Noun complement clauses as referential modifiers

A number of recent analyses propose that so-called noun complement clauses should be analyzed as a type of relative clause. In this paper, I present a number of complications for any analysis that equates noun complement clauses to relative clauses, and conclude that this type of analysis is on the wrong track. I present cross-linguistic evidence showing that the syntactic behavior of noun complement clauses does not pattern with relative clauses. Patterns of complementizer choice and complementizer drop as well as patterns involving main clause phenomena and extraction differ in the two constructions, which I argue is unexpected under a relative clause analysis that involves operator movement. Instead I present an alternative analysis in which I propose that the referentiality of a noun complement clause is linked to its syntactic behavior. Following recent work, I claim that referential clauses have a syntactically truncated left-periphery, and this truncation can account for the lack of main clause phenomena in noun complement clauses. I argue that the truncation analysis is also able to accommodate complementizer data patterns more easily than relative clause analyses that appeal to operator movement.


Introduction
A number of authors have recently advanced the claim that so-called Noun Complement Clauses (NCCs) are actually Relative Clauses (RCs) in disguise (Kayne 2008;Arsenijević 2009;Haegeman 2012;among others). While each of these authors implements their proposals differently, they all share the general claim that NCCs like (1) should be analyzed as a type of RC construction as in (2).
(1) a. the fact that it is raining b. the claim that the government monitored phone conversations (2) a. the fact that the scientists discussed b. the claim that the inventor made The goal of this paper is to present some empirical challenges to existing NCC as RC proposals and to propose an alternative non-RC analysis for NCCs that avoids some of the pitfalls of the RC analyses. I show that NCCs behave syntactically differently from RCs in a number of cases cross-linguistically, making it unlikely that there is a universal RC structure for NCCs. Patterns of complementizer choice, the availability of complementizer drop and agreement patterns are presented that show that NCCs syntactically pattern more closely with complement clauses to verbs than with RCs, which is an unexpected result if NCCs were indeed a type of RC. Instead, I propose a truncation analysis for NCCs along the lines of de Cuba & Ürögdi (2009), who propose that referential complement clauses are structurally smaller than non-referential complement clauses, as in (3).
(3) a. Referential CP: [CP] b. Non-referential cP: [cP [CP]] The structures in (3) were proposed to account for restrictions on the appearance of the pronominal element azt in Hungarian to non-referential clauses as well as restrictions on Main Clause Phenomena (MCP) to non-referential clauses cross-linguistically. Their proposal is that referential complements lack the cP projection, leaving no position for pronominal azt in Hungarian or the relevant movements involved with MCP in a number of languages. The main proposals I present here are as follows: (A) Embedded clauses come in two sizes, as in (3). (B) NCCs are generally referential CPs, regardless of the head noun they are associated with. In this way they differ from clausal complements to attitude verbs, which can vary between referential and non-referential.
The proposal that NCCs are truncated referential CPs can help explain why NCCs uniformly block MCP like argument fronting, while complements to attitude verbs vary on this point, as illustrated in (4). 1 (4) Examples based on Hooper & Thompson (1973: 486) a. *John regrets that on the wall hangs a picture of Mao. b. John claimed that on the wall hangs a picture of Mao. c. *The regret that on the wall hangs a portrait of Mao weighs heavy on John. d. *The claim that on the wall hangs a portrait of Mao is still unsubstantiated.
I claim that all NCCs are referential because of the nature of the construction itself -in examples like (1), content nouns like fact and claim are co-referential with their associated CP clauses, much in the way that two nominals have been argued to share the same referent in cases of close nominal apposition (see Keizer 2007 for an overview). I argue that this rather simple proposal allows for an analysis that accounts for more data than the RC analysis. The paper is organized as follows. Section 2 briefly summarizes the NCC as RC proposals in Arsenijević (2009) Kayne (2008; and Haegeman (2012). Section 3 first presents the main empirical evidence put forward in the literature in favor of the relative clause analysis and then presents a number of cross-linguistic counterexamples involving morphological markers of relativization appearing in RCs but different morphological markers appearing in NCCs and verb complement clauses (VCCs). I argue that these counterexamples weaken the force of the main empirical evidence that has been presented in favor of a general NCC as RC analysis. Section 4 presents more problematic data for the NCC as RC view from Scandinavian and Basque, where relative markers show up in typical relative operator movement contexts like RCs, clefts and embedded questions, but do not show up in NCCs, which proponents of the NCC as RCC view claim share the same type of relative operator movement. Section 5 shows that RCs and NCCs also pattern differently with regard to when they allow or do not allow complementizer drop, yet another way in which these types of clauses differ. Section 6 adds more data to the discussion, showing that in addition to the complementizer patterns between NCCs and RCs, NCCs also differ from VCCs in extraction patterns and in whether or not they allow Main Clause Phenomena (MCP). These are patterns that previous NCC as RC analyses do not present an account for. In Section 7, I present a truncation proposal to account for the properties of NCCs which does not involve RC-style operator movement. I argue that the referential status of the complement clause is tied to the structural size of the complement clause. I also compare the truncation account to Haegeman's (2012) operator movement account, which unlike Arsenijević's (2009) and Kayne's (2008; proposals, does discuss MCP differences between NCCs and VCCs and provides an account. However, I argue that the truncation account has certain advantages over the operator movement account. I end the section discussing some potential counterexamples from the literature where MCP do appear to occur in some NCCs, and provide some speculation on how to account for these facts. Section 8 concludes the discussion.

Previous NCC as RC analyses
In this section I briefly summarize the NCC as RC analyses proposed in Kayne (2008;, Arsenijević (2009) and Haegeman (2012). While these analyses differ in the amount of details provided and the details of implementation (all three are speculative to some degree), they all share the intuition that all NCCs should be analyzed as variants of relative clause constructions. Kayne (2008; proposes that nouns differ fundamentally from verbs in that they do not take complements. In other words, for Kayne, nouns enter the derivation with no unvalued features, as he states in (5).
If this is the case, then phrases like (6a) must not be noun complement constructions. In order to account for apparent noun complement cases like these, he proposes that they are actually relative clause constructions like (6b). For Kayne, (6a) is derived from (7a) with a silent in, and its object fact relativized, as in (7b).
(6) a. the fact that they're here b. the fact that you mentioned (7) a. they're here in fact b. the fact i that they're here in t i In this way, the fact that there is no obvious gap in NCCs like (6a) is explained: fact is relativized from a silent adjunct PP. Since for Kayne, nouns can never take a complement, all NCCs are assumed to receive the analysis in (7).

Arsenijević (2009)
In a related analysis, Arsenijević (2009) proposes that constructions like (8a) (in which nouns like claim, belief, doubt and question take clausal complements) involve relativization of the contents of the specifier of a Rizzian (1997 et seq.) Force projection. Specifically, the specifier of ForceP contains a syntactic element with a variable feature, as in (8b). The possible values for the variable feature in the force head are [assert], [question] and [imperative]. 2 The nominal head of the NCC also involves a force feature, and in the case of claim it is [assert], as in (9a). Arsenijević proposes that the nominal expression saturates the lambda-abstracted force of the complement clause in the same way the DP head in a relative clause saturates the abstracted reference value of the relativized element of a RC, as in (9b).

Haegeman (2012)
The hypothesis for the potential analysis of NCCs presented in Haegeman (2012) is more programmatic, thus it does not involve a detailed proposal. Haegeman briefly explores the possibility that her intervention analysis of main clause phenomena (MCP) restrictions in conditionals, adverbial clauses and certain clausal complements to verbs could also apply to complements of nouns. For Haegeman (2012) (see also Haegeman & Ürögdi 2010), MCP are blocked in clauses involving operator movement from a TP-internal position (tentatively labeled F(unctional)P) to a CP position as in (10).

CP OP i C… [ FP t i [ TP … ]]]
Haegeman & Ürögdi (2010) propose that in factive clauses, intervention effects due to event operator movement block MCP like topicalization, as in sentences like (11). Nonfactive clauses do not have this event operator movement, so no intervention takes place and topicalization is fine, as in (12).
(12) a. John believes that this book Mary read. b. Mary claims that this book, John read.
c. Hooper & Thompson (1973: 486) *The claim that on the wall hangs a portrait of Mao is still unsubstantiated. Haegeman (2012) cites the RC-type analyses for NCCs in Arsenijević (2009), Kayne (2008) and Nichols (2003) as raising the possibility that a similar intervention analysis to (10) can account for the lack of MCP in NCCs. 4 She posits (Haegeman 2012: 284) that the "nominal character" of NCCs can be captured through operator movement, much in the way that it is captured in Haegeman & Ürögdi's (2010) analysis of factive clauses, which she also sees as having a nominal character (following Aboh 2005).

Summary of previous analyses
What Kayne (2008), Arsenijević (2009) andHaegeman (2012) share is that they all propose RC-type analyses involving operator movement for NCCs. 5 As shown above, each has a different technical formulation for the proposed RCs, as summarized in (14).
(14) a. Kayne (2008) the fact i that they're here in t i b. Arsenijević (2009)  In addition to the differences in technical implementation, there is another important difference between these analyses. Kayne (2008: 27;2010: 216) and Arsenijević (2009) share the intuition that not just NCCs but all declarative complements are RC-type constructions, while for Haegeman (2012), non-factive clauses do not involve relativization. I return to this difference in Sections 6 and 7. 6 If all or some declarative complement constructions are RC-type constructions, we might expect NCCs and VCCs to display morphosyntactic similarities to RCs. Concentrating now on NCCs, we might expect them to behave in a similar manner to garden-variety RCs. In the next section I present complementizer choice data which shows that that in a number of languages, NCCs do not clearly pattern with RCs.

A cross-linguistic look at NCCs and RCs
In this section I first present some of the data that has been used to argue for parallelisms between NCCs and RCs. In these cases, the same morphemes appear in NCCs and RCs across a number of languages, providing evidence that these types of clauses may be structurally similar. However, I then present cross-linguistic data where the opposite is the case: in many languages that have distinct relative and declarative clause markers, the declarative marker, 4 Nichols (2003) proposes a relative clause construction for NCCs involving an "event argument" (e), as in (i).
(i) (Nichols 2003: 157) [ 5 A reviewer worries that by putting these three proposals together, I am creating a straw man argument by both overexaggerating and oversimplifying elements of both the Kayne and the Arsenijević proposals discussed here. I should make clear that both of these proposals are attempts to unify the analysis of complementation structures, and both proposals acknowledge and present analyses attempting to account for many of the surface differences between RCs and other clausal complement structures. Each of these proposals (as well as Haegeman 2012) should of course be independently examined on its own merits. Arsenijević (2009) in particular, as the most fleshed out of the proposals discussed here, proposes an analysis that has the potential to cover many of the phenomena that I discuss below. However, I will argue below that the proposed analysis put forth here covers the data in a simpler manner and covers more data than all three of the previous analyses that I discuss. 6 Haegeman (2012) does not commit to a general FCC as RC analysis, as for her, operator movement does not occur in a number of finite clauses (non-referential/non-factive clauses and peripheral adverbial clauses, for example). She is however open to the possibility of a general NCC as RC account, which she briefly discusses (Haegeman 2012: 272-275).
not the relative marker, shows up in NCCs. I argue that this weakens the main empirical evidence that has been provided in the literature in favor of the NCC as RC analysis, and also makes any universal claim that NCCs are a type of RC more difficult to maintain.

Empirical evidence presented for the NCC as RC view
While Kayne (2008; focuses mostly on data from English, Arsenijević (2009) andHaegeman (2012) present cross-linguistic evidence for a parallel between relative clauses and NCCs: Arsenijević (2009) cites the following data as evidence that NCCs are a type of RC. In each case an element that appears in relative clause constructions also appears in NCCs: the same complementizer in (15), wh-word in (16) and adnominal in (17) Korean (Arsenijević 2009: 46, citing Cha 1998 a. John-i sakwa-lul kkak-un khal John-nom apple-acc peel-adn knife 'the knife with which John peeled an apple' b. John-i sakwa-lul mek-un sasil John-nom apple-acc eat-adn fact 'the fact that John ate an apple' Haegeman (2012: 273, citing Nichols 2003 provides similar examples from Burmese showing that like relative clauses (18a), NCCs contain the relative marker té (18b,c).

(18)
Burmese (Soe 1999, reported  In all of these cases, the fact that the same morphological marker appears in both RCs and NCCs is taken as evidence that they share the same type of structure. However, I show in Section 3.2 that there is also plenty of cross-linguistic evidence that points in the opposite direction.

Empirical evidence against the NCC as RC view
In contrast to the data presented in Section 3.1, in this section I show that there are a number of languages where NCCs do not seem to behave like RCs in regards to the choice of relative vs. declarative clause markers. This evidence raises some questions for the claim that universally all NCCs are a type of RC. At minimum, any universal NCC as RC analysis would need to account for the patterns here, which I argue are unexpected under a RC view.

Scandinavian
Kayne notes in a footnote (2008: 15, fn. 37) that the som vs. att contrast in Scandinavian languages needs to be elucidated. Indeed, as we will see in this section, complementizer data from Scandinavian does not fit as neatly with the relative clause analysis as the data in Section 3.1. Unlike English that, Scandinavian relative complementizers and declarative complementizers have different morphological forms. As seen in (19) and (20), relative clauses in Scandinavian take a relative complementizer (som in Swedish, sem in Icelandic), while declarative complements take a declarative complementizer (att in Swedish, að in Icelandic). 8 Since Kayne (2008; argues that NCCs pattern with relative clauses, one might expect the relative complementizer som/sem to appear, as opposed to att/að. As (21) and (22)  DeclC earth-the shall be round is interesting 'The fact that the earth is round is interesting.'

Basque
These patterns are not limited to the Scandinavian languages. Basque complementizers pattern the same way: In RCs, the relative complementizer -(e)n appears (23), while in declarative complements the complementizer is -(e)la (24). 9 As is the case in Scandinavian languages, the complementizer in noun complement clause structures is the same as in declarative complement clauses, not relative clauses (25) (2003) refer to -(e)n as "the interrogative complementizer", as it appears in a wider distribution than just relative clauses (see the discussion in Section 4 below). I will continue to call it the "relative complementizer" here, but nothing rests on this choice. 10 Urbina (2003: 147) hil dutelako kontua kill aux-DeclC-rpt report 'the report that he has been killed'

Bulgarian
In Bulgarian, the declarative complementizer če appears in declarative clauses (26a) while the relative complementizer deto appears in RCs (26b). As with Scandinavian and Basque, Bulgarian NCCs feature the declarative complementizer, not the relative complementizer (27) Krapova (2010Krapova ( : 1266 classifies the set of predicates that can take deto complements as "[…] a subset of 'true' factives, including emotives". In fact, Haegeman (2012) cites Krapova's (2010) data and analysis as support for her operator movement account of MCP restrictions in factive clauses. 11 Krapova (2010) analyzes these declarative deto clauses as "hidden" RCs headed by the pronoun tova, as in (29) (27a) and (27c), this is not the case. Here we have a language that actually allows the relative complementizer to appear in a set of some complements to verbs, yet it uniformly fails to appear in any NCCs.

(33)
Durban Zulu (Halpert 2012 As we saw in (32) and (33), the relative complementizer does not show up in NCCs and NCCs display a different agreement pattern from RCs. Both of these facts suggest that the syntax of NCCs differs significantly from the syntax of RCs, something that we might not expect if they share the same type of structure. In addition to coming up with a story for why the relative complementizer does not show up in NCCs, proponents of the NCC as RC analysis would also need to explain why agreement patterns differ between NCCs and RCs.

Finnish and Hindi
Finally, Finnish and Hindi display similar behavior to the other languages in this section when it comes to complementizer choice. In Finnish, joka is the relative complementizer and että is the declarative complementizer. We find joka in RCs (34) and että in declarative complements (35). 16  In Hindi, jo is the relative complementizer and ki is the declarative complementizer. We find jo in RCs (37) and ki in declarative complements (38). Declarative ki appears in NCCs (39).

Summary
In this section we have seen that across a number of unrelated languages, the morphological form marking a relative clause does not match the morphological form marking NCCs. This distribution is not as clean a fit for the general RC analysis for NCCs, as NCCs in Swedish, Icelandic, Basque, Bulgarian, Durban Zulu, Hindi and Finnish and all take the declarative complement clause marker, as illustrated in Table 1. This poses some complications for analyses that argue that NCCs are always RCs, since in a number of unrelated languages NCCs do not behave like RCs in complementizer choice. If they were indeed RCs, the fact that the relative complementizer is not selected in NCCs these languages would need further explanation from the proponents of the NCC as RC account.
We also saw that in Bulgarian, a language that sometimes allows the relative clause marker deto to show up in certain declarative complements to verbs, the relative marker still never shows up in NCCs. In addition, in Durban Zulu NCCs and RCs do not share the same agreement patterns. Thus, the general picture we get from the data presented in this section is that in some languages RCs and NCCs share similar morphosyntactic properties (Section 3.1), while in other languages they do not (Section 3.2). While none of these facts rule out the NCC as RC account, they do suggest that there are significant 17 Except of course for the factive cases from Krapova (2010) discussed above where deto appears. Crucially for the present discussion though, deto never appears as an NCC marker.

Complementizer choice and left-peripheral operators
As we saw in Section 3.2, in Swedish the relative complementizer som appears in RCs and the declarative complementizer att appears in declarative complements and NCCs. The same situation is found in Basque, where the relative complementizer -(e)n appears in RCs and the declarative complementizer -(e)la appears in declarative complements and NCCs. In this section I show that in both of these languages, relative complementizers, not declarative complementizers, also appear in clefts and embedded questions, constructions that are typically associated with operator movement to the left-periphery. Following the work of Vikner (1991), Stroh-Wollin (2002) and Franco & Boef (2015) for Scandinavian, and Artiagoitia & Elordieta (2016) for Basque, I assume the generalization that complementizer choice in these languages correlates directly with the presence or absence of an operator in the left-periphery: the relative complementizer shows up when an operator is present and the declarative complementizer appears in the absence of an operator. Given this generalization, the fact that the relative complementizer does not show up in NCCs in these languages casts some doubt on the claim that there is operator movement involved in these constructions.

Relative complementizers and operator movement in Swedish
In Swedish, the relative complementizer som can appear in embedded questions, as in (40). Som also appears in cleft sentences, as in (41).
(40) Swedish a. Johan undrade vem som/*att hade öppnat dörren. Johan wondered who RelC/DeclC had opened door-the 'Johan wondered who has opened the door.' b. Lisa vet vem som/*att köpte kakor. Lisa knows who RelC/DeclC bought cookies 'Lisa knows who bought cookies.' 18 For more ways that NCCs differ syntactically from restricted RCs, see Krapova & Cinque (2015), who instead analyze NCCs as reduced non-restrictive RCs, as in (i): Their non-restricted RC analysis actually correctly predicts that the relative complementizer deto should not appear in NCCs (c.f. (25)), since the complementizer that appears in these constructions heads a declarative phrase. However, the proposal that NCCs are non-restricted relative clauses is not without its problems. Platzack (2000) notes that non-restrictive relative clauses differ from restrictive relative clauses in that only non-restrictive relative clauses allow speaker-oriented adverbs like "by the way" or "incidentally", as in (ii).
(ii) a. The man who (*by the way) arrived yesterday (*incidentally) is tired. b. The man, who (by the way) arrived yesterday (incidentally), is tired.
These adverbs can appear in the full restricted relative forms like (i), but are degraded in NCCs. The relevant reading for the examples in (iii) involves the speaker-oriented adverbs modifying the embedded clause, not the main clause.
(iii) a. John spread a rumor that Bill (*incidentally) was in town (*by the way). b. Jason came to the conclusion (*by the way) that Gwen ate the cupcake (*incidentally). Depending on one's analysis of relative clauses, embedded questions and clefts, it is certainly plausible to conclude that som appears when some sort of relative operator resides or moves into the specifier of CP (relativized elements in RC constructions, wh-words in embedded questions and the clefted phrase in clefts). In fact, this analysis is provided by Vikner (1991: 120-123), who assumes that it is a lexical property of som in Danish that it requires an empty relative operator in its specifier, Franco & Boef (2015), who argue that som in Swedish and Norwegian, "lexicalizes a C head that bears an OP feature: [OP]." (Franco & Boef 2015: 67), and Stroh-Wollin (2002: 301-305), who describes som in Swedish as appearing in various clauses when there is some visible or invisible constituent in Spec,CP (with these pre-complementizer constituents appearing in relative clauses, clefts and embedded wh-clauses). 19 The embedded wh-questions in (40) clearly each have a displaced wh-operator (vem) directly to the left of the complementizer som and a gap in subject position. Clefts are often analyzed as being a type of restrictive relative clause with a relative operator in CP, as in Reeve's (2007) proposal, shown in (42). 20 In (41), the clefted constituents (min idé and han respectively) are also directly to the left of the complementizer.
(42) Reeve (2007: 158) On the other hand, att appears in declarative sentences where no operator appears in the specifier of CP. If NCCs in Swedish were indeed RC constructions with operator movement to the left-periphery, we might expect them to behave in the same manner as other types of constructions with operator movement and have the relative complementizer som. However, as shown in (21), repeated here in (43), this is not the case. 19 Note that there is some variation among the Scandinavian languages when it comes to the appearance of som/sem in embedded questions. The Swedish pattern is the same in Norwegian, as shown in (i).
(i) Norwegian (Thráinsson 2007: 448) Han spurte hvilken buss som/*Ø gikk til sentrum. he asked which bus RelC/Ø went to centre-the 'He asked which bus went to the centre of town.' Holmberg & Platzack (1995: 73) also report that in several Norwegian dialects, som can also appear in matrix questions.

(ii) Vad som ikke er sandt?
what that not is true 'What is not true?' The facts in Danish and Icelandic differ from Swedish and Norwegian. In Icelandic, the relative complementizer sem cannot appear in these constructions. As shown in (ii), the complementizer must be null.
(iii) Icelandic (Thráinsson 2007: 449) Ég veit ekki hver *sem/Ø kemur. I know not who RelC/Ø comes 'I don't know who will come.' In Danish, som cannot appear when the relative operator is overt. For details on Danish see Vikner (1991), for details on Icelandic see Thráinsson (2007) and for an overview of the Scandinavian languages see Stroh-Wollin (2002) and Thráinsson (2007). 20 For arguments that clefts are a type of restrictive relative clause, see Reeve (2007: 160-161 More evidence that these clauses do not involve relative operator movement comes from a NCC in Swedish that actually does take som instead of att, namely an interrogative complement of the noun "question". In (44), we have a clear case of an operator moving overtly to the left periphery, and in this case the relative complementizer som appears. If indeed we had operator movement in all NCCs, we might expect som in all cases, counter to fact. (44) Swedish ( is-answered later 'The question of who has won it is answered later.'

Relative complementizers and operator movement in Basque
Basque follows the same pattern as Swedish in regards to complementizer choice in relative clauses, embedded questions, and clefts. Embedded questions take the relative complementizer as opposed to the declarative complementizer (45), and the same goes for clefts (46). As in Swedish, Basque NCCs take the declarative complementizer (47).
(45) Basque a. de Rijk (2008: 442) Ez dakit ezagutzen duzun. not know know aux-RelC 'I don't know whether you know him.' b. de Rijk (2008: 448) Hazaeiek galdetu zion ea zertako negar egiten zuen. aff-is-DeclC-rpt proof one 'a proof that God exists' b. de Rijk (2008: 462) adiskide garelako ezaugarritzat friend are-DeclC-rpt sign 'as a sign that we are friends' c. Hualde & Ortiz de Urbina (2003: 147) hil dutelako kontua kill aux-DeclC-rpt report 'the report that he has been killed' I take this pattern as evidence that complementizer choice correlates with the presence or absence of a relative operator in the left-periphery in Basque, and thus that there is no evidence for relative operator movement in Basque NCCs. 21 This analysis is in line with Artiagoitia & Elordieta (2016), who correlate the -(e)n complementizer in Basque with the presence of a left-peripheral operator. They state, "[…]-en generally spells out a complementizer that agrees with an operator in its specifier; this operator may be a wh-phrase in the case of indirect questions and wh-exclamatives, or a null (wh) operator in the case of yes/no indirect questions, relatives and comparatives." (Artiagoitia & Elordieta 2016: 396-397). Finally, the Basque example in (48) behaves the same as the Swedish example in (44).

(48)
Basque (Hualde & Ortiz de Urbina 2003: 524) [Nor ettori ez den galderari] ez diot erantzunik emango. who come not aux-RelC question not aux answer-prt give-fut 'I won't provide any answer to the question of who didn't come.' 21 A reviewer objects to my taking the lack of the presence of the relative complementizer in NCCs as evidence that there is no operator movement in these constructions. The reviewer states, "the entire model involving operator movement is a theoretical construct, a model, which has empirical predictions, but cannot be immediately empirically measured or identified […] no clear argument is provided that operator movement does not take place in NCCs." I would certainly agree that I have not presented any definitive evidence here against the possibility of there being operator movement in NCCs. Much like trying to disprove the existence of God, it is impossible to disprove the existence of a covert operator in NCCs. I am relying on the previous literature (Vikner 1991 , where it has been independently argued that there is a correlation between the presence or absence of operators in the left-periphery and the choice of complementizer, and using this indirect evidence to suggest (not prove) that operators are not involved in NCCs. With that said, it is impossible to rule out that a very special type of relative operator exists that, unlike other relative operators, syntactically leaves no gap and has no effect on the form or presence of complementizers. However, the analysis I present below also makes the correct predictions for the complementizer data in Swedish and Basque complement clauses (as well as main clause phenomena data discussed in Section 6) without any special stipulations, while an operator movement account would need extra machinery to get the same results. On this last point, the reviewer goes on to state, "Another model may be proposed which does not postulate operators, or movement, or operator movement in the first place, or which does, but does not use them to derive any clausal constituent. And such an account might as well be simpler and with a broader empirical coverage than the operator movement account. The author's argument, however, accepts the adequacy of the operator movement model for relative clauses, but rejects it for NCCs. There is a clear reductionist advantage of the approach which assumes a general operator movement in subordinate clauses, as it says: clausal subordination is the surface effect of the modeling notion of operator movement and operator movement may be joined by different structural peculiarities to derive different types of subordinate clauses." Thus, the reviewer argues that by accepting operator movement in one type of sentential complementation and not another, that, "This renders subordination a non-uniform phenomenon, which makes the theory more complex." I am sympathetic to the theory that there are operators involved in relative clauses, mainly because there is empirical evidence for this in the form of wh-words appearing overtly in the left periphery in many cases presented in the literature, as well as gaps left behind in the positions where many of these operators are postulated to move from. On the other hand, I have yet to see any convincing evidence (direct or indirect) presented in the literature in favor of operator movement in NCCs. The question of the relative complexities of competing theories is an interesting one, but I will leave a comparison of the relative complexity of each of the proposals discussed here to the reader. I will say that if subordinate clauses vary as to whether or not they contain operator movement, then this brings them in line with matrix clauses, where it is widely believed that some matrix clauses have operator movement (say wh-questions) and others don't (say simple declarative sentences).

Complementizer drop
In this section I examine how NCC as RC analyses handle complementizer drop (C-drop) facts, and argue that it is not clear how proposal in Kayne (2008; would adequately explain the different C-drop patterns in these constructions. Arsenijević (2009) does provide a story for the patterns of complementizer drop in English, but I argue that the nature of his proposal forces him into a rather complex technical solution for the different patterns in NCCs and RCs, and leaves questions on C-drop patterns in Verb Complement Clauses (VCCs) (I leave an alternative analysis of C-drop until the main proposal in Section 7). 22 As was the case in previous sections, NCCs display different behavior than RCs with operator movement.

Complementizer drop in English RCs
It's a well-known fact that relative complementizers can be optionally dropped under certain conditions in English. Specifically, the complementizer is obligatory in subject relative clauses (49a), but optional in non-subject relative clauses (49b-c).
(49) a. I saw the man *(that) ___ ate the pizza. b. I saw the pizza (that) the man ate ___. c. I like the way (that) they solved the problem ___.
I leave aside a full analysis of this pattern of C-drop and simply use it here as a diagnostic for RC constructions. Since this is the general pattern of C-drop for RCs in English, if NCCs were indeed RCs, we might expect NCCs to behave in the same way with respect to C-drop.

Complementizer drop and Kayne (2008; 2010)
Recall from Section 2.1 that in Kayne's (2008; proposal, fact is a relativized complement to a silent in, as in (7), repeated here as (50).
(50) a. the fact that they're here b. the fact i that they're here in t i Kayne proposes that (50a) derives from (51a). In addition, Kayne (2008: 14, fn. 35) proposes that other possible placements for in fact, as in (51b-c).
(51) a. They're here, in fact. b. They're in fact here. c. In fact, they're here.
Wherever in fact originates, it is clear that it is not subject position: the subject is occupied by they in all of the sentences in (51). Given that Kayne argues that phrases like (50a) are relative clauses, we would expect them to behave in the same manner with respect to C-drop. However, the data in (52) do not bear this out, as the complementizer is obligatory.
(52) a. The fact *(that) John is here infuriates Mary. b. I resent the fact *(that) Mary left.
In other words, RCs allow C-drop with non-subject relatives, but NCCs do not. This is unexpected under Kayne's analysis, as what is being relativized is 'in fact' a non-subject. 23 It is also unexpected for Haegeman's (2012) proposal, where the relativized operator originates in a functional projection in the TP-field and moves to CP. Neither of these relativized items are subjects, so we would expect C-drop to be licit, counter to fact (52).

Arsenijević (2009) on complementizer drop
Arsenijević (2009) does take up the issue of C-drop, and he provides a speculative analysis. As he notes, in addition to RCs (53a), C-drop is also possible in verbal complement clauses (VCCs), as in (53b), but not NCCs (53c).
(53) (Arsenijević 2009: 43) a. John saw the cup (that) Mary bought yesterday. b. John said (that) Mary was ill. c. The fact ??(that) Mary was ill surprised them all.
In his discussion, Arsenijević refers to NCCs and VCCs both as FCCs (finite complement clauses), a point I will return to below in Section 6. In his system, both VCCs and NCCs are types of RC, with the contents of the Spec,ForceP relativized. For the availability of C-drop in VCCs, Arsenijević appeals to a version of Bošković & Lasnik's (2003) C-drop 23 There are other issues in Kayne's proposal, two of which I note here. The first has to do with the interpretation of in fact vs. the fact. To my ear, when we use in fact, there is an assumption that the proposition being commented on is not presupposed. To me, it's much like by the way. These modifiers are degraded in the fact that clauses (i) and relative clauses in general (ii). Note that the relevant reading is with in fact and by the way modifying the embedded clause, not the main clause. (iii) the rumor/news/story/reason that Elvis left the building (iv) a. They're here, in fact.
b. They're in fact here. c. In fact, they're here. If these are relative clause constructions analogous to the fact that constructions, we might expect derive from sentences like the ones in (v).
(v) a. *Elvis has left the building, in rumor.
b. *Elvis in story has left the building. c. *Elvis has in news left the building. d. *In reason, Elvis left the building. Kayne's (2008) analysis relies on sentences like those in (v), which are ungrammatical on their own, to be the source of the RC constructions.
analysis where that is argued to undergo affix-hopping to a position adjacent to the verb. In Arsenijević's analysis, incorporation of the nominal head into the verb (and the absence of any other intervening material) is a precondition for affix hopping, and thus that-deletion.  24 Arsenijević proposes that RCs can undergo affix hopping to a noun while NCCs cannot and that this is due to their differing feature matrixes.

Problems for Arsenijević (2009)
There are a few potential problems for Arsenijević's that-deletion analysis. First, the combination of affix-hopping for that-deletion along with the proposal that there are different feature makeups for that in NCCs and RCs that allow or disallow affix hopping is rather stipulative. Arsenijević is forced into this move as his claim is that NCCs and RCs are variants of the same construction, and thus one would expect the complementizer in both constructions to behave the same.
A second potential problem for Arsenijević's (2009) C-drop analysis is that the analysis glosses over important differences within the class of what he refers to generally as finite complement clauses (FCCs). Within this class there are subclasses that display different behavior from each other. In the domain of VCCs, non-factive complements generally allow C-drop, while emotive factives complements (55) and manner-of-speaking complements (56) are much more resistant to it.
(55) a. John said (that) Mary was ill. b. John regretted *(that) Mary was ill. c. John resented *(that) Mary was ill. d. John hated *(that) Mary was ill.
(56) a. John said (that) Mary was ill. b. John whispered *(that) Mary was ill. c. John groaned *(that) Mary was ill. d. John shouted *(that) Mary was ill.
In the case of emotive factives and manner-of-speaking verbs it is not clear why C-drop should not happen, given Arsenijević's analysis of FCCs and the fact that we could plausibly provide the same type of analysis proposed in (54) to (55b) or (56c), as illustrated in (57). Ruling out C-drop in (55b-d) and (56b-d) in Arsenijević's system would require adding more machinery to an already complicated system. I take up a third potential problem for Arsenijević's system with regards to main clause phenomena and extraction next in Section 6.

Summary
In Sections 3 and 4 we saw that the distribution of relative clause markers vs. declarative clause makers in a number of languages does not provide clear support for a RC analysis of NCCs. In this section we have seen that NCCs in English do not have the same properties of C-drop as RCs. This data provides yet another case of NCCs and RCs diverging in their syntactic behavior. These facts suggest at minimum that equating NCCs to RCs creates the need for a number of non-trivial explanations for the empirical differences between these constructions. This leaves the door open for an alternative analysis of the NCC facts presented thus far, which I will discuss in Section 7. Before this though, I widen the scope of inquiry to main clause phenomena and extraction in NCCs, VCCs and RCs.

Two accounts for main clause phenomena and extraction in clauses
In this section I explore at the availability of main clause phenomena (MCP) and extraction in NCCs, RCs and verb complement constructions (VCCs). The proposals in Kayne (2008; and Arsenijević (2009)

Main clause phenomena and extraction
Returning to the discussion of Arsenijević's (2009)   (ii) (Chomsky 1986: 31) a. Who did they leave before speaking to? b. To whom did they leave before speaking?
I have nothing interesting to say about this contrast, but in terms of the present analysis, in Section 7 I analyze NCCs as adjuncts, so the fact that the same pattern holds in (i) and (ii) fits in with the general picture. 26 Kayne (2008; does not address extraction or MCP restrictions. Arsenijević (2009: 45) briefly discusses extraction from FCCs, claiming that extraction out of FCCs is normally possible, but is blocked in NCCs when there the head noun (claim, fact, etc.) does not incorporate, as in the case of NCCs, or when a demonstrative is present. He does not address MCP restrictions.
the factive VCC vs. non-factive VCC asymmetries. I discuss these analyses in the Sections 6.2 and 6.3 and eventually conclude in Section 7.5 that the operator movement account, while accounting for the factive vs. non-factive VCC extraction pattern, has more of a challenge handling NCCs than the truncation account does.

Operator movement: Haegeman & Ürögdi (2010)
Haegeman & Ürögdi (2010) do not take up NCCs, but they do provide an analysis for the VCC data in (60-63). They claim that VCCs that resist MCP and are weak islands for extraction are referential. They propose that only factive VCCs (referential clauses in their system, following de Cuba & Ürögdi 2009) have event operator movement, and this operator movement causes intervention effects that block MCP and create weak islands for extraction. Since non-factive VCCs (non-referential clauses in their system) do not have event operator movement, there are no intervention effects and thus no restrictions on MCP or extraction. 27 Haegeman & Ürögdi (2010) propose that the event operator is a null wh-operator which moves from a TP-related position to a position in the left periphery. Their structure is repeated here in (64). Haegeman & Ürögdi claim that referentiality is tied to this event operator movement, and that this movement causes intervention effects that block extraction and MCP movements. Following de Cuba & Ürögdi (2009), they define referential and non-referential CPs as follows.
(65) (Haegeman & Ürögdi 2010: 137; definitions to be revised below) a. Referential CP: a referential entity that denotes a proposition without illocutionary force (a sentence radical in the sense of Krifka 1999); a semantic object encoding a proposition/question which the complex sentence (the embedding context) positions in the dynamics of conversation. As such, an RCP in itself does not constitute a speech act and cannot be used as an utterance. RCPs can be embedded under both factives and non-factives.
b. Non-referential CP: a non-referential semantic object denoting a speech act with illocutionary force, i.e., one which involves a conversational move. An NCP can thus be a matrix sentence, or an embedded clause subject to various restrictions. Factive verbs cannot embed speech acts due to conflicting semantic requirements.

Truncation: de Cuba & Ürögdi (2009) and the motivation for truncation
Although their analysis also claims that referentiality (which they define similarly) is the important distinction, de Cuba & Ürögdi (2009)  McCloskey proposes that factive verbs select a single CP structure, while wonder-type predicates select a recursive CP. 28 McCloskey claims that the availability of the complex structure under a predicate like wonder (and its unavailability under a predicate like find out) derives from the fact that the complement of a question predicate like "wonder" is a different semantic object from the complement of a resolutive predicate like "find out", albeit both are realized as embedded questions. For the particulars of the analysis, I refer the reader to McCloskey's paper. What is important for us here is that semantic complexity corresponds to syntactic complexity on this analysis. As McCloskey notes, his proposal is reminiscent of the CP-recursion analysis of embedded verb-second (EV2) constructions in Scandinavian languages (Iatridou & Kroch 1992;Watanabe 1992;Holmberg & Platzack 1995;Vikner 1995; among others; see Heycock 2006 for a summary). In Mainland Scandinavian, clausal complements of 'bridge verbs' can optionally exhibit verb-second (V2) word order (indicated by the post-verbal position of negation), which is not generally allowed in an embedded clause. Typically, EV2 order is impossible under a factive (68b), but available under a non-factive (69b). 29 The CP-recursion analysis allows for verb-movement to (the lower) C in the presence of an overt complementizer (69b), but since factives do not license CP-recursion, (68b) is ruled out. As with McCloskey's account, the CP-recursion analysis postulates a more complex syntactic structure associated with non-factives as opposed to factives. 31 Haegeman (2006) also argues for a more articulated CP structure under non-factives. In a discussion focusing primarily on adverbial clauses, she adopts (and adapts) a Rizzi (1997) style CP-field, with "peripheral adverbial clauses" and non-factive complement clauses having a full left periphery (like root clauses), and "central adverbial clauses" and factive complements having an impoverished left periphery. 32 This structural difference is exploited to account for the fact that peripheral adverbial clauses allow Main Clause Phenomena (MCP) such as topicalization and speaker oriented adverb placement, while central adverbial clauses do not; the positions designated for these phenomena are present in (70a) and missing in (70b). Haegeman then speculates that factive complements, like central adverbial clauses, are structurally impoverished. She cites data from Hooper & Thompson (1973) and Maki et al. (1999), showing that factives are also resistant to MCP like topicalization.
(71) a. Maki et al. (1999: 3), their (2c) *John regrets that this book Mary read. b. Hooper and Thompson (1973: 479), their (109) *I resent the fact that each part he had to examine carefully.  (69)) is influenced by information structural concerns. She states, "Without exception, native-speakers who were asked to assess the acceptability and significance of MSc embedded V2 clauses responded by making reference to considerations of informational salience." (Biberauer 2002: 46). She reports that EV2 clauses emerge when the speaker wants to express "a strong assertion". She reports the same pattern for Modern Spoken Afrikaans (MSA) EV2. Biberauer's claim that "considerations of heavy informational salience" condition the acceptability of EV2 in MSc and MSA support the claim that semantic complexity corresponds to syntactic complexity in these cases. 32 Note that Haegeman (2006) replaces "Force" with "Speaker Deixis". 33 Note that Bentzen et al. (2007) do not characterize the semantic differences in predicate types as factive vs. non-factive. Instead, they appeal to a Hooper & Thompson (1973) division between clauses selected by assertive and semifactive predicates (Class A, B and E for Hooper & Thompson), which have the structure in (70a), and clauses selected by non-assertive and factive predicates (Class C and D for Hooper & Thompson), which have the structure in (70b). (2009) argue that Hungarian sentential embedding constructions also exhibit different syntax depending on the information structure of the complex sentence, but this time with the availability of an extra morphological element. They present a robust pattern that surfaces under non-factives and one under factives. In a neutral sentence (i.e. a sentence without contrastive focus or negation), non-factive verbs feature azt in the preverbal position, while factive verbs are not possible with azt.

De Cuba & Ürögdi
(72) Hungarian a. Péter (*azt) sajnálja hogy havazik. Peter Dem-acc regrets comp snows 'Peter is sorry that it's snowing.' b. Péter azt mondta (hogy) havazik. Peter Dem-acc said comp snows 'Peter said that it's snowing.' Following the intuitions of the other truncation accounts discussed above, they propose that the unavailability of azt in neutral factive complements is a result of truncation -the position that houses azt is unavailable in truncated clauses. 34 They also argue that MCP restrictions and extraction facts can be captured by appealing to the presence or absence of the cP structure in (66): the extra structural position is argued to provide a landing site for MCP, and the referential character of referential CPs disallows non-specific wh-phrases from extracting through a referential CP (accounting for weak island effects in referential complements).
The key difference between the analyses in de Cuba & Ürögdi (2009)  To sum up, there is a large body of work that ties clausal truncation to a semantic/pragmatic contrast in embedded clauses (Holmberg 1986;Platzack 1986;Suñer 1991;Iatridou & Kroch 1992;Watanabe 1992;Holmberg & Platzack 1995;Vikner 1995;de Villiers 1999;Krifka 1999;Grewendorf 2002;Lahiri 2002;Benincá, & Poletto 2004;Haegeman 2006;McCloskey 2006;Bentzen et al. 2007;de Cuba 2007;Basse 2008;de Cuba & Ürögdi 2009;Demonte & Fernández-Soriano 2009;de Cuba & MacDonald 2013; among others; for the view that non-factive type clauses are the truncated ones, see Kiparky & Kiparsky 1971 and a host of papers that followed it). The present analysis will follow in this tradition of tying semantic/pragmatic differences to syntactic differences between embedded clause types. Following de Cuba & Ürögdi (2009), I claim that the semantic/ pragmatic difference here is in the referentiality of the clause. I modify the definitions for referentiality in the next section.

A modified definition for referentiality
A reviewer points out that there is a potential problem with the definitions for referential and non-referential CPs provided in de Cuba & Ürögdi (2009) and above in (65) from Haegeman & Ürögdi (2010). In both sets of definitions, ±referentiality can be seen as being tied to ±factivity. One potential problem with this characterization lies in behavior differences between complements to "semifactives" and "true factives". Unlike true factives, semifactives can lose their factivity in questions, if embedded in the antecedent of a conditional, and under certain modals (Karttunen 1971). As discussed by Hooper and Thompson (1973), semifactive complements (their class E predicates) pattern with assertions (classes A & B) in allowing embedded main clause phenomena (MCP). True factives (class D) do not allow MCP. This has also been demonstrated more recently for other languages. For example, Wiklund et al. (2009) show that in Scandinavian languages embedded V2 (both non-subject topicalization and V-Neg word order) is possible in Hooper and Thompson's strong assertions (Class A), weak assertions (class B) and semifactives (class E), but disallowed in non-assertions (class C) and true factives (class D). Most relevant to the present discussion, semifactive predicates are demonstrated to be factive and presupposed even when they display embedded MCP, as shown by Wiklund et al. (2009) (73). (73a) shows a clause with non-V2 word order embedded under semifactive discover and (73b) shows a V2 clause (non-subject topicalization) under the same verb. Both sentences presuppose the truth of (73c). This presupposition remains under negation (73d), even with a V2 embedded clause (73e).

(73)
Swedish (Wiklund et al. 2009(Wiklund et al. : 1924(Wiklund et al. -1925 a. De upptäckte att han läste den bloggen varje dag, #men det they discovered that he read that blog-the every day but that gjorde han inte. did he not 'They discovered that he read this blog every day #but he didn't.' b. De upptäckte att den bloggen läste han varje dag, #men det they discovered that that blog-the read he every day but that gjorde han inte. did he not 'They discovered that this blog he read every day #but he didn't.' c. Han läste inte den bloggen varje dag.
he read not that blog-the every day 'He didn't read this blog every day.' d. Vi upptäckte faktiskt inte att han inte läste den bloggen varje dag. we discovered actually not that he not read that blog-the every day 'We actually didn't discover that he didn't read this blog every day.' e. Vi upptäckte faktiskt inte att den bloggen läste han inte varje dag. we discovered actually not that that blog-the read he not every day 'We actually didn't discover that this blog he didn't read every day.' This provides further evidence that factivity is not the relevant distinction for whether that-clauses allow embedded V2. In order to fully capture these patterns, the definitions for referentiality needs to be refined. In a discussion of referentiality at the phrasal level, Cinque (1990: 16) defines referentiality in regards to wh-phrases as "the ability to refer to specific members of a set in the mind of the speaker or pre-established in the discourse". Based on this, and along the lines of de Cuba & Ürögdi (2009), I propose the following modified definitions for referential CPs and non-referential cPs. 35 (74) a. Referential CP: an accepted (or pre-established) proposition in the existing discourse which has no illocutionary force.
b. Non-referential cP: a speech act which introduces a proposition (or an open question) which is not yet accepted (or pre-established) in the existing discourse.
By removing any discussion of factivity from these definitions, it removes the expectation that all complement clauses with factive interpretation should behave similarly, which is certainly in the spirit of de Cuba & Ürögdi (2009). In addition, as noted by the reviewer, the definition in (74) has similarities to the notion "main point of the utterance" proposed in Wiklund et al. (2009;following Simons 2007). This allows semifactives like know and discover to be both factive and non-referential at the same time. The difference then between semifactives (class E predicates) and true factives (class D predicates) then is that while both are factive, only semifactives can introduce a new proposition into the discourse. 36 35 A reviewer worries that the definitions in (74) do not correctly pick out all of the relevant CPs in the two groups, specifically Hooper & Thompson's (1973) Class C predicates (non-assertive, non-factive predicates like doubt and deny, which block MCP) and Class E predicates (semi-factives, which do not block MCP). I'll briefly discuss each of these in turn. First, the reviewer states that Class C predicates don't have illocutionary force, which is in line with the definition, but they are not accepted or pre-established propositions in the discourse. However, the status of Class C as a separate class is not entirely clear to me. As noted in Heycock (2006: 192-3), in Mainland Scandinavian, "inherently negative verbs (doubt, deny, regret) or verbs taking irrealis complements do not license embedded V2; and again, even the subset of bridge verbs that allow embedded V2 do not license it when negated or modalized." She also notes (2006: 190) that Andersson (1975), based on his investigation of Swedish MCP, proposes that the non-negative predicates in Class C should actually be in Class B, while negated Class B predicates should be in Class C. In other words, the presence of matrix negation/irrealis is what blocks MCP both in Class C and in negated/irrealis Class A and Class B. So for me the issue is, why does negation/irrealis generally block MCP with predicates we would otherwise expect MCP to occur? There are some previous analyses that propose solutions for the negation/irrealis blocking effects, such as Iatridou & Kroch (1992) and de Cuba (2007), both of which follow the idea of negative complementizers proposed in Laka (1990). For my purposes here I will assume that these types of analysis are on the right track and leave negation/irrealis MCP blocking in non-referential CPs as a separate issue which certainly deserves more attention. Second, the reviewer points out that semi-factive Class E predicates, which are presupposed, allow MCP, despite the fact that they are "accepted/pre-established" in the discourse. For me, there is a clear difference between Class D true factives (regret, resent, hate, love, etc.) and Class E semi-factives (discover, notice, know, realize, etc.) in that semi-factives can much more easily present new information that is not part of the common ground. For example, while (i) is completely natural out-of-the-blue, (ii) is only natural if the hearer already knows about the labyrinth. In other words, semi-factives can introduce CPs that are "not yet accepted or pre-established in the discourse". Thus, even though the speaker presupposes the truth of the embedded CP, there is no necessary commitment from the speaker that the information is known by the hearer. By separating speaker presuppositions of truth from speaker beliefs about what is in the common ground, I believe that the definitions provided can still capture the data. For more discussion of MCP occurring in factive contexts, see Section 7.6. 36 It might be the case that even true factives can sometimes introduce a non-referential complement. I'm thinking specifically of the oft discussed example of a complement of regret unexpectedly allowing topicalization, a MCP, as discussed by Haegeman (2006): Haegeman (2006Haegeman ( : 1666 I regret that those details, I cannot reveal to non-members.

More on the nature of referentiality
In this section, based on de Cuba & Ürögdi (2009) and de Cuba & MacDonald (2013), I provide examples from discourse contexts and do so-replacement in English, as well as sentential referential properties of it in English to support the idea that referentiality is important to morphosyntax. The definition in (74a) describes a referential CP as a proposition that is accepted in the existing discourse. In other words, it forms part of the common conversational ground (i.e. the ground shared by the speakers). Consider the following discourse contexts and felicitous and infelicitous uses of referential and nonreferential complements in light of this characterization (discussion based on de Cuba & MacDonald 2013: 129-132).
In the first discourse context, a parent and a teacher discuss a theft of lunch money that occurred at school. The teacher states (75a) and the parent responds with (75b). The parent's response uses the verb regret, which typically embeds a referential CP. A referential CP refers back to a resolved proposition, one that forms part of the common ground, which in this case is the proposition introduced by the teacher; namely, the sentence in (75a). By contrast, in the same discourse context and with the same original statement from the teacher from (75a), if the parent responds as in (76), the result is infelicitous.
(76) Parent: #I think that my son stole the lunch money.
In this case, think embeds a non-referential complement. Thus, it cannot refer back to the proposition introduced (under normal intonation) by the teacher. The result is infelicity. Stated differently, think introduces a proposition for acceptance into the common ground, and it is odd to introduce a proposition if it is already accepted as part of the common ground shared by the speakers.
In out-of-the-blue discourse contexts, we find the inverse patterns of felicity with respect to referential and non-referential clauses. In a context in which a parent walks up to a teacher and initiates a discourse where there is no previous mention of stealing lunch money nor that the parent's son is involved, the parent's out-of-the blue statement in (77) is infelicitous.
(77) Parent to teacher: #I regret that my son stole the lunch money.
Here the proposition my son stole the lunch money is not part of the common ground shared between the speakers. This conflicts with the nature of the embedded sentential complement of regret, since regret typically requires a referential complement. However, assuming the same out-of-the blue discourse context as above, there is no infelicity when the verb think replaces regret, as in (78). This is because think can typically take a nonreferential complement.
(78) Parent to teacher: I think that my son stole the lunch money.
In addition to cases of felicitous and infelicitous complements, we see other correlations between the referential status of a sentential complement and the availability of different morphosyntactic phenomena. de Cuba & Ürögdi (2009) provide syntactic arguments for the differing referential status of some English embedded clauses. Observe in (79a) that do-so replacement targets the VP, a predicational element, while it-replacement targets referential arguments, as illustrated in (79b).
(79) a. Bill tried the cake, and John did [ VP so] too b. Bill tried the cake, and John tried [ DP it] too Now consider the contrast exhibited between factives and non-factives in (80), provided by Kiparsky & Kiparsky (1971: 362). Under a non-factive verb, as in (80a), the phrase that Bill had done it can be replaced with so (just like the VP ate a cake in (80a) The evidence here suggests that the embedded CP here patterns with referential DPs (rather than predicative elements).

Summary
In this section I introduced MCP and extraction data into the discussion of NCCs and VCCs and noted that the general FCC as RC analyses of Arsenijević (2009) and Kayne (2008;, which did not focus on MCP and extraction, nonetheless do not provide an obvious account for the syntactic differences between referential and non-referential VCCs. At minimum, these analyses would need non-trivial modifications to cover the MCP and extraction data. On the other hand, the analyses of VCCs in Haegeman & Ürögdi (2010) and de Cuba & Ürögdi (2009) both account for the referential/non-referential asymmetries in VCCs. I also introduced the idea that referentiality is the relevant notion for dividing the behavior of so-called factive and non-factive complement clauses, and provided some motivation for linking referentiality to a structural difference in referential and non-referential clauses. In the next section I extend the referential analysis to NCCs, and argue that while both the truncation approach of de Cuba & Ürögdi (2009) and the operator movement account Haegeman & Ürögdi (2010) can reasonably handle the MCP and extraction facts in NCCs, the truncation account has advantages over operator movement. 37 Note that for some speakers, so is the only grammatical option in (80a). 38 As noted by Moulton (2015), Bhatt (2010) shows there are naturally occurring examples of semifactive know selecting so: (i) (Bhatt 2010: 177-178) a. I knew you would be angry enough about that madam, or I should have told you before; and he knew so too... (from 'Mysteries of Udo lpho', AnnRadcliffe, 1794: 554) b. Rooney knew he was special from a young age. And those who nurtured a talent that comes along rarely in any sport knew so too. (www.dailymail.co.uk/sport/article-389647/Walking-miracle. html, 2006) Given the revised definitions for referentiality in (74), this data is not surprising. As discussed, factivity is not the relevant notion, so it is unsurprising that semifactives like know can occur with non-referential cP and thus allow so-replacement. As discussed above, know and other semifactives can also allow embedded V2 and other MCP. The difference between semifactives and true factives then would be that semifactives allow new information clauses more freely than true factives.

A simple proposal: NCCs are truncated referential clauses
Given the potential complications discussed throughout this paper for previous analyses of NCCs (Kayne 2008;Arsenijević 2009), it seems that there is room for an alternative analysis. Table 2 summarizes the types of complement clauses discussed and the properties they have. In this section I propose that the truncation analysis for VCCs presented in de Cuba & Ürögdi (2009) can be extended to account for the patterns in Table 2, including the behavior of NCCs.

The proposal
One important difference between NCCs and VCCs is that NCCs always have an antecedent in the discourse, unlike VCCs. In other words, the content noun (claim, fact, etc.) and its associated CP refer to the same entity. Thus, my claim is that in this sense all NCCs are referential since they are co-referential with their content noun.
The content noun in NCCs (claim, fact, idea, etc.) can be thought of as the antecedent for the NCC. I follow Hawkins (1978) in claiming that NCC constructions like those in (83) are " referent-establishing" and involve "close apposition". Under this type of analysis, all NCCs are referential, as opposed to the split that exists between referential and non-referential VCCs. Syntactically, I adopt the truncation analysis of de Cuba & Ürögdi (2009) for referential CPs, as in (66) above. Despite the terms "noun complement clauses", "NCCs" and "complements of N" that are often used for these structures, I treat these CPs as modifying adjuncts. 40 In the next section I present some motivation for treating all NCCs as referential CPs.

The referential status of NCCs
In a discussion of Danish NCCs, Mikkelsen (2014: 6), following Hankamer & Mikkelsen (2012), notes that what she calls "direct structure" constructions like (84b) have an intuitive similarity to "close nominal apposition" constructions like (84b). 41 39 I return to some exceptions to the generalizations made here regarding MCP in NCCs and VCCs in Section 7.4. 40 See Stowell (1981: Ch. 3, Section 7) for arguments that so-called "Complements of N" are not in fact complements. 41 Mikkelsen (2014), following Hankamer & Mikkelsen (2012), divides Danish NCCs between "relational structures", which involve a preposition (i) and "direct structures", which do not (ii).   She notes that in (84a) the second element is a proper name that identifies the referent and the common noun labels that referent with a particular property. For her, direct structure NCCs are very similar in that a CP like that everyone participates in (84b) provides the primary semantic content of the fact. Mikkelsen (2014: 10), following Hankamer & Mikkelsen (2012), proposes that direct structure NCCs are "referent-establishing" (in the sense of Hawkins 1978: 130-149). Content nouns with NCCs have the unusual property of being able to be felicitously uttered with a definite article at first-mention (without previous introduction with an indefinite article). Interestingly, content nouns cannot be used as a first-mention without the CP. Thus, under this view one can certainly argue that the CP in (84b) has referential properties. 42 Keizer (2007: 22-60) provides an extensive discussion of close appositions and notes that in most analyses of close nominal apposition (e.g. Haugen 1953: 169;Hockett 1955: 101;Sopher 1971: 27), both parts of the construction are taken to refer to the same entity. Keizer also notes that Hawkins (1978: 146) finds the constructions in (85) to "have a lot in common with" the NCC constructions in (86).
I don't like the color red. b.
I can't stand the name Algernon. c.
The number seven is my lucky number.
(86) a. Bill is amazed by the fact that there is so much life on earth. b.
The philosophic aphasic came to the conclusion that language did not exist.
According to Hawkins (1978: 147), one way in which they are similar is that in both constructions the success of definite reference depends on the presence of the modifier. In other words, in (85) the color, the name and the number fail to refer when used as a first-mention (without previous indefinite descriptions) if they appear without the "modifiers" red, Algernon and seven respectively (87). The same situation holds for the fact and the conclusion in (86); they cannot be used as a first-mention without the CP modifiers in (88).
(87) a. I don't like the color. b. I can't stand the name. c.
The number is my lucky number. (all decidedly odd if used as a first-mention in a discourse) (88) a. Bill is amazed by the fact b. The philosophic aphasic came to the conclusion (both decidedly odd if used as a first-mention in a discourse) Hawkins argues that in both of the sets of constructions in (85) and (86) the modifier "takes over the role of previous discourse, and enables the hearer to identify some set of objects within which he is to locate the referent." (Hawkins 1978: 148). 43 Keizer (2005) also discusses the discourse functions of close appositions. Among these uses is the descriptively identifying use in which "[…] the descriptive element provides information which allows the hearer to relate the referent of the construction as a whole to her 'knowledge base', or, more specifically to anchor the referent in the discourse situation" (Keizer 2005: 449). In regards to close apposition in general, Keizer (2007) concludes that, "[…] the whole point in using an apposition consists in the fact that through the combination of a proper noun and a descriptive element one can produce a referring expression which is felicitous in a given context" (Keizer 2007: 60). Under the present analysis, I take this as evidence in favor of the idea that the "modifying" CP in NCC constructions is referential.
The referential view of NCCs I am taking here is differs from the view taken by Arsenijević (2009)  In looking at the propositions in (89), it is not clear that they have illocutionary force. Typically, an assertion is defined as a speech act in which the speaker puts forth a proposition as being true (i.e. to be included in the common ground). For example, Krifka (1999; takes the view that in an assertion, the speaker takes on the social commitment that the content of the assertion is true. By this definition, the NCCs in (89a-b) and (89e) are not asserted, since they can be uttered felicitously by a speaker without any commitment to the truth of the proposition. Krifka argues that there is a speech act operator ASSERT added to what he calls a sentence radical, which for him is a proposition without illocutionary force. Along the same lines, a question is usually seen as a speech act that obligates the addressee to respond with an answer. Krifka (1999; sees the illocutionary act of questioning as going beyond the denotational 43 A reviewer worries about claims of referentiality, given that the content noun in an NCC construction can be indefinite, as in (i).
(i) a. a claim that MMR is a defective product b. a statement that the earth is flat However, in the nominal domain indefinite NPs can be antecedents to anaphors without problem: (ii) A man shot himself.
Likewise, the antecedent in the NCCs in (i) need not be in the previous discourse. The important part in the present analysis is the referentiality of the CP modifier, not the content noun. Hawkins (1978) actually analyzes sentences like those in (86) as deriving from sentences with the indefinite article (a fact, a rumor) and the modifier S (CP in current terms) introduces the referent. meaning a question. He takes the view of Hamblin (1973) that questions denote sets of propositions, namely the set of possible answers. But for Krifka, this is simply the denotational meaning of questions, which for him is a sentence radical. In order for such sets of propositions to be used to ask a question, the illocutionary operator QUEST must be added. Again, the sentences in (89c-d) do not seem to be question speech acts, as they are not explicit requests for information. Finally, an imperative is generally seen as a speech act in which there is an obligation expressed by the speaker for the addressee to perform some act. Krifka claims that the illocutionary operator DIRECT is involved in these speech acts. Once again, (89f) can be uttered felicitously without any imperative force from the speaker. In sum, Krifka (1999; claims that there are sentence radicals, which denote propositions, and speech acts, which are formed when illocutionary operators are applied to sentence radicals. He argues that speech acts are distinct from regular semantic objects and that this greatly restricts (but does not completely rule out) the embedding of speech acts. For Krifka, sentence radicals do not involve illocutionary force. He also notes that sentence radicals have more syntactic restrictions, citing examples of MCP being restricted to contexts following predicates which typically allow associated clauses with illocutionary force operators. He also notes that the three speech act operators he discusses (ASSERT, QUEST and DIRECT) are often grammaticized in languages. 44 In contrast to the proposal in Arsenijević (2009), my definition for referential CPs explicitly states that they do not have illocutionary force, which is in the spirit of Krifka (1999;. I believe that this characterization fits the data in (89) better, as arguably none of these NCC constructions involve illocutionary force in the sense discussed above.

Relative Clauses
Unlike the previous accounts discussed in this paper, I do not tie the structures of NCCs to the structure of RCs. Specifically, I claim that NCCs to not universally involve relative operator movement. Thus, I make no prediction that the kinds of morphosyntactic behaviors that are related to operator movement in RCs should occur NCCs and VCCs, as for me these clauses do not involve operator movement. Given the numerous examples of NCCs and VCCs not behaving like RCs that I have provided throughout this paper, I see this as an advantage for the referential truncation account over a NCC as RC account. Since the present proposal focuses on NCCs and does not involve RCs, I leave aside any analysis of the complementizer drop, complementizer choice, extraction and argument fronting properties of RCs. 45 On the other hand, Arsenijević (2009) andHaegeman (2012) do provide examples (see Section 3.1) that show some similarities between RCs and NCCs (i.e. the same morphological items showing up in both constructions). To explain these similarities, I appeal to an obvious similarity in my analysis of NCCs and most analyses of RCs: the presence of a CP. Given the fact that complementizers and wh-phrases are commonly found in 44 As discussed by Heycock (2006: 190), Andersson (1975) argues that the relevant semantic distinction between embedded clauses that allow MCP in Swedish and those that do not is that MCP clauses make a statement, ask a question, or give a command (Andersson's class of 'semantically main clauses') and non-MCP clauses perform none of these functions (Andersson's 'semantically subordinate clauses'). This characterization of the facts fits well with Krifka's (2014) semantic analysis. 45 See Taraldsen (1986) and Stroh-Wollin (2002) for discussion of C-drop in relative clauses. left-peripheral positions in CP, it is not surprising that there may be overlap between morphological items in both types of construction (as well as VCCs). 46

Extraction
NCCs are strong islands for extraction, disallowing both argument and adjunct extraction as shown above in (58). If they are adjoined modifiers, as I claim, we would expect them to be strong islands, as other adjuncts are. As for referential VCCs (weak islands, see (60)) and non-referential VCCs (non-islands, see (62)), the truncation account of de Cuba & Ürögdi (2009) that I adopt here can account for these facts. The referential status of truncated CP disallows extraction of non-referential wh-phrases. Non-referential cPs are not referential by definition and therefore are not subject to the restriction against extraction of non-referential wh-phrases. 47

MCP
I analyze all NCCs as being referential clauses, so they should always have the truncated CP structure. Thus, MCP are correctly predicted to be banned from all NCCs (59) given the lack of the cP projection which provides a position for MCP (66a). Referential VCCs are obviously also analyzed as being referential, so they also have the truncated structure and block MCP (61). Non-referential clauses have the full structure (66b) and thus allow MCP (63).

Complementizer drop
I pointed out some of the problems the NCC as RC account might have with the C-drop facts in Section 5. Here I sketch out a plausible analysis that has the advantage of being simpler because I do not tie the RC C-drop facts to the FCC C-drop facts, given the fact that for me, they are structures with different properties. 48 This avoids some of the complications facing Arsenijević's (2009) C-drop account which I discussed in Section 5. That leaves us with the FCC C-drop facts: NCCs (which I have argued are referential) and referential VCCs resist C-drop, while non-referential VCCs allow it. This is a clean division in that referential clauses as a class resist C-drop while non-referential clauses as a class allow C-drop. Thus, complementizer drop is yet another morphosyntactic difference (like 46 As for adnominals like Korean -un presented by Arsenijević (2009) (example (17) above), it is not so clear what their status is. In the source paper for the Korean data presented, Cha (1998) argues that constructions like (17a) and (17b) differ in a number of syntactic and semantic properties, and that the adnominal -un should be analyzed as having two lexical entries, one as a relativizer as in (17a) and one as a complementizer as in (17b). In addition, Kim (2015) claims that many languages, including Japanese and Korean have what are called "generalized noun-modifying clause constructions" (GNMCC). Kim argues that Korean NCCs are among these constructions, and that Korean also has RC constructions with a different syntax from NCCs. In addition to RCs and NCCs, Kim's "Perception or Event-Related Clauses" (phrases like "smell of fish grilling") also take the adnominal. Finally, note in the Basque examples the adnominal -ko appears in the NCCs in (25) but not the RCs in (23). Additionally, -ko in Basque appears with other phrases, "converting postpositional phrases into adjectival ones." (de Rijk 2008: 90). This fits with the general picture presented here of the CP in NCCs being a modifier. 47 See Szabolcsi & Zwarts (1993), who argue that there can be no variable left unbound within a referential complement, ruling out extraction of non-referential extractees and giving "factive" complements (referential complements in the present proposal) their weak island status. See also Szabolcsi (2006) for discussion of referentiality and weak islands. 48 A reviewer asks about the status of so-called "subject contact relatives" as illustrated in (i).
(i) a. My friend's got a brother Ø used to be in the school.
b. There's a train Ø goes without stopping.
It is clear is that the pattern of C-drop here is different from NCCs, which generally resist C-drop. Therefore, such structures are outside the scope of the present analysis. For discussion of subject contact relatives, see den Dikken (2005). Interestingly, den Dikken analyzes these constructions as topic-comment structures which do not involve relative operator movement (and he argues that they should not be called relatives at all). the availability of MCP and extraction possibilities) that tracks along the referential/nonreferential division. 49

Complementizer choice
As seen in Section 3, there is a great deal of variation in complementizer choice cross-linguistically. Some languages use the same complementizer for relative clauses and NCCs (Section 3.1), some languages use the same complementizers for declarative clauses and NCCs (Section 3.2), and some languages, like English, use the same complementizer for all three of these constructions. The truncation account proposed here does not makes any general predictions as far as complementizer choice goes. However, it crucially does not make the wrong prediction for languages like Swedish and Basque that have a relative complementizer that appears when a relative operator is present in the left-periphery. 50 I return to this point below in Section 7.5 in a comparison of the truncation account and the operator movement account.

Extending the proposal to it-complement clauses
In this section I present another construction that seems to share some behavior with NCCs. Note that it-complement clauses (ICCs) can be analyzed in the same way as NCCs. 51 Like NCCs, ICCs are degraded with complementizer drop, as in (90).
(90) a. I regret it *(that) I overslept yesterday. b. Eddie resents it *(that) his brother bet against him. c. The bosses noticed it *(that) John left early.
(i) a. He said (that) he would leave.
(class A) b. I thought (that) he would leave.
(class B) c. He denied *(that) he had done it.
(class C) d. He regretted *(that) he had done it.
(class D) e. He discovered *(that) she had done it.
(class E) However, to my ear the facts are not so clear. While some semifactives do seem to resist complementizer drop, this is not always the case: for example the sentences in (ii) seem fine to me with the complementizer omitted.
(ii) a. I notice (that) you are wearing your class ring today. b. You know (that) I always like to wake up early. c. I was in my bedroom when I discovered (that) my wallet was gone.
The fact that semifactives are less resistant to complementizer drop than true factives mirrors the difference in the availability of MCP between the two classes. Of course I have argued that referentiality is the relevant notion for different syntactic behavior between the classes of complement clause, not factivity, so a difference in behavior between true factives and semifactives is not unexpected. The prediction made by the present analysis would be that MCP like EV2 would be allowed in non-referential semifactive contexts, which has been confirmed by Wiklund et al. (2009). As for complementizer drop, the prediction would be that non-referential semifactives would be less resistant to complementizer drop than referential semifactives. I leave a comprehensive examination of the complementizer drop facts to future work. 50 Krapova & Cinque (2015: 12) make the following remarks: "Note that for languages such as Italian, French, and Bosnian-Croatian-Serbian, which use the same subordinator for both declarative and ordinary relative clauses, it can be maintained that this subordinator neutralizes the difference found in languages like Bulgarian, which distinguish the two. The only real counterexample would be to find a language which distinguishes the subordinator used in clausal complements of verbs from the relative subordinator, and uses the latter also in the clausal complements of Ns, as in an imaginary variety of Bulgarian where only the relative subordinator deto were to appear in clausal 'complements' of nouns." I have presented numerous cases where the complementizer for NCCs and VCCs differs from the complementizer of RCs. However, in my (admittedly incomplete) investigation of languages thus far, I have yet to run into a language with the complementizer properties of the "imaginary variety of Bulgarian" described by Krapova & Cinque, where the complementizer of the RC and the NCC are consistently the same but the complementizer for VCCs is consistently different. 51 Note that ICCs are sometimes referred to as "object extraposition" structures in the literature.
ICCs are also strong islands for extraction like NCCs but unlike referential VCCs (91).
(91) a. *Who does Mary regret it that she saw t? (ICC) b. *Who does Mary regret the claim that she saw t?
(NCC) c. Who does Mary regret that she saw t?
(referential VCC) It is interesting to note that in English, ICCs are typically selected by predicates that typically select referential complements, as shown in (90). Predicates that typically take nonreferential complements tend to avoid ICCs, illustrated in (92).
(92) a. *I think it that I overslept yesterday. b. *Eddie supposes it that his brother bet against him. c. *The bosses claimed it that John left early.
It is tempting then to analyze ICCs in the same manner as NCCs, since they share similar properties. 52

Operator movement vs. truncation
As mentioned above, Haegeman (2012) does not provide a fleshed-out proposal for operator movement in NCCs, but one could imagine some form of operator movement being proposed to block MCP and extraction from NCCs. 53 In addition, she might come up with a similar story to mine for the C-drop facts where it is operator movement that blocks C-deletion. If one accepts my claim that all NCCs are referential, she could claim that operator movement would occur in all NCCs. This raises the question as to whether or not one of these referential accounts has any advantages over the other. However, I will suggest here that the truncation analysis has certain advantages over a potential operator movement account. 54 Recall that Haegeman's operator movement proposal for NCCs is a RC-style account in the spirit of Arsenijević (2009) and Kayne (2008), repeated here in (93).
(93) a. Kayne (2008) the fact i that they're here in t i b. Arsenijević (2009)  I presented evidence in Section 4 that Swedish and Basque are languages where the morphological form of the relative complementizer is specifically tied to the presence of a relative operator in the left-periphery of the clause (since the relative complementizer shows 52 Complements clauses to manner-of-speaking verbs are also a class that could potentially fall under a similar analysis, as they are resistant to C-drop in the same way as NCCs and ICCs (see Section 5.4). Manner-ofspeaking verbs have been claimed to involve cognate objects, as in John whispered a soft whisper (see Zwicky 1971). One could imagine an analysis of manner-of-speaking CCs similar to that of NCCs and ICCs where a relationship of close apposition holds between a null cognate object and the manner-of-speaking CC, with the manner-of-speaking CC being a referential CP that resists C-drop. I leave the details of this speculative analysis to future work. 53 Though this operator would need to be different from the event operator she proposes for VCCs, since only factive/referential VCCs have the event operator in her system, not non-factive/non-referential VCCs. In addition, the event operator only causes weak island effects, and NCCs are strong islands. See discussion below. 54 Haegeman (2012: 189-193) provides a series of general arguments against the truncation account for central adverbial clauses, and claims that these also apply to truncation accounts for factive/referential that-clauses such as de Cuba & Ürögdi (2009). However, de Cuba (2014) argues that Haegeman's stated problems for truncation analyses in general are specific to her (2006) truncation account and do not necessarily rule out other truncation accounts for factive/referential that-clauses.
up in relative clauses, indirect questions and clefts in these languages: all constructions that have been widely analyzed as involving a relative operator in the left periphery). Given that the declarative complementizer, not the relative complementizer shows up in NCCs in these languages, the operator movement account, which specifically proposes that there is relative operator movement in NCCs, makes the incorrect prediction that the relative complementizer should appear. In contrast, since the truncation account does not propose relative operator movement, it makes no prediction that NCCs should behave like RCs. Since there is no operator movement proposed, the Basque and Swedish data can be explained: there is no operator movement to the left periphery of NCCs, so the declarative complementizer should appear, not the relative complementizer. For this reason, despite the fact that both operator movement and truncation accounts can potentially handle the referential/non-referential distinctions in VCCs, the truncation account fares better, at least for NCCs, since it handles the complementizer choice facts in Swedish and Basque better.
In addition, the present proposal is more general regarding different types of structures than the operator movement account. One notion, referentiality, is proposed here to account for the behavior of different types of embedded clauses. In contrast, different types of operators have been proposed for operator movement in different types of embedded clauses. Haegeman (2012: 273) cites Nichols (2003) in support of her RC proposal for NCCs. Nichols proposes that NCCs are actually adjuncts, not complements, and that they are formally RCs. Nichols argues that these structures involve event operator relativization, as in (94) For Nichols, all attitude nominals, both factive and non-factive, involve event operator relativization. 55 On the other hand, Haegeman (2012) cites as predecessors to the event operator movement account in VCCs both Aboh (2005), who claims that movement of an event operator results in factivity, and Melvold (1986;, who claims that complements of factive predicates are "event arguments" that contain an operator in CP licensed by the functional element definiteness which binds "an open event-position". There seems to be a mismatch here between when we see an event operator in NCCs vs. VCCs. The interpretation of NCCs is not factive or definite in the complement of N examples in (95), given the non-factive/non-referential nature of the head nouns claim and story.
(95) a. Many people have made the claim that the earth is flat. b. Other people believe the story that vampires walk the earth.
In other words, in order to block MCP in NCCs we would need to follow Nichols and propose operator movement for all attitude nominals, including non-factives. However, in order to account for the MCP facts in VCCs we would need to propose event operator movement in factive VCCs to block MCP but no event argument movement in non-factive VCCs which allow MCP. Of course one could propose that there are two different operators at work in NCCs and VCCs that appear for different semantic reasons, but this would need to be spelled out. On the other hand, the referential truncation account presented Other examples come from Germanic, where Heycock (2006) provides an embedded verb-second (as opposed to verb-final) NCC example from German in (101), and Julien (2010) provides an embedded verb-second (finite verb preceding negation) NCC example from Norwegian Bokmål in (104).
(101) German (Heycock 2006: 194, citing Beatrice Santorini, p.c.) der Glaube, die Erde sei flach the belief the earth is(SBJ) flat 'the belief that the earth is flat' (102) Norwegian Bokmål (Julien 2010: 14) Så trekker han konklusjonen at annet er ikke å vente. then draws he conclusion-the that other is not to expect 'Then he draws the conclusion that nothing else is to be expected.' The existence of these types of examples is obviously unexpected for my analysis of NCCs, given my claim that all NCCs are truncated referential CPs.
Here I will present some speculation on how to account for these types of examples. One possibility is that in languages that generally allow MCP in NCCs (i.e. the British speakers in Maki et al. 1999 who allow examples like (100)) there is dialectal variation as to whether or not NCCs are seen as co-referential with the head noun. In varieties where NCCs are not treated as co-referential with the head noun, the possibility of MCP in NCCs would open up. However, this availability can still be constrained by the referential status of the particular NCC. One thing to note about the MCP examples in (99-102) is that in all of these cases the head noun is non-factive as opposed to factive, mirroring the behavior regarding the availability of MCP in factive and non-factive VCCs discussed in Section 6. Discussing example (102), Julien (2010: 15) reports that, "the nouns that take V2 clauses as complements are morphologically and semantically related to verbs of saying and of cognition, and their complements represent assertions in the same way as complements to the related verbs." 58 Thus, in varieties that do not treat NCCs as co-referential with the head noun, NCCs will pattern the same as VCCs in terms of MCP.
Another important thing to note here is that there is also some reported variation on the possibility of MCP in factive VCCs. Twelve out of fifteen of Bianchi & Frascarelli's (2010) English informants accepted (103), Maki et al.'s (1999) British informants, who accepted (100), also accepted (104), and Kiparsky & Kiparsky (1971: 348, note a) report that for some speakers, even NCCs with "fact" as a head noun can allow non-presupposed NCCs.
(103) Bianchi & Frascarelli (2010: 10) I am glad that this unrewarding job, he has finally decided to give up.
Additionally, Green (1976) reports that MCP in factives are better with a first person subject than those with a third person subject, as in (105).
(105) Green (1976: 388) a. I regret that never before has such a proposal been made. b. *But Bill regrets that never before has such a proposal been made.
This kind of variation (and many other examples) lead Green (1976: 397) to the conclusion that, "[…] the embedability of so-called MCP is influenced not only by syntactic forms and semantic functions, but also by pragmatic functions, by what a speaker is trying to do in using a particular syntactic form to express something." The referentiality proposal in this paper is undeniably pragmatic in nature, with the speaker of an utterance having the power to present an embedded clause as referential or non-referential. Under the current proposal this choice is implemented syntactically, with non-referential complement clauses having a full left-periphery and referential complement clauses a truncated left periphery. While the presence of different types of head nouns in NCCs and head verbs in VCCs (factive or non-factive) often signal the referential status of a complement clause, this is not achieved by syntactic selection. The referential status of the complement clause itself (as decided by the speaker) is what matters. 59 If this is the case, we would expect to see some of the variation discussed in this section. For example, the speakers who allow (103) and (104) could consider the embedded clause to be factive, but not referential in a pragmatic sense (giving a "regret to say" or "regret to inform" reading, as discussed by Haegeman 2006Haegeman : 1666. For the speakers who generally allow non-presupposed clauses in "fact" NCCs (discussed by Kiparsky & Kiparsky 1971), a similar explanation can be given. For these speakers, "fact" can introduce a non-referential clause. The contrast in (105) could be tied to the fact that in (105a) the speaker is both the utterer of the sentence and the subject of the sentence, and in this case the speaker has the flexibility to mark the embedded clause as non-referential, thus allowing topicalization.

Conclusion
In this paper I've presented arguments against the NCC as RC analyses presented in Kayne (2008;, Arsenijević (2009) andHaegeman (2012) and indeed any other analysis that proposes that NCCs generally involve operator movement cross-linguistically. Instead, I've proposed an analysis of NCCs that argues they are referential clauses with a truncated left-periphery (in the spirit of de Cuba & Ürögdi 2009), equivalent to referential complement clauses to verbs (referential VCCs) and different from relative clause structures. I have argued that the analysis presented here is more amenable to cross-linguistic NCC data than the NCC as RC approach regarding complementizer choice facts in languages like Swedish and Basque, which tie complementizer choice to the presence or absence of an operator in the left-periphery, and complementizer drop facts from English. In addition, it covers the MCP and extraction facts that were not touched on by Kayne (2008;, and Arsenijević (2009). Given the many cases presented here and elsewhere where the morphosyntactic behavior of RCs and NCCs differ significantly, an analysis which treats them as different structurally avoids the many complications that arise in trying to treat them as the same animal. As for Kayne's (2008) general claim that nouns do not select clausal complements, I remain agnostic here. I treat NCCs as adjuncts but leave open to future research the adjunct vs. complement status of NCCs. The claim here is simply that however these clauses are syntactically associated to the head noun or head verb, it is not by a hidden relative clause structure, at least not as a universal property of human language. The relatively simple alternative referentiality proposal for NCCs that I have presented avoids the problematic claim that NCCs involve relative operator movement while remaining amenable to the data I have presented here. Abbreviations acc = accusative, adn = adnominal, aff = Affirmative sentence particle, aug = augment, aux = auxiliary, comp = complementizer, conj = conjunction, DeclC = declarative complementizer, def = definite, dem = demonstrative, det = determiner, eval = evaluative, fut = future, int = interrogative sentence particle, itr = intransitive auxiliary, loc = locative, neg = negation, nom = nominative, O = object agreement, pfv = perfective, prt = partitive, pst = past, RelC = relative complementizer, RelM = relative marker, rpt = relational particle, s = subject agreement, sg = singular