The pro-head analysis of the Japanese internally-headed relative clause

The objective of this paper is two-fold. One is to advance the view (call it the pro-head analysis) that the so-called internally-headed relative clause (IHRC) in Japanese is a nonrestrictive relative clause whose external head position is occupied by pro functioning as an E-type pronoun. It demonstrates that this view provides a natural framework in terms of which sundry and significant phenomena associated with Japanese IHRCs can be accounted for, including their characteristic properties contrasting with the corresponding change relatives and the doubly-headed relatives. The pro-head analysis contradicts a currently popular claim propagated by a series of papers that rely on null operator movement leading to the thesis that: (i) Japanese IHRCs may exhibit a “change sub-variety” (i.e., “Change IHRC”) that is not reducible to gapless externally light-headed relatives; (ii) Japanese IHRCs are strictly island-sensitive; and (iii) they disallow definite referential semantic heads. The other objective of this paper, therefore, is to present this long overdue challenge by arguing that their claims (i) ~ (iii) are not empirically sustainable, although their claim (iii) presents some difficult issues that defy clear-cut treatment as yet.


Objective
The purpose of this paper is two-fold. The first is to advance a view that the so-called "internally-headed" relative (IHR) in Japanese is a nonrestrictive relative clause whose external head [ dp [ np pro i ] D] contains phonologically null pronoun pro as in (1) acting as an E-type pronoun. 1 (1) [ dp/ihrc … [ cp/ihr … NP i … V] no [ dp [ np pro i ] D]] Θ-Assigner Where the Q-Assigner is either a syncategorematic noun (Kuroda 1999: 421-423, e.g., an inalienably possessed noun such as sippo 'tail', kekkan 'defect') 1 or a verbal predicate. (The construction marked by slanted brackets consisting of both IHR and the external head will be referred to as IHRC.) I refer to the view represented by (1) as "the pro-head analysis" of Japanese IHRCs. 2 I intend to show that the pro-head analysis model provides a natural framework in terms of which a number of empirical phenomena, including the IHRCs' distinct behaviors vis-àvis those of the corresponding change relatives (Tonosaki 1998) and the doubly-headed relatives (Erlewine & Gould 2016), can be naturally accounted for. More importantly, in proposing the pro-head analysis, this paper presents the first substantial challenge against a claim recently put forward by a series of papers, including Grosu and Landman (1998), Grosu (2010), Grosu & Landman (2012), , Landman (2016), and Grosu & Hoshi (2018). These works essentially share a claim that Japanese IHRCs involve a null operator movement cum lambda abstraction over the operator's trace. This core claim concerning the properties of IHRCs is expressed earlier in terms of the "maximalizing relative" thesis (Grosu 2012: 7) and has not been challenged in any critical manner since. The second objective of this paper is thus to show that there are some empirical data which are consonant with the pro-head analysis but in fact disprove their null operator movement hypothesis. In this endeavor I especially focus on  (henceforth "G&H"), as G&H profess that a principal goal of their endeavor is to put this particular null operator movement thesis "on a firmer basis" (Grosu & Hoshi 2016: 2). For, what happens upon closer examination is that careful review of G&H's argumentation actually has an effect of integrating the two objectives of this paper, as their empirically testable approach at the level of high clarity provides concrete venues to support the correctness of the pro-head analysis.
What G&H attempt to show is that the following three points are empirically sound: (2) a. Japanese IHRCs are strictly island-sensitive; b. They disallow definite referential heads (e.g., proper names); and c. The "change" sub-variety they refer to as "Change IHRC" is a genuine IHRC.
I will take up each of those issues at some length. The outline of the paper is as follows. In Section 2, I identify the notion "cospecification" based on Sells's (1986) view that the external head pro is a structurally conditioned E-type pronoun. Section 3 offers an account on the discourse function of the IHRC contrastive to that of the corresponding nonrestrictive "externally-headed relative clause construction" (henceforth EHRC), so as to shed light on the unique discourse characteristics of Japanese IHRCs. Sections 4 ~ 6 take up G&H's empirical claims articulated in (2). Section 4 examines G&H's assertion (2c) (i.e., "change IHRCs" are genuine IHRCs) to demonstrate that the relevant facts disprove their claim. Section 5 concerns G&H's point (2a), i.e., the claim that the Japanese IHRC is strictly island-sensitive. I argue that, while many native speakers exhibit the pattern G&H note, there are those who accept the kind of counter-examples that cannot be dealt with by G&H's analysis. Specifically, I show that the analysis that is called for is one that has a natural recourse in accounting for "repairs" that are possible even against the restrictions imposed by syntactic island constraints. The pro-head approach is shown to have such a recourse, whereas a G&H style analysis does not. Section 6 engages G&H's claim (2b), namely, that the IH cannot be a definite referential head (in particular, a proper name). I present clear counter-examples to this claim, although the core issue involved turns out to be difficult to fathom properly, as we will see. Section 7 offers a brief account on what Erlewine & Gould (2016) call "doubly-headed" relative clauses, to examine how the phenomenon of this particular construction type relates to our pro-head analysis of IHRCs. Section 8 concludes the paper.

Preliminary note on the external head pro
Given our representation of a Japanese IHRC as in (1), I follow, for the treatment of no and pro, Kitagawa & Ross's (1982) independently motivated analysis of Chinese and Japanese prenominal modification markers (i.e., Chinese de and Japanese no) that particle no, a semantically vacuous pre-nominal modification marker, is introduced by an independently motivated PF rule, as follows: Rule (3i) specifies that, if the nominal head X is not lexically represented, MOD no appears in Japanese in the surface form. In (1), pro is not lexically represented; hence, no is present. 5 Given (1), pro may well be replaced by a phonologically null Nominalizer or Complementizer.

Cospecification
It is generally understood that there is no clear syntactic device to distinguish nonrestrictive from restrictive relatives in Japanese (see Kuno 1973;Inoue 1976, among others). Kuroda makes the semantic distinction that (i) a restrictive relative clause narrows the notion expressed by the head noun phrase by adding attributes deriving a more complex noun phrase that identifies an entity, or entities, in a more limited domain of possible reference; that (ii) a nonrestrictive relative clause simply supplies "additional information to an entity, or entities, already identified by the referential function of the head noun phrase" (1992: 121); and that (iii) Japanese IHRCs are "nonrestrictive in a certain semantic sense" (1975-1976: 95; 1999: 427). Shimoyama (1999: 169) notes that IHRCs and nonrestrictive EHRCs do "share the property that an embedded clause is interpreted as if it were an independent sentence and is not composed with the rest of the DP it occurs in," and that the two constructions "share another property" -that of "cospecificational/E-type anaphora relation" -with respect to the type of anaphoric relation involved. Sells (1986: 435) claims that the cospecificational anaphora interpretation of the external head is "inescapable for a nonrestrictive relative clause," enforcing the "maximality." I accept this characterization by Sells on the inevitability of the cospecificational/E-type anaphora relation associated with nonrestrictive relatives. And I will henceforth adopt the term "cospecification" to describe the E-type anaphora relation involved in the Japanese IHRCs, assuming that the Japanese IHRC, as specified in (1), is a nonrestrictive relative, and that its external head is pro (or, more specifically, [ dp [ np pro] D]). Sells (1985) notes that cospecificational anaphora is only licensed when the antecedent is available in the discourse structure. I maintain that, for Japanese IHRCs, the relevant discourse structure is defined in terms of theta relation. Following Shimoyama (1999;2001), I take the following points to be obvious: (i) the antecedent (IH) must be internal to the IHR; (ii) the IH must bear a thematic role within an eventuality described by the IHR; (iii) it is "in contrast to discourses, where the antecedent needs not be in the sentence that immediately precedes the anaphor"; and (iv) the antecedent needs not be explicitly present in earlier discourse at all. 6 Grosu (2010: 234) correctly points out that Shimoyama's (1999) and Kim's (2007) E-type anaphora analyses "are constructed in a way that enables them to deal only with situations where the IH is in the highest clause of the relative." 7 I take this characteristic to be an inherent property of Japanese IHRCs. As a natural extension of this, I also incorporate into our analysis Grosu & Hoshi's (2018: 7) semantic condition that the relative and the matrix clauses "share a thematic participant." 8 What all this means is that: (i) an IHRC takes the structural form represented as in (4a); (ii) the argument in the IHR (the highest clause in the IHRC) serving as the antecedent (IH) must necessarily be a theta-participant within that clause; and (iii) the cospecification operation does not extend beyond this linkage. (4b) exemplifies (4a).
(i) a. Dono gakusei-mo i [soitu i -ga/pro i kongakki paper-o 3-bon kaita]-no]-o kesa every student (s)he-nom/pro this.semester paper-acc 3-cl wrote.nm-acc this.morning teisyutusita. turned.in 'Every student wrote three term papers this semester and turned in the term papers he or she wrote this semester this morning.' b.
[ The N-position within the NP in (ib) is occupied by an unpronounced "proform that is a predicate that denotes an n-place property," and remaining as a free variable it receives "its value from the context of utterance." The bound variable pro "saturates the first argument of the relational noun" (Shimoyama 1999: 168). NM no is semantically non-vacuous, its function being "similar to that of the definite article the in English' (Shimoyama 1999: 169). 7 For the record, Kim similarly states her Interpretability Condition for Korean kes IHRC (Kim 2008: 108, K's (14)) as follows: In order for an IHRC to be interpretable, for events e and e' the embedded and the embedding clauses describe, respectively, and a temporary state s, an individual k, and thematic relations P and Q, and a relation r, (k, e'); and (iii) r <s, e'>, where r ∈ {circumstantial, cause-effect, concessive}.
To quote Kim (2008: 101) further: "Informally, this formalization means that, for a sentence instantiating the IHRC construction to be judged acceptable, the embedded clause's event e must have a stative subpart s, which temporally overlaps with the embedding event e', and the relative clause's semantic head noun's denotation k must bear a thematic role (e.g., Agent or Theme), both in s and in e'." The interpretation of the external head pro in (4b) works as follows: (i) gesuto 'guest' is a theta-participant in the IHR theta-assigned by syookaisitekureta 'introduced'; (ii) the external head pro is a theta-participant in the matrix clause theta-assigned by syncategorematic noun hurumai 'behavior'; and (iii) the external head pro is linked to IH gesuto 'guest' by means of cospecification with the IH being a theta-participant in the IHR, the highest clause of the IHRC. 9 Notice that, with (4c) too, if the external head pro should be cospecified with Ayaka (with the sense 'I marveled at Ayaka for the elegant way she introduced three guests to Jun'ya'), it may serve as the IH, just like gesuto 'guest' in (4b), being an argument that is thetaassigned in the highest clause (IHR) of the IHRC. As it stands, however, the external head pro in (4c) with its intended sense cannot be cospecified with the IH gesuto 'guest' because it is not an argument that is theta-assigned in the highest clause (IHR) of the IHRC. It is critical to note that in this sense the cospecificational interpretive operation is island sensitive; it cannot "see" into a syntactic island. We may trace the failed interpretive process for (4c) with its intended sense as follows: (i) gesuto 'guest' is theta-assigned by syookaisitekureta 'introduced'; (ii) the external head pro is theta-assigned by hurumai 'behavior'; (iii) the external head pro looks for the individual available in the event discussed by the highest relative (IHR) within the IHRC, bearing some theta-role R with salient property P; but (iv) there is no such individual present in the relevant structural context; (v) hence, the interpretive effort fails (we will return to this issue of syntactic islands in Section 5).
It is in this manner that, in our pro-head analysis of Japanese IHRCs, the external head [ dp [ np pro] D] in (1), an anaphoric definite, picks out a unique linguistic antecedent, be it a case of co-variation involving a QP or an individual of type e. 10 To emulate Kim's (2007) brand of E-type strategy, what the cospecificational computation system would "see" is 9 Grosu (2010), referring to Watanabe (1992;, notes that an IH may be embedded at an arbitrary depth, citing (i) (=(39a) in Watanabe 2003, reformulated to suit our format). Here, I take it that the Complementizer (comp) to is transparent for cospecificational operation. The required interpretation proceeds as follows: (i) the IH kasetu 'hypothesis' is theta-marked by teiansi-ta 'proposed'; (ii) the external head pro is theta-marked by syncategorematic noun kekkan 'defect'; (iii) the external head pro is linked to the IH via cospecification. In this operation, the presence of comp is transparent, and not a factor. 10 If the antecedent of a pronoun is a referring expression, the configuration given in (1) may be considered to incur a violation of Condition C of the Binding Theory, which says that referring expressions must be free, meaning they should not be both c-commanded and coindexed by a noun phrase in the same binding domain. I follow Kameshima (1989) in taking advantage of Chomsky's (1986) multi-segmented category notion, given as follows: (i) (a) X c-commands Y iff X excludes Y and every element that dominates X dominates Y; and (b) X excludes Y iff no segment of X dominates Y.
the unique individual that bears some thematic role R in the event described by the clause that embeds the IHRC, having some salient property P. So this cospecificational operation picks out any individual that matches the description for pro as the IH, whether it is a proper name or a QP. Once some individual is picked out, pragmatic knowledge will come into play and the whole logical structure will be judged to make sense or not, depending on whether or not the resulting truth-conditions agree with our knowledge of the world. As noted, there are two important E-type analyses of Japanese IHRCs, Shimoyama (1999;2001) and Kim (2007). For the purpose of this paper, there is one critical difference between them. Shimoyama predicts that proper names do not appear as IHs, while Kim makes no such prediction. For Shimoyama, the use of proper names constitutes an empirically feasible test to show the superiority of her E-type anaphora analysis over the pro-head model in accounting for the Japanese IHRC. She states (1999: 174): "the fact that proper names do not sound good as internal heads falls out automatically in my analysis, in which what is recovered is a property, by means of N-proforms, but crucially not by means of pros, given the descriptive generalization that a restriction for the definite description in the matrix clause should obligatorily include a predicative part of the internal head." Thus, while Shimoyama acknowledges as a non-controversial fact that "pros in Japanese are known to have varying semantic functions including an E-type pronoun use" (1999: 170), she makes a firm stance that pro cannot be the external head of a Japanese IHRC.
Her treatment of the proper name is quite similar to G&H's, essentially ascribing to (2b) "Japanese IHRCs disallow definite referential heads (e.g., proper names)." Grosu & Hoshi (2016: 21-22) note that their analysis, while significantly different from Shimoyama's in a number of ways, nonetheless makes a comparable assumption, namely, that the definiteness operator σ (which in Shimoyama's analysis is denoted by no and in Grosu & Landman's 2012 analysis is denoted by the null D which heads the complex DP) applies to a property formed by intersecting properties internal to the IH with the property restricted by the remainder of the relative clause. Empirically, however, as we will see in Section 6, their claims are not entirely feasible. I therefore follow the brand of E-type strategy adopted by Kim on this particular issue of proper names. 11 3 Discourse function of Japanese IHRCs Shimoyama (1999: 179, footnote 28) states that, despite the similarities such as those pointed out in the preceding section, IHRCs cannot be equated with nonrestrictive EHRCs, because the content of an IHR supplies, albeit indirectly, a relevant restriction for the matrix clause, while a nonrestrictive relative only supplies additional information to its antecedent. Hoshi (1995: 127, fn.7) similarly argues that the property of a Japanese IHRC is "different from that of the nonrestrictive relative clause," and asserts that the IHRC "provides necessary information to identify the relevant entity or entities as the reference of the head noun phrase, which is itself a kind of anaphora lacking its inherent reference." So viewed, the external head DP in (1) constitutes a segment of a double-segmented category, thus obviating the IH from the Condition C violation. 11 The following factor however, needs to be kept in mind. A number of recent publications on Japanese and Korean IHRCs (e.g., Kim 2007;Grosu et al. 2013) assume that the IHRC formation in Japanese and Korean exhibit "the same characteristic properties" (Grosu et al. 2013: 1). One important fact about the Korean IHRC, however, is that its external head position is occupied by a grammaticalized formal noun kes (derived from fully lexical kes 'thing') acting as a quasi-pronominal. While I believe that Kim's (2007) analysis is accurate for Korean IHRCs, I speculate, based on Horie (1993), that: (i) this quasi-pronominal kes is not entirely vacuous semantically but is associated with residual semantic properties affecting their anaphoric strength, thus following Hopper's (1991) persistent principle, which says that when a lexical item undergoes grammaticalization to serve a grammatical function, the meaning and function of the derived item may be related to those of the source lexical item); (ii) kes, being not pro in a strict sense, cannot participate in repairing the island-sensitivity of an IHRC by means of coreference to be discussed in Section 5.4.
He comments that, in this sense, what is going on in the case of Japanese IHRC is "just the 'opposite' of that in the case of the nonrestrictive relative clause as defined by Kuroda." I propose to deal with the issue raised here by appealing to the difference in discourse function played by the occupants of the external head position. Thus, consider IHRC (5) in contrast with nonrestrictive EHRC (6). 12 Y-nom escape-try-attempt-pres no no shoulder-acc wasizukami-ni si-te, hikimodosi-ta. 12 eagle.hold-by do-ing pull.back-past 'Yuuji tried to get away, and the man clasped Yuuji's shoulders with both hands and pulled him back.' escape-try-attempt-pres Yuuzi no shoulder-acc wasizukami-ni si-te, hikimodosi-ta. eagle.hold-by do-ing pull.back-past 'The man clasped the shoulders of Yuuji, who tried to get away, with both hands and pulled him back.' In IHRC (5) the external head is occupied by anaphoric definite pro, whereas in nonrestrictive EHRC (6) it is occupied instead by non-anaphoric Yuuzi. I suggest that the difference in discourse function played by these two external heads fits the pattern delineated by Johnson's (2018: 7) following contrastive account (though he makes no reference to IHRCs): What non-anaphoric definites do is presuppose the existence of some individual that has a property. What anaphoric definites do is presuppose a property that their referent has. When you use a non-anaphoric definite out of the blue, you are inviting your listeners to put in the context that identifies it. When you use an anaphoric definite out of the blue, you are inviting listeners to put into the context a property, by which your referent can be identified.
To apply Johnson's (7) to example (5) and (6): (8) What anaphoric definite pro in the external head position of an IHRC like (5) does is presuppose the property (e.g., 'trying to get away') that its referent (e.g., Yuuzi) has. By the use of an anaphoric definite out of the blue, it invites listeners to put into the context such a property (e.g., 'trying to get away'), by which the referent (e.g., Yuuzi) can be identified.
Let me briefly describe our native speaker consultants. Consultants D, E, H, I, N, O and R are veteran Japanese language instructors teaching at colleges in the U.S. and Korea with either Masters or Doctorate degrees, some with considerable bodies of publication in the field. Consultant A is a computer programing specialist, whose Ph.D. dissertation in Computer Science has to do with a retrieval program involving transitivity alternation in Japanese. The rest are all professors in linguistics in the generative traditionexcept one, Consultant Q -at colleges and universities in Japan and the U.S. Consultant Q is the foremost authority in the research field of Japanese IHRCs in a non-generative orientation. Consultant M is a formal semanticist, who is also one of the co-authors of the best selling Japanese language textbook in the world today. And, Consultant K is undoubtedly the foremost authority in the field of English linguistics in Japan.
(9) What a non-anaphoric definite in the external head position of a nonrestrictive EHRC like (6) does is presuppose the existence of some individual that has a property (e.g., 'trying to get away'). By the use of a non-anaphoric definite out of the blue, it invites listeners to put in the context all that may identify what this non-anaphoric definite can be.
I suggest that it is in this manner that the IHRC in Japanese is distinct in its discourse function from that of the corresponding nonrestrictive EHRC counterpart.

Change IHRCs and gapless light-headed EHRCs
G&H claim that the construction type they call "change IHRCs" are genuine IHRCs, and need to be viewed as "existing independently of homophonous gapless light-headed EHRCs" (Grosu & Hoshi 2016: 26). In this section I argue against that view, taking the position that what they call the "change IHRC" is nothing but an instance of the gapless light-headed EHRC.

Do "Change IHRCs" exist?
Following Hoshi (1995: 122, fn.5), G&H consider a construction marked by slanted brackets in (10) to be structurally ambiguous between the two different constructions: (i) the "change IHRC" in which the rightmost "nominalizer" no is semantically vacuous; and (ii) the "gapless light-headed EHRC" in which the rightmost no is a semantically non-empty "pronominal no" that may in principle be replaced by a nominal form such as yatu 'guy, stuff', mono 'thing', or, in the context of (10), zyuusu 'juice'.
M-nom morning-in apple-acc squeeze-past no-acc afternoon-in hitoikide nomihosi-ta. in.a.gulp drink.up-past 'Mary squeezed apples (in the morning), and John drank it [= the juice produced by squeezing the apples] in a gulp (in the afternoon).' G&H's term "change IHRC" is due to the fact that its denotatum is characterized not by a syntactically represented nominal within the relative, but rather by the result of a process of change described by the relative: e.g., (10) does not say that John drank apples. For Tonosaki (1998: 154), who first called our attention to this construction type with the nomenclature "change relative," the salient entity (e.g., 'apples' in (10)) within the relative undergoes some property change denoted by the clause internal predicate; she maintains that the change relative is a pseudo-IHRC (see Matsumoto 1997 for a comprehensive discussion on the pseudo-relative constructions in Japanese), not an authentic IHRC. Grosu & Hoshi (2016: 5) claim, however, that the "change IHRC" (which on the surface may look like a regular gapless light-headed EHRC) is a variant of the true IHRC, though involving a change-of-state process. It is this assertion by G&H that we challenge in this section.
There are two assumptions, represented below as (11) and (12), that G&H rely on to make that claim: (11) (i) Adjectival modification directly preceding no works as a diagnostic tool, because it clarifies the categorical nature of no. As Grosu & Hoshi (2018: 5) put it, adjectival modification of no is "appropriate for pronominals but not for adverbial markers, nominalizers or complementizers"; and (ii) The "pronominal no" is intrinsically associated with a derogative sense, which renders it highly inappropriate to represent a venerable human.

(12)
The coordination test is sensitive to the categorical distinction separating the semantically empty Nominalizer no involved with an IHRC from the semantically non-empty pronominal no involved with a gapless light-headed EHRC.
G&H, assuming an acceptability contrast such as between sentences like (13) and (14), offer a tight-knit empirical argument to support their Change IHRC thesis, which I summarize in (15) before offering my rebuttal.
no]]-o (san-bai) non-da. squeeze-past delicious.looking one-acc three-cl drink-past 'John served wine and Mary, the delicious looking stuff that she got by squeezing apples, and Bill drank{it, three glasses of them altogether} (= the wine and apple juice).' J-nom wine-acc serve-past-nml and M-nom apple-acc sibottekure-ta] -no]]-o (san-bai) non-da. squeeze-past -nml-acc three-cl drink-past 'John served wine and Mary squeezed apples, and Bill drank {it, three glasses of them altogether} (= the wine and apple juice).' (15) (i) A regular EHRC and a gapless light-headed EHRC may conjoin (see G&H's (54)); (ii) Two non-change IHRCs may conjoin (see G&H's (55)); and (iii) Non-change IHRC may not conjoin with a gapless light-headed EHRC as the ill-formed status of (13) (G&H's (57)) demonstrates. In (13) the first conjunct is a non-change IHRC and the second conjunct is gapless light-headed EHRC (because no in the second conjunct is modified by adjectival oisisoona 'delicious looking'). (iv) Sentence (14) (G&H's (58)) is well-formed. Its first conjunct is a non-change IHRC and the second is exactly the same as that of (13) -except that it lacks the adjectival modification. This fact should point to "the conclusion that the second conjunct can only be a Change IHRC," because if it were a gapless lightheaded EHRC, the coordination test would have marked (14) to be illegitimate as is the case with (13) (Grosu & Hoshi 2016: 26).
But, empirical facts clearly do not support the above argumentation. Like so many cases of IHRC examples in Japanese, the acceptability judgments involved are subject to many contextual pragmatic factors, as Kuroda (1998: 19) points out (recall our discussion in Section 1.2). In the case of (13), the construction seems to improve in acceptability if both conjuncts appear to coordinate to enhance a particular theme such as 'the loving care taken by the hosts to entertain the guests' as in (16) no]]-o non-de yatto kuturoi-da. 13 squeeze-past delicious.looking no-acc drink finally relax-past 'John served wine and Mary the very delicious looking stuff that she made by squeezing apples, and we relaxed finally.' Here is another example, which is also comparable to (13)  These examples demonstrate that G&H's critical assessment (15iii) ("a non-change IHRC may not conjoin with a gapless light-headed EHRC") is on the wrong track. The upshot of all this is that, contrary to G&H's claim, the coordination test does not necessarily tease out the construction type "change IHRC" as an authentic IHRC distinct from the gapless lightheaded EHRCs. Accordingly, there is no proof that a distinct category "change IHRC" exists.
To pursue this issue a little further, consider example (18) 15 blanket-by wrap-ing bury-past 'The villagers wrapped the horrendously shocking remains of their beloved village chief, who had been tortured and killed by terrorists, and buried (the corpse).' In (19), which clearly identifies the 'village chief to be loved and respected,' the presence of a modifying adjective mattaku hisanna arisama-na 'so horrendously shocking' occasions no conflict in register with the external head pro (their "pronominal no"). Note that in (19), pro can be replaced by (sono (sontyoo no)) igai '(that (village chief's)) corpse', but not by sono sontyoo '(that) village chief'. 16 So, the construction in question in (19) is a change 14 "OK": A、F、J、M.
15 "OK": A、H、L、M、N、Q. "?": B、C、D、F、E、I、O、P、R. "*": G、J、K. 16 With sono sontyoo 'that village chief' in the external head position, the sentence would constitute what Erlewine & Gould (2016) call "doubly-headed relative," which shows remarkable similarities to IHRCs in many significant respects. We return to this issue in Section 7.
relative, an instance of the gapless light-headed EHRC, which involves a change-of-state process. And, its well-formed status shows that the external head pro, whether or not it is modified by an adjective, is not intrinsically associated with a derogatory sense. This fact pre-empts G&H's point expressed in (15iv). For, there is in fact no basis for the second conjunct of the pivotal example (14) to be accorded any special status as a "change IHRC."

Pronominal no and the derogatory sense
Let us now turn to the source of the derogatory sense itself that G&H associate with their "pronominal no." Consider the following pairs:  (20) and (21) has been widely assumed (e.g., Ito 1986;Hoshi 1995;Tonosaki 1998) to be a semantically non-empty pronominal intrinsically associated with a derogative sense which becomes obvious when used to represent a human as it creates dissonance with the use of honorific as in (20b) and (21b). A disconcerting aspect of this category "pronominal no" is its ad hoc character: i.e., the sole supporting evidence for its categorical identity is that it conveys a derogative sense when it refers to a human. I therefore follow Kitagawa (2005Kitagawa ( : 1259Kitagawa ( -1260 here, assuming that semantically empty "prenominal modification marker" (MOD) no is introduced as in (3i). So, the source of the derogatory sense in question must be associated somehow with the use of pro in the particular context such as provided by (20) and (21). What is critical with these pro's is that they are associated with arbitrary references (i.e., equivalent to English indefinite pronoun one). It is precisely in such contexts that pro pragmatically acquires a derogatory sense. As Kuroda perceptively notes elsewhere (1965: 73, fn.3): "when one uses an indefinite noun to refer to an object, one conceives of that object as an arbitrary representative of the category denoted by that noun." The derogatory connotation in question is thus a pragmatic effect of using an indefinite pro to refer to a specific human individual (or individuals). The pro's that appear in the external head position in (20) and (21) are, therefore, merely pro's with arbitrary references, far from being, for instance, "non-anaphoric definite." What is at work, therefore, is the following pragmatic principle: A sense of deference cannot be read into pro if it represents an arbitrary reference.
Kitagawa: The pro-head analysis of the Japanese internally-headed relative clause Art. 62, page 13 of 31 This explains why the anaphoric definite [ dp [ np pro]] in the external head of an IHRC is never associated with any such derogatory sense. Nor is there any mystery that the same situation obtains with a gapless light-headed EHRC such as (19), wherein the external head pro is interpreted with a specific and unique identity.

Two types of pros in the external head positions
With the issue of the derogatory sense out of the way, we may ask a new question: What kind of creature is the "light-head" pro in the external head position of a gapless lightheaded EHRC? How can we account for some critical difference between the pro of this kind and the pro in the external head of an IHRC? With the light-head pro there is no formal link. And this pro may be replaced by such lexical items as yatu 'stuff', mono 'thing', or any contextually suitable lexical item like zyuusu 'juice' in (10) or bakamono in (20), thus contrasting radically with the case of the pro head of an IHRC. Hoshi (1995) dealt with the light-headed gapless EHRC involving no formal link and, following Cooper (1979), postulates a free variable in his E-type strategy, the value of which is to be determined by the context of use, in terms of which the exact identification of the "salient entity" in question can be made precise without employing such a device as co-indexation. This approach may be proper for change relatives. Notice, however, that with this "Cooperian E-type strategy" the requisite property of the external head pro is that it works as a receptor -and not as an anaphor -for the "salient entity." Thus, in the theoretical framework identified in (1) with the formal link clearly specified, Hoshi's Cooperian E-type strategy cannot be recognized as representing cospecificational/E-type anaphora operation involved with Japanese IHRCs. In this I concur with Tonosaki in maintaining that the change relative is not an authentic IHRC. And I have nothing to say as to whether or not the change relative could be characterized as a nonrestrictive relative.
With this preamble, let us compare the following two sentences: K-nom apple-acc squeeze-past no thing juice-acc non-da. drink-past 'Ken squeezed apples, and Naomi drank {it, the stuff, the juice}.' I maintain that there are two different types of pros involved here as follows: (25) a. The external head pro of an IHRC is anaphoric definite; b. The external head pro of a gapless light-headed EHRC is non-anaphoric definite.
The pro in (25b) is a part-whole anaphora in Schwarz's (2009) sense (I got into a car. The steering wheel was broken. Car ⇒ steering wheel), which uses the weak definite article in German. In Schwarz's analysis, this type of bridging is non-anaphoric. 17 Applying Johnson's (2018: 7) dictum (7) to this pair, we have the following two distinct stories, differentiating the IHRC and the gapless light-headed EHRC: (26) What the anaphoric definite (i.e., pro in IHRC (23)) does is presuppose a property (i.e., 'Ken's buying (it)') that its referent (i.e., ringo 'apples' = pro) has. By the use of an anaphoric definite out of blue, it invites listeners to put into the context such a property ('Ken's buying'), by which the referent ('the apples') can be identified.
(27) What the non-anaphoric definite (i.e., pro in gapless light-headed EHRC (24)) does is presuppose the existence of some individual (i.e., 'apples') that has a property (i.e., Ken's squeezing (them)). By the use of a non-anaphoric definite out of the blue, it invites listeners to put in the context all that may identify what this nonanaphoric definite can be.
The IHRC and the gapless light-headed EHRC can thus be distinguished in terms of the different types of the external pro-heads involved. This contrastive analysis comes naturally with our pro-head analysis of IHRCs and gapless light-headed EHRCs, the fact of which enhances the credibility of our approach.

Island constraints
A construction such as marked by the slanted brackets in (28) (see G&H's (2a)) is structurally ambiguous, as G&H observe: it may be an IHRC as in (29a), or it may be a string-wise homophonous adverbial as in (29b)  With (i), involving the Dative Case marker ni that is requisite in a selected group of verbal predicates including taiatari-o-kurawase 'tackle' (see Kuroda 1978: 42), it is abundantly clear that there are undeniable prosodic and semantic differences between the IHRC (ia) and the adverbial (ib). There are many cases (particularly those involving accusative case marker o), however, where some formal diagnostic tests are called for to isolate real IHRCs from the corresponding adverbials masquerading as IHRCs. Kitagawa: The pro-head analysis of the Japanese internally-headed relative clause Art. 62, page 15 of 31 b. Anthony-wa [[doroboo-ga huta-ri nige-ru]-no]-o pro tukamae-ta. ⇐ Adverbial 'As two thieves were running away, Anthony caught them.' G&H maintain that homophony arises due to the fact that the suffix -no, which has many functions in Japanese but functions as a semantically empty nominalizer (NML) in IHRCs, typically bearing a Case marker in such cases, sometimes gives rise to affixal sequences homophonous with adverbial markers. The core idea espoused by G&H is proposed by Grosu & Landman (2012) and Landman (2016) as follows: the IH is coupled with a 'co-argument', which plays, in the set of eventualities as the IH, the same thematic role, so that the two co-arguments define in effect a single thematic participant. This co-argument is contained in a PP adjoined to the IP that most immediately contains the IH, and constitutes the syntactic position from which the null operator is launched. Given the local relation between the two co-arguments, it follows that with respect to any island either both are internal to that island, or both external. Therefore, deviance is automatically predicted whenever the IH is internal to an island, because the movement of the null operator will necessarily cross that island's boundary. (Grosu & Hoshi 2016: 17-18) It takes me too far afield to explain the formal apparatus Grosu & Landman (2012), Landman (2016) and G&H assume in characterizing Japanese IHRCs. For the purpose of this paper, let me take advantage of their claim that there are two kinds of telltale signs -which are identified as (2a) and (2b), now repeated as (30a, b) -that would unmistakably lead us to the "true Japanese IHRC." (30) a. The IHRC is strictly sensitive to island constraints -in particular Complex NP Constraint (CNPC) and Adjunct Island Constraints (AIC) (Grosu & Hoshi 2016: 14-18); b. The IH needs to be (existentially or otherwise) a quantified nominal with relative-internal scope. It cannot be a definite referential expression (in particular a proper name) (Grosu & Hoshi 2016: 21-24).
I examine the claim made in (30a) in this section, leaving the case of (30b) to Section 6. If these claims should turn out to be empirically unfeasible, the theoretical stance upon which these claims are made must also be unfeasible.
As G&H rightly point out (and I follow their lead), we need to identify the context in which only IHRCs are allowed, in order to identify the formal characteristics of IHRCs. They claim (Grosu & Hoshi 2016: 12) that there are two diagnostic tests to tease out the genuine IHRC unmistakably: Only an IHRC can materialize as a "genitive argument" (this, in our terms, is the case of an IHRC licensed by a syncategorematic noun); and b. Only an IHRC can occur as a "split headed relative construction" (SHRC).
Since G&H maintain that either of these tests would identify a real IHRC, I will not consider the (31b) option. The term "split" there refers to the fact that a numeral that may in principle occur within an IH may also occur outside CP immediately to the right of the Case marker. As a diagnostic test with regard to proper names (to be discussed in Section 6), however, it is of questionable value. For, with specification of proper names, the number of persons involved is obvious; an SHIC case would violate the Gricean maxim of quantity ("Be parsimonious"). Landman (2016: 5) rightly notes that there are many IHRCs whose acceptability status is unstable, judgments ranging between totally acceptable, somewhat odd, and totally unacceptable. He asserts, however, that "there is no such variability" with respect to an example like (32) (L's (7b)) where the IH atarasii kasetu 'new hypothesis' is inside a syntactic island (complex noun phrase): it is "judged infelicitous by everybody" ( Landman 2016: 5). Landman's assertion is based on his view that the IHRC involves "null operator movement cum lambda abstraction over the operator 'trace'," as cited earlier (Grosu & Hoshi 2016: 17-18). In (33), where the IH is isolated in a syntactic island, such movement operation is blocked. That is to say, given the theoretical construct he assigns for Japanese IHRCs, the syntactic structure of the following form (now represented in our format), where the Q-Assigner is a syncategorematic noun, should be invariably illicit; the conclusion, he states, is fully collaborated by G&H who present extensive discussion eliminating challenges, showing "beyond doubt" that a Japanese IHRC thus violating an island constraint is invariably illicit (Landman 2016: 5-6).

Empirical facts
Landman's assertion, however, is not borne out empirically. The fact is that Japanese IHRCs violating an island constraint are not necessarily judged to be illicit "beyond doubt." Rather, the result of questionnaires received from eighteen linguistically sophisticated native speakers' judgment of IHRCs bearing the structural pattern (33) reveal that these IHRCs' acceptability status is unstable, subject to idiosyncratic variation, judgments ranging between totally acceptable, somewhat odd, to totally unacceptable. While many native speakers indeed support G&H's injunction (2a) ("Japanese IHRCs are strictly island-sensitive"), there are those who accept some counter-examples to (2a) -thus revealing an interesting phenomenon that requires explanation. What is called for is an analysis that may account for how these "repairs" are made. And, in this respect, the pro-head approach proves to have such a recourse, whereas a G&H style analysis does not.
IHRCs given in (34) ~ (36), for example, which bear the structural pattern (33) involving island constraint violation, are NOT "judged infelicitous by everybody" (Landman 2016: 5). They are subject to idiosyncratic variation, as our statistics in footnotes (19) ~ (21) Kitagawa: The pro-head analysis of the Japanese internally-headed relative clause zituyoosei-ni imasaranagara kantansi-ta. 19 no wide.ranged applicability-at again marvel.at-past 'I again marveled at the wide-range applicability of that (seemingly) preposterous solution, which the engineer, who is now a big shot in an American company, proposed.' ]] no tane]-ga continue.must-pres.and suffering.be-pres no no seed-nom haittei-mas-u. 21 contain-polite-pres 'Contained in this bag are the seeds of cotton that do not (prematurely) germinate, for which the farmers must pay a huge amount of money every year to the company that developed them.' Notice, in this regard, that a case like (37), which does indeed involve WH-Operator movement in a structural context of the CNPC, is illicit for an overwhelming number of people, presenting sharp contrast with the comparable IHRC (38). 192021

(37) *Sorede, [[kyoozyu-ga [ Complex np [dare-ga kai-ta]
ronbun]-o hometei-ta] so.then professor-nom who-nom write-past paper-acc praise-past no]]-ga kondo zyosyu-de saiyoos-are-ru kotoninat-ta no? 22 no-nom now instructor-as hire-pass-pres come.to.be-past Q (Intended: 'Who the Professor praised the paper he had written has been appointed as an instructor now?') ronbun]-o professor-nom that grad.student-nom write-past paper]-acc hometei-ta] no [pro i ]]-ga kondo zyosyu-de saiyoosare-ru kotoninat-ta. praise-past no -nom now instructor-as hire.pass-pres come.to.be-past 'That graduate student of whom the professor praised the paper he had written has been appointed as an instructor.' The Adjunct Island Constraint (AIJC) presents a similar picture. The following example in which the IHRC is a "genitive argument" may be odd and taxing to the imagination but not "judged infelicitous by everybody." 2223 no student-nom that important.be hypothesis-acc kokusaikaigi-de happyoosi-te koohyoo-o hakusi-ta] node] international.conference-at present-ing popularity-acc gain-past because Yano-sensei-ga sukkari otikondesimat-ta] no [pro i ]] no syooryaku-nasi no Y-pro.-nom greatly get.depress-past no no omission-free no purinto-ga kore des-u. 23 copy-nom this be-pres 'This is a complete copy of the important hypothesis, concerning which Prof. Yano was so greatly depressed because his rival's student presented it at an international conference and received favorable responses.' To reiterate, the point I want to make is not that these sentences are perfectly acceptable, but that their felicity status is not consonant with Landman's (2016: 5-6) assertion that native speakers would invariably consider them to be illicit "beyond doubt." The fact 22 "OK": "?": J、O. "??": M. "*": A、B、C、D、F、G、E、H、I、K、L、N、P、Q、R.
The following example, in which the IH is replaced by nani 'what' is illicit, as expected: no student-nom that what-acc international.conference-at present-ing koohyoo-o hakusi-ta] node] Yano-sensei-ga sukkari otikondesimat-ta] no [pro i ]] no popularity-acc gain-past because Y-prof.-nom greatly get.depress-past no no syooryaku-nasi no purinto-ga kore des-u. omission-free no copy-nom this be-pres (Intended: 'This is a complete copy of what, concerning which Prof. Yano was so greatly depressed because his rival's student presented it at an international conference and received favorable responses?') that these sentences exhibit "a systematic variation among speakers" with respect to their acceptability status indicates quite clearly that, with Japanese IHRCs, we are NOT dealing with null operator movement as Landman and G&H claim.

How repairs are done
The question we should ask now is why these IHRCs are acceptable at all. Recall that we specified that the IH isolated in a syntactic island cannot be part of our cospecificational operation. Consider again the structural representation of IHRC in (34) and (36) repeated here as (40) and (41) zituyoosei-ni imasaranagara kantansi-ta. wide.ranged applicability-at again marvel.at-past 'I again marvel at the wide-range applicability of that (seemingly) preposterous solution, which the engineer, who is now a big shot in an American company, proposed.' farmers-nom that germinate-not-pres cotton-acc kaihatusi-ta] kaisya]-ni mainen tagaku no okane-o haraidevelop-past company-to every.year huge no money-acc paytuzuke.nakereba.narazu, kurusinde.i-ru] no [ DP pro i ]] no tane]-ga continue.must-pres.and suffering.be-pres no no seed-nom haittei-mas-u. contain-polite-pres 'Contained in this bag are the seeds of cotton that do not (prematurely) germinate, for which the farmers must pay a huge amount of money every year to the company that developed them.' The external head pro's here are theta-marked by syncategorematic noun zyuuyoosei 'applicability' in (40) and tane 'seed' in (41). But, concerning the referent of pro's, the computation system cannot "see" either the IH sono toppina kaiketuhoo 'that preposterous solution' in (40) nor the IH sono hatugasi-nai wata 'the cotton (the seeds of) which do not (prematurely) germinate' in (41), because those IH's are isolated in a syntactic island. An IHRC, however, still requires that a formal link be established between the IH and the external head, to meet the specification of (1). So, some repairs are necessary for those IHRC cases to make any sense. And, the acceptability of (40) and (41) reveals that such a repair is available for those native speakers who consider (40) and (41) to be not necessarily illicit "beyond doubt." The same story goes with (35) as well.
So, how is the repair done? I suggest that it is accomplished by means of "coreference" (Reinhart 1983), which is not encoded in syntax. We can test this hypothesis because referential properties are prerequisite for coreference (Dechaine & Wiltschko 2002: 420). If they are cases of coreference, the replacement of the referential antecedents by indefinite expressions would render those examples incoherent. This prediction is borne out, as exemplified by such "indefinite" versions of (34)/ (40) These reasonably felicitous examples, though certainly subject to idiosyncratic variation, contradict G&H's claim (45b). 27 I must admit, however, that there are two types of native speaker reactions to examples like those given above. Two of my native speaker consultants, L and N, have commented to me that, while they give "OK" (i.e., "the sentence is perfectly acceptable") or "?" (i.e., "the sentence is somewhat odd but more or less acceptable") to (5)/(47), (48) and (49), they would normally be inclined to reject sentences involving proper name IHs. On the other hand there are definitely native speakers, including Consultant M (p.c.) and the present author, who accept proper name IHs without any qualms. So, we have an issue to deal with: How can this inconvenient fact be accounted for?
Let me first of all, however, present the argumentation employed by G&H concerning the contrast between IHRC (50) and the corresponding nonrestrictive EHRC (51) (see Grosu & Hoshi 2016: 21-23, their (44b)/(46b) and (48) G&H contend that the IHR in (50) in its entirety has the same denotation as the IH, namely 'Lucky' or 'that gray cat'. They thus suggest that the infelicity of IHRC (50) "is due to the definite referential status of IH, which makes the contribution of the property defined by the remainder of the relative clause vacuous" (p.22). To quote G&H (2016: 22-23) further: What this means in effect is that the IH could have been used instead of the relative, the remaining material within the latter having at most the status of a presuppositional check. Such a meaning can be straightforwardly expressed by means of an appositive EHRC [e.g., (51)], but not by means of an IHRC. I cite their claim (Grosu & Hoshi 2016: 22, their (47)) in (52): 27 "OK": M、Q.
(52) CP minus the internal head must contribute in a non-vacuous way to the building up of the DP interpretation. Hence, the infelicity of IHRC (50), which "is due to the definite referential status of IH." A natural question that presents itself here is the following: Why, given this condition, aren't examples like (5)/(47), (48), and (49) utterly infelicitous? I do not profess to have a satisfactory explanation for this as yet. Let me take the following position for the present. There are two native speaker types, I and II. With the native speaker type I, represented by native speakers such as Consultants L and N, the discourse roles of the protagonists associated with the proper names as in (5) (52) nor the concept of discourse role aspect of proper names figure as a critical factor in the acceptability judgment of those sentences. And Kim's (2007) brand of E-type strategy adopted in Section 2 applies straightforwardly.
7 Doubly-headed relatives Erlewine and Gould (2016;E&G hereafter) present an interesting analysis of a heretofore understudied construction type they refer to as the "doubly-headed relative clause" (DHRC hereafter). In their view, Japanese relative clause constructions come in three types as in (53)  E&G claim that, syntactically, their "novel DP head-raising derivation for relative clauses that takes advantage of the Copy Theory of movement and the late-merger of relative clauses" allows for a unified account of these three types of relative clause constructions. In this section, I present a brief review of their claim. I will then examine some empirical facts about the DHRCs, to see how this construction type relates to our pro-head analysis of IHRCs.

c. Sue-ga [[koinu no Poti i -ga ikakus-ru
Mikeh j -o kamawootosu-ru] sono S-nom puppy no P-nom threaten-pres M-acc engage.try-pres that Poti i to Mikeh j ]] no sippo-o tukan-de hikihanasi-ta. P and M no tail-acc grab-ing separate-past 'Pochi, a puppy, tried to engage Mikeh, who threatened him, into a play, and Sue separated them by grabbing Pochi's and Mikeh's tails.'

Structural similarity between the IHRC and the DHRC
Recall that felicitous IHRC examples discussed in Section 5.4 (e.g., (34) ~ (36), (39)) all take the form given in (64a) (cf. (33)), while the corresponding DHRC would assume the form as in (64b) More generally, we may represent the contrastive picture as follows without limiting ourselves to the cases where the presence of syntactic island factors in prominently: (65) a. IHRC (see (1) It is tempting to suggest that the Japanese IHRC (65a) is the pronominal external head version of the DHRC (65b), granting at the same time that the DHRC has its own discourse function distinct from the IHRC. Any responsible characterization of DHRCs visà-vis IHRCs, however, must await more rigorous future investigations. Nonetheless, it is clear even at this point that there exists a very significant meaning and acceptability parallelism, coupled with distinct discourse functions, between the IHRCs and the corresponding DHRCs. And this parallelism is what would be predicted by our pro-head analysis of IHRCs, whose external head pro is anaphoric definite, as discussed in particular in Sections 2, 3, and 5.4.

Conclusion
I advanced in this paper the view that the Japanese IHRC is a nonrestrictive relative clause whose external head position is occupied by pro as represented in (1) acting as an E-type pronoun. In this view, as argued in Section 2, the external head [ dp [ np pro] D] is an anaphoric definite, to pick out via cospecification a unique linguistic antecedent, be it an individual of type e or a case of co-variation involving some quantificational context. The IHRC's nonrestrictive status is defended in Section 3 by appealing to Johnson's (2018) contrastive account given in (7) regarding the anaphoric definite status of the IHRC's external head pro and the non-anaphoric definite status of the nonrestrictive EHRC's lexical external head. The pro-head analysis model of a Japanese IHRC fits well into this picture, indirectly supporting the correctness of the model itself. Following this line of analysis, I demonstrated in Section 4 that the difference between the IHRC and the gapless lightheaded EHRC is reducible to that between the anaphoric definite pro (for the IHRC) and the non-anaphoric definite pro (for the gapless light-headed EHRC) both occupying the external head position. To further promote the pro-head analysis, I presented arguments in Section 5 against the view represented by G&H and Landman (2016) that Japanese IHRCs, involving a null operator movement, are strictly island-sensitive. I have shown in Section 5.4 that the pro-head analysis is the only model available at present to account for the coreference phenomenon, which may, even crossing a syntactic island barrier, relate the external pro head to the IH. In Section 6, I attended to the problem of proper names occurring as the IH of an IHRC, suggesting directions where future investigations may profitably follow. And, in Section 7, I presented a brief account on the meaning and acceptance parallelism between IHRCs and the corresponding DHRCs -the parallelism that would be predicted by our pro-head analysis of an IHRC. In pursuing some of those issues, I focused primarily on G&H because their empirically testable approach at the level of high clarity provides concrete venues to examine sundry and significant phenomena associated with Japanese IHRCs.