined over dominance relations created by regular Merge. The semantic effect of Par can be compared to Potts’s (2005) “comma operator.” In G&dV, Par is used transitively: its complement is an ARC and its specifier is the ARC’s anchor, as illustrated in (30).
(30) [ParP Jill [Par′ Par [ARC who is my neighbor]]]
Because Par and the ARC in (30) are concatenated by Par-Merge, neither the antecedent nor any other constituent in the host clause can c-command into the ARC. This generates the required opacity effects.
As Ott (2016) correctly points out, the G&dV analysis is less parsimonious than the orphanage approach, at least with respect to opacity effects.15 We show in section 3.3 and previous sections that orphanage has serious empirical and theoretical disadvantages in other respects, so the question remains: can we really do without a stipulated concept such as Par-Merge? Notably, it constitutes a possible integrational solution to the opacity problem, but not necessarily the only or best solution. Let us consider what may be a conceptually viable alternative.
Recent developments in the study of the syntax-pragmatics interface engender a contemporary integration analysis that captures the opacity effects exhibited by ARCs (and other parentheticals) in a promising way. To be specific yet succinct, we will adumbrate these developments ([Ds]) in a declarative manner.
[D1] The ascription and determination of illocutionary force is encoded in linguistic expressions (e.g., Speas and Tenny 2003, Haegeman 2014). Assertoricity (for instance) is encoded in the syntactic head Assert (Krifka 2014), to which are relativized all the operators and variables that constitute the propositional denotation that Assert takes as an argument (31) (Koev 2013).
(31) [ForceP Assertp [TP Jillp is nowp my bossp]]
[D2] Operator-variable chains with links that are relativized to distinct illocutionary operators are treated as ill-formed (Koev 2013). In other words, binding variables across distinct speech acts is impossible.16
[D3] A predicate that selects a speech act α must exhibit the same relativization index as α. Put differently, embedded speech acts and the matrix clauses that embed them are indistinct with respect to their illocutionary contribution (32) (Krifka 2001).
(32) [ForceP Assertp [TP Ip thinkp [ForceP Assertp [[to Hollandp]1 [TP Violetp wentpt1 ]]]]]
(33) [ForceP Assertq [TP whoq is my neighborq]]
In themselves, developments [D1–4] do not entail an integration analysis of ARCs, as each one could be independently adopted by an advocate of orphanage. If we do integrate, the result for (28), to take an example, is represented in (34).
(34) [ForceP Assertq [TP [DP Jillp, [ForceP Assertq [TP whoq is my neighborq]]], is nowp my bossp]]
The merger of the antecedent and the ARC makes it possible to (indirectly) determine the required precedence relations. In addition, the analysis accounts for the prosodic isolation of ARCs, as Force phrases are always mapped to intonational phrases (see Güneş 2015, Truckenbrodt 2015, and references therein).
The integration analysis now also straightforwardly accounts for the opacity effects: in unacceptable utterances such as those in (35) (in which irrelevant details are omitted), the links that constitute the long-distance dependency chains are relativized to different illocutionary operators, a process that is independently banned (see [D2]).
In short, orphanage and integration analyses can capture opacity effects with equal efficacy, depending on background assumptions.
Importantly for the current debate, we have demonstrated that a possible conceptual disadvantage of one particular integration analysis of ARCs does not logically invalidate integration altogether, contrary to what Ott (2016) seems to suggest.
3.3 Integration vs. Orphanage
Aside from their approach to opacity effects, the orphanage and integration analyses make different predictions. Integration accounts predict that utterances that contain ARCs may display products of grammatical operations that fit the template in (36), since grammatical operations (but not discourse instructions) contribute to the formation of linguistic expressions. Conversely, orphanage predicts that utterances that contain ARCs display no such products.
input α, β ⇨ output γ
(where α refers to the ARC (or a constituent thereof) and β refers to the host clause (or a constituent thereof))
Evidence that the merger of ARCs fits the schema in (36) can be obtained from Right Node Raising constructions such as (37a–b), where again small capitals represent pitch accents.18 Here, the phrase a little animal is structurally shared between two clauses (see, e.g., Barros and Vicente 2011).
I hate Sam, who WOUNDED, and you hate Sue, who even KILLED, a little animal.
I hate Sam, who WOUNDED, and you hate a girl that even KILLED, a little animal.
According to the general ideas underlying Distributed Morphology, morphophonological processes are also grammatical operations: they apply on the PF branch of the Y-model of grammar (Embick and Noyer 2001:566). Assuming that this is true, the integration approach predicts that products of morphophonological grammatical operations that fit the template in (36) can be observed in utterances that contain ARCs. This prediction is borne out by the examples in (38), which show that Saxon genitives and contracted auxiliaries cliticize to ARCs. Note also that ARCs cannot be bypassed when cliticizing Saxon genitives and contracted auxiliaries, as (39) shows.
*King Alphonso’s, who ruined the party, mother left early.
*Jack’s, who I can’t stand, been given a promotion.
If one maintains that syntactic Merge and morphophonological operations such as cliticization are grammatical operations, the utterances in (38) are incorrectly predicted to be unacceptable on the orphanage analysis. The only plausible means by which orphanage can account for these data is to claim that Merge and cliticization contribute not only to the formation of linguistic expressions, but also to the formation of discourse instructions. Problematically for this scenario, Merge appears not to be utilized in the formation of discursive instructions more generally. If it were, sisterhood relations would occur across sentences. They do not. For instance, Moby-Dick cannot be interpreted as the direct object (i.e., the syntactic sister) of read in (40).
(40) John read. And Mary bought Moby-Dick.
Cliticization appears not to be utilized in the formation of discursive instructions more generally, either. If it were, cliticization would occur across sentences. It does not. For instance, cliticization is prohibited in so-called split utterances, as (41) shows.
B: will be here at two.
B′: *’ll be here at two.
That Merge and cliticization are not utilized in the formation of discourse instructions more generally (and hence are not discourse operations) greatly diminishes the possibility that the multidominant sisterhood relation observed in (37) and the cliticization observed in (38) are the products of discourse operations. This in turn provides opposition to orphanage analyses of ARCs and thus provides indirect support for the integration analysis.
Alongside these empirical problems, orphanage has conceptual disadvantages. In Minimalism, it is implicitly assumed that precedence effects can be established by two means: (a) grammatical operations (i.e., Merge and Linearization)19 and (b) the passage of time (which is not specific to language, of course).20
Harry’s an artist. His work’s good.
an > artist = established by grammatical operations in PF
artist > his = established by the passage of time
Because it claims that ARCs and their hosts constitute distinct linguistic expressions, orphanage cannot appeal to the passage of time or to a grammatical operation such as Linearization to account for the observation that Jill immediately precedes who in (43) (repeated from (28)), where the ARC is interpolated.21
(43) Jill, who is my neighbor, is now my boss.
Instead, orphanage must stipulate that an additional operation—a discourse operation—establishes this precedence relation.22 Moreover, the function of this discourse operation must be well-specified, as it must prohibit utterances from being linearly interlaced in more complex ways.
*Jill – who’s – is a good woman – my boss.
(Intended: Jill, who’s my boss, is a good woman.)
On the other hand, such stipulations are unnecessary on integration analyses, which adhere to the orthodox conjecture that all precedence relations (aside from those created by time’s passage) are established by grammatical operations. Moreover, the interlacing observed in (44), which could only be derived by movement on integration analyses, is straightforwardly excluded by [D2], which precludes the establishment of long-distance dependencies across distinct ForcePs (see section 3.2), as the representation in (45) shows.
Another disadvantage of orphanage is its inability to capture in a conceptually attractive manner the restrictions on which linear positions ARCs may occupy within their hosts. (Recall from section 2 that ARCs either occupy a position linearly adjacent to their anchors, or occupy an extraposed position.) Because integration analyses maintain that ARCs and host clauses are contained within one linguistic expression, it is possible to appeal to grammatical operations and constraints to explain the linear distribution of ARCs. For instance, one may claim that, while ARCs are normally first-merged with their anchors, they may be extraposed, provided that the Right Roof Constraint (Ross 1967) is respected. This explanation aligns ARCs with RRCs, whose positional freedom is restricted in exactly the same manner as that of ARCs (see section 2.2).
Because orphanage claims that ARCs and their host clauses are generated as distinct linguistic expressions, advocates of orphanage cannot appeal to syntactic notions such as coordination, adjunction, extraposition, clausehood, or constituency to explain the linear distribution of ARCs. Instead, orphanage must appeal to “prosodic and pragmatic factors governing the organization of the discourse” (Ott 2016:588). In particular, Ott suggests that the emergence and disappearance of potential questions (Onea 2016) during the flow of conversation plays a crucial role in restraining the otherwise complete positional freedom that ARCs possess. However, we remain skeptical that an adequate pragmatic explanation of the distribution of ARCs that appeals to potential questions can be fashioned. Consider, for instance, the examples in (47) and (48), assuming the slightly simplified pragmatic constraint in (46).
(46) An assertion α can respond to a potential question β only if the propositional explicature that triggered β is still being, or has only just been, articulated.
[That John – and he’s a good friend of mine – has been fired] is upsetting.
[That John has – and he’s a good friend of mine – been fired] is upsetting.
[That John’s been fired] – and he’s a good friend of mine – is upsetting.
*[That John’s been fired] is upsetting – and he’s a good friend mine.
[That John – who’s a good friend of mine – has been fired] is upsetting.
*[That John has – who’s a good friend of mine – been fired] is upsetting.
[That John’s been fired] – who’s a good friend of mine – is upsetting.
*[That John’s been fired] is upsetting – who’s a good friend of mine.
In the examples in (47), the parenthetical assertion and he’s a good friend of mine answers the implicit potential question Who is John?, which is raised when John is articulated. The examples in (47a–b) obey the pragmatic condition in (46) because the parenthetical assertion is uttered somewhere within the propositional explicature that contains John, that is, the bracketed unit in (47). (While the exact position of the parenthetical assertion within this proposition is pragmatically irrelevant, potential positions may be disfavored for prosodic reasons; see Ross 1984.) The example in (47c) obeys the pragmatic condition in (46) because the parenthetical assertion is uttered immediately after the proposition that contains John is articulated. The example in (47d) is thus unacceptable because it violates the condition in (46): the parenthetical assertion is uttered neither within nor immediately after the proposition that contains John is articulated.
On the assumption that the condition in (46) provides a reasonable approximation of how the flow of conversation might potentially restrict the interpolational freedom of alleged orphans, it appears that one must appeal to an additional constraint to explain why (48b) is unacceptable. In other words, the difference in acceptability between (47b) and (48b) is unexpected on an analysis that utilizes only pragmatic and prosodic conditions to constrain the linear interpolation of orphans. This is because the ARCs in (48) and the parenthetical assertions in (47) are alike from both a pragmatic perspective (as both are assertoric responses to the potential question raised by John) and a prosodic one (as both are intonational phrases that are inserted into the same prosodic “niches”).
Consequently, when coupled with the observation that orphanage analyses must appeal to distinct classes of operations to explain the linear restrictions on ARCs and RRCs (discursive vs. grammatical operations, respectively) even though their linear distributions are identical, the observation that pragmatic constraints alone appear unable to capture the linear distribution of ARCs casts further doubt on orphanage’s conceptual plausibility. In contrast, integration analyses can capture the linear distribution of ARCs in a straightforward and familiar manner, by appealing to well-known syntactic constraints (in addition to other factors that may play a role).23
To summarize: We have shown that, from a broader perspective, the integration approach to ARCs is conceptually and empirically superior to the orphanage approach, contrary to Ott’s claim.
In Griffiths and De Vries 2013 (G&dV), we appeal to the impossibility of nonconstituent deletion to explain the behavior of appositive relative clauses (ARCs) in elliptical environments. Ott (2016) presents a four-pronged critique of G&dV. Three of these criticisms target the specifics of the G&dV analysis. These are that (a) the analysis ignores the import of extraposition, (b) a simpler analysis of the data that appeals to “prominence” is available, and (c) the mechanism invoked in G&dV for capturing the opacity effects that ARCs display is conceptually unattractive. We demonstrated in section 2 that (a) and (b) are invalid: extraposition has no bearing on the G&dV analysis and no simpler explanation of the relevant data is available. We showed in section 3.2 that (c), while valid, is not generally true for integration accounts, and we sketched an alternative solution. Ott’s final criticism is that the approach adopted in G&dV, the integration approach to parenthesis, is conceptually and empirically inferior to the orphanage approach to parenthesis. We demonstrated in section 3.3 that, in reality, the converse situation obtains, and that the orphanage approach exhibits a number of conceptual and empirical shortcomings that no integration approach exhibits.
1 In the recent generative literature, among others De Vries (2012) and Griffiths (2015) argue that all parentheticals are integrated; Haegeman (2009), Döring (2014), and Ott (2016) argue that all parentheticals are orphans; and Arnold (2007), Cinque (2008), and Del Gobbo (2017) argue for a mixed approach. See De Vries 2006, Dehé and Kavalova 2007, and Dehé 2017 for more references and literature reviews.
2 In much of the literature on ARCs, an anchor is the host clause constituent with which an appositive relative pronoun/operator corefers.
3 For concreteness, we treat extraposed elements as first-merged to the right, linearly following the intervening material. Nothing hinges on this analysis of extraposition, however.
4 Two random examples are (i) and (ii).
[Bill Wyman]i was phoning me, whoi I’d known in the 1970s.
Did you talk to Billi yesterday, whoi knows Hillary Clinton?
5 We set aside some exceptional cases. Under well-defined conditions that are irrelevant for current purposes, ARCs can precede or interpolate within a clausal anchor; see (i) and (ii).
Their apparently similar, sharply segmented body plan either arose more than once or, whichi is also more than possible, [it is very primitive]i.
?[Ben has—whichi has started to annoy his wife—developed a tendency to whistle in the shower.]i
6 Note that the restrictive interpretation obtained for the relative clauses in (8a–b) is conveyed by their prosodic integration and intersective semantics. As a result, the use of the relative pronoun who in each relative clause is purely coincidental. The same restrictive interpretation is obtained if the relative pronoun is substituted for the complementizer that in each case.
7 Binding effects (using quantifiers, for instance) show that subjects scope over extraposed object-related phrases, and so on. If there is multiple extraposition, a nested pattern is obligatory (mirror effects); see De Vries 2002:chap. 7 and references there.
8 We provisionally adopt a “stacking” analysis of multiple fragments (Merchant 2004:711), as the conclusion reached in the main text (namely, that non-wh constituents cannot escape ellipsis via extraposition) militates against a theory of multiple fragments that treats one remnant as leftward-moved and the other as rightward-moved. (Lasnik (2014:14), who outlines such a theory for multiple sluicing, concedes that his analysis cannot be extended to multiple-fragment-answer constructions, a concession that Ott (2016:583) appears unaware of.)
9 An anonymous reviewer wonders whether a deeper explanation is available for the fact that extraposed phrases cannot escape ellipsis sites. Because this phenomenon is a locality effect, we suggest that phase theory (Chomsky 2001) may offer fruitful insights. For instance, the data discussed in section 2.2 appear to indicate that extraposition is phase-bound. If ellipsis targets entire phases, as Fox and Pesetsky (2005) and Aelbrecht (2012) claim, then a deeper explanation is obtained. Investigating the tenability of such an explanation requires an in-depth discussion of the syntactic nature of ellipsis, however, which cannot be undertaken here.
10 Example (i) shows that no unrelated phrase may intervene between an ARC in a nonextraposed position and its anchor.
Mary (*probably), who has a lot of money, bought a new car.
Example (ii) shows that even if the ϕ-features of the adjacent DP do not match the requirements of the ARC, it is not the case that the linearly more distant, compatible DP can be interpreted as the anchor.
*The teacher[SG] provided the students[PL], who[SG] knows[SG] a lot, with advice.
11 The possibility of interpreting presentationally focused distant candidate anchors as coreferring with appositive relative pronouns in extraposed ARCs varies somewhat in English, as a comparison of (20) and (ia–b) demonstrates (note that, for the second author, the Dutch equivalent of (a) is perfectly acceptable). An explanation for this variation—which is potentially related to the elaborative or continuative nature of the discourse relation between the host clause and the ARC (see Loock 2007)—is beyond the scope of this article. We thank an anonymous reviewer for drawing our attention to the fact that variation is observed in this domain.
A: Who stole Mary’s car?
B: ?JOHNi stole Mary’s car, whoi’s a notorious thief.
B: ?JOHNi did, whoi’s a notorious thief.
12 Let us back this up with a Dutch minimal pair of examples, which shows a very clear contrast. In (i), where the ARC is sentence-medial, only the adjacent minister can figure as the anchor. In (ii), where the ARC is extraposed across the final participle, there is ambiguity: both the closest DP and the focused DP are possible anchors.
De corrupte PRESIDENTk heeft de ministeri, diei/*k gefraudeerd had, ontslagen.
the corrupt president has the minister who committed.fraud had fired
De corrupte PRESIDENTk heeft de ministeri ontslagen, diei/k gefraudeerd had.
the corrupt president has the minister fired who committed.fraud had
‘The corrupt president fired the minister, who committed fraud.’
13 In the example in (i), the relative pronoun refers to two anchors simultaneously. An anonymous reviewer remarks that one of these anchors is (obviously) not linearly adjacent to the relative pronoun.
Kim bought Sandy a booki, and Sam bought her a penk, [whichi+k they gave her for Christmas].
Importantly, however, such “split antecedent” interpretations of appositive relative pronouns are only available if the ARC occupies an extraposed position, as (ii) and (iii) demonstrate. So constructions like those in (i) are not counterexamples to the generalization in (24).
*Maryi gave Johnk, whoi+k are my siblings, a gift.
*Kim heeft voor Sandy een boeki (gekocht) en Sam heeft voor haar een penk, diei+k ze haar met de
Kim has for Sandy a book bought and Sam has for her a pen DEM they her with the Kerst willen geven, gekocht.
Christmas want.to give bought
‘Kim bought Sandy a book, and Sam bought her a pen, which they intend to give her for Christmas.’
14 Ott suggests that relative pronouns behave like d-pronouns in German, which are said to be antitopical (see Bosch and Umbach 2007), and similarly for Dutch (see Van Kampen 2010), we could add. However, an example like (i) shows that the parallel breaks down. Here, the d-pronoun and the personal pronoun in the second sentence preferably take complementary referents (under a regular intonation pattern). Unlike the demonstrative, the relative pronoun (also a d- form) can—and must—refer to deze boef ‘this villain’. More generally, relative clauses can easily be attached to subjects, but demonstrative d-pronouns usually do not refer to a subject, unless it is focused or contrasted.
Janj hoorde dat deze boefb, die*j/b/*v een zwarte pet droeg, zijn vriendv had beroofd. Hijj/b/?v/Die*j/*b/v
Jan heard that this villain who a black cap wore his friend had robbed he/DEM
‘Jan heard that this villain, who wore a black cap, had robbed his friend. He/That man was angry.’
15 For an additional critique of Par-Merge, see Griffiths 2015:219–222. Still, we would not characterize it as a “construction-specific . . . operation” (Ott 2016:588): Par-Merge turns anything into a parenthetical with respect to the syntactic context, and we know that there are a large number of different kinds of parentheticals (see www.let.rug.nl/paracrawler).
16 This derives from Koev’s (2013:205) No Free Variables condition, which states that “all variables in a ForceP are bound within that ForceP.”
17 Aside from [D4], it should be emphasized that the developments listed in the main text are supported by observations unrelated to ARCs. For example, the existence of a syntactic reflex of illocutionary force is supported by research on embedded assertions and questions (de Cuba 2007, Krifka 2001, 2014), while the relativization of semantic content to illocutionary acts provides a straightforward account of deixis- and perspective-shifting phenomena (Koev 2013, 2015).
18 Note that [D2] prohibits scope-related dependencies across parenthetical boundaries but not other grammatical operations, including Merge itself. This important distinction cannot be captured under any orphanage approach.
19 We abstract away from particular proposals about linearization.
20 That is, utterances are necessarily processed sequentially and hence positioned after each other in discursive time.
21 If orphanage appealed simply to the passage of time, one utterance would be placed after the other, resulting in an extraposed position for the ARC.
22Ott (2016:588) rejects this familiar critique against orphanage (citing De Vries 2007)—namely that, given the Y-model of grammar, it is unclear how a parenthesis could be pronounced (in particular: in interpolated positions, we should add) if it were to be added beyond the LF interface—and he calls it a “misunderstanding of the orphanage approach.” As we make explicit in the main text, although the earlier critique can be made more precise, it is in fact entirely valid.
23 Another (indirect) argument for integration emerges from work by Del Gobbo (2003:sec. 3.6.2), who shows that the parallelism between independent sentences and appositives breaks down when a variety of quantified antecedents is taken into consideration.
*[Most/Many students]k, whok were late, came to the party with their parents.
[Most/Many students]k came to the party with their parents. Theyk were late.
Thanks to a reviewer for reminding us of this.
The first author wishes to acknowledge the financial support of the European Commission and the Zukunftskolleg Institute of Advanced Study at the University of Konstanz.