Available online at www.sciencedirect.com

ScienceDirect

Saturating syntax: Linkers and modification in Tagalog*

www.elsevier.com/locate/linguaLingua 149 (2014) 17--33

Gregory Scontras, Andreea C. NicolaeDepartment of Linguistics, Harvard University, Boylston Hall Third Floor, Cambridge, MA 02138, USA

Received 5 December 2012; received in revised form 3 May 2014; accepted 11 May 2014Available online 27 June 2014

Abstract

Not all instances of composition are saturating in the sense of functional application (McNally, in press). For example, intersectivemodification with adjectives or relative clauses requires a non-saturating mode of composition (cf., e.g., Chung and Ladusaw, 2004). Toaccount for such semantic configurations, composition rules like Predicate Modification (Heim and Kratzer, 1998) or Restrict (Chung andLadusaw, 2004) have been posited. In Tagalog, wherever we find instances of non-saturating composition, we also find the element na/-ng, known in the Austronesian literature as LINKER. Sabbagh (2009, fn.31) points out that LINKER may be analyzed either as a semanticallyvacuous element serving as a ‘‘morphological flag’’ for non-saturating composition (in the style of Chung and Ladusaw, 2004, or as anoperator of type het, eti that composes with a predicate and adds the ‘‘instruction’’ to compose its output with another predicate via a non-saturating composition rule. Both options leave us with the need for a specialized composition rule in addition to functional application sothat two properties may compose in the semantics. Alternatively, Rubin (1994) proposes that the grammar supplies a functional head oftype het, het, etii that composes with two predicates and returns a single predicate denoting the intersection of both -- a head that does thework of modification. Following Rubin, we argue that in Tagalog this element is realized overtly as LINKER. Adopting this latter proposalallows us to simplify the compositional mechanism so that it need not rely on specific processes for modification in addition to functionalapplication. It also makes strong predictions, consistent with the data, about the environments in which we expect LINKER to appear. Weexplore the possibility of expanding the predictive power of our account of LINKER such that modification is implicated wherever it surfaces.© 2014 Elsevier B.V. All rights reserved.

Keywords: Modification; Linker; Semantic composition; Tagalog

1. Introduction

Modification poses problems to all sub-disciplines of linguistics (in addition to the issues raised by the papers collectedin this volume, see McNally, in press, for a recent overview of the difficulties associated with modification, includingdefining the term itself). Here, we focus on a particular problem modification presents to theories of semantics, syntax, andtheir interface. Specifically, we concern ourselves with the modes of composition modification necessitates, not only interms of meaning composition but also the phrase structure needed to generate these meanings (e.g., Chung andLadusaw, 2004). As we show, accounting for modification in any cohesive theory of grammar requires a tradeoff betweenparsimony and complexity across the various modules of language involved: either we burden the syntax with additionalstructure so as to simplify our inventory of composition rules in the semantics, or we augment the semantics with non-saturating modes of composition so as to trim down our syntax.1 Deciding the outcome of this tradeoff, both withinlanguages and across grammar more generally, need not bottom out in an argument around preferences: our theory of

* Corresponding author. Tel.: þ1 312 305 0394.E-mail addresses: scontras@fas.harvard.edu (G. Scontras), nicolae@fas.harvard.edu (A.C. Nicolae).

1 We do not intend for this to be an exhaustive list of the options available to theorists as they allow for modification. For example, we make nomention here of lexical strategies that assign to modifiers an argument-taking semantics (e.g., McNally, in press, Section 1.3); but see Section 2.1below.

http://dx.doi.org/10.1016/j.lingua.2014.05.0050024-3841/© 2014 Elsevier B.V. All rights reserved.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3318

modification should find its foundations in empirical facts. To this end, we offer the following case study of modification inTagalog. Throughout our investigation, we are guided by the assumption that if modification proceeds on the basis ofadded functional structure, there ought to be a reflex of that added structure in all instances of modification.

We begin by considering what modification is and how it is handled in theories of formal semantics. Our discussionfocuses on the process of non-saturating composition, which many accounts of modification assume yields the meaningthat results when a modifier combines with another term. We then turn to the data from Tagalog, showing that wheneverwe find modification LINKER surfaces. This observation leads us to consider the syntactic, semantic, and morphological roleof LINKER in Tagalog and linking particles more generally. We conclude that the evidence supports a view of LINKER underwhich it projects a Mod(ification)P in the syntax and, solely on the basis of saturating composition, performs the semanticrole of modification by combining with the elements that flank it (Rubin, 1994, 1997, 2003). This approach has theadvantage of reducing the modes of composition in Tagalog to just Functional Application, supporting the original intuitionfrom Frege that all composition proceeds in a saturating manner. Establishing this precedent, we may then begin towonder whether all composition within Language more broadly proceeds in a similar, saturating way.

2. Theoretical background: composition

The composition of the meaning of a whole out of the meanings of its parts is often viewed as a process of completion.Assuming that composition is binary, the meaning of one of the composing elements, often the predicate, is incomplete, orunsaturated; it combines with a complete, or saturated expression, and the meaning that results is a function of themeaning of the two parts.

This talk of meaning composition in terms of functions, saturation, and completeness finds its beginnings in the writingsof Gottlob Frege. In his essay ‘‘Function and Concept’’ (Frege, 1891), Frege says of the elements involved in meaningcomposition that ‘‘a function by itself must be called incomplete, in need of supplementation, or unsaturated,’’ whereas theargument to which the function applies is ‘‘a whole complete in itself, as the function is not.’’ Here is where we find thefamous example Caesar conquered Gaul, which Frege separates into the parts Caesar and conquered Gaul; the latter heviews as a predicate function which applies to the individual argument Caesar and returns a truth value determined bywhether or not Caesar conquered Gaul.

Modern theories of formal semantics build on Frege’s characterization of meaning by assuming that semanticcomposition proceeds in terms of functional application (e.g., Montague, 1973): a predicate, viewed as a function, appliesto its argument in order to saturate its otherwise unsaturated meaning.

(1)

Functional Application (Heim and Kratzer, 1998): If α is a branching node, {β, γ} is the set of α’s daughters, and vβb is a function whose domain contains vγb,then vαb = vβb(vγb).

Assuming the predicate conquered Gaul corresponds to β in (1) and the subject Caesar corresponds to γ, we cancompose the meaning for Caesar conquered Gaul by applying the predicate to its argument. Functional application mapsintuitively onto theories of argument structure in syntax: a head takes its complement (and then its specifier) as anargument.

Unfortunately, our theory is incomplete: not all instances of composition are saturating in the sense of functionalapplication. Furthermore, head complementation is not the only option for combining elements syntactically. Consider, forexample, what happens when two unsaturated, incomplete elements compose, say two predicates like American andcitizen. Examining our intuitions on meaning, a phrase like American citizen denotes a hybrid of the two properties thatconstitute it: for something to hold the property American citizen, it must independently hold the properties American andcitizen. Assuming both elements are property denoting functions, neither may serve as an argument of the other. As we shallsee, all such instances of semantic modification ostensibly preclude functional application and head-complement structures.

2.1. Allowing modification

What we need is an account of modification, but before getting there we first must consider what the role of a modifier is.According to McNally (in press), modifiers satisfy the property in (2).

(2)

Modifier (McNally, in press): An expression that combines with an unsaturated expression to form another unsaturated expression of thesame type.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 19

To make sense of the definition in (2), one must understand meaning within a Montague-style system of semantic types. Inthe absence of modifiers, a one-place predicate of type he, ti (i.e., a function from individuals to truth values) composeswith an individual-denoting argument, type e, via functional application; the resulting meaning, type t, is a function ofwhether the individual holds the relevant property. When two elements of type he, ti compose to form a new predicate oftype he, ti we must evoke modification: something unsaturated combines with an expression and the result has the sameunsaturated semantic type. But with functional application as our only mechanism by which to compose meaning, wehave no way to perform modification semantically.

We encounter three styles of approaches to the semantics of modification.2 In the first, which we term the LEXICAL

approach, all composition is saturating in the sense of functional application. The approach is lexical in that it burdensmodifiers themselves with the semantics of modification: either modifiers like American are ambiguous betweenproperties (type he, ti) and property modifiers (type het, eti), or a process of semantic type-shifting changes them from oneuse into the other. It is important to leave open the possibility for modifiers to have both a modificational and a propertysemantics given their ability to serve as both modifiers and predicates.

(3)

2 For

to McNa

a.

a muchlly (in

the American citizen

fuller discussion of the first two apress).

modifier

b. The citizen is American. predicate

A LEXICAL approach to modification has the benefit that all composition is saturating in the sense of functional application,but this move is not without its cost: we have hypothesized a very rich and complex lexicon.

The second approach to semantic modification, which we call COMPOSITIONAL, posits a distinct mode of composition formodification. Under this approach modifiers, like predicates, are properties of type he, ti. Thus, the approach avoids theproblem of rampant ambiguity and/or type-shifting inherent to the LEXICAL approach, but as McNally notes, it introducescomplexity elsewhere into the grammar: ‘‘a smaller inventory of composition rules requires more complex assumptionsabout lexical items; a more parsimonious lexicon requires a richer inventory of modes of composition’’ (McNally, inpress:13).

Different authors have spelled out what modification requires in a mode of composition (e.g., Larson, 1998; Heim andKratzer, 1998; Chung and Ladusaw, 2004). Here we present the version from Heim and Kratzer (1998). The goal is toallow two property-denoting nodes to compose and form a new property whose extension is the intersection of thedenotations of its daughter nodes. In this way, we account for the intuition that American citizen names the property ofbeing American and being a citizen. The rule Predicate Modification (Heim and Kratzer, 1998) accomplishes just that: theresult of combining two elements of type he, ti is itself of type he, ti (satisfying the definition of a modifier in (2) above), and itdenotes the intersection of the two composing predicates.

(4)

Predicate Modi ¯cation (Heim and Kratze r, 1998 )If α is a branching node, {β; γ } is the set of α’s daugh-ters, and [[β]] and [[γ ]] are both in D e;t , then[[α]] = λx∈De. [[β]](x) = [[γ]](x) = 1.

αe; t

βe; t

γe; t

By assuming an invariant predicate-denoting semantics for modifiers the COMPOSITIONAL approach allows for their predicativeuses (cf. (3b)); modification uses arise when predicates compose with other predicates. While this style of approachavoids the word-level complexity of the LEXICAL account, it is not without complexity of its own: we must abandon Frege’sintuition that composition is functional application and instead augment our semantics with a special rule like predicatemodification.

The third type of approach to modification has received relatively little attention in the syntax and semanticsliterature. In it, composition is functional application and modifiers receive predicate denotations. Instead of burdeningthe lexicon or the semantics, this approach assumes that modification is carried out on the basis of a functional headwhich takes the composing predicates as arguments. Because of the added structure required, we call this approachSTRUCTURAL.

Rubin (1994) (but see also Rubin, 1997, 2003) provides such a STRUCTURAL approach to modification. He motivates theexistence of a functional head Mod, which projects a Modification Phrase. Mod receives a semantics under which it takestwo predicates as arguments and returns their intersection, thus performing the role of predicate modification on the basisof purely saturating composition.

pproaches (LEXICAL and SEMANTIC) and their relative strengths and weaknesses, we refer the reader

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3320

(5)

[[Mod]]=a. λP e;t .λQ e;t .λx.P(x)=Q (x)=1b. XP

e; t

XPe; t

ModPet; et

Mod0

et; et;etPrede; t

Mod0 first composes with a predicate, type he, ti, and returns a predicate modifier, type het, eti. The predicate modifier thencomposes with the second predicate and returns yet a new predicate denoting the intersection of Mod’s two arguments.We thus retain the benefits of the LEXICAL approach, that is, a single mode of composition, as well as those of theCOMPOSITIONAL approach, that is, an unambiguous property semantics for modifiers, without the discouraging complexityinherent to both. Under this STRUCTURAL approach, the complexity associated with modification is attributed to the syntax.

Before attempting to decide in favor of one of these approaches over the others, we first consider data from Chung andLadusaw (2004) highlighting the special morphological status of modification, which they cash out in a version of theCOMPOSITIONAL approach. Although Chung and Ladusaw intend for the data they present to support a COMPOSITIONAL

approach, we show that they are in fact consistent with each style of approach outlined above.

2.2. Marking modification

In their discussion of Maori, Chung and Ladusaw (2004) observe that speakers have a choice between two indefinitearticles, he and tētahi. Chung and Ladusaw cite differences in behavior between these two articles as evidence that they‘‘flag,’’ or implicate different modes of composition: tētahi behaves as a normal indefinite in that it is existentially closed andsaturates an argument position of the predicate; he, on the other hand, composes with the predicate in a non-saturatingmanner, leaving the argument position it targets free to be filled by another element.

Chung and Ladusaw begin with the assumption that indefinites are predicative, so that they may not compose directlywith a predicate before an operation like existential closure applies. In support of this view of indefinites as predicates, theauthors show that neither he nor tētahi can be anaphoric, as they both introduce new referents.

(6)

a. Kua tae mai [he manuhiri] ki taku āinga. TENSE arrive to.here a guest to my house ‘Some visitors came to my house.’

b.

Ka tae mai [tētahi taraka tino nui]. TENSE arrive to.here a truck very big ‘A huge truck came.’

Furthermore, both he and tētahi allow narrow scope with respect to negation and other scope-bearing elements.

(7)

a. Kaore anoø kia kite-a e [tētahi tangata] [he rongoā mo te mate parekore]. TENSE.not yet TENSE discover-PASS by a person a cure TENSE of the AIDS ‘No one has yet discovered a cure for AIDS.’

b.

Ka kite-a [he tohora] e [tētahi kaititiro], ka whakatū-ria te haki o tōna kāinga. TENSE see-PASS a whale by a lookout TENSE raise-PASS the flag of his house ‘If a lookout spotted a whale, he would rase the flag of his house.’

However, only tētahiþNP allows wide scope with respect to negation and other logical operators in quantificationalconstructions.

(8)

a. Kāore tētahi tangata i waiata mai. TENSE.not a person TENSE sing to.here ‘A person didn’t sing.’ (= ‘There was a person who didn’t sing’)

b.

Kāore he tangata i waiata mai. TENSE.not a person TENSE sing to.here ‘Nobody sang.’ (≠ ‘There was a person who didn’t sing’)

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 21

According to Chung and Ladusaw, the only difference between these two articles lies in the mode by which theycompose semantically with other elements in the structure. That is, given the amount of overlap in their distribution, thechoice between the two elements cannot be assumed to correlate with either specificity, referentiality or even scope. Theauthors claim then that tētahi signals a saturating mode of composition, akin to functional application, while he flags thatthe mode of composition is non-saturating, akin to predicate modification. Specifically, they take tētahi to type-shift theindefinite from type he, ti to e so that it may serve as the argument of a verb and compose via functional application. He, onthe other hand, induces no type shift but instead signals that the type he, ti indefinite must compose with its predicate sisterin a non-saturating manner via Restrict, a rule similar in nature to predicate modification. Thus, according to Chung andLadusaw (2004), a morpheme may inform our composition mechanism by flagging which mode of composition must beused. Note that these morpheme-specific composition rules are conceptually distinct from Heim and Kratzer’sCOMPOSITIONAL account of modification in (4): the latter is meant to be a global rule handling modification on the basis of thesemantic types of the composing elements, whereas the former operate at the lexical level depending on whether he ortētahi appears. For our purposes it suffices to note the process of morphological flagging Chung and Ladusawhypothesize, and how this process is meant to interact with an expanding inventory of modes of composition.

3. Empirical foundation: LINKER in Tagalog

Although it is arguably the world’s largest family of languages, Austronesian has received relatively little study, owinglargely to the Indo-European roots of linguistics as it emerged from the European practice of philology. Without a tradition ofdescriptive work on which to base investigations of theory, these languages are too often overlooked in the pursuit ofUniversal Grammar. Here, we join much recent work in the endeavor to change this trend: we examine modification inTagalog.

Tagalog, a member of the Austronesian language family (Mayalo--Polynesian: Western Malayo--Polynesian), isspoken by approximately 20 million people, primarily in the Philippines (Gordon, 2005). Like many languages (not justthose within Austronesian; cf. den Dikken and Singhapreecha, 2004; Rubin, 1994) Tagalog has a linking particle thatseems to appear in the context of modification. As we will show, this particle provides empirical support for a STRUCTURAL

approach to modification.As noted in Foley (1976), Tagalog possesses a particle which surfaces as enclitic -ng on words ending in a vowel, and

as na elsewhere. Note that this particle is distinct from the adverb na ‘already’; furthermore, in its enclitic form (-ng), it isimportant not to confuse this particle with the standalone non-subject marker ng, which is phonologically distinct ([-ŋ] vs.[naŋ]). In what follows, we examine the distribution of this particle, henceforth referred to as LINKER.

3.1. The distribution of linker

We begin with adjectival modification.3 Recall that under the assumption that adjectives are predicates, type he, ti, theycannot straightforwardly compose with nouns, which are also taken to be predicates, in a saturating manner. In (9), we seethat when a noun composes with an adjective in attributive position, LINKER is obligatory, (9a).4 Furthermore, the relativeorder of the property-denoting terms flanking LINKER is flexible; compare (9a) and (9a0). If, on the other hand, the adjectiveappears in predicative position, LINKER is prohibited. In (9b), where we ascribe the property of being beautiful to the house,the use of LINKER is inappropriate. In other words, when adjectives are used as predicates LINKER is obligatorily absent.

(9)

3 For

4 The

a.

a discu glossin

bahay

ssion of sg conven

*(na)

imilar fations us

maganda

cts, see Nicoed here are

house

LK beautiful ‘beautiful house’ Rubin (1994:118)

a0.

maganda-ng bahay beautiful-LK house ‘beautiful house’

b.

Maganda(*-ng) ang bahay.

lae anas follo

beautiful-LK

S house ‘The house is beautiful.’ Rubin (1994:118)

Nominal modifiers also surface with LINKER, whereas predicative nominals do not. That is, the combination of NþN inargument position requires LINKER, (10a), but when one noun serves as the predicate of another noun LINKER cannotappear, as in (10b). Again, the relative order of the elements flanking linker is flexible.

d Scontras (2011).ws: S = Subject, NS = Non-subject, LOC = Locative, LK = Linker, COMP = Complementizer.

(10) a. ang doktor *(na) babae

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3322

S

doctor LK woman ‘the woman--doctor’ (i.e., the female doctor)

a0.

ang babae-ng doktor S woman-LK doctor ‘the doctor--woman’ (i.e., the woman, who is a doctor)

b.

Doktor (*na) ang babae. doctor LK S woman ‘The woman is a doctor.’

Adverbial modification surfaces with LINKER as well, similarly to modificational adjectives and nominals. In (11a), we modify

the event of door opening by Fred with the adverb bigla ‘suddenly’ and LINKER appears between the adverb and the modifiedclause. However, when the adverb serves as a predicate of the clause, LINKER is prohibited, as in (11b). The relative order ofthe adverb and the clause it modifies is flexible, as shown in (11a0), like with nominal and adjectival modifiers.

(11)

a. Bigla*(-ng) binukasan ni Fred ang pintuan. sudden-LK be opened NS Fred S door ‘Fred suddenly opened the door.’

a0.

Binuksan ni Fred ang pintuan na bigla. be opened NS Fred S door LK sudden ‘Fred opened the door suddenly.’

b.

Bigla(*-ng) ang pagbukas ni Fred ng pintuan. sudden-LK S opening NS Fred NS door ‘The opening of the door by Fred was sudden.’

We find this same contrast in the distribution of LINKER in the clausal domain. Observe that when we modify a noun with a

relative clause, LINKER obligatorily intervenes between the head noun and the property-denoting relative clause, as in (12).

(12)

a. bahay *(na) nakita ko house LK saw I ‘house that I saw’

b.

ang babae*(-ng) nagbabasa ng diyaryo S woman-LK read NS newspaper ‘the woman who is reading the newspaper’

When the same clause serves as a standalone proposition, i.e., a matrix clause as in (13), LINKER may no longer appear.

(13)

a. Nakita ko (*ng) ang bahay. saw I LK S house ‘I saw the house.’ Rubin (1994:117)

b.

Ang babae(*-ng) ay nagbabasa ng diyaryo. S woman-LK is reading NS newspaper ‘The woman is reading the newspaper.’

Once again, relative order is flexible. Consider the following pair from Aldridge (2004).

(14)

a. libro-ng binili ni Maria book-LK bought NS Maria ‘the book that Maria bought’ (= ‘the book bought by Maria’)

b.

binili ni Maria-ng libro bought NS Maria-LK book ‘the book that Maria bought’

The puzzle we are faced with at this point is what characterizes the distribution of LINKER. More precisely, we ask what role

LINKER serves such that it obligatorily appears in the positions in which we observe it and is prohibited elsewhere. Reviewits distribution in Table 1.

We see that LINKER appears wherever we expect to find non-saturating semantic composition, namely whenever itsflanking elements are both property-denoting. In other words, -ng/na surfaces in the context of modification (cf. Rubin,1994; Sabbagh, 2009; Author, 2011). When we have two property-denoting elements, type he, ti, LINKER is present

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 23

Table 1The distribution of LINKER.

U LINKER � LINKER

Attributive adjective Predicative adjectiveAdverbial modifier Predicative adverbialNominal modifier Predicative nominalRelative clause Matrix clause

between them, potentially flagging that the semantics needs to resort to a compositional mechanism other than functionalapplication, such as predicate modification.5 Alternatively, one can view the contribution of LINKER as that of symmetrybreaking: whenever two elements of the same type need to compose, LINKER intervenes (with or without semantic contentof its own). For the purposes of the immediate discussion, we need not distinguish between an analysis that takes LINKER tobe a functional element performing the role of Predicate Modification (type het, het, etii; cf. Rubin, 1994), and one underwhich LINKER merely serves as a morphological flag of the fact that this non-saturating composition is taking place(cf. Chung and Ladusaw, 2004). Regardless of the approach we take, we make strong predictions about the distribution ofLINKER: whenever modification is implicated we expect LINKER to appear. In the next section we review the facts fromSabbagh (2009) that lead to a similar conclusion.

3.2. Tagalog existentials

As a basic introduction to existential constructions in Tagalog, consider the examples in (15).6

(15)

(17)

5 Anoth(2009) fo

6 All da

a.

er modr discuta in th

May

e of nonssion.is subs

malaki-ng

may

-saturating se

ection come fr

disyerto

m

mantic com

om Sabbag

sa

ano

positio

h (200

Australya.

k

n that could b

9).

e signa

exist

big-LK desert LOC Australia ‘There is a big desert in Australia.’

b.

May babae-ng darating sa bahay ko.

led by

exist

woman-LK came LOC house NS.1SG ‘There was a woman (who) came to my house.’

c.

May-roo-ng ila-ng mga dahilan kung bakit atrasado

LINKER is R

ang

estrict

mga

(Chung

bayad.

exist-there-LK some-LK PL reason COMP why late S PL payment ‘There are a few reasons why the payments are late.’

Existential sentences are formed by may, an existential predicate, followed by a noun phrase, the pivot. Note that may

sometimes appears with roon, which we glossed as ‘there’. We return to this issue below.

As in English, these existential sentences assert the non-emptiness of the set denoted by the pivot. They optionallycontain a locative PP or some other phrase following the noun phrase. Only indefinite, property-denoting pivots areallowed in existential constructions. According to Sabbagh (2009), the restriction on the pivot of a Tagalog existential isthat it must be predicate-denoting, type he, ti, which is directly in line with the Definiteness Restriction of Heim (1987):definite pivots are precluded from existential constructions. For Sabbagh, this restriction follows straightforwardly from thesemantics he gives to the existential predicate may, which he takes to be of type hhet,ti,ti.

(16)

vmayb = λPhe,ti .9x P(x)

Maya.exist

manokchi cken��

sa

LOC

bahay.house

`Th ere’s a chi cken in the house.'

b. t: ∃x chi cken(x)

et; t : P. ∃x P(x) e; t : λλ x. chi cken(x)

and Ladusaw, 2004); see Sabbagh

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3324

As we see in the semantics in (16), may requires that its argument be predicate-denoting. Thus, when the complement ofmay is not predicate-denoting, say of type e, the derivation crashes, as shown in (18).

(18)

*May-(roo-ng)a.exist-there-LK

(si/ni)S/NS

PabloPablo

saLOC

handaa nparty

ko.NS.1SG

(‘Th ere was Pabl o at my party.’)

b. ?? ?

et,t : λP. ∃x P(x)

may

e: Pablo

Pabl o

(Crash)

Now, consider the data in (19) where we see both the particle roon and LINKER. Sabbagh (2009) analyzes roon as asemantically inert ‘there’ (cf. Moro, 1997:Chapter 3), and provides the semantics in (20) for it.

(19)

a. May-roo*(-ng) manok sa bahay. exists-there-LK chicken LOC house ‘There is a chicken in the house.’

b.

May-roo*(-ng) malaki-ng disyerto sa Australya. exists-there-LK big-LK desert LOC Australia ‘There is a big desert in Australia.’

(20)

vroonb = λx. there(x)

According to Sabbagh, the reason LINKER appears in the presence of roon is just as we have stated above: the pivot is

predicate-denoting, type he,ti, and so is roon; in order to compose, we need to resort to non-saturating composition,signaled by LINKER. In (21), we illustrate how this composition proceeds for a sentence such as (19a). Crucially, where non-saturating composition is required between roon ‘there’ and the pivot manok ‘chicken,’ LINKER appears. (Having not yetcommitted to a COMPOSITIONAL or STRUCTURAL approach to modification, we abstract away from the syntax of LINKER in (21).)

(21)

t: ∃x [there(x) ∧ chi cken(x)]

et,t : λP. ∃x P(x)

may

e,t : λx.[ there(x) ∧ chi cken(x)]

e,t : λx.there(x)

roon

LK e,t : λx.chi cken(x)

manok

Lastly, note that when roon is absent, LINKER is prohibited, as in (22).

(22)

May *(-ng/na) manok sa bahay. exist LK chicken LOC house ‘There is a chicken in the house.’

Thus, we find that our prediction that LINKER surfaces whenever non-saturating composition is required extends beyond the

run of the mill cases of modification discussed in the previous subsection: its appearance in existentials with roon isexpected on the basis of the non-saturating composition that yields these constructions.

4. Account: the role of LINKER

Recall the three strategies for accounts of modification that we outlined in Section 2.1. The first was the LEXICAL approach,wherein the burden imposed by modification was levied against the lexicon: either predicates are ambiguous between truepredicate uses and predicate modifier uses, or type shifting derives one use from the other. The second style of approach wasCOMPOSITIONAL: the compositional system itself handles the task of modification on the basis of a special mode of compositionjust for it (e.g., Predicate Modification as in (4)). Lastly, there was the STRUCTURAL approach, wherein modification proceeds viasaturating composition and is mediated by a functional head performing the role of set intersection.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 25

On balance, the STRUCTURAL approach to modification fares best, avoiding the ambiguity required under the LEXICAL

approach and the expanded inventory of composition rules required under the COMPOSITIONAL approach. However, withoutempirical support for added functional structure, the STRUCTURAL approach faces the problem of unnecessarily complicatingthe syntax. The facts about LINKER in Tagalog provide this empirical support for the STRUCTURAL approach to modification:whenever modification is involved LINKER appears.

What we have, then, are evidence of a Modification head in Tagalog; that head is LINKER (Rubin, 1994). LINKER

contributes a functional semantics, composing with two predicates and yielding their intersection, as in (23).

(23)

vLINKERb = λP.λQ.λx. P(x) = Q(x) = 1

We follow Rubin (1994) in assuming that LINKER projects ModP, composing first with one predicate to form a predicate

modifier, type het,eti, which then composes with the second predicate to yield a property, type he,ti. (24) XP

e; t

XPe; t

ModPet; et

Mod0

LINKER

et; et; et

Prede; t

Note that we have thus far placed no restrictions on the order of LINKER’s arguments (cf. den Dikken and Singhapreecha,

2004); all that is required is that the arguments denote predicates. We thus predict flexibility in the order of the elementsthat flank LINKER as in (9), repeated below in (25).

(25)

a. babae-ng maganda woman-LK beautiful ‘beautiful woman’

b.

maganda-ng babae beautiful-LK woman ‘beautiful woman’

Crucially, reversing the order of the elements flanking LINKER in (25) does not affect the meaning of the resultingexpression, which is what we would expect given the intersective semantics for LINKER in (23). Syntactically, ModP appearsto adjoin either to the left or to the right of the modified element -- the linearization process that injects LINKER between themodified and the modifier would be post-syntactic (Nunes, 2004). However, the order of the elements in a LINKER

construction is not so free as the facts in (25) would have us believe. We turn to this point when we consider a broaderrange of LINKER constructions in the next section.

5. Extending the account

Having found LINKER in the context modification, we next consider what an account would require if we push thebidirectional claim that whenever we have LINKER we have modification, and whenever we have modification we have LINKER.So far the examples we encountered have been consistent with these predictions, but we have only looked at clear cases ofmodification. In this section we consider constructions where we have LINKER and yet modification, at least at first glance, doesnot occur.

One such instance is in constructions with complement clauses, such as (26).

(26)

a. Kailangan ni Pedro *(-ng) magbasa ng libro. need NS Pedro LK read NS book ‘Pedro needs to read a book.’

b.

Kailanga*(-ng) magbasa ng libro si Pedro. need-LK read NS book S Pedro ‘It needs to be the case that Pedro reads a book.’

If we view LINKER as always signaling modification, its obligatory co-occurrence with complement clauses suggests thatthese clauses should not be viewed as true complements of the embedding verbs. Rather, these complement clauses

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3326

would serve instead as modifiers of the embedding predicate/clause. Incidentally, this reasoning is in line with recenttheorizing on the syntax of clausal complements, wherein CPs do not saturate argument positions but rather serve asmodifiers; for more discussion on this topic we refer the reader to, e.g., Moulton (2011, 2012); see also Koster (1978).

LINKER also surfaces with demonstratives. Just as with adjectives, the relative order of the terms flanking LINKER is flexible.

(27)

a.

*

ito-ng

libro this-LK book ‘this book’

b.

libro-ng ito book-LK book ‘this book’

As with clausal complements, we may take LINKER’s co-occurrence with demonstratives to signal that these elements arenot taking the nominals as arguments. Instead, demonstratives in Tagalog would be adjectival; that is, modificational(cf. Giusti, 1997, 2002; Brugè, 2002; Grohmann and Panagiotidis, 2004; Umbach and Gust, in press for analyses ofdemonstratives as modifiers).

LINKER surfaces in the context of so-called ‘‘privative’’ adjectives like ‘fake’ and ‘former,’ as illustrated by (28) and (29).While the speakers we consulted expressed a weak preference for the ADJ-N word order for privative adjectives, bothconfigurations are possible.

(28)

a. dati-ng senador former-LK senator ‘former senator’

b.

senador na dati senator LK former ‘former senator’

(29)

a. peke-ng alahas fake-LK jewelry ‘fake jewelry’

b.

alahas na peke jewelry LK fake ‘fake jewelry’

These data suggest that privative, intensional adjectives are modificational (cf. Larson, 1998; Partee, 2010) and thuscompose in a non-saturating, restrictive manner; such a view of privative adjectives would account for LINKER’s appearancewith them.

If we take LINKER as a marker of modification, we may use it as a diagnostic for non-saturating composition in othercontexts. For example, not all adverbials appear with LINKER. Contrast (30a) with (11a), repeated below in (30b).

(30)

a. Umalis (*na) na ang babae. left LK already S woman ‘The woman left already.’

b.

Bigla*(-ng) binukasan ni Fred ang pintuan. sudden-LK be opened NS Fred S door ‘Fred suddenly opened the door.’

Assuming the bidirectional claim regarding LINKER, the presence/absence of LINKER can inform the way we classify the

meaning of adverbial elements into those that are modificational and those that are not. In fact, an anonymous reviewersuggests that na ‘already’ in (30a) is not an adverb at all, but rather an aspect particle or clausal head.

Lastly, LINKER appears with quantificational elements like kuanti ‘few’ and isa ‘one’.

(31)

a. kaunti-ng bata few-LK child ‘few kids’

b.

bata-ng kaunti child-LK few (‘few kids’)

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 27

(32)

7 We w8 For a9 Additi

similar w

(i) a.

b.

a.

*

*

*

ould lik discusonally,

ord ord

ito-nthis-‘thismagbeau‘this

isa-ng

e to thansion of nuan anonyer restrict

g magaLK beaut

beautiful

anda-ng

tiful-LK

beautiful

babae

one-LK woman ‘one woman’

b.

babae-ng

k anmermouions.

nda-iful-Lwombabwomwom

isa

woman-LK one (‘one woman’)

Here is the first case where the relative order of the elements flanking LINKER is not flexible: when one of those elements isquantificational, only the Q--N order is acceptable. In the next subsection we elaborate on the syntax and semantics thatcould yield this word order restriction while preserving the modification role of LINKER.

5.1. Word order matters

Throughout the paper we have assumed that the order between the two constituents flanking LINKER is free, based ondata such as (25), involving an adjective and a noun, and (33), which involves a demonstrative and a noun.

(33)

a. ito-ng babae this-LK woman ‘this woman’

b.

babae-ng ito book-LK this ‘this woman’

This generalization breaks down, however, once we turn to quantifiers and numerals.7 In (34) we see that the onlypossible combination is one in which the quantifier precedes the noun, while in (35) we see that numerals too mustprecede the noun.8 In other words, the order of a quantifier/numeral and its noun phrase is not flexible in the same way thatthe order between an adjective/demonstrative and a noun is.

(34)

a. kaunti-ng bata few-LK child ‘few kids’

b.

bata-ng kaunti child-LK few (‘few kids’)

(35)

a. isa-ng babae one-LK woman ‘one woman’

b.

babae-ng isa woman-LK one (‘one woman’)

Based on these data, it would appear that the order of the elements flanking LINKER is not as free as we might have thought.Review the facts: in a noun-adjective or demonstrative-noun configuration relative order is free, but in a nominal with aquantificational element like kuanti ‘few’ or isa ‘one,’ the order is restricted.9 Moreover, in LINKER constructions phrasal

anonymous reviewer for bringing the data in this subsection to our attention.als as modifiers, see Link (1987), Carpenter (1995), and Landman (2003), among many others.s reviewer points out that nested modification structures involving the demonstrative ito ‘this’ are constrained by

ng babaeK womanan’ae-ng itoan-LK thisan’

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3328

stress occurs on the rightmost element. Perhaps LINKER’s appearance with quantificational elements does not signalmodification, but performs instead a different role. Gone would be our bidirectional claim: LINKER appears wherever wesuspect modification, but modification does not occur whenever we have LINKER. In what follows we consider a potentialexplanation for these word order facts that would preserve the bidirectional claim. We see, however, that this move hasfar-reaching consequences for the grammar of Tagalog.

We have followed Rubin (1994, et seq.) in assuming that ModP, headed by LINKER, adjoins to the element it modifies.10

We have further claimed that adjunction applies either to the right or to the left of the modified element, accounting for theobserved flexibility in word order. Suppose instead that adjunction happens to the left and ModP is head-final, as in (36).Note that the LINKER construction would be one of the only instances of a head-final structure in Tagalog, requiring a mixed-headedness classification for the language. Alternatively, we could take ModP to be head-initial and attribute the wordorder to post-syntactic linearization.

(36)

c.

d.

Howevethe interto future10 See

and numflagging11 Thiskuanti (cit shouldassumin

kuanti-n g bata `few child ren'NP

ModP

XP

kuanti `few'

Mod0

LK

NP

bata `child'

Here we remain agnostic about the syntactic category of the weak quantifier kuanti ‘few’ (and similarly the numeral isa‘one’); for our purposes it suffices to assume that whatever this category is, it is not that of a noun. Now, continuing to takeadjunction to occur on the left in Tagalog, consider the structure for the illicit noun-LK-quantifier construction:

(37)

*bata-n g kuanti (`few child ren')XP

ModP

NP

bata `child'

Mod0

LK

XP

kuanti `few'

Assuming that modifiers adjoin to the left of the modified element, we have a plausible explanation for the word orderrestriction on quantificational structures (cf. (34) and (35)): the noun must appear on the right because the resultingstructure must be a nominal; were the quantificational element to appear on the right the resulting structure would nolonger be nominal, presumably precluding it from argument positions and accounting for its illicit status. Similar reasoningapplies to isa ‘one.’11 We may therefore settle on the structures in (36) and (37) for modification constructions and derive

ito-ng babae-ng maganda

this-LK woman-LK beautiful‘this beautiful woman’

*babae-ng maganda-ng itowoman-LK beautiful-LK this(‘this beautiful woman’)

r, Himmelmann (in press) reports judgments on nested modification structures involving ito and signals that their acceptability depends onpretation speakers get for these phrases. Given that we are unable to do full justice to the status of ito in modification structures, we leave it

work to uncover the facts about these constructions.Nuger (2010) for a similar discussion of LINKER in Palauan; like we do here, Nuger takes LINKER’s presence in the environment of quantifierserals to suggest an adjunction structure. However, Nuger ascribes no semantic content to LINKER (cf. the process of morphological; Chung and Ladusaw, 2004). line of reasoning does not attribute ungrammaticality to noun-LK-quantifier constructions. In fact, we do not predict that the string bata-LK-hild-LK-few) is itself ill-formed, just that it will not be used to refer to few children: it is itself a (complex) quantificational element. Therefore, enjoy uses as a quantificational element. In other words, a string like bata-LK-kuanti-LK-babae (child-LK-few-LK-woman) should be fine,g it has a plausible interpretation.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 29

the word order restrictions for quantificational elements: only when the phrase is headed by a lexical item of theappropriate (i.e., intended) syntactic category will the construction be licit.

But if modified nominals must be headed by a noun, we are left with a puzzle concerning the free order of adjective-noun constructions presented in (25): the order noun-LK-adjective is perfectly acceptable, despite the fact that a nominal isostensibly headed by its rightmost element, the adjective. At play here is a process of nominalization transforming theadjective into a nominal. Crucially, this process cannot apply to quantificational elements like kaunti ‘few’ or isa ‘one,’ asevidenced by the ungrammaticality of nominals headed by these elements. Interestingly, this nominalization processappears to apply to demonstratives like ito ‘this’ in (33), lending further support to our idea above that Tagalogdemonstratives are in fact adjectival. To see this proposed nominalization process active elsewhere in Tagalog, considerthe examples in (38) where maganda ‘beautiful’ and ito ‘this’ serve as (nominal) subjects.

(38)

12 Note

clauses (clauses amodified

a.

that thee.g., Cnd othnoun.

Nadito

adjunctioomrie, 199er noun-m

ang

n ana8): theodifyin

maganda

here S beautiful ‘Here is a beautiful (one).’

b.

makulit ang

lys ag

ito

naughty S this ‘This (one) is naughty.’

Next, we consider a consequence of the left-adjunction syntax of LINKER for the formation of relative clauses.

5.2. Relative clauses

To derive the word order restrictions for quantificational elements, we have considered the structure in (39) for LINKER.The phrase is headed by the right-most element, XP; ModP, itself headed by LINKER, adjoins to the left of this element andmodifies it.

(39)

XP

ModP XP

This structure commits us to the claim that the right-most element of a LINKER construction heads the resultingphrase. With the added assumption that nominals may be derived from adjectives, we correctly predict the wordorder restrictions on modified nominals in Section 5.1: only nouns or adjectives may serve as the right-most elementof a nominal LINKER construction; quantificational elements like kuanti ‘few’ or isa ‘one’ are not (and cannot be) nominaland therefore cannot occur on the right edge, i.e., they cannot head the phrase. Consider now the implicationsthis proposal has for Tagalog relative clauses, which have LINKER intercede between the noun and the clause thatmodifies it.

So far the relative clause constructions we have considered have been postnominal as in (14a), repeated in (14a).Tagalog also allows prenominal relative clauses, (40b).

(40)

a. libro-ng binili

is

rguco

ni

of Mmennstru

Maria

odP -- hets of LINK

ctions (c

book-LK

bought NS Maria ‘the book that Maria bought’ (= ‘the book bought by Maria’)

b.

binili ni Maria-ng libro

f

bought

NS Maria-LK book ‘the book that Maria bought’ (Aldridge, 2004)

Given what we have said about the structure of modification, Tagalog prenominal relative clauses should not immediatelyconcern us: the phrase is headed by the right-most element (i.e., the noun) which is modified by the clause to its left;the resulting construction is nominal and behaves as such.12 The problematic case is the postnominal relative clause,(40a).

re the modifying clause -- does suggest a non-movement analysis of prenominal relativeER are derivationally independent. Moreover, we postulate no difference between relative. Japanese; e.g., Matsumoto, 1997): all such constructions feature ModP adjoined to the

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3330

Consider the structure that would result for a postnominal relative clause like (40a) given the proposed syntax in (39).

(41)

13 Tagadiscussi

libro-LK binili ni Maria `the book that Maria bought'RC

ModP

NP

libro � book'

Mod

LK

RC

binili ni Maria`bought by Maria'

We intentionally leave the resulting category label for the phrase in (41) opaque, signaling only that the category of theclause ‘bought by Maria’ determines the category of the entire structure: RC. Crucially, it is this clause, and not thesupposed head noun ‘book,’ that is modified. Here we face a problem. Conceptually, it is the clause (‘bought by Maria’)that modifies the noun (‘book’): among the relevant books the phrase specifies the one that Maria bought. However, in (41)the inverse holds: among the things that Maria bought, the phrase specifies just the book.

More importantly, the entire phrase in (41) behaves as a nominal. Yet whatever category we attribute to the pre-theoreticallabel ‘‘RC,’’ it seems implausible that this category will be nominal. However, there is precedence for treating the supposedmodifying clause in a Tagalog relativization structure like (41) as a nominal. Nagaya (1998, et seq.) provides evidence thatthe verbs in Tagalog relative clauses may serve as referential expressions (taking markers otherwise associated with nouns,i.e., case, number, and/or definiteness markers), may invert in ways that nouns but not matrix verbs can, and may assigncase like nominals (in a way that matrix verbs cannot). In other words, Nagaya demonstrates the nominal properties of themodifying clause in (41).13 Following Starosta et al. (1982), Nagaya proposes that the focus affix on the clause-internal verbserves as a nominalizer. RC is thus nominal, a predicate of individuals (type he, ti); what we have been glossing as ‘bought byMaria’ is not sentential, but rather a derived nominal denoting the set of things Maria bought. That such a nominal could bemodified by another noun proves unproblematic: ‘book’ restricts this denotation to those things that Maria bought that arebooks. We have already seen instances of noun-noun modification constructions like doktor-LK-babae ‘doctor-woman’ in(10a). That LINKER should intercede between the noun and nominalized clause is predicted: the noun modifies the derivednominal; in order to perform this modification ModP must be projected. In addition to prenominal and postnominal relativeclauses, Tagalog also allows what have been classified as head-internal relative clauses (e.g., Aldridge, 2004). We see noway at present to treat these head-internal relative clauses within our left-adjunction proposal.

5.3. On balance

In this section we have considered a broader range of LINKER constructions. Our aim has been to extend the claim regardingLINKER’s contribution: not only does LINKER surface whenever modification occurs, but LINKER surfaces only when modificationoccurs. We would then arrive at the bidirectional claim: if modification, then LINKER; and if LINKER, then modification.

We saw that complement clauses, demonstratives, and privative adjectives would implicate modification in theirsemantics. We also saw that there is precedent for treating such constructions with modification. More problematic arequantificational elements like kuanti ‘few’ and isa ‘isa’. Here we saw the only case of a word order restriction involvingLINKER: in quantifier-noun constructions, the quantifier must precede the noun. To derive this word order, we hypothesizedthat ModP adjunction only occurs to the left of the modified elements. Quantifiers cannot appear on the right of a nominalLINKER phrase because they cannot head that phrase. The costs of this move are many, all of them associated with thepredicted right-headed-ness of LINKER constructions.

While we believe there is hope for the bidirectional claim regarding the contribution of LINKER, at present we hold only tothe weaker claim: whenever modification occurs, LINKER surfaces. The semantics of LINKER, together with the syntax itprojects, perform modification solely on the basis of saturating composition and adjunction structure. We leave it to futurework to explore more fully the possibility of modification in the constructions discussed in this section.

6. LINKER AND ANO

With an analysis of LINKER in hand, we now confront one last potential problem for our approach, which centers aroundthe issue of wh-words in Tagalog existentials. Consider the contrast in (42), where we see that ano ‘what’ but not sino‘who’ may serve as the pivot of an existential.

log is not alone in employing this nominalization-as-relativization strategy. See Comrie and Thompson (1985), Comrie (1998) foron.

(42) a. Ano ang mayroon t sa bahay ni Juan?

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 31

14 An anbe able todenoting

Juan ‘wh

*

*

onymo compoexpresat is the

us reviese direcsion likere in Ju

wer sutly with

roon inan’s ho

ggests that the the existentia

the pivot of ause?’ (Sabbag

tral prn eh,

ce of

edicatexistent2009:3

ano could may, itseial. Howe9). We th

be mlf typver, rerefo

odificatioe het, ti. Ioon’s prere conclu

what

S exist.there LOC house NS Juan ‘What is there in Juan’s house?’

b.

Sino ang mayroon t sa bahay? who S exist.there LOC house (‘Who is there in the house?’) (Sabbagh, 2009)

This contrast is similar to the facts in English, where we see that what can serve as the pivot of an existential but who isless acceptable (Heim, 1987, judgments attributed to Safir, 1982).

(43)

a. What is there in Austin? b. ? Who was there in the room when you got home?

In the discussion of Tagalog sino ‘who’, Author (2011) follows Heim (1987) in assuming that ‘what’ may leave behind aproperty-denoting trace of type he, ti. With a trace of this type, ‘what’ does not violate the Definiteness Restriction banningindividual-denoting pivots. Unlike ‘what,’ ‘who’ leaves behind an individual-denoting trace, thus its inability to serve as apivot of an existential.

As we saw in Section 3.2, in the presence of roon, property-denoting (type he, ti) pivots surface with LINKER realized onroon. Assuming that ano ‘what’ moves from the pivot position and leaves behind a trace of type he, ti, we might expect tofind LINKER signaling the composition of roon and this trace. Interestingly, LINKER does not surface with ano (or with sino, forthat matter) as we see in (44).

(44)

a. Ano ang may-roon(*-g) t sa bahay ni Juan?

nal

n othsende t

what

S exist-there-LK LOC house NS Juan ‘What is there in Juan’s house?’

b.

Sino ang may-roon(-g) t sa bahay? who S exist-there-LK LOC house (‘Who is there in the house?’) (Author, 2011)

The absence of LINKER in (44a) presents a puzzle: either we have mischaracterized the distribution of LINKER and it does notsurface whenever we have modification, or Author (2011) misidentified the semantic type of the trace of ano ‘what’ suchthat in fact it may compose in a saturating manner with roon. We consider this second tact first.

In order for the trace that ano leaves behind to saturate the argument of roon, it would have to be individual denoting,type e.14 Ignoring the fact that this sort of trace violates the Definiteness Restriction presented above, once roon has itsargument saturated the resulting structure is propositional in type, and so it cannot compose with the existential predicate;we expect the derivation to crash when the existential predicate tries to compose with the unit that results from thecomposition of roon and an individual-denoting expression, the trace. We illustrate this point in the tree below.

(45)

???

e; t ; t : λP.∃x P(x)

may

t: there(x)

e,t : λx.there(x)

roon

e: x

(Crash)

Without changing our assumptions about the semantic type of the trace left behind by ano, we retain the characterizationof the contrast between sino and ano from Author (2011): the semantic types of the traces left behind by these wh-wordsdiffer such that ano’s trace, type he, ti, satisfies the Definiteness Restriction whereas sino’s trace, type e, does not and so itcannot serve as the pivot of an existential.

just like ModP, type het, eti. However, such a trace would noter words, ano would always need to appear with a predicate-ce with ano is optional, as in ano ang may(-roon) sa bahay nihat a modificational analysis of the trace of ano is untenable.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--3332

Why, then, is LINKER absent in an ano-existential with roon, as in (44a)? By characterizing its presence in terms ofwhether or not we have modification, and by assigning both roon and the trace left behind by ano predicate-denotingtypes, our analysis predicts that LINKER should be present; how else would roon and the trace of ano compose? While theissue requires further, more detailed study, at this point we hypothesize that LINKER is pronounced only when the elementsparticipating in modification overtly flank it. Because the elements involved in the case of an ano-existential are roon and atrace, that is, one of the elements is covert and thus not phonologically realized, either the phonological or the syntacticpressure to break up these elements is absent. Under our analysis, LINKER would still be present in the underlying structure,but given the lack of overt material flaking it, it would not be phonologically realized.

7. Conclusion

We began with the observation that not all composition is saturating in the sense of functional application, at least not atfirst blush. The difficulty lies in accounting for modification, by which two unsaturated elements compose to form a newunsaturated element. We outlined three possible approaches to modification, LEXICAL, COMPOSITIONAL, and STRUCTURAL. Weshowed how a STRUCTURAL approach allows for the most parsimony in terms of the lexicon and the inventory of compositionmodes it necessitates. However, a STRUCTURAL approach requires empirical support, lest we posit unnecessary functionalstructure.

Following Rubin (1994), we find empirical support for a STRUCTURAL approach to modification in the Austronesianlanguage Tagalog. In Tagalog, the particle LINKER surfaces whenever modification is implicated. We thus have support fora Mod(ification)P headed by LINKER. ModP adjoins to either side of a modified phrase, deriving the flexible word order weobserve for modification structures. The semantics assigned to LINKER takes it to be an element that performs the role ofnon-saturating composition on the basis of solely saturating composition. We further considered the possibility thatmodification always occurs when LINKER surfaces. Word order restrictions concerning quantificational elements led us toabandon this line of reasoning; at present we treat LINKER and the homophonous particle in quantificational structures asseparate items. We have, however, highlighted a plausible approach to aligning these two elements that restricts ModPadjunction to the left of modified phrases.

There remains the possibility that LINKER does not perform modification, as the STRUCTURAL approach proposes, butrather flags it in the sense of Chung and Ladusaw (2004). We could have a COMPOSITIONAL approach under which there is aseparate mode of composition that is flagged by the presence of LINKER. While we see no way of deciding between thisflagging approach and our STRUCTURAL approach, the latter avoids the trouble associated with specifying how the processof morphological flagging interacts with the other modules of grammar. Were we to adopt the flagging approach andanalyze LINKER merely as a morphological cue for predicate modification, we would be faced with a morpheme that has apredictable distribution and yet no semantic content. Thus, we take the evidence from Tagalog to suggest that predicatemodification is not a necessary composition rule, and that where it has previously been posited, we in fact have afunctional head responsible for modification. In this way, we do away with the need for non-saturating semanticcomposition and the flagging thereof, and we simplify our compositional inventory: functional application is sufficient, atleast in Tagalog. The task now is to investigate linking particles and modification across language more broadly to seewhether non-saturating composition is ever required.

Acknowledgments

We gratefully thank our Tagalog consultants, Justine Santa Cruz, Henrison Hsieh, and Shaz Kim, for taking time out oftheir busy schedules to work with us. Special thanks go to Maria Polinsky for her guidance and support. Julie Li Jiang,Joey Sabbagh, Tsunekazu Moriguchi, Edward Rubin, and three anonymous reviewers provided insightful comments andhelpful suggestions. Finally, we thank Berit Gehrke and Elena Castroviejo-Miro, together with the audience at theWorkshop on Modification held at CSIC-CCHS in Madrid.

References

Aldridge, E., 2004. Internally headed relative clauses in Austronesian languages. Lang. Linguist. 5, 99--129.Brugè, L., 2002. The positions of demonstratives in the extended nominal projection. In: Cinque, G. (Ed.), Functional Structure in DP and IP: The

Cartography of Syntactic Structures. Oxford University Press, New York, pp. 15--53.Carpenter, B., 1995. Distribution collection and quantification: a type-logical account of plurality. In: Carpenter, B. (Ed.), Lectures on Type Logical

Semantics. MIT Press, Cambridge, MA.Chung, S., Ladusaw, W.A., 2004. Restriction and Saturation. MIT Press, Cambridge, MA.Comrie, B., 1998. Rethinking the typology of relative clauses. Lang. Des. 1, 59--86.Comrie, B., Thompson, S.A., 1985. Lexical nominalization. In: Shopen, T. (Ed.), Language Typology and Syntactic Description, Grammatical

Categories and the Lexicon, vol. III. Cambridge University Press, Cambridge, UK, pp. 349--398.

G. Scontras, A.C. Nicolae / Lingua 149 (2014) 17--33 33

den Dikken, M., Singhapreecha, P., 2004. Complex noun phrases and linkers. Syntax 7, 1--54.Foley, W., 1976. Comparative Syntax in Austronesian (Ph.D. thesis). University of California, Berkeley.Frege, G., 1891. Function and concept. In: Beaney, M. (Ed.), The Frege Reader. Blackwell, Oxford, pp. 130--148.Giusti, G., 1997. The categorical status of determiners. In: Haegeman, L. (Ed.), The New Comparative Syntax. Longman Publishing Group,

New York, pp. 95--123.Giusti, G., 2002. The functional structure of noun phrases: a bare phrase structure approach. In: Cinque, G. (Ed.), Functional Structure in DP and

IP: The Cartography of Syntactic Structures, vol. 1. Oxford University Press, New York, pp. 54--90.Gordon, R.J., 2005. Ethnologue: Languages of the World, 15th ed. SIL International, Dallas, USA.Grohmann, K., Panagiotidis, P., 2004. Demonstrative doubling in Greek. In: Chandra, P., Fujii, T., Soltan, U., Yoshida, M. (Eds.), University of

Maryland Working Papers in Linguistics, vol. 13, pp. 109--134.Heim, I., 1987. Where does the definiteness restriction apply? In: Reuland, E., ter Meulen, A. (Eds.), The Representation of (In)definites. MIT

Press, Cambridge, MA, pp. 21--42.Heim, I., Kratzer, A., 1998. Semantics in Generative Grammar, Blackwell Textbooks in Linguistics. Blackwell Publishers Inc./Blackwell Publishers

Ltd., Maiden, MA, USA/Oxford, UK.Himmelmann, N.P., in press. Notes on noun phrase structure in Tagalog. In: Ewing, M., Musgrave, S. (Eds.), Special issue of Australian Journal of

Linguistics. To appear in a Festschrift (in press).Koster, J., 1978. Locality Principles in Syntax. Foris, Dodrecht.Landman, F., 2003. Predicate-argument mismatches and the adjectival theory of indefinites. In: Coene, M., D’Hulst, Y. (Eds.), From NP to DP: The

Syntax and Semantics of Noun Phrases. John Benjamins, Amsterdam, Philadelphia, pp. 211--237.Larson, R., 1998. Events and modification in nominals. In: Strolovitch, D., Lawson, A. (Eds.), Proceedings from SALT VIII. CLC Publication, Ithaca,

NY, pp. 145--168.Link, G., 1987. Generalized quantifiers and plurals. In: Gärdenfors, P. (Ed.), Generalized Quantifiers. D. Reidel, Dordrecht, pp. 151--180.Matsumoto, Y., 1997. Noun-modifying Constructions in Japanese. John Benjamins, Amsterdam.McNally, L., in press. Modification. In: Aloni, M., Dekker, P. (Eds.), Cambridge Handbook of Semantics. Cambridge University Press, Cambridge

(in press).Montague, R., 1973. The proper treatment of quantification in ordinary English. In: Hintikka, K.J.J., Moravcsik, J.M.E., Suppes, P. (Eds.),

Approaches to Natural Language: Proceedings of the 1970 Workshop on Grammar and Semantics. Reidel, Dordrecht, pp. 221--242.Moro, A., 1997. The Raising of Predicates: Noun Phrases and the Theory of Clause Structure. Cambridge University Press, Cambridge, UK.Moulton, K., 2011. CPs Don’t Saturate: Explaining the Distribution of Clausal Complements. Talks Presented at McGill University (Jan 21),

University of Ottawa (Jan 24), Göttingen University (Feb 22), University of California, Santa Cruz (Feb 6).Moulton, K., 2012. Clause Positions (unpublished MS). University of California, Los Angeles.Nagaya, N., 1998. Tagalog relativization as a lexical process.In: Talk Presented at the 15th Annual Meeting of the Austronesian Formal Linguistics

Association. The University of Sydney.Nicolae, Andreea C., Scontras, Gregory, 2011. How does who compose? In: Clemens, L.E., Scontras, G., Polinsky, M. (Eds.), Proceedings from

AFLA 18. Cambridge, MA, pp. 125--139.Nuger, J., 2010. Architecture of the Palauan Verbal Complex (Ph.D. thesis). University of California, Santa Cruz.Nunes, J., 2004. Linearization of Chains and Sideward Movement. MIT Press.Partee, B.H., 2010. Privative adjectives: subsective plus coercion. In: Bäuerle, R., Reyle, U., Zimmermann, T.E. (Eds.), Presuppositions and

Discourse: Essays Offered to Hans Kamp. Emerald Group Publishing, Bingley, UK, pp. 273--285.Rubin, E.J., 1994. Modification: A Syntactic Analysis and its Consequences (Ph.D. thesis). Cornell University, Ithaca, NY.Rubin, E.J., 1997. The transparent syntax and semantics of modifiers. Proc. West Coast Conf. Form. Linguist. 15, 429--440.Rubin, E.J., 2003. Determining pair-Merge. Linguist. Inq. 34, 660--668.Sabbagh, J., 2009. Existential sentences in Tagalog. Nat. Lang. Linguist. Theory 27, 675--719.Safir, K., 1982. Syntactic Chains and the Definiteness Effect (Ph.D. thesis). Massachusetts Institute of Technology.Starosta, S., Pawley, A.K., Reid, L.A., 1982. The evolution of focus in Austronesian. In: Wurm, S.A., Carrington, L. (Eds.), Papers from the Third

International Conference on Austronesian Linguistics 2. Tracking the Travelers, pp. 145--170.Umbach, C., Gust, H., in press. Similarity demonstratives. Lingua (in press).