More appositives in heaven and earth than are dreamt of in your Linguistics

1. Mandarin Chinese and Japanese appositive relative clauses belong to the class of fully integrated appositives; 2. English appositive relative clauses are diametrically different from the fully integrated ones. I classify them as non-integrated appositives; 3. Italian and French have two types of appositives: semi-integrated and non-integrated; 4. Reduced relative clauses can also be appositive: if pre-nominal, they are fully integrated; if post nominal, they are semi-integrated.


Introduction
A well-known distinction in the literature on relative clauses is the one between restrictive and appositive (or nonrestrictive) relative clauses (see De Vries 2002; for an overview).Thesetwotypesofrelativeclausesarecharacterizedbyspecificsyntacticand semantics features, the most important of which states that while a restrictive relative Glossa general linguistics a journal of Del Gobbo, Francesca. 2017. More appositives in heaven and earth than are dreamt of in your Linguistics. Glossa: a journal of general linguistics 2(1): 49. 1-38, DOI: https://doi.org/10.5334/gjgl.14 Del Gobbo: More appositives in heaven and earth than are dreamt of in your Linguistics Art.49, page 2 of 38 clausefurther"restricts"thereferenceofthenominalitmodifies,anappositiverelative clause does not, as it simply adds information about that nominal: 1 (1) the man that I like (restrictive) (2) Jeff,whoIlike (appositive) In (1) above, the relative clause that I like further restricts the reference of the nominal the man;in(2),theappositiverelativeclausespecifiesthatIlikeJeff,butthecontentof the relative is not necessary in order to identify the individual referred to by the proper name, hence the appositive provides additional information. In this paper, I use the term appositive relative clause to refer to those relative clauses that do not semantically restrict the reference of the head that they modify, or in other words, to those relative clauses that can modify proper names. A recent focus on these constructions has brought to light the fact that we are not facing a homogeneous set. For example, Potts (2002) distinguishes between what he calls asparentheticals and which-appositives: (3) Adapted from Potts (2002) a. Americans should get cheap oil, as the whole world knows. b. Americans should get cheap oil, which the whole world knows. Potts (2005) brings those two types of appositives under the same umbrella when he claims that, together with other constructions, they trigger conventional implicatures. Specifically,hestudiesbothsupplementsandexpressives.Andamongthesupplemental expressions, he includes as-parentheticals, appositive relative clauses, appositive nominals as well as parenthetical adverbs: (4) a. Ames was, as the press reported, a successful spy. b. Ames, who stole from the FBI, is now behind bars. c. Ames, the former spy, is now behind bars. d. Cleverly, Beck started his descent. e. Luckily, Beck survived the descent. Stowell (2005) brings to our attention the following construction, which he calls a parenthetical restrictive relative: (5) Stowell (2005) The guy next door (that I sold my car to) was arrested today.
Finally, Cinque (2006;2008), on the basis of evidence from Italian, distinguishes between "integrated" and "non-integrated" appositive relative clauses. Against this background, my goal in this paper is to lay out a proposal for a typology of appositive relative clauses, or ARCs. I identify three types of appositive relative clauses: non-integrated, semi-integrated and (fully)-integrated and account for their behavior by introducing two parameters (one linked to the presence or absence of the intonational break and one linked to the relative pronoun).
The paper is structured as follows: in Section 2 I lay out the empirical domain that I will use for my investigation. In Sections 2.1, 2.2, 2.3 and 2.4, I analyze how the 2 The empirical domain I investigate three sets of languages, whose appositive relative clauses are distinguished by whether they are prenominal or not, whether they are introduced by a relative pronoun or not and whether or not there is an intonational break between the appositive and the phraseitmodifies.
Inthefirstsetoflanguages,theappositiverelativeisprenominal,thereisnorelative pronounandtherelativeclauseisnotsetoffintonationallyfromthematrixsentencethat contains it. Two languages where appositive relatives behave this way are Chinese and Japanese, illustrated respectively in (6) and (7): Chinese (adapted from Lin 2003) [ CP Xianglai jiu bu ai du shu de] Xiaoming xianzai ye kaishi du always then not love study book mod Xiaoming now also begin study qi shu lai le. begin book come asp 'Xiaoming, who does not love to study, now also has begun to study.'

(7)
Japanese (adapted from Yuasa 2005) [ CP Shuuron-o kaite i-ru Iwasaki-san]-ga sono master's.thesis-acc write be-pres Iwasaki-hon-nom the gakkai-de happyo shi-ta. conference-at presentation do-past 'Mr. Iwasaki, who is writing a master's thesis, presented a paper at the conference.' Thesecondsetoflanguagesischaracterizedbyappositiverelativeclausesthataresetoff intonationally from the matrix. Within this set, it is possible to further distinguish those relative clauses that are introduced by the complementizer (what Cinque 2006; 2008 calls the "integrated" appositive relative clauses), from the ones that are introduced by a relative pronoun (these are called by Cinque 2006; 2008 the "non-integrated" appositive relative clauses). Among the languages that are characterized by these types of appositiverelatives,wefindItalianandFrench(Cinque1982;2008).Thefollowingtwoexamples illustrate appositive relative clauses introduced by the complementizer che in Italian (example (8)) and by the complementizer queinFrench(example(9)): Italian (adapted from Cinque 2008) Inviterò anche Giorgio, [ CP che abita qui vicino]. (I) will.invite also Giorgio that lives here close 'I will invite also Giorgio, who lives nearby.' (9) French(adaptedfromCinque1982) Ma soeur, [ CP que le magistrat avait convoquée pour le lendemain], … my sister that the magistrate had summoned for the next.day 'My sister, who the magistrate had summoned for the next day, …' The example in (10) shows an appositive relative clause introduced by the relative pronoun il quale in Italian, and the example in (11) shows an appositive relative clause introduced by the relative pronoun laquelle in French: 2 (10) Italian ( The third set of languages is represented by English and Romanian (Cinque 2008). In these languages, appositive relatives need to be introduced by a relative pronoun (modulo the differenttypediscussedbyStowell2005)andtobesetoffintonationallyfromthematrix: Romanian ( Appositives in these three sets of languages behave in a substantially different way with respect to, among other phenomena: illocutionary independence from the matrix; 3 the ability to take split antecedents; the categorial nature of the antecedent; and binding phenomena.

Illocutionary independence
English and Romanian allow appositive relative clauses to be illocutionary independent from the matrix: 4 2 NoticethatIassumehere,followingKayne(1976)andCinque(1982)thatque in French and che in Italian are complementizers, and not (weak) relative pronouns. But this assumption is not crucial for my proposal.
For a relevant discussion on this topic, see Sportiche (2011 Inthepreviousexample,twodifferentstrategiesforquestion-formationhavebeenused. In (18a), the auxiliary is followed by its negated copy, and in (18b), the question particle ma is added. None of the examples result in grammaticality. Notice though that it may be misleading to take the examples in (18) to show that a relative clause cannot be interrogativeinMandarinChinese.Thisisbecausefinalparticlessuchasma (in example (18b)) are a root phenomenon, and are never allowed in embedded clauses: in other words yes-no questions with ma in Mandarin Chinese are restricted to matrix clauses. As for A-not-A questions (the form used in (18a)), we know that they appear both in directandindirectquestions,asshowninthefollowingexamplesfromHuang,Liand Li (2009) The sentence in (21) is ungrammatical because in order to get the wide scope reading, the A-not-A form should undergo movement at LF, and in (21) this is blocked by the ECP. InordertoembedtheA-not-Aforminsideanisland, Huang,LiandLi(2009) DylanTsai(p.c.)observesthatitisunlikelythatthewh-expressioninvolvedin(24)isanindefinite,since there is no intentional context to license polarity interpretations. Rather, the construction in (24) may be related to the so-called "wh-placeholder" construal in Chinese, which typically appears within the scope of adefinitedeterminersuchasnage in (i): (i) Na-ge shei dique lai guo zheli. that-cl who indeed come asp here 'That whatshisname indeed came here before.' The examples that more strongly show that an appositive relative clause in Mandarin Chinese cannot be interrogative are the ones in (24) through (26). We know that whwords can have scope over the matrix clause or only over the embedded clause, as the followingexamplesfromHuang,LiandLi (2009) If an appositive relative clause in Mandarin Chinese could be interpreted as interrogative, then in the sentences (24) through (26) the wh-word should be able to take scope over the relative clause, but we saw that this is not possible. I therefore conclude that the evidence here provided shows that appositive relative clauses in Mandarin Chinese cannot beinterrogatives,differingthereforeinacrucialwayfromappositivesinRomanianand English (Cinque 2008). 6 Italianbehaves differently dependingon whether theappositiveisintroducedby the complementizer or by a relative pronoun. If introduced by the complementizer, Italian appositive relatives behave like the Chinese ones in not allowing an illocution that is differentfromtheoneofthematrix.Ifintroducedbytherelativepronoun,theybehave like appositive relatives in English and Romanian, allowing the appositive to have an illocutionthatisdifferentfromtheoneofthematrixclause: Italian (Cinque 2006) *Tuo padre, [che potrà mai perdonarci per quello che abbiamo fatto?], your father that will-be-able ever forgive-us for that that have done non si sarebbe mai comportato così. not himself would-be ever behaved this.way (30) Italian (Cinque 2006) Tuo padre, [il quale potrà mai perdonarci per quello che abbiamo your father the which will-be-able ever forgive-us for that that have fatto?], non si sarebbe mai comportato così. done not himself would-be ever behaved this.way 'Yourfather,bywhomwillweeverbeforgivenforwhatwehavedone?,would have never behaved like that.'

Split antecedents
In English and Romanian, appositive relatives allow split antecedents: 6 A reviewer wonders whether the prenominal position of the relative clause may play a role in the ungrammaticality of interrogative relative clauses in Mandarin Chinese. I exclude that to be the case, given theexamplein ( It is simply impossible to allow split antecedents with relative clauses, even if they are appositive. So, in (34), the operator inside the relative clause can only refer to the individual denoted by the "head" of the relative, that is, Zhangsan, and it cannot in addition refer to the individual denoted by the subject of the second sentence, that is, Lisi.
It is important to observe that there can be different types of split antecedent constructions. Take the following Italian example from Cinque (2008). In this example, thetwo antecedentshavedifferentthetaroles: Italian Se Carlo i non amava più Anna j , i quali i+j d'altra parte non si if Carlo not love.past any-longer Anna the which of other side not rec erano mai voluti veramente bene, una ragione c'era. were ever wanted really well a reason there was 'If Carlo was no longer in love with Anna, who at any rate never really loved each other, there was a reason.' RelativeclauseswithsplitantecedentswerefirstnoticedbyRossandPerlmutter(1970).
The following examples are from Zhang (2007): a. Mary met a man i and John met a woman j who i+j knew each other i+j well. b. A man i came in and a woman j left who i+j knew each other well. c. The house has a room i and the shop has a cellar j which i+j are joined by a small underground passageway. Zhang (2007) claims that the two antecedents of relative clauses with split antecedents are originally two conjuncts of a coordinate nominal. She proposes that each has undergone a sideward movement, involving a move from the original working site to a new one. The two nominals take part in the construction of a coordinate clausal complex. In the old working site, a complex nominal is constructed, in which the relative clause takes the remnant coordinate nominal as its antecedent. Finally, the complex nominal adjoins to the coordinate clausal complex. Crucially, the construction that Zhang (2007) studies has a number of restrictions, the most important of which is that the two antecedents need to share the same theta-role. More in general, in order for sentences with split antecedents to be grammatical, they have to be parallel, in a sense that I will not make precise here (for details, see Zhang 2007; Section 5). Del Gobbo (2010) cites the following as a potential example of split antecedents with an appositive relative clause in Mandarin Chinese: (38) Chinese Fenbie dou ai guo Xiaoyu de Zhangsan jiagei le Wangwu, Lisi separately all love asp Xiaoyu mod Zhangsan marry asp Wangwu Lisi jiagei le Houliu. marry asp Houliu. 'Zhangsan i married Wangwu and Lisi j marriedHouliuwho i+j both had loved Xiaoyu.' Zhang's (2007) account could be applied to such sentences, but what is relevant for us here is that it is the other type of split antecedent relative that is not allowed in Mandarin Chinese. We have seen that in Italian it is grammatical for an appositive relative clause to have two split antecedents, even when they do not share the same theta-role (see example (36)). Under these conditions, similar examples in Mandarin Chinese turn out to be ungrammatical. Let me start with two sentences in a piece of discourse: Chinese a. Ruguo Zhangsan bu ai Mali le, jiu yinggai you yi-ge liyou. if Zhangsan not love Mary asp then must have one-cl reason 'If Zhangsan doesn't love Mary any longer, there must have been a reason.' b. Shishi-shang tamen conglai mei-you zhenzheng de ai guo duifang. reality-on they ever not-have really mod love asp the.other 'At any rate, they never really loved each other.' Ifwetrytoconveywhatisexpressedbythetwosentencesin(39)usingasinglematrix sentence and a relative clause -as we did with the Italian example in (36) -the outcome is ungrammatical:  (40) and (41) above are ungrammatical, in the intended reading of 'If Zhangsan doesn't any longer love Mary, who at any rate never really loved each other, there must be a reason'. The ungrammaticality of the sentences in (40) and (41) shows that appositive relative clauses in Mandarin Chinese are not able to take split antecedents, when these do not share the same theta-role. Italian again shows a split. Appositives introduced by the complementizer behave like the Chinese ones in not allowing split antecedents, while those introduced by the relative pronoun behave like the English and Romanian ones in allowing the pronoun inside the relative to pick up its reference from two antecedents: Italian ( Mary is touchy that her sister certainly not be Lit. 'Mary is touchy, that her sister surely is not.'

Binding
When we consider binding phenomena, the situation we have been delineating so far no longer holds. We would expect the Italian appositive relatives introduced by the complementizer to behave à la par with the Chinese ones, but this surprisingly does not happen.Bindingofapronouninsidetheappositivebyaquantifiednominalinthematrixis not allowed in Italian, regardless of whether the appositive is introduced by the relative pronoun or by the complementizer. In other words, both types of appositives in Italian behave like appositives in English 7 and Romanian: 8,9 (i) a. *[Every Christian] i forgives John, who harms him i . (=(54) in the text) b. [Every Christian] i forgives a man who harms him i . Jackendoff(1977:176)hadalsomadethesameobservation: (ii) a.
[Everyone] i bough a suit that suited him i . b. * [Everyone] i bought a suit, which suited him i . Thereareexceptionstothispattern,asobservedbySafirhimself(1986:ft.9)andSells(1985b): (iii) [Every Christian] i prays to God, who forgives him i . Sells(1985b)showsthattheseexamplesdonotinvolvesyntacticbinding,butatypeofdiscourselinking called"cospecification".Cospecificationisonlypossiblewithcertainoperatorsinacontinuativediscourse, andwhenthereistemporalormodalsubordination.Intheexamplesinthetextwherethequantifiednominalcannotbindthepronouninsidetherelativeclausesuchconditionsarenotfullfilled. 9 A reviewer is suggesting the following two as counterexamples to the claim that it is not possible to bind within appositives: (i) Tutti gli studenti hanno perdonato Anna, che li ha aiutati molto. all the students have forgiven Anna, that them has helped a lot 'All the students forgave Anna, who helped them a lot.' (ii) Every student forgave Anna, who helped them a lot.
Thesetwoexamplesfallinthesamecategoryasthecospecificationones,i.e.theyareexamplesofdiscourse anaphora. As a matter of fact, they can be transformed into pair of sentences in a piece of discourse, as follows: (iii) Tutti gli studenti hanno perdonato Anna. Lei li ha aiutati molto. all the students have forgiven Anna she them has helped a lot 'All the students forgave Anna. She helped them a lot.' (iv) Every student forgave Anna. She helped them a lot.
Compare (iv) with the following example, which is ungrammatical with the reading where him is bound by every student: (v) *Every student forgave Anna. She helped him a lot. Tosummarize,wefindthatChineseappositiverelativeclausescannotbeillocutionary independent of the matrix clause, they cannot have split antecedents, they can only modify nominals(typeNPorDP),andtheydoallowaquantifiednominalinthematrixtobind a pronoun in the appositive. I classify them as "integrated" appositive relative clauses, following the terminology established in Cinque (2006;2008). In Italian, appositive relative clauses introduced by the complementizer differ minimally from the "integrated" appositives of Chinese, insofar as they do not allow binding. Because in all other respects they do behave like the "integrated" ones, I classify them as "semi-integrated" appositive The example in (v) is ungrammatical because every student cannot bind him,sincetheyareintwodifferent sentences. The sentences in (i) and (ii), and their discourse counterparts in (iii) and (iv), are grammatical because the plural pronouns do not need to be bound, they can instead refer back to tutti gli studenti, 'all the students' and every student via discourse anaphora. 10 For additional examples from Mandarin Chinese, I refer the reader to Del Gobbo (2001;2003a;. Del Gobbo: More appositives in heaven and earth than are dreamt of in your Linguistics Art.49, page 14 of 38 relative clauses. The other type of appositive relatives in Italian, the one that is introduced bytherelativepronoun,behavesdiametricallydifferentfromthe"integrated"type:itdoes allow illocutionary independence, it can have split antecedents, it can modify "heads" of different categories, and finally, it does not allow binding. Along with the English and Romanian appositive relative clauses, it belongs to the class of the "non-integrated" appositives (Cinque 2006;2008). The empirical facts are summarized in Table 1: I conclude that there are more types of appositives than originally thought by Cinque (2006;2008).Morespecifically,Iemphasizethatthebindingfactsforceustotakeinto consideration establishing a third category of appositives, what I have called the "semiintegrated" one. In the following section, I provide an account to explain the variation just observed.

Proposal
My proposal is based on three core elements. First, I assume that the syntax of non-integrated appositive relative clauses (henceforth, ARCs) involves a CommaP projection and aForcePprojection(asinKoev2013;Griffiths2015),whilesemi-integratedARCscontain only a CommaP. Second, I provide a semantics for ARCs crucially built on the idea that therelativepronounisE-type(Sells1985a;b;Demirdache1991;DelGobbo2003a,and following work). 11 Finally, I claim that it is the syntax of prenominal relative clauses that prevents them from licensing overt relative pronouns.
I propose two parameters to account for the variation described in Section 2 of the paper.ThefirstparametertakestheroleoftheintonationalbreakinARCsveryseriously, and in this I am very close to Potts' (2005) approach: (59)

Intonational break
If there is an intonational break, the ARC projects a CommaP.
The second parameter focuses on the presence versus absence of the relative pronoun: (60)

Presence of the relative pronoun in ARC
If the pronoun is present, it is an E-type one and the ARC is a non-integrated one. If the pronoun is absent, the ARC is either integrated or semi-integrated.
Let's now focus more on the details of the proposal, by spelling out how each component works and how the whole system accounts for the empirical variation we observe. For the syntax of post-nominal restrictive relative clauses, I follow Rebushi's (2005) proposal. Rebushi (2005) hypothesizes that since nominal modification is property conjunction at the semantic level, maybe it is its counterpart (some type of coordination) atthesyntacticleveltoo.HethereforeproposesthatafunctionalprojectionConjunctionP 11 Griffiths (2015) claims that appositive relative pronouns are referential. For an explanation as for why appositive relative pronouns need to be E-type and not just referential, see footnotes 14, 15 and references cited there.
For appositive relative clauses, I propose that instead of a Conjunction Phrase (ConjP), there is a Comma Projection that mediates between the "head" of the relative and the relative itself. Notice that the category Comma is also contemplated by Potts (2005) in his system, buthere,differentlyfromwhatheproposes,Iclaimthattheintonationalbreakisaterminal node that projects its own maximal category. This is very much in accordance with de Vries' (2007) notion of Specifying Coordination, that is, &:P. Notice that both in the case of the restrictive relative clauses and in the case of the appositive relative clauses, the "head" of the relative clause and the relative itself are not of the same syntactic type. Following de Vries (2007), we need to assume that we are dealing with an instance of unbalanced coordinate structure, whose existence needs to be assumed for independent empirical reasons: Moreover there is the question of the external visibility of the Conjunction Phrase. Given that no lexical head subcategorizes for a sheer ConjP, the complement phrase must also be identifiedasaDP,anNP,oraPP.Wethereforeneedamechanismallowingthecategorial features of one of the conjuncts to percolate to the Conjunction's maximal projection. Rebushi (2005)observesthattheconfigurationweneedisjustifiedbyJohannessen's(1998)theory, accordingtowhichitistheelementthatoccupiesthespecifierpositionintheConjPthat transmits its relevant features to Conj, (under Spec-head Agreement), whence they percolate toConjP(themaximalprojection).Thus,in (61),theresultingfullyspecifiedphrasewillbe a [+Conj,+N]P. This mechanism is assumed for all the appositive relative clause structures with CommaP, but I will abstract away from it in the tree diagrams that follow. I furthermore propose that a ForceP projection is at play in non-integrated appositives. Koev(2013)andGriffiths (2015)proposetotreatARCsasForcePhrases(inthesenseof Rizzi1997)andadjuncts.AccordingtoKoev(2013),aForceheadhostsanoperatorthat introducesapropositionalvariableforitsconstituents.Heassumesthatoperatorsandpredicates are adorned with propositional variables. Thus, operators can bind predicates while 12 SubP stands for Subordinate Phrase. I use this label, instead of CP, for the following reasons: 1. I follow Rizzi (1996), and therefore adopt the view that the CP projection is split into multiple projections; 2. Since restrictive relative clauses do not have a Force of their own, they should not project a ForceP.
SubP"servessimplytosubordinateaclause"(Haegeman2003:335),to"makeitavailableforcategorial selectionindependentlyofitsforce"(Rizzi1997:fn.6). higher operators can bind lower operators. Force heads bind into the lexical expressions in their syntactic scope and such lexical expressions are relativized to their respective propositional variables. It follows from this that appositives cannot be bound from outside. Koev (2013) also maintains that the operator in Force hosts features for speech acts. Given the evidencefromItalianARCs,andpreciselythedifferentbehaviorshownbynon-integrated ARCs and semi-integrated ones, I propose to allow each operator to be hosted by its own functionalprojection.Specifically,Commaistheheadthatintroducesapropositionalvariable for its constituents. Force, instead, hosts the speech act operator. 13 13 OneofthereviewerraisesconcernsaroundmyproposaltodivorceKoev's(2013)propositionalvariables (which my account links to CommaP) from illocutionary independence (which my account links instead to ForceP). Notice that Koev's (2013) proposal addresses mainly the following two aspects of ARCs: (i) Non at-issue status; (ii) Projection behavior, i.e. the ability to escape scope.
Regarding (i.), for Koev (2013), the fact that appositive proposals are usually introduced before main clause's proposals explains why ARCs are often not at-issue: all proposals associated with a sentence are silently accepted except the one introduced last, which is at-issue. If we assume that the propositional variable is introduced by Comma, and not by Force, we then expect that both semi-integrated and nonintegrated ARCs were to behave the same with respect to answerhood capabilities, and this is indeed borne out. The examples in (iii)  Non-integrated appositives, that is, English, Romanian and Italian il quale-appositives, have therefore the following structure (63) In (63),the"head"oftherelativeclauseisintheSpecifierpositionoftheCommaPhrase projected by the intonational break. The actual appositive relative clause is in the com-plementpositionoftheCommaPhrase. FollowingGriffiths(2015:149)andKoev(2013, IproposethatARCscontainForcephrases,inthesenseofRizzi(1997).Non-integrated ARCs can be illocutionary independent because ForceP allows them to contribute adifferentillocutionaryforcethantheoneofthehostclause.IproposethattheComma projection introduces a variable for the content of the constituent in its scope, as it is proposed by Koev (2013) for Force, thereby preventing any operators from the host clause to bind into the appositive clause. Following Del Gobbo (2007), I claim that the relative pronoun in the appositive relative clause, who in (63), is an E-type pronoun. It is necessarilycoindexedwithitsantecedent,the"head"oftherelative. FollowingHeim(1990),this means that at LF it is a copy of the antecedent. 14 For the syntactic structure in (63), I propose the LF in (6) Notice that in each of the examples above the ARC projects, regardless of whether it is introduced by che (semi-integrated), or by la quale (non-integrated). 14 Notice that, as pointed out by one of the reviewers, the situation is more complex, and the interpretation of the E-type pronoun varies depending on its antecedent. Del Gobbo (2007) proposes that if the antecedent (or"head"oftherelativeclause)isdefiniteorgeneric,atLFthepronounisanidenticalcopyofitsantecedent.Iftheantecedentisquantificational,atLFtheE-typerelativepronounisadefinitedescription,whose NP denotation is obtained from the matrix clause. See Del Gobbo (2007) for additional details. See also Heringa(2011),who,followingDelGobbo(2003a,proposesthatappositionsalsocontainacovert E-type pronoun.
Del Gobbo: More appositives in heaven and earth than are dreamt of in your Linguistics Art.49, page 18 of 38 The proper name JohnthatoccupiesthespecifierpositionofForcePisacopyoftheproper name John in the spec of CommaP, by virtue of the fact that it is an E-type pronoun. 15 As a relative pronoun, it originates in the object position of the FinP and moves to the Spec of ForceP, by predicate abstraction. In this respect, it behaves like a relative pronoun in a restrictive relative clause, where -taking as an example the tree diagram in (61) above -the FinP John saw is of type t, the pronoun is not assigned a denotation of its own, but by virtue of its movement, by predicate abstraction, it allows the entire relative SubP to become a predicate, hence to obtain a denotation of type <e,t>. Since in (64) the relative pronounisE-type,itiscontentfulandnotvacuous,andspecificallyitcombineswithits sister Force' and yields a ForceP of type t.
This proposal explains why in non-integrated ARCs (i.e., in appositives in English and Romanian, and in Italian il quale-appositives) split antecedents are allowed, why nonintegrated ARCs can be illocutionary independent of the matrix and why they can modify any syntactic category. I provide an explanation in what follows.
We know that split antecedents are allowed with pronouns, but not with operators. Ross andPerlmutter(1970)firstobservedthatsomerelativeclausescanhavesplit antecedents, as in the following example: RossandPerlmutter (1970) A man entered the room and a woman went out who were quite similar.
Notice that, although not explicitly addressed in the literature, this type of relative clause cannot be introduced by a complementizer (and hence contain an operator): (66) *A man entered the room and a woman went out that were quite similar.
Ontheotherhand,arelativepronouncanrefertoasplitantecedentjustlikeareferential pronoun can: (67) A man entered the room and a woman went out. They were quite similar.
Thus, for reasons that as far as I know are still unclear, it is not possible to have split antecedents if, instead of a relative pronoun, the relative clause is headed by an operator. Therefore it is the presence of the relative (E-type) pronoun that allows non-integrated ARCs to take split antecedents. 16 The presence of the Force projection allows the ARC to be illocutionary independent ofthematrix,sinceviaForcePitcancontributeadifferentillocutionaryforcethanwhat specifiedbythematrix.
We know that E-type pronouns can pick up the reference of any syntactic category (Sells1985a;b;Potts2002),whilepropertiesandpropositionsarenotrelativizedinthe absence of an E-type relative pronoun that can pick up their denotation. Specifically, 15 For an explanation as to why we need an E-type vs. a simple referential pronoun, I refer the reader to Del Gobbo(2003a)aswellasSells(1985b)andDemirdache(1991. 16 A reviewer asks if it is possible for little pro to license split antecedents. This seems to indeed be the case: Un uomo entrò nella stanza, ed una donna uscì. pro erano piuttosto eleganti. a man entered in.the room and a woman left were rather elegant 'A man entered the room, and a woman left. They were rather elegant.' I will not try to explain here why little pro licenses split antecedents, but an empty relative operator does not. I will limit myself to observe that what a regular pronoun, the E-type one and little pro have in common is the fact that they are referential, while the relative operator is not. Sells(1985a)claimsthatinwhich-relatives, properties and propositions are individuals in the semantic interpretation, and Potts (2000) treats which-appositives traces as individual-denoting (i.e., nominalized propositions). We can infer from this that the relative pronouns which in English and il quale in Italian are able to denote nominalized properties or propositions, but operators (in our raising structures, the raised element denoted by identity with the "head") are not able to function the same way. 17 We therefore conclude that it is the presence of an E-type pronoun, such as which or il quale, within nonintegrated ARCs that allows this type of appositives to be able to modify any kind of syntactic phrase.
We also explain the binding facts by assuming, following Koev (2013) in spirit, that Comma introduces a variable for the content of the constituents in its scope, thereby blocking binding by any element of the matrix clause into the appositive. In other words, variablesintheseappositivesarenotboundbyquantifiersinthematrixclausebecause thetwoelementsarerelativizedtodifferentpropositionalvariables. 18 Theanalysisofsemi-integratedARCsdiffersminimallyfromtheoneproposedforthe non-integrated ones, but in a crucial way. For semi-integrated appositives, following Haegeman's(2003)treatmentofcentraladverbials,IproposethattheprojectionForceP is absent, and CommaP takes as its object a simple SubP (Subordinate Phrase), as illustrated in the tree diagram in (68) below. Such SubP is headed by a category Sub, which "servessimplytosubordinateaclause"(Haegeman2003:335),to"makeitavailable forcategorialselectionindependentlyofitsforce"(Rizzi1997:fn.6).Also,theelement that raises from within the FinP to Spec of SubP is a DP that is identical and coindexed with the DP "head" of the relative. This allows it to delete both in the LF and in the PF component. As a result of this, that is, because of the absence of a contentful E-type pronoun, the appositive relative clause maintains its status as a property, and it does notswitchtoapropositionstatus,asshowninthetreediagramin(69)below.Inother words, up to the SubP node, the semantics of the semi-integrated appositive is identical to the semantics of a restrictive relative clause. The Comma node is responsible for preventing binding into the semi-integrated ARC, just as it does in the structure for nonintegrated ARCs. 17 I use the term "operator" here and elsewhere in the paper as an abbreviation for "the raised element denoted by identity with the "head" of the relative clause", and because so much of the traditional literature on relative clauses distinguishes between relative operators and relative pronouns. Strictly speaking, though, there are no operators in my account, as what used to be called an operator is the DP that raises from within the relative FinP to Spec of SubP (see diagram (68) for an example), and deletes by identity with the c-commanding head (see Del Gobbo 2003a andCinque 2003;2008 for similar accounts). 18 One of the reviewers pointed out to me recent work by Simons et al. (2010) on projective meaning andat-issueness.Simonsetal.(2010)maintainthatprojectionisaunifiedphenomenaandthatdiverse expression types (including ARCs) project because they all share the pragmatic property of being notat-issue. They propose that the common property of projective meanings can be characterized in terms of the notion of at-issueness utilizing the concept of relevance to a QUD (Question Under Discussion). In their proposal, projection is intimately related to the structuring of information in discourse. In their own words, "It is a consequence of the fact that in the totality of information conveyed by an utterance, some is central to the speaker's conversational goals, and some is peripheral. The peripheral projects". (Simons et al. 2010: 325). If I were to follow Simons et al. (2010), I would have to maintain that ARCs project because they are never at-issue, and the projection is the result of operators "ignoring" some of the content triggered in their scope. This alternative proposal would also account for the binding phenomena I describe in the paper, as far as we can make the reasonable assumption that the intonational break is the PF signal that at LF we are dealing with non-at-issue, or projection material. Because the proposal by Simons at al. (2010) is tied to information structure in the discouse, it has the potential of explaining the exceptionalcasesof"cospecification",asmentionedinfootnote8andasrecentlydiscussedbyAmaral, Roberts and Smith (2007).
This analysis accounts for all of the properties of semi-integrated appositive relative clauses. Specifically,sinceno(E-type)pronounispresent,weexplainwhysplit antecedentsare not allowed. The appositive cannot be illocutionary independent because no ForceP is present. Finally, the absence of the contentful E-type pronoun explains why with this kindofappositivetheonlymodifiablesyntacticphrasesareDPs.Thisisbecause,aswe mentioned before, only E-type relative pronouns are able to denote nominalized properties or propositions, which, syntactically, can be any type of phrases, such as VPs, APs, PPs and even IPs. In the semi-integrated appositive relative, the raised element denoted by identity with the "head" is not able to function the same way. This analysis also explains why the semi-integrated ARCs are identical to the non-integrated ones in their blocking of binding from the matrix clause. Comma introduces a variable for the content of the constituents in its scope, and this prevents binding from constituents in the host clause (Koev 2013). Fully integrated appositives, which are found in Chinese and Japanese, are characterizedbycompletelydifferentsyntaxandsemantics.
ThestructurefortherelativeclauseinChineseisexemplifiedbelow.Withinthe relative clause, we have movement of Zhangsan from inside FinP to the spec of SubP. I follow Cheng (1986;andPaul (2009)inconsideringthemodificationparticlede as a non-root complementizer or subordinator. The "internal head" deletes because it is identical and coindexed with the "external head". 19 Since SubP is adjoined to the "external head", this one cannot c-command SubP and anything that SubP dominates. It follows from this that the "internal head" cannot be a relative pronoun, as it would not be c-commanded, as it has to, by the "exernal head". 19 A reviewer asks what are the exact conditions under which deletion under identity happens. According to Sauerland(2000),c-commandisnotnecessary.HeintroducesthetermRelative Deletion to refer to the process that renders the internal head of matching relatives unpronounceable: (i) Relative Deletion: In matching relatives, the internal head must not be pronounced. Furthemore, the external head must be the antecedent of the internal head.
Since it cannot be a relative pronoun, the "internal head" also cannot be an E-type pronoun, forcing the relative SubP to be semantically a property, and not a proposition.
The "head" of the relative clause, Zhangsan, and the relative itself concatenate via predicate modification.Iassumeherethatpropernamesareassignedthesemantictypepredicate, i.e. <e,t>, with the understanding that they are predicates true of just one individual (seeQuine1939;Chierchia1998).
In fully integrated appositives (as in the Chinese ones), the absence of the relative E-type pronoun explains why they cannot have split antecedents, and why they can only modify nominals. The absence of the intonational break explains the binding phenomena: the appositive is always in the scope of the matrix operators, since there is no Comma availabletointroducedifferentpropositionalvariables.Finally,sincethereisnoForceP, the relative cannot be illocutionary independent.
If we compare restrictives with integrated appositives, we find that their behavior is very similar. In restrictives there is no intonational break, hence no CommaP, therefore binding into them is possible. They are predicates, not propositions, their relative pronoun is not E-type, and there is no ForceP, thereby explaining why they can only modify nominals, why they cannot take split antecedents and why they cannot be illocutionary independentfromthematrixclause.Thecrucialdifferenceisthatrestrictives semantically restrict the reference of the "head" they modify, while appositives do not, and can therefore modify proper names.

Appositives in Japanese
AmongthelanguageswhoserelativeclausesbehavesimilarlytoChinese,wefind Japanese. In Japanese, as in Chinese, relative clauses are prenominal, they are not separated from the modified "head" by an intonational break, and they are not introduced by a relative pronoun. Given that they share these features with relative clauses in Chinese, it is worthwhile to investigate whether appositive relative clauses in Japanese also show the same behavior with respect to illocutionary independence, the possibility to have split antecedents, binding and the categorial nature of the antecedent.
As we have seen for Chinese, in Japanese as well the appositive relative clause cannot have an illocutionary force that is independent of and different from the matrix one. Specifically, the appositive modifying Taroo in the following example cannot be interrogative (ka is the interrogative typing particle in Japanese): (72) *[Wareware-o kessite yurusa-nai dearoo ka] Taroo-wa we-acc never forgive-neg may q Taro-top kono yoo-ni-wa si-nakat-ta daroo. this way-in-top do-neg-past may Int. 'Taro, who will he ever forgive us, would have never behaved this way.' The second empirical domain to test involves split antecedents. Recall that in English the following sentences are grammatical: TheQueenservesmuffins i , and Prince Charles serves scones j , which i+j they buy atHarrods.
(74) A man i entered the room and a woman j went out, who i+j were quite similar.
When native speakers of Japanese are asked to provide sentences similar to the two examples above, they necessarily come up with two independent conjoined sentences: (75) Erizabesu Zyoo-wa mafin i -o dasi-te, Tyaaruzu kootaisi-wa Elizabeth queen-top muffin-acc serve-ing Charles prince-top sukoon j -o dasu-ga, (sorera-o) karera-wa Harozzu-de ka-u. scone-acc serve-C (they i+j -acc) they-top Harrods-at buy-pres. 'QueenElizabethservesmuffinsandPrinceCharlesservesscones,theybuy thematHarrod's.' In the previous example, ga, glossed as 'c', is a declarative conjunctive marker, simply connecting two sentences without establishing any particular causal relation between the two propositions. Notice that the pronominal object is optional. The following example shows similar results: we can only convey the desired meaning by constructing two independentsentences.Theonlydifferencebetweentheexamplein(75)andtheonein (76) is that the pronominal subject is required in (76): (76) Taroo i -wa heya-ni hair-i, Hanako j -wa heya-o de-ta-ga, Taro i -top room-in enter-ing, Hanako j -top room-acc leave-past-c karera i+j -wa totemo ni-te i-ta. they-top quite resemble-ing be-past 'TaroenteredtheroomandHanakolefttheroom,theyarequitesimilar.' When we attempt to use the relative clause to modify both nominal phrases, that is, when we try to establish the split antecedency, we run into ungrammaticality, or at best, we can allow the relative clause to modify only one of the two potential antecedents: (77) *[Op i+j Hanako-o kirat-te i-ru] Taroo i -ga hait-te ki-ta Hanako-acc hate-ing be-pres Taroo-nom enter-ing come-past ga, Ziroo j -wa de-te it-ta. though Jiro-top leave-ing go-past Theonlyinterpretationavailableforexample(77)isthatthepersonwhohatesHanakois Taro (alone), in other words, the split antecedent interpretation is impossible.
Another trademark feature of an appositive relative clause is the categorial nature of the antecedent. We know that in English, for example, non-integrated relative clauses (94) Krause (2001)  The fact that standard reduced relatives enforce subject relativization has been linked to thefactthatintheseclausesthesubjectpositiondoesnotreceiveCase(cf.Kayne1994; Bhatt1999).Anyderivationinwhichanon-subjectconstituentisrelativizedwouldleave behind a subject that has no Case inside the reduced relative, triggering a violation of the Case Filter and causing the derivation to crash. Relativization of the subject prevents such a violation, because once the subject is relativized, it can get case from the verb of the superordinate clause. I mentioned before that reduced relatives do not project up to the CP node, the question thenbecomes:whatfunctionallayerintheclausalstructuredotheyprojectto?Notice that the fact that they are smaller than CP accounts for the fact that they do not allow relative pronouns and complementizers. According to Burzio (1986) and Bhatt (1999), reduced relative clauses project up to the IP level (i.e. in our terms, the FinP level). According to Krause (2001), they are smaller than FinP, but need to be at least vP. Krause (2001) also concedes that there may be crosslinguistic variation in this respect.
It is interesting to investigate if standard reduced relatives can be appositive, and if they can, what type of appositive are they? In other words, would they be fully-integrated, semi-integratedornon-integratedappositives?InwhatfollowsIwillanswerthesequestions for both types of standard reduced relatives, i.e. for both prenominal and postnominal ones. Let's start with the postnominal reduced relatives. 22 The following example contains a postnominal reduced relative clause, which we can describe as appositive given 22 Areviewerpointsoutthattheleft-edgedeletionprocessknownas"whizdeletion"(Ross1967)createsin English post-nominal reduced ARCs that contain participle verbs. Such reduced relatives cannot be used as pre-nominalmodifiers: (i) a. John, (who was) looking secure in his job until recently, will be made redundant.
b. *Looking secure in his job until recently John will be made redundant.
(ii) a. John, (who was) bitten by a million mosquitoes, retreated indoors. b. *Bitten by a million mosquitoes John retreated indoors.
As there are several arguments in the literature against a "whiz deletion" approach to reduced relatives (see Huddleston1971;Berman1973;Hudson1973;Williams1975),suchapproachisnotfollowedinthispaper.
Prenominal reduced relatives are not illocutionary independent because they are not fullclauses,i.e.ForceP.TheydoallowsplitantecedentsbecausetheyhostaPROthatcan pick up the reference of more than one antecedent. They do not allow other antecedents beside nominal ones because they are not introduced by E-type pronouns. But they do allow binding, because they are fully integrated with the matrix clause (as they are not setoffintonationally)andnoCommaPispresent.
As it is clear from Table 4, appositive prenominal reduced relatives pattern with fully integrated ones (like in Chinese and Japanese), while appositive postnominal reduced relatives pattern with the semi-integrated ones (like in Italian and French), modulo the differentbehaviorwithrespecttosplitantecedents.

Conclusion
Ourempiricalinvestigationhasledtothefollowingfindings: 1. Chinese and Japanese appositive relative clauses are not illocutionary independent, they do not allow split antecedents, the antecedent can only be nominal, and they allow binding, hence they belong to the class of fully integrated appositives.

2.
EnglishandRomanianappositiverelativeclausesarediametricallydifferent from the integrated ones, insofar as they can be illocutionary independent, allowsplitantecedents,haveantecedentsofdifferentcategorialnatureand not allow binding. I classify them as non-integrated appositives. 3. Italian and French have two types of appositives. The semi-integrated ones share with the integrated ones the following features: they are not illocutionary independent, they do not allow split antecedents, and their antecedents can only be nominal;butdifferentlyfromtheintegrated appositivesofChineseandJapanese, these semi-integrated appositives do not allow binding. The second type behavesjustliketheEnglishones,andhenceareclassifiedasnon-integrated. 4. Reduced relative clauses can be appositives. This is summarized in   The empirical data I presented is accounted for by a proposal that is based on the following core elements: 1. The syntax of non-integrated and semi-integrated appositive relative clauses involves a CommaP such projection is missing from integrated ARCs. 2. The non-integrated appositives' relative pronoun is E-type. 3. The syntax of prenominal appositive relative clauses prevents them from licensing an overt relative pronoun, and therefore an E-type pronoun. 4. The intonational break is the PF instantiation of the Comma head, which by introducing a propositional variable, prevents binding from the host clause. 5. The ARC can be illocutionary independent only if it contains a Force projection.
The empirical data analyzed and the account proposed leave us with an interesting fine-grainedtypologyofappositives,therebyconfirmingtherecenttrendintheresearch on appositives pointing to the conclusion that we are dealing with a more variegated set of constructions than previously thought. What is even more interesting is that the conclusions reached with this study leave us with precise directions to investigate, whether thatisenrichingthedatasetofthelanguagesinvestigatedtoconfirmordisconfirmthe findingshereestablished,ordeterminingwhatothertypesofappositiverelativeclauses there could be. To give an example, we may wonder if there could be a language with appositive relative clauses with E-type relative pronouns but no intonational break, as hinted at in the last row of Table 5 above. Finally, a more detailed study of the parameters that are responsible for the typology of appositive relative clauses, paired with an investigation on a higher number of languages, would shed light on a crucial question for parametric theory, specifically whether parameters are implicational or not (see Longobardi, Gianollo & Guardiano 2008).