1.1. Scrambling: the data

The problem posed by scrambling in Japanese is not merely the apparently free permutability of the sequences of noun phrases that occur before the rigidly ordered final verb complex (as in (1)–(3)),Footnote ⁹ nor even that there is long-distance dependency potential, but rather that there can apparently be more than one such expression arbitrarily far from the verb to which they have to be correlated to receive interpretation – as in (5)–(6), both of which allow an interpretation in which both the object and the indirect object expression are interpreted as arguments of the embedded verb:Footnote ¹⁰

1. (5) shorui-o supai-ni keisatsu-ga jaanarisuto-ga watashita to koohyooshita document-acc spy-dat police-nom journalist-nom handed comp reported ‘The police reported that the journalist had handed the document to the spy.’

1. (6) supai-ni shorui-o keisatsu-ga jaanarisuto-ga watashita to koohyooshita spy-dat document-acc police-nom journalist-nom handed comp reported ‘The police reported that the journalist had handed the document to the spy.’

While (4) showed that there may be one long-distant dependent term construed from a left-periphery position in Japanese, (5)–(6) show that there may be more than one such NP, and in either order. Moreover, whenever there is more than one such NP construed cross-clausally as removed from some source site, these are subject to a stringent additional restriction: they have to be construed as local to each other, despite being an arbitrary distance away from any verb which can provide the head to which either of them is argument. Thus in (5) and (6), the dative NP can be interpreted either relative to the matrix clause (relative to the clause of which keisatsu-ga is subject), or relative to the subordinate clause. Of these two interpretations, it is only the latter that is relevant to the issue of multiple long-distance dependency. And in such cases, there is no possibility of interpreting supai-ni and shorui-o as in distinct clauses: they have to be interpreted together. Thus in the following examples, there is no interpretation in which the dative NP is interpreted relative to the subordinate clause at the first level of embedding, while the object NP is interpreted relative to the second level of subordination:

1. (7) supai-ni shorui-o keisatsu-wa jaanarisuto-ga watashita to itta to koohyooshita spy-dat document-acc police-top journalist-nom handed comp said comp reported ≠‘The police reported that the journalist said to the spy that he had handed the document to them.’

1. (8) shorui-o supai-ni keisatsu-wa jaanarisuto-ga watashita to itta to koohyooshita document-acc spy-dat police-top journalist-nom handed comp said comp reported ≠‘The police reported that the journalist said to the spy that he had handed the document to them.’

That is, leaving aside the irrelevant matrix construal of the dative NP in each case, the only additional interpretation which (7) and (8) can have is that ‘The police reported that the journalist said (to them) that he had handed the document to the spy’. And this is so even if, as in (7), the dative expression is at the left periphery, hence higher than the following object-marked expression, under normal assumptions. Thus if two left-peripheral NPs are taken not to be arguments of the matrix verb but of some subordinate predicate, they must both be taken to be co-arguments of the same predicate. That this is indeed a structural condition is made clear by the fact that despite the plausibility of the precluded interpretation, it is signally not available.

These facts are essentially the same in Korean. The short scrambling data are identical in the two languages; we give two examples in both, by way of illustration.

1. (9) Jina-ka pizza-lul Mina-hanthey Pizza-Express-ese sa-cwuessta [Korean] Jina-nom pizza-acc Mina-dat Pizza-Express-at buy-gave ‘Jina bought-to-give pizza to Mina at Pizza Express.’

1. (10) pizza-lul Mina-hanthey Jina-ka Pizza-Express-ese sa-cwuessta pizza-acc Mina-dat Jina-nom Pizza-Express-at buy-gave ‘Jina bought-to-give pizza to Mina at Pizza Express.’

1. (11) Jina-ga piza-o Mina-ni Pizza-Express-de katte-ageta [Japanese] Jina-nom pizza-acc Mina-dat Pizza-Express-at buy-gave ‘Jina bought-to-give pizza to Mina at Pizza Express.’

1. (12) piza-o Mina-ni Jina-ga Pizza-Express-de katte-ageta pizza-acc Mina-dat Jina-nom Pizza-Express-at buy-gave ‘Jina bought-to-give pizza to Mina at Pizza Express.’

There is also long-distance scrambling in the two languages, both for a single expression and for a pair of expressions (as illustrated in the multiple long-distance scrambling of (13)); again we provide data from both languages:

1. (13) pizza-lul Mina-hanthey Yuna-ka Jina-ka Pizza-Express-ese sa-cwuessta ko haysseyo [Korean] pizza-acc Mina-dat Yuna-nom Jina-nom Pizza Express at give-bought comp said ‘Yuna said that Jina bought-to-give pizza to Mina at Pizza Express.’

1. (14) piza-o Mina-ni Yuna-ga Jina-ga Pizza-Express-de katte-ageta to itta [Japanese] pizza-acc Mina-dat Yuna-nom Jina-nom Pizza-Express-at buy-gave comp said ‘Yuna said that Jina bought-to-give pizza to Mina at Pizza Express.’

The NPs in such a scrambled pair (pizza, Mina) may occur in either order without any essential difference in the interpretation of the pair as a multiple long-distance dependency sequence (which, as we shall see, is a problem for Koizumi Reference Koizumi2000); and again there is the same locality constraint imposed on multiple long-distance scrambling. The problem posed by this phenomenon is that non-locality of the dependency itself would seem to impose the necessity of positing independent processes correlating each such expression with the site of its construal; but this would leave unexplained the locality relative to each other which these expressions invariably display.

So far, the facts are relatively familiar (Koizumi Reference Koizumi2000, Takano Reference Takano2002, Saito Reference Saito2003).Footnote ¹¹ However, there is the additional complication: the asymmetry between Korean and Japanese judgements of long-distance scrambling. First, there is the general problem that Japanese speakers regularly report such data to be of borderline status. In Korean, however, no such borderline judgement is reported (Kiaer Reference Kiaer2007). This is because prosodic factors can be straightforwardly used as a disambiguation device, with syllable lengthening (a characteristic feature of Seoul Korean: see Jun Reference Jun2000) indicating a structural break between the expression so lengthened and whatever follows. The differences between the flexibility of scrambling in the two languages are, however, more than merely prosodic: there is a clear difference in terms of licensing for long-distance discontinuity, which affects the potential for multiple long-distance dependency pairs. ga-marking of subject expressions severely limits Japanese scrambling to only local variation, as reported by Saito (Reference Saito1985, Reference Saito1992). Korean subject expressions, on the other hand, freely allow long-distance effects. Thus (15), unless highly marked emphatic intonation is used, allows only the implausible reading in which the child reports the teacher's illness, despite the fact that this is contraindicated by the honorific form ossyatta, which is an inappropriate form for describing actions by a child.Footnote ¹² The Korean analogue (16), to the contrary, given the honorific form of the matrix verb, unproblematically allows the reading in which the leftmost subject is interpreted as the subordinate subject:

1. (15) ??sono kodomo-ga sensei-ga byooki-de gakkoo-ni ko-nai to ossyatta-hon [Japanese] that child-nom teacher-nom illness-of school-to come-not comp said ‘The child said that the head-teacher had not come to school because of illness.’

   ?≠‘The child, the head-teacher said because of illness had not come to school.’

1. (16) Jina-ka sensengnim-i apase hakkyo-e mot-wassta ko kure-si-ess-eyo [Korean] Jina-nom teacher-nom sick school-at neg-came comp say-hon-past-decl ‘The teacher said that Jina couldn't come to school because of illness.’

This difference affects the potential in the two languages for multiple long-distance dependency effects. Korean freely allows subject expressions to occur as one of a left-peripheral pair of expressions needing long-distance construal, with the pair occurring, as in local scrambling, in either order:

1. (17) tu hwanca-ssik-ul se kanhosa-ka Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo [Korean] two patient-dist-acc three nurse-nom Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses looked after two patients each all night.’

1. (18) se kanhosa-ka tu hwanca-ssik-ul Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo three nurse-nom two patient-dist-acc Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses looked after two patients each all night.’

Japanese, however, does not. There is no analogue to these Korean data in Japanese; speakers report that the only available interpretation of (20) is that the nurses say that Dr Kim looked after two patients all night,Footnote ¹³ whether the subject NP in question is initial in some putative long-distance dependent cluster of NPs, (19), or not (20):Footnote ¹⁴

1. (19) san-nin-no kangohu-ga kanjya-o hutari-zutsu Kim-sensei-ga hitoban-jyuu kanbyooshita to itta [Japanese] three-class-gen nurse-nom patient-acc two-dist Kim-Dr-nom night-during nursed comp said ‘Three nurses said that Dr Kim looked after two patients at different times all night.’

1. (20) kanjya-o hutari-zutsu san-nin-no kangohu-ga Kim-sensei-ga hitoban-jyuu kanbyooshita to itta patient-acc two-dist three-class-gen nurse-nom Kim-Dr-nom night-during nursed comp said ‘Three nurses said that Dr Kim looked after two patients separately all night.’

Described as a movement from some canonical ordering, the lack of permutability of a subject-marked expression out of an embedded position thus appears to be sustained in multiple long-distance dependency, and furthermore sustained irrespective of the relative ordering of the paired expressions. Even for those Japanese speakers for whom such an interpretation is marginally acceptable, the data are of sharply reduced acceptability compared to the non-inverted form of construal. So whatever the basis of the relative locality restriction in multiple long-distance dependency, the requirement of locality imposed on a subject-marked expression cannnot be defined merely in terms of its immediately containing environment. As we shall see, not only are these multiple long-distance dependency structures and the locally required pairing up of their construals not expressible in conventional frameworks other than by stipulation, but any move to assign such sequences to a nonstandard constituent, in order to retain the assumption of their movability and relative locality, faces the problem that there will no longer be any basis from which to predict that it is subject expressions within such left-peripheral structures that are precluded in Japanese.

1.2. Scrambling: recent analyses

This last problem is pertinent in evaluating two recent accounts of Japanese multiple long-distance scrambling within the minimalist framework: Koizumi (Reference Koizumi2000) and Takano (Reference Takano2002).Footnote ¹⁵ In an attempt to capture the relative locality constraint on multiple long-distance scrambling for the left-peripheral sequenced NPs, Koizumi (Reference Koizumi2000) argues for an analysis involving vacuous verb movement, with subsequent leftward movement of the ‘remnant’ propositional structure. Against this account, noting its problematic violation of the so-called Proper Binding Constraint, Takano (Reference Takano2002) proposes an alternative adjunct-plus-head analysis of the left-peripheral sequence of NPs, arguing that the leftmost NP in the pair (or triple) is first moved from some canonical ordering of arguments in the subordinate structure by a so-called process of Oblique Movement, to create an adjunct modification on the rightmost of the pair. As a result of this operation, the two NPs come to be contained within a structure immediately dominating both whose category matches the NP at the landing site, with regular A′ Movement then applying to the composite adjunct structure. The oblique movement process is distinct from other movement processes in creating a superstructure, so that neither of the NP daughters within this super-structure c-commands the remainder string (see Takano Reference Takano2002).

Neither of these analyses is unproblematic. The Koizumi account involves vacuous verb movement, a stipulation directly contrary to minimalist assumptions, effectively an unmotivated re-bracketing, with no feature-trigger either for this or for the process of so-called Remnant Movement that it feeds. Furthermore, as Koizumi himself notes, it predicts wrongly that anything other than canonical ordering of argument expressions in such left-peripheral pairs is not well-formed. Such a permutation of the canonical order could only be achieved by a sequence of three steps: (i) movement out of the clause of what is in the surface sequence the rightmost element in the sequence of NPs, (ii) vacuous movement of the verb, (iii) movement of the containing ‘remnant’ VP or IP across that already-moved NP. Yet this very type of configuration is known to need debarring because of cases such as (21); it was in order to exclude structures like this that Saito (Reference Saito1992) introduced the Proper Binding Condition.Footnote ¹⁶

1. (21) [*Hanako-ga t _i iru to]_j Sooru-ni_i Taroo-ga t _j omotteiru Hanako-nom be comp Seoul-in Taroo-nom think ‘[That Hanako is t _i]_j in Seoul_i Taroo thinks t _j.’

Amongst recent attempts to defuse this issue in cases where application of Remnant Movement apparently needs to be posited, Grewendorf (Reference Grewendorf2003) has argued that Move α is in fact a set of rules, each distinguished by a feature, and that Remnant Movement is possible as long as the second apparent application of Move α is feature-distinguishable from the first. Koizumi, noting that his own account wrongly predicts that non-canonical orderings in multiple long-distance dependencies should be very much less acceptable than sequences in which the expressions are in canonical order (see (7)–(8)), invokes just such a Grewendorf-style analysis. But this powerful auxiliary assumption that a core rule of the grammar is in fact an open-ended set of such rules, combined with the non-minimalist assumption of vacuous application of rules, is, at best, a challenge to develop a better account, which Takano (Reference Takano2002) purports to provide. To add to its problematic nature, the Koizumi account signally fails to provide a basis for capturing the restriction that subject expressions in Japanese must be interpreted locally as in (19)–(20): what is moved is a remnant category (VP, TP, or CP), so there are no grounds for precluding the incorporation of a subject NP within that long-distance-moved constituent in either order, as it is not the NP constituent itself but the containing constituent that is subject to movement.

The alternative adjunct analysis of Takano (Reference Takano2002) might seem to be more principled, but in the event, it is not. Even in its own terms, it is no more than a formal solution, as the supposed adjunct displays no adjunct-like behaviour other than that within the structure for which it is defined. In particular, as an adjunct, one would again expect that licence would thereby be created for pairs of subject and non-subject expressions just in case the subject is assigned adjunct status. Indeed, Takano's major argument for this adjunct form of analysis is that it provides a basis for explaining why the adverbial naze ‘why’ loses what are otherwise locality constraints on its construal when it occurs within an object–naze sequence. But this leaves this approach open to the same problem facing the Koizumi analysis. Any subject expression that undergoes such Oblique Movement would as a result be nested within the adjunct structure, and accordingly should no longer display locality restrictions specific to being in a subject configuration. And, according to the Takano analysis, both canonical and noncanonical ordering of the NP sequence must be subject to Oblique Movement in order to create the required composite adjunct category. It is, indeed, precisely this analysis which Takano proposes for the composite focus given by the canonically ordered sequence in (22) (with first the direct object NP moving to become an adjunct to the indirect object, then that composite adjunct DO–IO being adjoined to TP, and finally, the subject NP moving to become an adjunct to the DO–IO adjunct to yield the order S–DO–IO at the left periphery as itself a composite adjunct):Footnote ¹⁷

1. (22) ageta-no-wa John-ga hon-o Mary-ni da gave-nm-top John-nom book-acc Mary-dat is ‘JOHN gave the BOOK to MARY.’

Accordingly, (23) should similarly be able to have the interpretation as a fronted embedded clausal sequence, by the same compounded adjunct sequence of operations:

1. (23) John-ga hon-o Mary-ni Hiromi-ga ageta to itta John-nom book-acc Mary-dat Hiromi-nom gave comp said *‘Hiromi said that John gave a book to Mary.’ ‘John said that Hiromi gave a book to Mary.’

But (23) only allows the interpretation in which John-ga is the matrix subject: it cannot be interpreted with Hiromi-ga construed as the matrix subject. Quite generally, as a single moved constituent, any left-peripheral sequence of NPs that is construed as involving multiple long-distance dependency has to be analysed as involving successive applications of Oblique Movement. Yet, as in the Koizumi analysis, such a derivation provides no basis for predicting the unacceptability of any construal in Japanese in which a subject-containing constituent in a multiple long-distance dependency structure is reconstructed into a subordinate structure because such a subject has been re-analysed as an adjunct. In any case, as Takano himself notes, the account is incomplete, since, contrary to the strict relative locality observed by all such left-peripheral sequences, nothing in his adjunct account prevents an adjoined NP from further movement once long-distance movement of the created adjunct structure has taken place, thereby predicting that a clausal adjunct may intervene between two such fronted arguments and nevertheless be interpreted as modifying the matrix verb:

1. (24) ringo-o kinoo Bill-ni John-ga Mary-ga ageta to kiita apple-acc yesterday Bill-dat John-nom Mary-nom gave comp heard ‘John heard that Mary gave apples to Bill yesterday.’ *‘John heard yesterday that Mary gave applies to Bill.’

Because this interpretation is not available, with only the unstarred interpretation being possible, Takano suggests that it is merely processing factors that determine that only a subordinate clause interpretation of the adjunct is possible in cases like (24). Yet the judgement that matrix construal of the adjunct is precluded is no less robust than with other multiple long-distance dependency data for which he proposes the adjunct form of analysis.Footnote ¹⁸

This is by no means the end of the puzzle: there are semantic problems that are not predicted by either the Koizumi or the Takano account. In both languages there is the same high degree of variability as to what order of NPs is allowed within a single clausal sequence, and, the greater stringency of ga-marking in Japanese apart, whatever variability there is in the correspondence between local scrambling and interpretation is replicated exactly in the variable correspondence between NP ordering and interpretation within pairs of NPs in multiple long-distance dependency which the two languages allow. These data are the semantic counterpart to the problem for Koizumi that a switch in the order of NPs in such a left-peripheral sequence, contrary to his analysis, does not lead to a sharp drop in acceptability.

First, with object–subject order in both Japanese and Korean, there is ambiguity irrespective of choice of quantifying expression: for this sequence, there is more than one interpretation, both that in which the left-peripheral expression apparently takes wide scope over the remainder of the string, and that in which it does not:

1. (25) hotondo-no uta-o dareka-ga utatta most-gen song-acc someone-nom sang ‘Most of the songs, someone sang.’ (ambiguous: indefinite narrow/wide scope)

With subject–object ordering, by contrast, there is much less flexibility, with the additional complication that individual quantifiers impose additional restrictions. A canonical ordering of subject–object–verb appears to be ambiguous if the object expression is a pure indefinite (such as dareka (=‘someone’)) as in (26), but not if the subject expression is a pure indefinite and the object expression is incontrovertibly quantificational as in (27) and (28) (Kuroda Reference Kuroda1971, Kuno Reference Kuno1973, Hoji Reference Hoji1985, Saito Reference Saito1992, Tada Reference Tada1993):

1. (26) daremo-ga dareka-o aishiteiru everyone-nom someone-acc loves ‘Everyone loves someone.’ (ambiguous, narrow-scope preferred)

1. (27) dareka-ga hotondo-no uta-o utatta someone-nom most-gen song-acc sang ‘Someone sang most of the songs.’ (unambiguous)

1. (28) nwukwunka-ka taypwupwun-ui nolay-lul pwulessta [Korean] someone-nom most-gen songs-acc sang ‘Someone sang most of the songs.’ (unambiguous)

However there is a subtle interaction of pragmatic constraints, which shows that this restriction is not a rigidly structural one.Footnote ¹⁹ Speakers report (29) and its analogue in Korean to be ambiguous even with the subject–object ordering: (29) and (30) allow both an interpretation reflecting linear order and one departing from it, the availability of the inverse interpretation being clearly triggered by contingent information about the circumstances described that preclude the interpretation matching linear order, despite the default prevalence of context-dependent construal for left-placed nominals (see footnote 2).

1. (29) kangohu-ga subete-no kanjya-o monshinshita [Japanese] nurse-nom every-gen patient-acc interviewed ‘A nurse interviewed every patient.’

1. (30) kanhosa-ka motun hwanja-lul myenciephayssta [Korean] nurse-nom every patient-acc interviewed ‘A nurse interviewed every patient.’

The similarity between the two languages continues with plural quantification, though the syntactic facts are somewhat different. Both languages have a distributivity marker that fixes scope relativity; the minor syntactic differences between them do not disturb the generalisation that once scoping of a quantified expression relative to another such expression is encoded, the dependency does not require linear order to bring out the requisite interpretation. Thus both (31)–(32) and (33)–(34) require the interpretation in which two patients are construed relative to three nurses so that there are six patients in all.

1. (31) tu hwanca-ssik-ul se kanhosa-ka pamse-dongan tolpasseyo [Korean] two patient-dist-acc three nurse-nom night-during nursed ‘Three nurses looked after two patients all night.’

1. (32) se kanhosa-ka tu hwanca-ssik-ul pamse-tongan tolpasseyo three nurse-nom two patient-dist-acc night-during nursed ‘Three nurses looked after two patients all night.’

1. (33) kanjya-o hutari-zutsu san-nin-no kangohu-ga hitoban-jyuu kanbyooshita patient-acc two-dist three-class-gen nurse-nom night-during nursed ‘Three nurses looked after two patients all night.’ [Japanese]

1. (34) san-nin-no kangohu-ga kanjya-o hutari-zutsu hitoban-jyuu kanbyooshita three-class-gen nurse-nom patient-acc two-dist night-during nursed ‘Three nurses looked after two patients all night.’

In the absence of such a distributivity marker, in both languages, speakers contrarily strongly prefer to attribute group interpretations to paired numerical expressions, and again linear order appears not to affect this, so that all of (35)–(37) imply that there are just two patients whom just three nurses looked after all night:

1. (35) tu hwanca-lul se kanhosa-ka pamse-dongan tolpasseyo [Korean] two patient-acc three nurse-nom night-during nursed ‘Three nurses looked after two patients all night.’

1. (36) se kanhosa-ka tu hwanca-lul pamse-tongan tolpasseyo three nurse-nom two patient-acc night-during nursed ‘Three nurses looked after two patients all night.’

1. (37) san-nin-no kangohu-ga hutari-no kanjya-o hitoban-jyuu kanbyooshita [Japanese] three-class-gen nurse-nom two patients-acc night-during nursed ‘Three nurses looked after two patients all night. ’

1. (38) hutari-no kanjya-o san-nin-no kangohu-ga hitoban-jyuu kanbyooshita two patients-acc three-class-gen nurse-nom night-during nursed ‘Three nurses looked after two patients all night.

Bringing long-distance dependency considerations now into the picture, in both languages there is apparently conflicting evidence from quantifier construal as to the relationship between short and long scrambling. On the one hand, there is asymmetry between the interaction of quantifier and pronoun construal in short and long scrambling, suggesting that these must be distinct processes. If a quantifying expression precedes a pronominal in a non-canonical ordering within a simple clause, then that quantifier may but need not be construed as binding the pronoun as in (39) (the verb syomeishita in Japanese requires a dative-marked object). However, for the very same sequence of words, if the quantifying expression is part of a long-distance scrambling structure, a bound-variable interpretation is apparently precluded as in (40) (Saito Reference Saito2003, Reference Saito, Sabel and Saito2005):Footnote ²⁰

1. (39) dono hon-ni-mo sono hon-no tyosya-ga syomeishita every book-dat-also that book-gen author-nom autographed ‘Every book_i that book's_i author autographed.’ (indexical/bound-variable interpretation of sono both available)

1. (40) dono hon-ni-mo sono hon-no tyosya-ga Hiroto-ga syomeishita to itta every book-dat-also that book-gen author-nom Hiroto-nom autographed comp said ‘That book_i's author said that Hiroto autographed every book_j.’ (indexical interpretation of sono only)

Essentially the same distribution occurs in Korean, as illustrated by (41), which only allows the indicated interpretation:

1. (41) motun haksayngtul-ul ku haksayng-ui emma-ka Kim-sensengnim-i simhake ttayriesta ko cwucanghayssta [Korean] every student-acc the student-gen mum-nom Kim-teacher-nom severely treated comp insisted ‘That_i student's mum insisted that Teacher Kim treated every student_j badly.’ (indexical interpretation of ku only)

It might seem that this distribution can only be explained if some basis for asymmetry between short- and long-distance dependency forms of construal is posited. Yet, contrary to this, in multiple long-distance dependency the paired NPs at a long-distance remove from the verb with which they are construed are subject to exactly the same form of construal as in short-distance scrambling, the one wrinkle about Japanese ga-marking aside: in both languages, whatever flexibility or inflexibility there may be with pairs of left-placed NPs in simple clauses is replicated exactly in multiple long-distance scrambling. In Korean, this parallelism can be illustrated with any pairs of NPs whatsoever, in particular, given the scope-fixing properties of the distributivity particle -ssik (Choe Reference Choe1987). Thus, just as the distributivity marker determines narrow scope of the expression it is suffixed to in both (31) and (32), so it does in (42) and (43):

1. (42) tu hwanca-ssik-ul se kanhosa-ka Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo two patient-dist-acc three nurse-nom Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses each looked after two patients all night.’

1. (43) se kanhosa-ka tu hwanca-ssik-ul Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo three nurse-nom two patient-dist-acc Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses each looked after two patients all night.’

Equally, the construal of the paired NPs in (35) and (36) as independent of each other and with a group interpretation carries over directly to (44) and (45):

1. (44) tu hwanca-lul se kanhosa-ka Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo two patient-acc three nurse-nom Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses looked after a group of two patients all night.’

1. (45) se kanhosa-ka tu hwanca-lul Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo two patient-acc three nurse-nom Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses looked after a group of two patients all night.’

In Japanese, where the parallelism between short scrambling and multiple long-distance dependency scope effects can only be tested with non-subject pairs, in a simple clausal sequence, as we've seen, if the singular indefinite follows a quantified expression, both narrow and wide scope construal for the indefinite are freely available:

1. (46) san-nin-no supai-ni shorui-o jaanarisuto-wa watashita [Japanese] three-class-gen spy-dat document-acc journalist-top handed

   1. (a) ‘The journalist handed the three spies a document.’

   2. (b) ‘The journalist handed a document to the three spies.’

If, furthermore, in such a pair, the singular indefinite precedes the quantified expression as in (47), despite a preference for interpretations that follow linear order, this is easily set aside in cases where contingent information dictates the plausibility of one interpretation over another. For example, though the preferred interpretation of (47) is that one document is handed over to a group of three spies, (48) and (49) allow distributed and wide scope interpretations for the singular indefinite equally, in virtue of what is presumed about spies and their illegal activities:

1. (47) shorui-o san-nin-no supai-ni jaanarisuto-wa watashita document-acc three-class-gen spy-dat journalist-top handed ‘The journalist handed one document to three spies.’

1. (48) nise-no pasupooto-o san-nin-no supai-ni CIA-wa watashita forged-gen passport-acc three-class-gen spy-dat CIA-top handed ‘The CIA handed a forged passport to three spies.’

1. (49) san-nin-no supai-ni nise-no pasupooto-o CIA-wa watashita three-class-gen spy-dat forged-gen passport-acc CIA-top handed ‘The CIA handed three spies a forged passport each.’

This pattern strikingly replicates itself in multiple long-distance scrambling structures. Just as there is a preferred interpretation of just one document in (47), so there is in (50):

1. (50) shorui-o san-nin-no supai-ni keisatsu-wa jaanarisuto-ga watashita to koohyooshita document-acc three-class-gen spy-dat police-top journalist-nom handed comp reported ‘The police said that the journalist had handed just one document to three spies.’

And when this preference for linear order is over-ridden, as in (48), so in (51), it is also the reverse form of interpretation which is the preferred interpretation:

1. (51) nise-no pasupooto-o san-nin-no supai-ni keisatsu-wa jaanarisuto-ga watashita to koohyooshita forged-gen passport-acc three-class-gen spy-dat police-top journalist-nom handed comp reported ‘The police reported that the journalist handed a forged passport to each of three spies.’

This parallelism extends across all such pairs, with both (52) and its long-distance dependent analogue (53) having, as preferred interpretations, a wide scope construal of the indefinite:

1. (52) ronbun-o gakusei-subete-ni sensei-wa setsumeishita article-acc student-every-dat professor-top explained ‘An article to every student the professor explained.’

1. (53) ronbun-o gakusei-subete-ni sensei-wa jyosyu-ga setsumei-suru-beki-da to kangaeta article-acc student-every-dat professor-top assistant-nom explain-pres-modal-copcomp thought ‘An article to every student the professor thought the assistant should explain.’

Even wh-questions, with their reported strict sensitivity to linear order, display obligatory wide scope for the wh-term for the following quantified expression, both in simple clauses (Saito Reference Saito1992, Tada Reference Tada1993) and in the corresponding multiple long-distance dependency structures:Footnote ²¹

1. (54) dono ronbun-o gakusei-subete-ni sensei-wa setsumeishita-ka which article-acc student-every-dat professor-top explained-Q ‘Which article to every student did the professor explain?’

1. (55) dono ronbun-o gakusei-subete-ni sensei-wa jyosyu-ga setsumei-suru-beki-da to kangaeta-ka which article-acc student-every-dat professor-top assistant-nom explain-pres-modalcomp thought-Q ‘Which article to every student did the professor think the assistant should explain?’

As we have already seen with (19)–(20), which display subject-marked expressions that could putatively be construed as multiple long-distance dependency but which in Japanese are not, the parallelism between local permutability and permutability of pairs in multiple long-distance dependency breaks down in Japanese when one of them is a subject expression. Nevertheless, the parallelism between short and multiple long-distance scrambling otherwise holds equally for both languages.

This leaves us with an anomalous situation in which there is both indication of apparent asymmetry between short scrambling and long-distance scrambling on the one hand, as displayed in (39)–(41), but nevertheless symmetry between short scrambling and multiple long-distance scrambling on the other (42)–(55). The asymmetry between short and long scrambling has been used as evidence within the minimalist literature to justify analysing them as unrelated processes (A Movement on the one hand, A′ Movement on the other). But this leaves unexplained the parallelism between short scrambling and the freedom of order of the two long-distance-moved terms in multiple long-distance dependency (involving either Remnant Movement or Oblique Movement). There is, indeed, on that account, no basis for anticipating such parallelism. And the range of variability in quantifier construal is an unrelated puzzle without any obvious structural basis. However, we shall argue that, once sensitivity to the left–right dynamics of parsing is incorporated into the grammar formalism itself, the various phenomena fall into place. The phenomenon of multiple long-distance dependency will reduce to a feeding relation between the long-distance-dependency-creating mechanism and the local-discontinuity-creating process: the data are directly predicted, as is the obligatory attendant relative locality restriction of the paired NPs. The minor variation between Japanese and Korean subject marking is explicable diachronically as an encoding of routinisation in the one language of what had earlier been a pragmatically based choice. The variability of quantifier-scope construal is analysed as a systematic variability in the construal of indefinites as having to be dependent on some other term, a pragmatic choice that can be made freely with respect to whatever precedes, but only restrictedly with respect to what follows. And this account of specificity effects as applied to Japanese and Korean interacts with both the general tree-growth principles and the idiosyncratic Japanese subject marking to give us the range of data that needs to be predicted. Thus the very data which constitute a puzzle for grammars that debar any feeding relation between performance considerations and articulation of grammar architecture will be seen to emerge from the shifted set of assumptions, without structure-specific stipulation.

2.1. Lexical specifications

As in other frameworks, the verb is the major projector of structure, for which actions are defined that induce some, or even all, of the propositional template they express. In Japanese and Korean, both full pro-drop languages, verbs project full propositional structure, with individual argument nodes decorated with placeholders that stand for some value to be assigned either from context or from the construction process. Such place-holders are represented as metavariables of the form U, V,.. of type e:Footnote ²⁶

1. (57) Result of carrying out the lexical actions of yom- ‘read’ [Japanese]:

   

The effect is that verbs have a lexical specification inducing a sequence of actions which might equivalently be expressed by discrete words. The decoration of argument nodes with a metavariable, for example, is the intrinsic property of pronouns, underspecification with respect to content being their hallmark. Whether from a parsed pronoun, or from decorations intrinsic to the verb, all such placing-holding devices must be provided with an assigned value (notice the requirement for a fixed value, ?∃x.Fo(x)). Different types of anaphoric expressions can then be defined according to different constraints which they impose on the process. Of these, reflexives have to be updated within a given locally defined propositional structure, pronouns outside such locally defined structure, and so on (see section 2.3.1 for a tree-theoretic concept of locality). On the other hand, metavariables projected as part of the intrinsic specification of the verb lack any such restriction, and can be defined either locally or from some more general context. The account is accordingly one of intrinsic lexical underspecification, with update to these metavariables occurring either from the partial tree under construction, or from context as a step of pragmatic substitution.Footnote ²⁷ Though external to the grammar formalism itself, this substitution process must interact with grammar-internal processes to ensure compositionality of content as defined over the resulting tree, for successful decoration of each non-terminal node, by definition, depends on having had all requirements on its daughters satisfied.

It might seem that this system cannot provide a representation that is transparent with respect to content because of quantification: the level at which to express syntactic generalisations across all noun phrases is very generally presumed not to coincide with the level needed to express scopal dependencies for quantifying expressions.Footnote ²⁸ However, in the epsilon calculus, quantified expressions are treated as name-like, with all the force of quantification expressed as part of the evaluation of the constructed quantifying terms, a matter we return to shortly (section 2.2).Footnote ²⁹ So names and quantifying expressions are all assigned actions that introduce a term of type e. In a language such as Japanese or Korean, with determiners being optional, nouns are defined as projecting the necessary conceptual structure to yield a composite type e term (like verbs projecting considerably more structure than is morphologically made explicit). The internal structure of such terms involves three parts: a binder, for example the epsilon operator ∊ analogous to the existential quantifier; a variable that it binds; and a restrictor of that variable, such as Shorui′ ‘document’ (Japanese), which in the bare noun case is simple, but which may be arbitrarily complex:

Such a lexical specification for a sample Japanese noun illustrates a general property of lexical specifications: the contribution of any one word to interpretation is considerably more than just the provision of some suitable logical expression. Thus, what a mere noun projects in Japanese (and Korean) is information more commonly associated with a determiner-noun sequence in a language such as English.

In general, words are taken to provide meta-linguistic instructions about the progressive setting out of structure, along with the conceptual Formula value that they provide for a node decoration, all expressed in the same tree-growth vocabulary as general structure-building operations. Indeed, some words may provide little more than such instructions; and this is a characteristic property of affixes. Final-placed affixes have a critical role to play in verb-final languages, since they signal the edges of phrasal constituents in the build-up of interpretation. The Japanese past-tense verbal suffix -ta, for example, takes as trigger a completed propositional formula of type t, with no requirements, to which a propositional operator denoting past time is added. Such a specification is no more than a transparent reflection of -ta's contribution to the semantic composition of the whole. However, its effect in addition is to signal the end of the entire sequence of steps constituting the interpretation process associated with the verb to which it is suffixed. This is because the triggering condition for the update which it provides is a completed formula of propositional type; and this can only be satisfied if all the aspects of interpretation needed to provide that formula have been resolved. This account notably requires the tense suffix to be processed last, and we derive verb-final ordering as a consequence.

This property of signalling the completion of structure for the word to which it is attached is not just an idiosyncrasy of the suffix -ta. To the contrary, it is a characteristic property of inflectional suffixes of languages which are systematically head-final; and case particles play a similar role, over and above their basic function. This basic function is to constrain the relative hierarchical position in the tree: for example, the Japanese object-marking suffix -o indicates that its mother node must be a predicate, i.e. it is defined as imposing a requirement ? 〈↑₀〉Ty(e→t). By definition, this is a constraint on tree growth: the mother of such a node has to be assigned a predicate value at some point in the construction process. But in addition to this, as the last morpheme in any subtask decorating a node with a type e term, case suffixes indicate that this task is now complete. That is, by requiring some completed type e term as the triggering condition for the update that they induce, they have the effect of closing off the internal structure of that term.Footnote ³⁰ In this way the compiling up of terms and propositions from their elementary parts is driven by the suffixing particles.

2.2. Quantifier scope

It might seem that this project of incrementally building up structure cannot be sustained without separating syntactic and semantic vocabulary because of the problem of scope: this aspect of interpretation just does need globally provided information. But this turns out to be unproblematic. Because the system adopted for representing semantic content is the epsilon calculus, scope is not represented in the tree itself but in terms of scope constraints, which are collected as they become available and are only implemented as part of a final step of evaluation of the tree once the parsing process is completed. These constraints take the form x<y, indicating that some introduced term binding variable x has scope over a discrete term binding y (see Kempson et al. Reference Kempson, Meyer-Viol and Gabbay2001: chapter 7 for formal details): these are introduced once the relative position of the term under construction is fixed (see section 3.1). This provides a basis for expressing idiosyncratic constraints on scope evaluation which words may impose: for example, indefinites, which have well-known wide-scoping specificity effects. These are taken to project a statement of scope dependency in which the first argument of the scope relation is a metavariable, U<y, for some scopal term binding a variable y (including terms denoting time). What this partial specification of a scope statement reflects is the fact that the choice of dependency for an indefinite is pragmatically driven (analogous to the way in which interpretation of pronouns is resolved contextually).Footnote ³¹

This characterisation of scope for indefinites correctly anticipates that indefinites can be construed as dependent on any term already constructed in the interpretation process (e.g. from a previously parsed quantifying expression). It also provides a natural basis for two observed phenomena which Japanese illustrates. On the one hand, it covers the prevailing cross-linguistic tendency for bare indefinites positioned to the left of some subsequent quantified expression to be interpreted independently of that expression, that is, as dependent only on some temporal variable as in (47) (repeated here):Footnote ³²

1. (47) shorui-o san-nin-no supai-ni jaanarisuto-wa watashita [Japanese] document-acc three-class-gen spy-dat journalist-top handed ‘The journalist handed one document to three spies.’

It also explains why choice of scope for a bare indefinite following another quantified expression invariably allows but does not enforce dependency on that preceding expression, as in (25), (26) and (46) (we give here (25) by way of illustration):

1. (25) hotondo-no uta-o dareka-ga utatta most-gen song-acc someone-nom sang ‘Most of the songs, someone sang.’

Because this choice is made on a pragmatic basis, subject to relevance (i.e. least effort) and other such considerations (Sperber & Wilson Reference Sperber and Wilson1986), we also expect that where such linearity considerations conflict with contingent knowledge of the situation described, threatening to yield an inconsistent interpretation, they can be set aside, as in (48) and (51):Footnote ³³

1. (48) nise-no pasupooto-o san-nin-no supai-ni CIA-wa watashita forged-gen passport-acc three-class-gen spy-dat CIA-top handed ‘The CIA handed a forged passport to three spies.’

Indeed, as we shall see, even in subject–object sequences of Japanese, where the subject-marking property of -ga forces immediate identification of a fixed subject relation, such inverted interpretations are possible if relative contingent knowledge renders the canonical interpretation implausible, as already seen in (29), which, like its English congener, allows an inverted scope interpretation:

1. (29) kangohu-ga subete-no kanjya-o monshinshita nurse-nom every-gen patient-acc interviewed ‘A nurse interviewed every patient.’

The restricted availability of this form of interpretation is predicted by the DS account, because there are not one but two points at which the place-holding metavariable associated with the subject expression can be identified. In (29), for example, the first point is when the expression kangohu-ga is parsed and an initial term with attendant scope constraint set up – i.e. at a relatively early step in the parsing process, given the order of the expressions. However, there is also a second point, and this is when the sister predicate value is completed and whatever aspects of underspecification were left open in the construal of that indefinite subject now have to be resolved in order for subject and predicate values to combine to yield a properly compositional interpretation of the whole. By this late stage, when the interpretation of the subject expression and then the entire proposition is finally being established, there may have been additional quantified terms added to the accumulating set of scope statements during the build-up of interpretation of the predicate. So, as a result, a broader range of choices will have become available on which to establish the dependency of the indefinite than was available at the earlier stage when the subject expression was first parsed. This strategy of interpretation is analogous to the construal of (subject) expletive pronouns whose interpretation has to be established before the final propositional formula can be compiled, with development of the subject node occurring after the construction of the attendant predicate:

1. (58) It's likely that I'm wrong.

This analysis of expletives is argued for in Cann et al. (Reference Cann, Kempson and Marten2005) (see also Kempson et al. 2009); and it is noteworthy that both expletive pronouns and inverted construal of indefinites are subject to the same tight locality constraint.Footnote ³⁴

As we would expect from such a pragmatically driven account, these facts carry over to Korean. Inverted construal of subject and object is equally available, and is determined by context or world knowledge. Thus, despite the variability of the data, the general assumptions of the framework imply that such inverted interpretations are expected to be possible if the first NP is indefinite, yet are predicted to be dispreferred, since they involve a selection which is not immediately resolved at the first parse point. And for all such cases where a choice is available but dispreferred, we do not define a grammar-internal mechanism for determining the choice, this being a consequence of relevance-driven considerations – though the account does provide the mechanism underpinning the availability of such interpretations. Indeed, as DS makes explicit, languages make available a range of strategies for interpretation build-up.

With so much freedom of choice in languages with very free NP ordering, it is little surprise that the languages might develop distributivity markers, narrowing down the choices that would otherwise be available, as displayed in (31)–(33). In the Korean case (ssik), this is defined as a nominal property, while in Japanese it is defined as a property of the constructed term as a whole, in both cases encoding the requirement of a co-occurring term in the same domain relative to which it is dependent; but given that the DS account of quantification allows idiosyncratic scope effects to be defined, such an encoding is not in principle problematic.Footnote ³⁵ Overall, then, we have a principled basis for anticipating flexibility of interpretation for indefinites, as in other languages, relative to whatever individual idiosyncrasies the language may lexically impose.Footnote ³⁶ As we shall now see, however, there are independent structural reasons for anticipating that linear order considerations do not always prevail.

2.3. Structural underspecification

So far in this exegesis of Dynamic Syntax, the primary focus has been on lexical specifications. But the system of interpretation is far from being exclusively lexicon-driven. There are general computational actions which reflect principles of semantic tree growth that the system licenses, their intrinsic incremental dynamism being a defining characteristic of this framework. The informal observation which the analysis seeks to reflect is that expressions parsed early on in the interpretation process may fail to fully determine the role that they are to play in the overall interpretation, this becoming available only later in the parsing process (at the site familiarly known as ‘the gap’). The formal analogue of movement is to define actions that license the construction of tree relations that are not fully determined, creating an underspecified tree relation which has an associated requirement for subsequent update, that must be satisfied during the ensuing construction process. Unlike the parsing analysis proposed for Japanese by Miyamoto, or more generally so-called D-tree grammar formalisms (Marcus Reference Marcus1980),Footnote ³⁷ the partial trees constructed by the DS formalism are part of the grammar specification. For example, long-distance dependency effects are expressed by the construction of a node from a top type-t-requiring node of some initiated logical structure, this new node being specified only as dominated by that top node, its position within the unfolding tree being otherwise unfixed at this point in the construal process. As indicated earlier, such nodes are annotated as 〈↑_*〉Tn(a). This is formally identical to the LFG concept of functional uncertainty (see Kaplan & Zaenen Reference Kaplan, Zaenen, Baltin and Kroch1989), but unlike that notion, in the present framework, because syntactic trees are expressed in the same terms as representations of interpretation, all such underspecification is defined with an update requirement as part of the construction of interpretation. This analysis will provide a second basis for expecting delay in scope assignments, as we shall see in due course.

2.4. Constructive use of case

Within this network of structural growth possibilities, case plays an important rule. We have so far introduced case as a filter on tree growth (e.g. accusative as requiring a predicate-node as mother). Because such specifications take the form of a requirement, they might not be satisfied until substantially later in the construction process. However, case can play a constructive role, and this is very simple to implement in this system as follows. Nothing dictates when a filter on output is met, and the filter can be used to induce the specific structural relation immediately upon decoration of the node in question, in anticipation of the relative tree position dictated by that filter. The effect is that structure may be built up progressively even before the verb is processed. For example in (62), the information provided by the case marker -o can be used to enrich some type-e node introduced by Local *Adjunction and so have the effect of updating the unfixed predicate relation with a fixed predicate relation, providing immediately a type-e tree node in the appropriate relation with which to instantiate the relevant output filter given by o:

1. (62) shorui-o jaanarisuto-ga yonda [Japanese] document-acc journalist-nom read ‘The journalist read the document.’

   

Furthermore, once this tree position is identified as an emergent object node in this derivation, we expect scope constraints to be set up directly, giving rise to the familiar preference for scope dependencies to follow the order in which the NPs occur (even in cases such as indefinites where there is flexibility and potential for delay).

It is this constructive use of case which provides a second point of comparison with Miyamoto (Reference Miyamoto2002) (though Miyamoto only considers ga-marking in any detail). Under both approaches, because this step is an enrichment of an underspecified ‘locally-dominated’ relation, the pointer can return to the locally dominating type-t-requiring node, allowing the process to take place all over again. Indeed, it is notable that only if such enrichment takes place is this possible: the constraint on there being only one unfixed node of a type at a time built from a given node would otherwise preclude the construction of any further node by Local *Adjunction. Indeed, if iterative applicability of Local *Adjunction is to provide the basis for flexibly building up type e terms, it is essential that there be some independent process which provides the update needed to allow such subsequent re-application. With case providing a condition on output, however, such a solution is available. The enrichment which is needed to ensure that this condition is satisfied can be induced at any time, subject to restrictions of pointer movement. It can therefore take place immediately upon the return of the pointer to the dominating node, after having constructed and decorated such a locally unfixed node. This form of update means that the NPs themselves can be processed in any order: each one will introduce the particular substructure needed for its own output, and each structural relation can then be fixed at once by enrichment. The actions of the verb then serve to complete the requisite structure.Footnote ⁴² There is no reflex of word order displayed by the resulting tree, so no matter what order the NP argument expressions occur in, the same result will obtain. Hence the essential correlation between free intra-clausal permutation of NPs and constructive use of case.

With case in Japanese seen as in principle playing both filtering and constructive roles, it is straightforward to specify the update induced by the stringent idiosyncrasy of subject marking in Japanese – in movement terms, reported as precluding long-distance movement (Saito Reference Saito1985). Seen from a parsing perspective, we can define -ga as having only a constructive case mechanism: the fixing of the structural relation is immediate, and the alternative of merely imposing a relatively weak output filter is not available. The effect is that -ga identifies a boundary edge for a local propositional domain wherever it occurs in an NP sequence. Accordingly, we define it as taking as input a node introduced by application of either *Adjunction or Local *Adjunction and fixing the relation to one of immediate dominance (note that the input condition is simpler than for the construction subcase of o, applying to any node which is unfixed and yielding a fixed subject relation):

1. (63)

   

This forced update applies to all nodes created by either Local *Adjunction or *Adjunction which -ga decorates, and accordingly yields a boundary-marking effect in all clausal sequences. This boundary-marking role will not be available to the other case-marking suffixes, since the dual function of other case specifications ensures that they will not themselves provide a deterministic parsing clue for identifying a propositional constituent boundary (though there might of course be other considerations giving rise to such an effect.

This tight correspondence between subject marking and immediate fixing of a local subject relation does not carry over to Korean, which freely allows the use of subject-marked expressions in long-distance dependency, as we saw earlier with (43):

1. (43) se kanhosa-ka tu hwanca-ssik-ul Kim-paksa-ka pamse-tongan tolpassta ko kuraysseyo three nurse-nom two patient-dist-acc Kim-Dr-nom night-during nursed comp said ‘Dr Kim said that three nurses each looked after two patients all night.’

Korean subject-marking thus retains the weaker output-filter characterisation of case-marking, shared by all other case-markers. Nonetheless, like all other case markers, enrichment of the introduced node decorated by the subject marker -ka has to take place yielding a fixed subject relation if some subsequent case-marked expression is to be processed as providing some local argument, and for just the same reason as in Japanese: only one unfixed node of a particular type can be constructed from a given node at any one time. Thus the only feature essentially differentiating subject marking in Japanese and Korean is that in Korean, such enrichment is a routinised strategy – often used, hence arguably a default, but overrideable. In Japanese, by contrast, such a strategy has become fully encoded.

A putative counter-example to this account of local scrambling, and in particular to the account of ga-marking and its enforced locality, are examples in which a left-peripheral -ga marked NP is interpreted as a subordinate subject immediately followed by a matrix subject marked with the topic-marker -wa, making it appear that the ga-marked expression should be taken to modify an unfixed node that is resolved into a subordinate structure by the same mechanism as other case-marked NPs:

1. (64) Saito-ga Chomsky-wa tottemo ii riron-o motteiru to omotteiru [Japanese] Saito-nom Chomsky-top very good theory-acc has comp thinks ‘Chomsky thinks that Saito has a very good theory.’

However, in other work (Cann et al. Reference Cann, Kempson and Marten2005, Kempson et al. Reference Kempson and Chatzikyriakidis2009), we have argued that topic-marked expressions decorate a so-called linked structure, a quasi-independent structure correlated with the primary structure solely through imposed identity of construal for the topic-marked expression and one of the arguments in the primary structure.Footnote ⁴³ Given such an account, the wa-marked expression in (64) can be associated with the construction of the linked structure from the initial subject-induced node decorated by Saito-ga.Footnote ⁴⁴ This strategy allows the interrupted sequence of Saito-ga totemo ii riron-o motteiru to be analysed as projecting a local array of arguments plus predicate at the appropriate level of embedding, through initial use of Generalised Adjunction (see section 3.2). On this analysis, the interpretation of (64) would be more faithfully reflected by the gloss ‘As for Chomsky, Saito has a very good theory, he thinks’.Footnote ⁴⁵ So the analysis of -ga as immediately determining a propositional boundary edge by fixing a subject relation can be preserved. The availability of this additional strategy applies equally in Korean, although since subject-marking in Korean is no different from other case relations, the preference for constructive case construal is due to general processing-cost considerations (Kiaer Reference Kiaer2007).

2.5. Complement clause construal

To extend the account of local projection of structure to the projection of subordinate structure, all that is needed is to define the suffixed complementiser -to as following the general pattern of inflectional suffixation, inducing the completion of the formula of propositional type, the end of whose construction it signals. This is formally reflected by imposing, as condition on its update, the necessity of having as input some completed propositional formula, with no subsequent revision allowed. To see this effect, we need to examine how the construal of a simple clausal sequence is incorporated at some arbitrary level of embedding.

1. (65) Hiroto-ga shorui-o yonda to itta Hiroto-nom document-acc read comp said ‘Hiroto said he read the document.’

As we saw earlier, Generalised Adjunction is a very general mechanism for initiating one structure from another of the same type requirement, with no indication of level of embedding of that introduced new node or even whether the propositional tree it introduces will be within the same tree as the root. It thus cannot provide any basis for application of enrichment by -ga to yield a fixed matrix subject relation: the only enrichment available will thus be to induce the subject node of the local platform type-t-requiring node which Generalised Adjunction introduces. With this now presumed as a possible initial step from the assigned goal, the sequence of actions for projecting some simple propositional structure in parsing a subordinate sequence can be seen as otherwise identical to that involved in processing an independent simple clausal sequence (suppressing the scope statement of the complement structure):

1. (66)

   

Thus -to has to be defined (in standard Japanese, where it is obligatory) in such a way as to determine the nesting of the propositional structure. This can be done in one of two ways: either by adding a further intermediate node locally dominating the node just completed, or by making use of the very weak structural relation already induced, returning the pointer there, and enriching the relation to that of local subordination (〈U〉 is the weakest subordination relation of all, indicating an arbitrary level of embedding, even across linked trees):

1. (67)

   

For present purposes, the primary significance of these actions is that, following the regular pattern of suffixes, the condition necessary for either of the two kinds of licensed update given -to is that the node should be decorated with a completed type t. The effect of this condition is to ensure the prior decoration of all non-terminal nodes which that node dominates (including nodes for which there is no explicit natural-language expression). What -to then imposes is obligatory subordination. The result of carrying out these actions is that some subsequent verb can project its actions; in (65) this is the verb itta, for which the propositional template already constructed from the parsed string Hiroto-ga shorui-o yonda provides its object argument:

1. (68)

   

Finally, in this particular parse of (65), the subject argument can be identified anaphorically as Hiroto′, to yield (as before, suppressing tense):Footnote ⁴⁶

This narrated sequence of actions may seem little more than a tutorial demonstration of a DS derivation; but there is more to it than this, for the so-called Proper Binding Condition effect, which, in movement accounts, has to be imposed as an additional constraint on processes of movement, now emerges as an immediate consequence, a mere side-effect of the account of head-marked suffixes in general as term-closure devices. The problem for minimalist, as for other movement accounts, is that data such as (69) (cited in section 1.2 as (21)) are on the face of it predicted to be well-formed since, given general principles of movement and no further auxiliary hypothesis, nothing precludes leftward movement of an NP out of its containing clause to an adjunction site, following by leftward movement over that adjunction site of what had been its containing clause, yielding (69):

1. (69) [*Hanako-ga t _i iru to]_j Sooru-ni_i Taroo-ga t _j omotteiru Hanako-nom be comp Seoul-in Taroo-nom think ‘[That Hanako is t _i]_j in Seoul_i Taroo thinks t _j.’

But (69) is sharply ungrammatical.Footnote ⁴⁷ Some additional assumption or set of assumptions thus had to be created – hence the added filter defined as the Proper Binding Condition. On the Dynamic Syntax set of assumptions, to the contrary, there is no question of ever generating the sequence indicated in (69): the parsing of the sequence ending with -to in (69) has to have been construed as a completed propositional formula in order to license the update provided by -to, and hence would have to have had the argument of iru provided in context. Parsing Sooru-ni following the parsing of to as modifying the embedded structure is precluded. There is no going back of the pointer: once that structure is completed, the only possibility would be to construe Sooru-ni as a dative argument to omotteiru, but this is independently excluded. In (70), by contrast, the full sequence of expressions needed to interpret the clausal sequence ending with -to allows a propositional structure to be routinely completed, and this then serves as the internal argument of omotteiru – all exactly as expected:

1. (70) Hanako-ga Sooru-ni iru to Taroo-ga omotteiru Hanako-nom Seoul-in be comp Taroo-nom thinks ‘Hanako is in Seoul, Taroo thinks.’

For similar reasons, the occurrence of a dative-marked NP following -to, as in (71), must be interpreted relative to the matrix subject, and not within the subordinate structure:Footnote ⁴⁸

1. (71) jaanarisuto-ga shorui-o yonda to keisatsu-ni supai-ga itta journalist-nom document-acc yonda comp police-dat spy-nom said ‘The spy said to the police that the journalist read the document.’ ≠‘The spy said that the journalist read the document to the police.’

The significance of this account for the ongoing debate over the Proper Binding Condition is that, according to the account provided here, the data are epiphenomenal – not because the data are contrary to what has been observed by many (paceHiraiwa Reference Hiraiwa2005), but because the effect of the Proper Binding Condition follows as an immediate consequence of the unfolding dynamics, in particular the effects in the closing stages of compiling some emergent structure as imposed by strict compositionality of the resulting tree.Footnote ⁴⁹ Moreover, on this account, this effect is not a constraint to be defined specific to clausal structure itself, or with respect to how that is built up, but is a consequence of the very general role of suffixes in a head-final language, that they define the end-point in a local term-construction process.

3.1. Cross-clausal quantifier construal

This account of long-distance scrambling gives rise to the following expectation. Given that such a left-peripherally constructed term is not initially assigned a specific structural position in the configuration of emergent semantic structure, any outstanding update requirements for the term decorating that node, which need a fixed tree position in order to be resolved, cannot be resolved until this requisite fixed tree-node position has been identified. This affects scope construal in particular. The assignment of relative scope dependency for a term decorating a node introduced by *Adjunction can only take place once the update process has occurred which fixes that node within the tree. So, in (40), the quantifying expression dono hon-ni-mo is first taken to decorate an unfixed node; and the position of this node must then be fixed before the scopal property making it a complete quantifying term can be determined:

1. (40) dono hon-ni-mo sono hon-no tyosya-ga Hiroto-ga syomeishita to itta every book-dat-also that book-gen-also author-nom Hiroto-nom autographed comp said ‘That book's author said that Hiroto autographed every book.’

However, the interpretation of the immediately subsequent sono hon-no tyosya-ga as matrix subject will be fixed immediately upon processing the suffix -ga because of the constructive use of case, so its containing anaphoric expression will also get fixed before the scopal property of the quantifying expression that would make it viable as an antecedent can be determined. So sono hon-no in (40) is interpreted indexically.Footnote ⁵³ Hence the cross-clausal reconstruction effects of (40) associated with quantifier construal for some left-peripheral quantified expression emerge as a consequence of the specific tree-growth perspective: while the unfolding of interpretation is strictly incremental with respect to the structural process itself, nevertheless, in defining systematic underspecification of relations and formulae within that perspective, there is leeway for systematic delay in compositionality of content in all such cases.

This analysis provides a basis for explaining the asymmetries between short- and long-distance scrambling environments in their licensing of scope construal for quantifying expressions. In individual clausal sequences (where NPs occur in any order), the node for each NP to decorate will be introduced as an unfixed node, but its tree relation will be updated as soon as the decoration of that node is complete – with the result, as we have seen, that assignment of scopal dependence for a quantifying expression takes place incrementally. There is a stronger result in Japanese when the subject expression is initial, as the underspecified unfixed node which it decorates will in Japanese always be updated immediately (ensured by the processing of -ga, which encodes this action directly). In such a case, we expect that unless there is reason to override the reliance on the linear order in which the words are presented, a possibility in any case only available with indefinite expressions, construal of quantifying expressions will follow linear order, yielding the already-noted judgement of (27) as unambiguous:

1. (27) dareka-ga hotondo-no uta-o utatta someone-nom most-gen song-acc sang ‘Someone sang most of the songs.’ (unambiguous)

However, when an expression with any case-marking other than subject-marking is initial, both *Adjunction and Local *Adjunction strategies will be available; and since the second of these operations licenses immediate fixing of the node constructed, and the first, conversely, licenses delay in fixing the node constructed, both interpretations are expected to be available. And so we get the much freer availability of non-inverted and inverted interpretations with object–subject ordering in local scrambling, as in (39) and (25):

1. (39) dono hon-ni-mo sono hon-no tyosya-ga syomeishita every book-dat-also that book-gen author-nom autographed ‘Every book_i that book's_i author autographed.’ (indexical/bound-variable interpretations of sono both available)

1. (25) hotondo-no uta-o dareka-ga utatta most-gen song-acc someone-nom sang ‘Most of the songs, someone sang.’ (ambiguous: indefinite narrow/wide scope)

In Korean, scope effects involving the subject are arguably weaker, being only a consequence of pragmatic pressures. But, in all cases, interpretations which go against linear order require the distinctive intonation indicating delay in construal of the quantified expression.

3.2. Implementing long-distance scrambling effects

In setting out our account of long-distance scrambling phenomena, we have not yet spelled out the mechanisms for constructing nested complement structure prior to the processing of the requisite verb. The problem that these structures appear to pose is that the one initially unfixed node which the left-peripheral expression decorates apparently has to be passed across what would seem to be a sequence of unfixed nodes for each of the argument expressions, since the verb will not yet have been parsed:

1. (4) shorui-o keisatsu-ga jaanarisuto-ga yonda to koohyooshita document-acc police-nom journalist-nom read comp reported ‘The document, the police reported that the journalist had read.’

At this point, the constraint of having only one unfixed node of a type at a time comes into play (see section 2.3.1). Though more than one underspecified tree relation can be constructed as long as a distinct adjunction rule is made use of for each one (since the three adjunction rules involve distinct modal relations), the same computational action cannot be used more than once. In this derivation, the expression shorui-o decorates an unfixed node, as introduced by *Adjunction, but this does not preclude parsing of keisatsu-ga as this is associated with introduction of a locally unfixed node which its subject specification duly fixes as matrix subject. The immediately following jaanarisuto-ga must then be taken to decorate a discrete subject node at some level of embedding. The problem is that *Adjunction has already been used in the construction of the node decorated by shorui-o, so it cannot be used to license the introduction of a node to host the embedded structure, as this would simply collapse with that still unfixed node, with no possible way of completing a well-formed derivation. The only solution available is to use the weakest form of transition, Generalised Adjunction, to construct the embedding relation. But this is not problem-free, since application of Generalised Adjunction, which introduces the very weakest of embedding relations, yields a structural relation which is too weak to license unification of the unfixed node originally introduced by *Adjunction with any subsequently introduced node:Footnote ⁵⁴

1. (72)

   

The problem is that Generalised Adjunction is not an operation which would introduce structure that could provide an update of the tree relation introduced by application of *Adjunction, here the node decorated by ∊, x, Shorui′(x), since the tree relation it introduces is so weak, weaker even than that introduced by *Adjunction.

This may seem to enforce a characterisation of all such strings as incapable of yielding a logical form as a result, hence ungrammatical. Yet there is a simple and monotonic repair process. The formal system is independently defined to allow interspersing of pragmatic substitution processes with building of partial structures – the enrichment of formula values by substituting a metavariable with some appropriate part term is taken to feed into the general construction process. All that is needed to ensure that the partial structures constructed up to this point lead to a well-formed result is to assume that pragmatic enrichment can apply to structural underspecification. This is hardly contentious: enrichment of stimuli is a general cognitive phenomenon (see Sperber & Wilson Reference Sperber and Wilson1986, among others), not one specific to a certain mode of representation. More specifically, what is required to yield a well-formed derivation for (4) is to introduce the requisite weak tree relation by Generalised Adjunction, and then, having done so, to enrich it to a fixed relation, transforming the emergent partial tree into one with a relation of immediate subordination between the structure in which the newly constructed term ∊, x, Jaanarisuto′(x) is contained (established by the processing of jaanarisuto-ga) and the matrix structure. Formally, the node identified as 〈U〉Tn(0) is updatable to 〈↑₀〉 〈↑₁〉Tn(0).Footnote ⁵⁵

The problem with this step of enrichment is that it is not morphologically triggered: it is an abduction step that is triggered solely by recognition that without some such step, no successful derivation will result – hence a meta-level step of reasoning. Being a pragmatic and optional process, any such choice should be expected to be associated with general cognitive constraints. And indeed, as an account invoking some intermediate step of abduction would anticipate, it is commonly reported that long-distance scrambling data are of reduced acceptability when considered in isolation, and are only possible if uttered in a particular type of context (e.g. in answer to a question with fronted wh-expression). Indeed, given that such front-positioning forces the evaluation of the unfixed node across a sequence of partial trees until its tree relation can be fixed, we expect this to be a marked option, and indeed it is invariably associated with a particular form of intonation (Saito Reference Saito1985, Reference Saito1992; Koizumi Reference Koizumi2000), using an Intonational Phrase boundary indicating a constituency break so as to disambiguate in favour of such a strategy.Footnote ⁵⁶

With this account of long-distance dependency in verb-final languages in hand, we return to the differential status of Japanese and Korean subject marking. That Korean subject marking freely allows long-distance dependency is captured in the fact that it identifies some subject expressions only in terms of an output filter and so is commensurate with an embedded construal:

1. (16) Jina-ka sensengnim-i apase hakkyo-e mot-wassta ko kure-si-ess-eyo [Korean] Jina-nom teacher-nom sick school-at neg-came comp say-hon-past-decl ‘The teacher said that Jina couldn't come to school because of illness.’

This option of delaying subject construal via use of *Adjunction is not open to Japanese because, although application of *Adjunction is available, the specification of -ga forces the construal of that term as subject within the very structure from which that unfixed tree relation was built (hence at the root as matrix subject), as we saw earlier.

This might seem to imply that Korean word order allows generally greater freedom of construal than Japanese, but apart from the stringency of subject construal in Japanese, the relative freedom of argument placement in the two languages is similar.Footnote ⁵⁷ This is because, while there is very great freedom in simple clauses, there is far less freedom as soon as there is co-construction of matrix and embedded structures. Take, for example, the construal of two nominative NPs in sequence reflecting matrix and subordinate subject respectively. This is always possible so long as selectional restrictions are appropriately satisfied, though it is in fact dispreferred, since it is much more natural to place the matrix subject and any attendant adjuncts so that this sequence of expressions immediately precedes the verb predicated of them. However, there are consequences to a marked choice of this sort. Because *Adjunction itself can only apply if there is no other structure already built from the initiating node, the only way to introduce the subordinate structure after the construction of a matrix subject node is via Generalised Adjunction introducing a radically unfixed node with a type t requirement, ?Ty(t). Once the pointer has moved to this node, there is no return to any superordinate structure until that substructure is complete: pointer-movement back from one node to another is possible only if the nodes in question are type-complete, and for all fixed structures there is a strict compositionality requirement that type-requirements on a mother node can only be met by type-deduction and functional application applying to their daughters. What this means in the derivation of embedded structures in a verb-final language is that all NP expressions following the introduction of the relevant substructure have to be construable as contained within this substructure until the verb that will enable it to be completed has been processed. For example, in (16), the adjunct cannot be construed as indicating that ‘The teacher said at the school that Jina was sick and not coming’. This restriction is quite general. With the single exception of the ability to build and decorate an unfixed node at the outset of creating some root structure, once there has been some indication of a break in local-structure building to start the construction of some appropriate subordinate structure,Footnote ⁵⁸ there is no return to higher levels of structure until that more subordinate structure is completed. Such restrictiveness is built into the licensing of pointer movement: there is strict compositionality of any sub-structure, and no licensing of switching to and fro, from matrix to subordinate structure and back again, within the setting out of that substructure. The only licensing for such apparent freedom is at the left periphery, where an underspecified tree relation can be constructed, and that too must have a secured type specification before the parsing process can proceed from the development of that node. The co-implementation of *Adjunction and Generalised Adjunction thus correctly imposes strict subordination on all remaining NP-expressions in the sequence. These facts are common the two languages. Thus, despite minor lexical variation in license of constituent order variation, the dynamics of how the major tree-building processes interact is identical.

4.1. Quantifier dependencies and multiple long-distance scrambling

Finally, we come to the challenge of explaining how it is that multiple long-distance scrambling effects could parallel those of short scrambling, whereas regular long-distance scrambling does not. Here DS provides a natural basis for explaining the difference, for it is only in the case of multiple long-distance scrambling that the unfixed node requires a type t node. This introduced proposition-requiring node then provides a platform for the construction of nodes on which to build the construal of sequences of NPs. So it will provide a basis from which scope-dependency choices can be made for the terms decorating the two introduced argument nodes, and the succession of actions involved can be identical to that in monoclausal sequences, even though the relevant node itself remains as yet unfixed in the overall structure. The account therefore leads to the prediction that whatever restrictions or flexibility there might be for construing sequences of NPs in local scrambling environments should be replicated in multiple long-distance scrambling environments. This is exactly the pattern we observed in section 1.1 with such pairs of examples as (47) and (50), both of them preferring an interpretation in which the indefinite is construed independently of the quantified expression that follows:

1. (47) shorui-o san-nin-no supai-ni jaanarisuto-wa watashita document-acc three-class-gen spy-dat journalist-top handed ‘The journalist handed one document to three spies.’

1. (50) shorui-o san-nin-no supai-ni keisatsu-wa jaanarisuto-ga watashita to koohyooshita document-acc three-class-gen spy-dat police-top journalist-nom handed comp reported ‘The police said that the journalist had handed just one document to the three spies.’

Exactly the same parallelism, we recall, was displayed in Korean. Furthermore, this parallelism between local scrambling environments and multiple long-distance dependency environments is predicted to extend across all cases, with the systematic exception of subject marking in Japanese.

This prediction of such strict parallelism between multiple long-distance dependency and local scrambling, modulo the idiosyncrasies of Japanese ga-marking, represents a clear advance of the DS approach over the Koizumi and Takano approaches, for on neither of these does the account of multiple long-distance scrambling bear any relation to the account of short-distance scrambling. To the contrary, Koizumi (Reference Koizumi2000) invokes vacuous verb movement, and Takano (Reference Takano2002) invokes Oblique Movement, which is distinct from both A and A′ Movement in yielding a position for the moved node which signally fails to c-command the containing structure out of which it has moved (see Takano Reference Takano2002). So on either of these accounts, this consistent parallelism is unexpected. Furthermore, the apparent paradox that multiple long-distance scrambling effects can display parallelism with short-scrambling effects with respect to sensitivity to linear order, even though simple long-distance dependency effects do not display any such sensitivity, is also resolved. It is precisely in the presence of a pair of locally constructed argument nodes that such dependency can be established, and it is this constructive process that takes place in both short- and multiple-long-distance scrambling. And even the difference between Japanese and Korean subject marking and their effects is unproblematic to formulate: the subject-fixing actions encoded in Japanese -ga are nothing more than the encoding of what had been a routinisation of common practice.

In sum, the system predicts a range of scope-dependency effects, but within limits. There is no overt or covert movement, no total suspension of incrementality. To the contrary, all interpretation involves building up interpretation via sequential steps of parsing; so, in general, interpretation is predicted to follow linear order, with two systematic exceptions. First, indefinites allow a systematic local delay in their final construal in virtue of the anaphoric nature of the basis for establishing their scope dependency, as in (29) (see the discussion in section 2.2):

1. (29) kangohu-ga subete-no kanjya-o monshinshita nurse-nom every-gen patient-acc interviewed ‘A nurse interviewed every patient.’

It is this which provides the underpinning to wide-scope effects for some quantified expression that follows the indefinite, as part of the projection of a single local propositional structure. Secondly, for any expression providing a term of type e to an unfixed node as in long-distance scrambling effects, there are structural reasons why interpretation may be delayed, as in (40):

1. (40) dono hon-ni-mo sono hon-no tyosya-ga Hiroto-ga syomeishita to itta every book-dat-also that book-gen author-nom Hiroto-nom autographed comp said ‘That book_i's author said that Hiroto autographed every book_j.’

In all such cases, interpretation will be definitively established only when the structural position of the unfixed node relative to its dominating propositional node is established (see section 3.1). Hence in long-distance scrambling, where quantified expressions may be at an arbitrary remove from their site of construal, inversion of quantifier-anaphora dependencies is expected. The mixed inversion effects of multiple long-distance scrambling arise because these in part display properties of long-distance scrambling, and in part properties of local scrambling. And the minor variation between Japanese and Korean subject-marking follows exactly as we would expect, given the idiosyncratic constructive-case construal induced by Japanese subject-marking. Unlike all other accounts of long-distance dependency, the facts of multiple long-distance dependency follow in toto from the major architectural properties of the model, specifically from the dynamics of incremental tree growth in real time.Footnote ⁶²