2.1.1 Semi-productivity and lexical gaps

As mentioned above, an SIP expression is somewhat idiomatic. For example, ko-kubi ‘small-neck’ does not have a compositional meaning which is the result of combining the meanings of ko- ‘small’ and kubi ‘neck’ in a straightforward manner. In fact, there is nothing that ko-kubi as a whole refers to, though the expression does indeed retain the core meaning and refers to a contextually appropriate inalienable possessor's neck.

Further, under some circumstances, both of the size-indicating prefixes ko- ‘small’ and oo- ‘big’ can form an SIP expression: e.g. ko-mata ‘small-groin’ and oo-mata ‘big-groin’. However, in other instances only one of the size-indicating prefixes can be employed, compare, for example, ko-kubi ‘small-neck’ but not *oo-kubi ‘big-neck’ and oo-guti ‘big-mouth’ but not *ko-guti ‘small-mouth’.Footnote ⁵

Furthermore, attaching the size-indicating prefixes ko-/oo- is not an option for all inalienable nominals. For example, none of *ko-/oo-momo ‘(Int.) small/big-thigh’, *ko-/oo-asikubi ‘(Int.) small/big-ankle’, and *ko-/oo-atama ‘(Int.) small/big-head’ is lexicalized.

Due to the fact that there are a limited number of nominals of the inalienable type to begin with (lexical items indicating ordinary non-technical body(-related) parts), SIP expressions that are derivative of these nominals are quite limited in number. The matter of lexicalization has to be taken into account as well. As Bauer (Reference Bauer2001) tells us, not all potential (complex) words are lexicalized. In many cases the factors determining the productivity (in terms of lexicalization) of a given word-formation process are difficult to discern. For example, according to OED, there are at least 473 English words ending in adjectival -ly (like friendly). However, there are only 10 adverbs formed with the derivational morpheme -ly that issuffixed to adjectives ending in -ly (as in friendlily). Similarly, for no apparent reason, the inventory (i.e. lexicalization) of SIP expressions has fluctuated historically. In the 13th century, for example (in Uji Shuin stories from the 13th century cited in Nihon Kokugo Daijiten), ko-tubaki ‘small-spit’, as in kotubaki hakite ‘small-spit spit.out, i.e. spitting slightly’ was used, but is not used currently. Why not? We can certainly engage in the act of spitting slightly even today.

I consider the formation of SIP expressions to be semi-productive/inactive in present-day Japanese. There seems to be no definite reason to deny the possibility for a new SIP expression to be added to the inventory modulo the need for lexicalization. The fact that a certain morphological process is inactive at one point does not necessarily mean that it is unproductive for good. In this connection, Bauer also reports that word-formation involving the derivational morpheme -ment in English (as in argument) historically shows two peaks in its productivity – one is around the early 17th century and the other is around the early 19th century, with a sharp drop in the middle, approximately between 1650 and 1800. In a similar way, SIP formation could turn out to be more productive in the future.

The issue of productivity is not confined to the formation of SIP expressions per se. It also stretches over to the relationship between an SIP expression and its verbal collocational partner. This aspect is considered in Section 2.1.3 below.

2.1.2 Lexicalized collocational dependency

There is an unpredictable collocational dependency between a given predicate and an SIP expression in that not all combinations of the two are acceptable. For example, while (1a) above is just fine with the combination of ko-kubi ‘small-neck’ and kasige ‘tilt’, the same SIP expression would be unacceptable with a different predicate, like araw ‘wash’, as (4a) tells us. After all, it should be possible for someone to wash his/her neck slightly. There is no problem for using the predicate with a regular inalienable possessive expression, as in (38b).

1. (4)

   1. (a) *Taroo-ga ko-kubi-o arat-ta.

      Taroo-nom small-neck-acc wash-past

      ‘(Int.) Taroo washed his neck slightly.’

   2. (b) Ziroo-ga kubi-o arat-ta.

      Ziroo-nom neck-acc wash-past

      ‘Ziroo washed his neck.’

Also ko-kubi used to have a slightly different collocational dependency in the past. In the mid-16th century (in Otogi Zoshi stories cited in Nihon Kokugo Daijiten), ko-kuhi ‘small-neck’ (ko-kuhi > ko-kubi) was used together with uti-katabuke ‘intensifier-tilt’ (katabuke > katamuke) instead of the present-day kasige. Though ko-kubi-o katamuke (the contemporary counterpart without the emphasis marker uti- on the verb) is listed in dictionaries today and sounds fine, it is not part of the present author's active usage/preference, reflecting individual differences in terms of lexicalization.

2.1.3 Localized collocational dependency

Collocational dependencies between predicates and SIP expressions are local in the sense that the two have to be within the minimal clause. This ‘clause-mate’ condition is seen by the contrast between (5a) and (5b, c).Footnote ⁶

1. (5)

   1. (a) Taroo-ga [Hanako-ga ko-kubi-o kasige-ta]-to ko-mimi-ni hasan-da.

      Taroo-nom Hanako-nom small-neck-acc tilt-past-comp small-ear-in insert-past

      ‘Taroo heard slightly that [Hanako tilted her neck slightly].’

   2. (b) Taroo-ga  [Hanako-ga kubi-o kasige-ta]-to ko-mimi-ni hasan-da.

      Taroo-nom Hanako-nom neck-acc tilt-past-comp small-ear-in insert-past

      ‘Taroo heard slightly that [Hanako tilted her neck].’ ≠ ‘Taroo heard slightly that [Hanako tilted her neck slightly].’

   3. (c) Taroo-ga [Hanako-ga ko-kubi-o kasige-ta]-to mimi-ni hasan-da.

      Taroo-nom Hanako-nom small-neck-acc tilt-past-comp ear-in insert-past

      ‘Taroo heard that [Hanako tilted her neck slightly].’ ≠ ‘Taroo heard slightly that [Hanako tilted her neck slightly].’

In (5a), ko-kubi-o ‘small-neck’ and kasige ‘tilt’ are paired in the embedded clause, and ko-mimi-ni ‘small-ear-in’ and hasam ‘insert’ are paired in the matrix clause. As shown by (5b), the fact that the size-indicating prefix ko- is available in the matrix clause alone does not give rise to a SIP interpretation (with the peculiar pattern of adverbial modification) for the combination of kubi-o ‘neck’ and kasige ‘tilt’ in the embedded clause. In addition the presence of the size-indicating prefix in the embedded clause alone does not give rise to a SIP interpretation involving mimi-ni ‘in (his) ear’ and hasam ‘insert’ in the matrix clause of (5c) either. (We note that kubi-o kasige-ru ‘tilt one's neck’ is possibly independent of the morpheme ko-, as seen in (1d) above.)

Continuing with locality, first, we observe that the combination of mini-ni ‘ear-in’ and si-ta ‘did’ is incompatible with the morpheme ko-, see (6a) vs. (6b) (cf. (5a) above). However, if locality is not an issue, we would expect that the size morpheme in the matrix clause can be exclusively associated with the inalienable nominal kubi in the embedded clause and (6c) would be possible.

1. (6)

   1. (a) Taroo-ga [Hanako-ga ko-kubi-o kasige-ta]-to mimi-ni si-ta.

      Taroo-nom Hanako-nom small-neck-acc tilt-past-comp ear-in do-past

      ‘Taroo heard that [Hanako tilted her neck slightly].’

   2. (b) *Taroo-ga  [Hanako-ga ko-kubi-o kasige-ta]-to ko-mimi-ni si-ta.

      Taroo-nom Hanako-nom small-neck-acc tilt-past-comp small-ear-in do-past

      ‘(Int.) Taroo heard slightly that [Hanako tilted her neck slightly].’

   3. (c) *Taroo-ga  [Hanako-ga kubi_i-o kasige-ta]-to ko_i-mimi-ni si-ta.

      Taroo-nom Hanako-nom  neck-acc tilt-past-comp small-ear-in do-past

      ‘(Int.)=(a) above’

2.1.4 Disunity with regular idioms

The empirical exposition so far seems to accentuate semi-idiomaticity and semi-productivity of the construction involving SIP expressions. It might appear that they can be brushed aside as mere idioms about which a general theory of grammar has nothing to say. However, such a conclusion is premature and not at all illuminating. First of all, as mentioned above, SIP expressions retain the core literal meaning of the head inalienable nominal. The expression ko-kubi ‘small-neck’, for example, retains its core meaning ‘someone's neck’ regardless of the size.

Secondly, the interpretation of a sentence containing an SIP expression and its verbal collocational partner is accomplished compositionally with the size-indicating morpheme ko- or oo- participating (albeit unusually) as a semantically regular predicate modifier.

Thirdly, and more significantly, the contrast between (7) and (9) below on the one hand and (10b) on the other renders the difference between the SIP construction and regular idioms incontrovertible.

1. (7)

   1. (a) Taroo-ga [_ADV  ryoo-te to  ko-waki-ni] hon-o  kakae-ta.

      Taroo-nom both-hand  and  small-under.arm-at  book-acc hold-past

      ‘Taroo [held some books with both hands] and [held others under his arm lightly].’ ≠ ‘Taroo [held some books with both hands lightly] and [held others under his arm lightly].’

   2. (b) Hanako-ga [_ADV  hadasi  ka oo-mata-de] arui-ta.

      Hanako-nom bare.foot or big-groin-by walk-past

      ‘Hanako [walked bare foot] or [walked with a big motion of legs].’

      ≠ ‘Hanako [walked with a big motion of bare feet] or [walked with a big motion of legs].’

We note that, for example in (7a), semantic relationship between the SIP expression ko-waki ‘small-under.arm’ and its partner kakae ‘hold’ remains the same as in the example without coordination, in (3k) above. While the SIP expression is the second conjunct NP here, it can exert its unusual adverbial modification. The adverbial modification, however, is limited to Taroo's act of holding books under his arm proper, and does not extend over to the act of holding books in both hands.Footnote ⁷ The same point is illustrated by (7b). As in (7a), oo- ‘big’ in the second disjunct retains the regular adverbial interpretation as in (3i) above and does not extend its adverbial effect to the act of walkingwith bare feet.

It should be noted that an apparent ‘periphrastic’ counterpart of (7a) – with a regular independent adverb but no size-indicating prefix – displays distinct adverbial scope. For example, the regular adverb tyokotto ‘lightly’ in (8) modifies both ‘holding with hands’ and ‘holding with an underarm’. This shows that there is no direct connection between the morphological adverbial modification unique to SIP and the regular ‘periphrastic’ adverbial structure.

1. (8) Taroo-ga [_ADV  ryoo-te to waki-ni] hon-o [_ADV  tyokotto]  kakae-ta.

   Taroo-nom both.hand  and  under.arm-at  book-acc lightly hold-past

   ‘Taroo [held some books with both hands lightly] and [held others under his arm lightly].’ ≠ ‘Taroo [held some books with both hands] and [held others under his arm lightly].’

In addition, coordination of regular and SIP complements (including subjects) is also possible, as seen in (9). Due to the more intimate nature of the dependency between complements and their predicates, care has to be taken to ensure their compatibility. (Hideki Kishimoto (p.c.) contributed (9a).)

1. (9)

   1. (a) Taroo-ga [_NP  ryoo-me to oo-guti-o] ake-ta. (cf. (3a))

      Taroo-nom both-eyes and  big-mouth-acc open-past

      ‘Taroo [opened his eyes] and [widely opened his mouth].’

   2. (b) Hanako-wa [_NP  me ka ko-te-ga] kik-u. (cf. (3n))

      Hanako-top eye or small-hand-NOM be.effective-PRES

      ‘As for Hanako, her eyes are keen or [her hands are dexterous slightly].’

If an SIP expression were part of an idiom, then we would expect other regular idioms to behave equivalently within the context of coordination. This expectation is not supported. In (10a) we have the expression asi-o hippar ‘pull (one's) leg’, which can be construed either literally or idiomatically in the absence of coordination. However, as (10b) shows, the expression in question does not at all retain an idiomatic meaning within the context of coordination.

1. (10)

   1. (a) Taroo-ga Ziroo-no asi-o hippat-ta.

      Taroo-nom Ziroo-gen leg-acc pull-past

      ‘(Lit.) Taroo pulled Ziroo's leg.’

      ‘(Idiom) Taroo derailed Ziroo's effort.’ [No body part is needed.]

   2. (b) Taroo-ga Ziroo-no [_NP  te to asi-o]  hippat-ta.

      Taroo-nom Ziroo-gen arm  and  leg-acc pull-past

      ‘(Lit.) Taroo pulled Ziroo's arm and leg.’ ≠ ‘(Idiom.) Taroo pulled Ziroo's arm and derailed Ziroo's effort.’

The fourth and final piece of evidence for separating SIP expressions and idioms comes from passives. The SIP expression ko-te-o kazas-u ‘small-hand-acc raise-pres (i.e. raise one's hand slightly (in a sheltering motion))’ seen in (3f) above retains the same meaning in an active sentence and its passive counterpart as seen in (11a–a′). However, the idiomatic expression te-o yaku ‘struggle with’ loses its idiomatic meaning when appearing in a passive sentence as shown by the contrast between (11b) and (11b′).

1. (11)

   1. (a) Hanako-ga ko-te-o kazasi-ta.

      Hanako-nom small-hand-acc raise-past

      ‘Hanako raised her hand slightly (in a sheltering motion).’

   2. (a′) Ko-te-ga kazas-are-ta.

      small-hand-nom raise-pass-past

      ‘(Her) hand was raised slightly (in a sheltering motion)’

   3. (b) Ziroo-ga konro-de/syukudai-de te-o yai-ta.

      Ziroo-nom stove-with/homework-with hand-acc  burn-past

      ‘(Lit.) Ziroo burned his hand with a stove.’ [i.e. selecting konro-de] ‘(Idiom.) Ziroo struggled with homework.’ [i.e. selecting syukudai-de]

   4. (b′) Te-ga konro-de/#syukudai-de yak-are-ta.

      hand-nom stove-with/homework-with  burn-pass-past

      ‘(Lit.) (His) hand was burned with a stove.’ ≠ ‘(Idiom.) (He) struggled with homework.’

The contrasts between (7)/(9) and (10b), and between (11a–a′) and (11b–b′) here highlight the disunity between the SIP construction and regular idioms – SIP expressions participate productively in a regular syntactic process (coordination and passives here). A proper account for SIP expressions needs to accommodate at least (i) lexicalized and localized collocational dependencies between SIP expressions and their partner predicates, and (ii) the peculiar adverbial modification – the instigator of morpho-semantic bracketing paradox – resulting in fairly literal, systematic, and compositional semantic interpretation. In the next two subsections, I outline a lexical account that not only offers solutions accommodating these dimensions but also enables us to remain faithful to morphological integrity and direct compositionality.Footnote ⁸

2.2.1 Bracketing paradox in constraint-based lexicalism

There have been lexical accounts for morphological bracketing paradox within the general framework of constraint-based lexicalism. Müller (Reference Müller2003) and Kubota (Reference Kubota2007) introduced below are such examples.

Müller (Reference Müller2003) deals with the German particle verb construction like das Herumgerenne ‘the running around’, where the nominalization morpheme ge- -e ‘repeated event’ is morphologically internal to the particle herum ‘around’, i.e. around′(repeated-event′(run′(x))), contradicting the semantic scope of the predicates involved. His solution consists of two components: (i) a lexical rule responsible for ‘productive’ particle-verb combination and (ii) another lexical rule for ge- -e nominalization. The former rule changes the combinatory property of a regular verb (with an empty VCOMP list – VCOMP is a feature to handle a verbal complex) and adds (to the VCOMP list) an adverbial particle as a subcategorized modifier of the verb. The latter rule applies to all verbs and gives rise to a noun stem.

To obtain Herumgerenne, rule (i) is employed first to create a verb renn-, which selects a particle via VCOMP. Then rule (ii) gives rise to a nominalized counterpart of this verb Gerenne-. The nominalized verb retains the VCOMP specification and combines with a particle herum to yield Herumgerenne. The importance of rule (i) is furnishing a place holder for the semantic contribution of an adverbial particle that is yet to be combined with the verb. Rule (ii) encapsulates the new verb meaning (along with the contribution of the particle) under its own semantic interpretation, as in λ ${\cal P}$.repeated-event′(${\cal P}$(run′)(x)) (${\cal P}$: a variable for a particle meaning) which eventually ends up being repeated-event′(around′(run′)(x)). Thus, the surface morphological combinatory order and the semantic scope of predicates that are mutually inconsistent are reconciled. (See also Ackerman & Webelhuth (Reference Ackerman and Webelhuth1998) and Egg (Reference Egg and Müller2004) for different accounts of German particle verbs that are also compatible with constraint-based lexicalism. Egg's observation regarding bracketing paradox in general and the nature (‘iconicity’) of syntax–semantics interface will be taken up again in Section 4.)

Turning our attention to Japanese, we find a similar morpho-semantic paradox, in (12), involving adverbial modification in the causative construction. In this case the adverb gakko-de ‘at school’ modifies either the main verb hasir ‘run’ alone or the combination of the verb and the causative suffix -sase.

1. (12) Hanako-ga Ziroo-o gakko-de hasir-ase-ta.

   Hanako-nom Ziroo-acc  school-at run-cause-past

   ‘Hanako made Ziroo run (and running took place) at school.’ ‘Hanako made Ziroo run (and the whole event took place) at school.’

Though there has been a long history of debate regarding mono- or bi-clausality of the causative construction, there is no disagreement that hasir-ase is a single unit morphologically speaking. (Yuhara (Reference Zwarts, Hamm and Hinrichs2008) offers a thorough survey of the issues in this domain stretching over the past five decades. See also Gunji (Reference Gunji1999), Manning, Sag & Iida (Reference Manning, Sag and Iida1999), and Uda (Reference Uda1992) for some observations regarding the clausality issues from the perspectives of constraint-based lexicalism.)

Kubota's (Reference Kubota2007) solution for this problem is allowing the causative lexical rule to be semantically flexible so that different components and sub-components of semantic composition can be accessed. Thus, instead of representing the semantics of the causative sentence (12) simply as CONTENT <cause′(hanako, run′(ziroo))>, it is rendered as CONTENT <λ ${\cal P}$.cause′(hanako, p), run′(ziroo)> (where the causative predicate is a relation between an individual and a proposition; CONTENT is a semantic feature). The latter is a list of semantic components that ultimately gives rise to the former after conversion. The order of the items in the list is important in that, for any two consecutive items on the CONTENT list, the right-hand element is a semantic argument for the one (a functor) to the left of it. An adverb combining with this lexical item (or the projection thereof) can pick any item from the list as its modificational target. If the first element is chosen from the above CONTENT list, the wide scope (‘whole event’) reading of the adverb is obtained, and the narrow scope reading for the other choice. (As noted by Kubota, however, this does not mean that all morphologically complex predicates display the same range of semantic flexibility: e.g. gakko-de hasiri-wasureru ‘forget running at school’ does not have the narrow scope reading where the adverb modifies the verb hasir alone. We ignore this additional dimension here.)

Müller's (2002) and Kubota's (Reference Kubota2007) approaches (or others similar to them) seem to be ineffective for handling the adverbial modification involving SIP expressions. First, the size-indicating prefixes oo-/ko- are not subcategorized for by inalienable head nouns (or vice versa). In fact, as we have seen above, there is no semantic relationship between the prefixes and inalienable nouns they attach to.

Secondly, there is no ambiguity with regard to the modificational range of the size-indicating prefixes. The adverbial modification associated with the size-indicating prefixes takes effect in an exocentric structural domain exclusively where they do not belong – they are properly contained within an NP but modify a verb outside of it. On the other hand, German particles and regular Japanese adverbs do their work in an endocentric structural domain (i.e. projection of a verb) where they belong naturally.

Related to this state of affairs is the fact that the modificational domain of the SIP prefixes is not the one that is ‘c-commanded’ by the prefixes. (The expression ‘c-command’ here is used for descriptive convenience.) Particles in German and regular adverbs in Japanese c-command the items they modify, i.e. the respective verbal stems/projections. The SIP prefixes do not c-command the verbs they modify. To c-command a verb that they modify, the prefixes have to escape their local (i.e. endocentric) domains, namely an NP, and attach to a position external (i.e. exocentric) to the NP from where they can c-command the verb.

This, however, is problematic for at least the following three reasons: (i) (Overt) extraction extraction from an NP is impossible in Japanese: *Dare-no _iHanako-ga [_NPt _isyasin-o] suki-ka ‘(Int.) Who_i does Hanako like [a picture of t _i]?’. Such extraction is unthinkable especially when an item to be extracted is a bound morpheme, as (2f) above has already demonstrated. (What if it is a ‘covert’ extraction? See Section 3 below for a review of such an account.) (ii) Would semantic flexibility of some sort be able to furnish a ‘semantic c-command’ effect? Such flexibility has to suspend compositional semantics for an NP containing a size-indicating prefix and mix semantic materials taken partially from two distinct noncontiguous semantic domains, respectively. It is doubtful how well-motivated and properly constrained such a partial semantic mixture approach can be for the SIP construction. (iii) Moreover, with regard to coordination, any approach that attempts to get the prefixes to do their work outside of SIP NPs either syntactically or semantically will have to face and explain the discrepancy between the patterns of modification, for example, with a regular adverb, as in (8) above, and modification employing a size-indicating prefix, as in (7).

What is needed, then, is an account where the effect of the unusual adverbial modification can be (lexically) encapsulated/localized within an SIP expression. An account that embodies such a localization effect is proposed in the remaining part of this section. In particular, in Section 2.3 below, it is proposed that SIP nominals are functors that encapsulate the adverbial effect of the nominals within themselves.

2.2.2 Capturing collocation

The central aim of this section is to demonstrate how the colloctional dependency between the SIP expressions and their host predicates is captured. Lexical definitions of the components needed for syntactic analyses are given in (13), employing a set notation. These (and other) lexical items – the central ingredients of the current lexical account – are combined by the syntactic rules mentioned immediately below.

1. (13)

   1. (a) Lexical entry of kasige ‘tilt’

      kasige: {HEAD verb; FORMkasige; SPR <NP_ga[COLLOC]>; COMPS <NP_o[COLLOC]>; SEM kasige′}

   2. (b) Lexical entry of kubi ‘neck’

      kubi: {HEAD noun; SPR <>; COMPS <>; COLLOC ∅; SEM kubi′}

   3. (c) Lexical entry of ko-kubi ‘small-neck’

      ko-kubi: {HEAD noun; SPR <>; COMPS <>; COLLOC kasige; SEM ko-kubi′}

   4. (d) Lexical entry of ko-waki ‘small-under.arm’

      ko-waki: {HEAD noun; SPR <>; COMPS <>; COLLOC kakae; SEM ko-waki′}

   5. (e) Lexical entry of -ni

      -ni: {HEAD adv; COMPS <NP[COLLOC]>; MOD IVP[FORM]; SEM ni′}

   6. (f) Lexical entry of ko-waki-ni ‘small-under.arm-at’

      ko-waki-ni: {HEAD adv; COMPS <>; COLLOCkakae; MOD IVP[FORM]; SEM ko-waki-ni′}

   7. (g) Lexical entry of waki-ni ‘under.arm-at’

      waki-ni: {HEAD adv; COMPS <>; COLLOC ∅; MOD IVP; SEM waki-ni′}

Just as other Japanese verbs, the verb kasige ‘tilt’, illustrated in (13a), comes with the lexical condition enforcing compatibility between the value of its FORM attribute and the values of its complements' COLLOC (collocation) attributes.Footnote ⁹ The attributes HEAD (part of speech), FORM (morphological form), COLLOC (collocation; defined for nouns only), SPR (specifier), COMPS (complements), and MOD (modifier) are head features and shared throughout the projection of the head unless noted otherwise. FORM and COLLOC are atom valued attributes whose values are phonetic strings. The latter can be unspecified for their values (indicated as ∅). Precise definitions of the SEM values (like kasige′) of the lexical entries are given in Section 2.3 below. We note that kasige in (13a) requires both NP_ga and NP_o for its subject (SPR) and object (COMPS) to be saturated and form a sentence. These two NPs are not to contradict the lexical condition above. Since regular nouns (and hence NPs) like (13b) are unspecified (‘∅’) for COLLOC, they are simply compatible with any verb as far as collocation is concerned. The shorthand notations, TVP, IVP, and S, stand for {HEAD verb; SPR <NP>; COMPS <NP>}, {HEAD verb; SPR <NP>; COMPS <>}, and {HEAD verb; SPR <>; COMPS <>}, respectively.

The relationship between an SIP noun (NP) like ko-kubi in (13c) and a verb is different. The SIP noun comes with the specification [COLLOC kasige] indicating its limited syntactic distribution. When combining with the verb kasige ‘tilt’ (13a), the compatibility requirement will be satisfied. As shown in Section 2.2.3, a collocational dependency between an adjunct and its partner predicate is handled differently.

The adverbial ko-waki-ni ‘at small-under.arm’ in (13f) is lexically derived from the SIP nominal ko-waki ‘small-under.arm’ in (13d) via suffixation of the postpositional affix -ni in (13e). The postposition passes up the COLLOC specification of its SIP complement. This is equivalent to what happens in prepositional phrases like to whom in English, where the wh feature specification of the complement NP daughter is inherited by the PP.

This adverbial SIP expression subcategorizes for nothing and contains the MOD feature whose value is specified as IVP[FORM kakae]. Adverbs select verbal projections in HPSG (through the MOD feature) and adverbial SIP expressions are not an exception. However, SIP expressions select VPs with a very specific morphological head shape – unlike regular adverb waki-ni in (13g), ko-waki-ni here requires a VP whose head is of the morphological form kakae (and/or hasam ‘insert, pinch’ depending on individual lexicalization). It should be noted that the COLLOC requirement for verbal arguments (including subjects) proposed above has no bearing on adjuncts.

The lexical items above are combined by the general syntactic rules below and sentences are formed (and interpreted compositionally, as illustrated in Section 2.3 below) reflecting properties of the lexical items. Since no innovation is intended by the current proposal in the domain of syntax, the syntactic mechanisms are not the central focus (with the distribution of the COLLOC attribute being an exception). The syntactic analyses are given simply as the bases for semantic composition. Consequently, I assume a simple set of rule schemata and principles within the general framework of constraint-based lexicalism. The schemata/principles presupposed here are the head feature principle (HFP), head-specifier rule (HSR), head-complement rule (HCR), head-modifier rule (HMR), and the coordination rule (CR) of Sag et al. (Reference Sag, Wasow and Bender2003) (with the semantic part of CR ignored for now. See Footnote footnote 16). Of particular relevance here is the COORDINATION PRINCIPLE, which regulates the distribution of attribute values in coordination. In this paper, (14b) is added to keep track of the morphological shapes of all the head daughters in a coordinated structure.Footnote ¹⁰

1. (14) Coordination Principle

   In coordination:

   1. (a) the values of the HEAD features of a mother node unify with those of the HEAD features of its daughters, unless noted otherwise;

   2. (b) the value of the FORM feature of a mother is the string concatenation (indicated with +) of the values of the FORM features of its head daughters.

Minimal and sufficient syntactic apparatus is employed in this paper just to cover the behavior of SIP expressions (and the additional data introduced in Section 4 below).Footnote ¹¹

2.2.3 Syntactic demonstration part 1: Basic structures

The syntactic analysis of (1a) above is given in (15).

1. (15) Taroo-ga ko-kubi-o kasige-ta. (=(1a))

   ‘Taroo tilted his neck slightly.’

The verb kasige ‘tilt’ selects NP_o[COLLOC kasige] in accordance with (13a). This IVP then combines with the subject NP_ga (Taroo-ga with an underspecified COLLOC value, making it collocationally compatible with any verb) to form V[SPR<>; COMPS<>] (=S). (Attributes and their values, for example the COLLOC specification for Taroo (NP_ga) in the SPR list of (15), are left out when they are not relevant for discussion at hand.)

In contrast, let us consider an ungrammatical example, presented in (16), where conflicting pieces of collocational information are present in a sentence.

1. (16) *Taroo-ga ko-kubi-o arat-ta. (=(4a)) ‘(Int.) Taroo washed his neck slightly.’

In this case, verbs like araw ‘wash’ (seen in (4) above) are incompatible with the SIP expression ko-kubi ‘small-neck’ due to the conflict between the FORM and COLLOC specifications of the verb and the NP, respectively. Unification fails and the S and VP nodes are left undefined (i.e. ‘∼’).

Next, the tree structure involving the adjunct ko-waki-ni ‘at small-under.arm’ is shown in (17). The SIP expression selects a regular IVP whose head is specified as [FORM kakae]. We note that the IVP is selected by the SIP adverbial rather than vice versa. In this way, an SIP adverbial is solely responsible for generating the SIP construction.

1. (17) Hanako-ga ko-waki-ni hon-o kakae-ta. ‘Hanako held books under her arm lightly.’

Since the IVP here is regular even after it combines with the SIP adverbial ko-waki-ni, it can be further modified by a regular adverb like tegiwayoku ‘skillfully’, as seen in (18).

1. (18) Hanako-ga tegiwayoku  ko-waki-ni hon-o kakae-ta.

   Hanako-nom skilfully  small-under.arm-at book-acc hold-past

   ‘Hanako skillfully held some books under her arm lightly.’

2.2.4 Syntactic demonstration part 2: Coordination

Let us now consider how coordinated instances of SIP expressions are treated. We begin with an instance involving an adverbial SIP in (7a), whose analysis is (19). (The symbol ∅ indicates an unspecified string value, rendering ∅ +kakae equivalent to kakae in (19).)

1. (19) Taroo-ga ryoo-te to ko-waki-ni hon-o kakae-ta. (=(7a)) ‘Taroo [held some books with both hands] and [held others under his arm lightly].’

This analysis is almost identical to the one for (17) above. Due to (14b) of the Coordination Principle, the value of the COLLOC attribute of the ADV mother is the concatenation of those of the ADV daughters, i.e. ∅+kakae. Since the regular noun ryoo-te (adverbial here) is unspecified for COLLOC (hence also for MOD) value, it is syntactically (but not necessarily semantically) compatible with an IVP with any FORM value.

Some readers may be uneasy about my treating ryoo-te ‘both-hands’ as adverbial here. Normally, to function as such, it needs the postpositional suffix -ni, as in ryoo-te-ni. Nevertheless, my adverbial analysis is not that far fetched. First, it is quite common that nouns (especially time nouns) render themselves adverbial: e.g. kyoo ‘today’, kinoo ‘yesterday’, and san-nen-mae(-ni) ‘three-year-before-at’. Secondly, as is well known, in the copula environment like that in (20a) below, bare ryoo-te ‘both-hands’ retains an adverbial interpretation without the postposition -ni ‘with’.

1. (20)

   1. (a) Taroo-ga hon-o kakae-ta-no-wa ryoo-te(-ni)-dat-ta.

      Taroo-nom book-acc hold-past-comp-top both-hands-with-cop-past

      ‘Taroo's holding books was carried out with both hands.’

   2. (b) [Taroo-ga ryoo-te(-ni)] sosite  [Hanako-ga kata-te-ni] hasi-o mot-ta.

      Taroo-nom both-hands-in and Hanako-nom single-hand-in chop.stick-acc hold-past

      ‘Taroo held chop sticks in both hands and Hanako in one hand.’

Finally, in the right-node raising (coordination) environment like that in (20b), a bare nominal adverb (without -ni here) is also possible.

Though I will not speculate about the nature of ‘postposition ellipsis’ seen here, what is noteworthy about these examples is that bare nominals remain adverbial when provided with the right kind of environment. In (19) above – another instance of coordination – there is some plausibility in analyzing ryoo-te as adverbial.Footnote ¹²

In contrast to adverbials seen above, coordinated SIP complements are analyzed differently, as is shown in (21).

1. (21) Taroo-ga ryoo-me to oo-guti-o ake-ta. (=(9a)) ‘Taroo [opened both his eyes] and [widely opened his mouth].’

In (9a) above, the regular nominal ryoo-me ‘both-eyes’ (with [COLLOC ø]) and the SIP nominal oo-guti ‘big-mouth’ (with [COLLOC ake]) are coordinated, and the entire unit serves as a complement. As we see in (21), coordinated NP as a whole has the COLLOC value of oo-guti due to (14b) of the Coordination Principle and satisfies the compatibility requirement imposed by the verb ake.

Continuing with coordination, we note that (22) is a consequence of the current syntactic account. SIP expressions with different collocational properties, such as oo-de ‘big-arm’ and oo-guti ‘big-mouth’, cannot be coordinated, and together become a complement of a single verb like hiroge ‘open’.

1. (22) *Hanako-ga [_NP  oo-de to  oo-guti-o] hiroge-ta.

   Hanako-nom big-arm and big-mouth-acc open-past

   ‘(Int.) Hanako [opened her arms widely] and [opened her mouth widely].’

The reason for this is that one daughter, oo-de, is specified as [COLLOC hiroge] (compatible with the verb hiroge) but the other daughter, oo-guti, comes with [COLLOC ake] (incompatible with hiroge). Due to (14b), the mother NP's COLLOC value is hiroge+ake, which is incompatible with the FORM specification of the verb hiroge (and the COLLOC specification of its complement NP_o). This state of affairs is in contrast with cases like (9a) analyzed in (21) above. The first conjunct NP, ryoo-me ‘both eyes’, comes with unspecified COLLOC value and creates no disturbance in coordination vis-à-vis oo-guti, with a specific COLLOC value.

Another consequence of the current syntactic analysis regarding coordination is that a verb that is compatible with an SIP argument cannot be conjoined with a verb that is not as seen in (23).

1. (23) *Taroo-ga ko-kubi-o [kasige (sosite) sawat-ta].

   Taroo-nom small-neck-acc tilt and touch-past

   ‘(Int.) Taroo [slightly tilted his neck] and [sligtly touched his neck].’

The unit with the two verbs ends up being [FORM kasige+sawar] due to (14b) of the Coordination Principle. However, the SIP expression ko-kubi ‘small-neck’ comes simply with the specification [COLLOC kasige] with which the conjoined verb unit is incompatible – (23) does not have the reading indicated here.Footnote ¹³

The following two factors are central to the syntactic account outlined in this subsection: (i) Due to the locality of both argument selection via SPR/COMPS lists and predicate selection by adjuncts, the locality requirement between SIP expressions and their host predicates is guaranteed. (ii) Since the directions of syntactic selection (based on the SPR/COMPS or MOD specification) are distinct, argument–verb relations and adjunct–verb relations receive distinct syntactic treatments. The opposition seen here is reflected in the semantic analyses outlined immediately below.

2.3.1 Semantics of complement SIP expressions

Let us begin with semantics for complement SIP expressions in (15) above. Semantic translations of the constituents in (15) and the one for the entire sentence are given in (24a) and (24b), respectively.Footnote ¹⁴

1. (24)

   1. (a) Translations for the constituents of tree diagram (15)

      kasige-ta ‘tilted’: λT ₁λT ₂.T ₂(λy.T ₁(λx.((kasigeta′(x))(y))))

      ko-kubi-o ‘small-neck’: λQ.ko′Q(iposs′(kubi′))

      Taroo-ga: λP.P(taroo)

      [N.B.: The translation of ko-kubi is lexically obtained by combining kubi′ (a common noun of the type <e, t>) and λPλQ. ko′Q(iposs′P)) (i.e. the translation for ko-). The function iposs′ picks out an individual from P that belongs inalienably to someone (the speaker or the referent of the subject of a predicate with an SIP object) who is contextually appropriate as a possessor. P (corresponding to a possessed body part) can contain plural individuals in the sense of Link (Reference Link1983). The mechanism of iposs′ is left vague. The case markers -ga and -o are identity functions semantically. TYPE(T)=NP′=<<e,t>,t>(i.e. a GQ); TYPE(P)=TYPE(Q)=<e,t>. Tense is ignored.]

   2. (b) Translation for (15)

      ko-kubi-o kasige-ta:

      λT ₁λT ₂.T ₂(λy.T ₁(λx.(kasigeta′(x))(y)))(λQ.ko′Q(iposs′(kubi′)))

      reduction:

      λT ₂.T ₂(λy.ko′kasigeta′(iposs′(kubi′))(y))

      Taroo-ga ko-kubi-o kasige-ta:

      λT ₂.T ₂(λy.ko′kasigeta′(iposs′(kubi′))(y))(λP.P(taroo))

      reduction:

      ko′kasigeta′(iposs′(kubi′))(taroo)

      [N.B.: TYPE(ko′)=<<e,t>,<e,t>>;TYPE(kasigeta′)= <e,<e,t>>.]

The SIP nominal ko-kubi-o ‘small-neck’ as well as the subject NP Taroo-ga are GQs (i.e. sets of properties) and kasigeta ‘tilted’ is a type raised predicate taking GQs as arguments. Lexically, the SIP expression comes equipped with the unusual adverbial modification ko′ ‘slightly’ already internalized. This happens when a lexical rule derives ko-kubi from kubi ‘neck’ ((24a) above) along with other changes, like the alteration of syntactic combinatory characteristics as seen in (13) above. Due to the fact that this SIP nominal takes a predicate meaning as an argument, the peculiar adverbial modification attributed to the size-indicating prefix ko- remains strictly local to the semantic content of the SIP nominal. It should be noted that ko′ is a functor that takes a one-place predicate meaning (i.e. kasigeta′ (iposs′(kubi′))) and returns a one-place predicate meaning.Footnote ¹⁵

Next, the translation for (21) involving coordination is given in (25b) which is just a slight variation of (24b).Footnote ¹⁶

1. (25)

   1. (a) Translations for the constituents of tree diagram (21)

      ake-ta ‘opened’: λT ₁λT ₂.T ₂(λy.T ₁(λx.((aketa′(x))(y))))

      ryoo-me ‘both-eyes’: λQ.Q(iposs′(ryoome′))

      oo-guti ‘big-mouth’: λQ.oo′Q(iposs′(kuti′)) [kuti → guti with sequential voicing]

      Taroo-ga: λP.P(taroo)

   2. (b) Translation for (21)

      ryoo-me to oo-guti-o (coordination of two NPs):

      λQ.Q(iposs′(ryoome′)) & oo′Q(iposs′(kuti′))

      ryoo-me to oo-guti-o ake-ta:

      λT ₁λT ₂.T ₂(λy.T ₁(λx.(aketa′(x))(y)))(λQ.Q(iposs′(ryoome′)) & oo′Q′(iposs′(kuti′)))

      reduction:

      λT ₂.T ₂(λy.(aketa′(iposs′(ryoome′))(y) & oo′aketa′(iposs′(kuti′))(y)))

      Taroo-ga ryoo-me to oo-guti-o ake-ta:

      λT ₂.T ₂(λy.(aketa′(iposs′(ryoome′))(y) & oo′akeeta′(iposs′(kuti′))(y))) (λP.P(taroo))

      reduction:

      aketa′(iposs′(ryoome′))(taroo) & oo′aketa′(iposs′(kuti′))(taroo)

      [N.B.: TYPE(oo′)=<<e,t>, <e,t>>; TYPE(aketa′)=<e, <e,t>>.]

2.3.2 Semantics of adjunct SIP expressions

Semantic interpretation is carried out in the following way when an SIP expression is an adverbial like ko-waki-ni ‘small-under.arm-at’ in (17). The translations of the constituents of the sentence are listed in (26a), and (26b) is the semantic translation for the whole sentence.

1. (26)

   1. (a) Translations for the constituents of tree diagram (17)

      kakae-ta ‘held’: $\lambda T_{\setnum{1}} \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.T_{\setnum{1}} \lpar \lambda x.\lpar {\bf kakaeta}^{\rm '} \lpar x\rpar \rpar \lpar y\rpar \rpar \rpar \rpar $

      hon-o ‘books’: λP.P(hon)

      ko-waki ‘small-under.arm’: λQ.ko′Q(iposs′(waki′))

      ko-waki-ni ‘small-under.arm-at’: λQλx.ko′ni′(iposs′(waki′))(Q(x))

      Taroo-ga: λP.P(taroo)

      [N.B.: ko-waki-ni is obtained by combining λTλQλx.T(λy.ni′ (y)(Q(x))) (i.e. the translation for -ni which combines with a GQ and returns an adverbial modifier) and the translation of ko-waki above. The notation $\vec{A}$ indicates that there are zero or more As. ‘Hon’ here is a plural individual. To focus on the issues of adverbial modification, I eschew the minute quantificational aspects of NPs in this paper. TYPE(ni′)=<<<e,t>,t>,<<e,t>,<e,t>>>; TYPE(T)=NP′; TYPE(A)=<<e,t>,<e,t>>; TYPE(P)=TYPE(Q)=<e,t>.]

   2. (b) Translation for (17)

      hon-o kakae-ta:

      $$\hskip-7pt \lambda T_{\setnum{1}} \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.T_{\setnum{1}} \lpar \lambda x.\lpar {\bf kakaeta}\prime \lpar x\rpar \rpar \lpar y\rpar \rpar \rpar \rpar \lpar \lambda P\hskip-1pt.P\lpar hon\rpar \rpar $$

      reduction:

      $$\hskip-92pt \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.\lpar {\bf kakaeta}\prime \lpar hon\rpar \rpar \lpar y\rpar \rpar \rpar $$

      ko-waki-ni hon-o kakae-ta:

      $$\lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.\lpar {\bf kakaeta}\prime \lpar hon\rpar \rpar \lpar y\rpar \rpar \rpar \lpar \lambda Q\lambda x.{\bf ko}\prime {\bf ni}\prime \lpar {\bf iposs}\prime \lpar {\bf waki}\prime \rpar \rpar \lpar Q\lpar x\rpar \rpar \rpar $$

      reduction:

      λT ₂.T ₂(λx.ko′ni′(iposs′(waki′))(kakaeta′(hon))(x))

      Taroo-ga ko-waki-ni hon-o kakae-ta:

      λT ₂.T ₂(λx.ko′ni′(iposs′(waki′))(kakaeta′(hon))(x))(λP.P(taroo))

      reduction:

      ko′ni′(iposs′(waki′))(kakaeta′(hon))(taroo)

      [N.B.: TYPE(ko′)=TYPE(ni′((iposs′(waki′)))=<<e,t>, <e,t>>; TYPE(kakaeta′)=<e,<e,t>>.]

The SIP adverbial ko-waki-ni is an endocentric modifier for a verb phrase meaning. It combines with an IVP′ and returns the same thing. The peculiar adverbial modification due to ko′ is already localized in the SIP's lexical translation and takes effect when the IVP hon-o kakae-ta ‘held books’ is chosen as a semantic argument by this endocentric modifier. This point is important when we consider coordination of an SIP adverbial and regular adverb in (27).

It should be noted that an optional place holder ($\vec{A}$) for an adverbial modifier is lexically inserted, which enables such a modifier to be combined into the right position in the translation. (The same item could be included in the translations above. However, due to its optionality and the absence of adverbs in those examples, it was not included.)

As the final semantic demonstration, I illustrate how the cases like (7)/tree (19) are handled involving coordination of regular and SIP adverbials. In (27a) we find the translations for the constituents in the example under consideration. A complete semantic translation for the entire sentence is given in (27b).

1. (27)

   1. (a) Translations for the constituents in example (7a)/tree diagram (19)

      kakae-ta ‘held’: $\lambda T_{\setnum{1}} \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.T_{\setnum{1}} \lpar \lambda x.\lpar {\bf kakaeta}{\prime} \lpar x\rpar \rpar \lpar y\rpar \rpar \rpar \rpar $

      hon-o ‘books’: λP.P(hon)

      ko-waki-ni ‘small-under.arm’: λQλx.ko′ni′(iposs′(waki′))(Q(x))

      ryoo-te ‘both-hands’: λQλx.ni′(iposs′(ryoote′))(Q(x))

      Taroo-ga: λP.P(taroo)

      [N.B.: TYPE(A)=<<e,t>,<e,t>>; TYPE(Q)=TYPE(R)= TYPE(S)=<e,t>.]

   2. (b) Translation for example (7)

      ryoo-te to ko-waki-ni (coordination of two ADVs):

      λQλx.(ni′(iposs′(ryoote′))(Q(x)) & ko′ni′(iposs′(waki′))(Q(x)))

      hon-o kakae-ta:

      $$\lambda T_{\setnum{1}} \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.T_{\setnum{1}} \lpar \lambda x.\lpar {\bf kakaeta}{\prime} \lpar x\rpar \rpar \lpar y\rpar \rpar \rpar \rpar \lpar \lambda P.P\lpar hon\rpar \rpar $$

      reduction:

      $$\lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.\lpar {\bf kakaeta}{\prime} \lpar hon\rpar \rpar \lpar y\rpar \rpar \rpar $$

      ryoo-te to ko-waki-ni hon-o kakae-ta:

      $$\eqalign{ \lambda \vec{A}\lambda T_{\setnum{2}}.T_{\setnum{2}} \lpar \vec{A}\lpar \lambda y.\lpar {\bf kakaeta}{\prime} \lpar hon\rpar \rpar \lpar y\rpar \rpar \rpar \lpar \lambda Q\lambda x.\lpar {\bf ni}{\prime} \lpar {\bf iposs}{\prime} {\bf \lpar ryoote}{\prime} \rpar \rpar \lpar Q\lpar x\rpar \rpar \cr {\rm\amp}\ {\bf ko}{\prime} {\bf ni}{\prime} \lpar {\bf iposs}{\prime} {\bf \lpar waki \prime} \rpar \rpar \lpar Q \lpar x\rpar \rpar \rpar \rpar \cr} $$

      reduction:

      λT ₂.T ₂(λx.(ni′(iposs′(ryoote′))(kakaeta′(hon))(x) & ko′ni′(iposs′ (waki′))(kakaeta′(hon))(x)))

      Taroo-ga ryoo-te to ko-waki-ni hon-o kakae-ta:

      λT ₂.T ₂(λx.(ni′(iposs′(ryoote′))(kakaeta′(hon))(x) & ko′ni′(iposs′ (waki′))(kakaeta′(hon))(x)))(λP.P(taroo))

      reduction:

      ni′(iposs′(ryoote′))(kakaeta′(hon))(taroo) & ko′ni′(iposs′(waki′))

      (kakaeta′(hon))(taroo)

The crucial step is the coordination of the SIP adverbial ko-waki-ni ‘small-under.arm-at’ and the regular adverb ryoo-te ‘both hands’. It should be noted that (a) the peculiar adverbial modification is already internalized to ko-waki-ni (as shown in (2a) above as well), and (b) both adverbials are endocentric modifiers and the VP meaning corresponding to hon-o kakae-ta ‘held books’ is distributed over the two modifiers. This renders the peculiar modification associated with ko-waki-ni (namely, that of ko′) applicable only for waki-ni hon-o kakae-ta ‘held books under (one's) arm’. This concludes the demonstration of the semantic analyses.