Content uploaded by Chiara Melloni
Author content
All content in this area was uploaded by Chiara Melloni on Nov 16, 2017
Content may be subject to copyright.
LINGUE
ELINGUAGGIO
Anno VIII, N. 1, Giugno 2009
SOMMARIO
Presentation: Words Don’t Come Easy
by Denis Delfitto and Chiara Melloni ...................................3
Universal Default Right Headedness and
How Stress Determines Word Order
by Joseph Emonds ....................................................5
Roots, Categories, and Nominal Concepts
by Paolo Acquaviva .................................................25
The Role of Constraints in Compound Formation:
The case of Bare-Stem Constraint
by Angela Ralli and Athanasios Karasimos ............................53
Compounds Don’t Come Easy
by Denis Delfitto and Chiara Melloni ..................................75
The Chinese Language and Some Notions
from Western Linguistics
by Bianca Basciano and Antonella Ceccagno .........................105
Compounding Adjectives
by Francesca Forza, Emiliano Guevara and Sergio Scalise .............137
PRESENTATION:
WORDS DON’T COME EASY
This issue of Lingue e Linguaggio is a monographic volume dedicated to some
fundamental subjects in current morphological theory.
The issue contains four papers presented at the International Workshop
“Words don’t come easy”, held in Verona on November 17-18 2008andor-
ganized by some members of LiVe, a group of linguists active attheDept.
of Linguistics, Literature and Communication Sciences of the University of
Ver o n a . Mo reov e r, t w o fur t h e r the m a tic a l l y r e l a ted a r tic l es are included in this
issue.
The papers collected within this volume explore the boundaries of mor-
phology and its interfaces with prosodic phonology, syntax and lexical seman-
tics, while challenging the traditional conception of morphology and its place
within the language design. A special attention is dedicatedtowordformation
and, in particular, to compounding, an acknowledged interface-phenomenon
for its contiguity with phrasal syntax.
In particular, Acquaviva’s contribution addresses the relation between nom-
inal concepts and nouns conceived as morphosyntactic objects, concluding that
nominal concepts are the value of nouns proper, rather than roots. Hence, the
paper supports the view of category-free roots, i.e. a livelydebatedissueincur-
rent morphological theory. Exploring the relation between syntax and phonol-
ogy and building on a varied set of morphological phenomena, Emonds’ paper
proposes that independently defined phonological patterns determine syntactic
word order, counter to the generative grammarian’s tenet that syntactic con-
structs can determine phonological patterns, but not vice-versa. The paper,
moreover, questions the status of morphology within the language design and
paves the way towards a ‘fully distributed morphology’ implying that no spe-
cial level in the language architecture is reserved for this module.
Further, two contributions are entirely dedicated to compounding phenom-
ena. Delfitto and Melloni’s article presents an original analysis of root com-
pounds in Germanic and Romance and proposes that their morpho-syntactic
and interpretive properties can be explained in compliance with narrow syntax
conditions on Merge and Projection, crucially related to Kayne’s Antisymme-
try model. This proposal contrasts in many respects with Emonds’, presenting
an alternative view of headedness issues in word formation. Ralli and Karasi-
mos’ paper, building on different types of compounds from Standard Modern
Greek and its dialects, explores the interaction between derivation and com-
pounding, claiming that the absence of derivational suffixeswithinModern
3
LINGUE E LINGUAGGIO VIII.1 (2009) 3–4
DENIS DELFITTO AND CHIARA MELLONI
Greek compounds is due to the Bare-Stem constraint, an operation applying to
output configurations.
This issue also includes two contributions on topics strictly related to the
issues of the Workshop. Basciano and Ceccagno’s contribution, on the ba-
sis of Chinese data, adds to the debate on roots and lexical categories (also
raised in Acquaviva’s paper) pointing out the categorial indeterminacy of lexi-
cal items in the Chinese language. Finally, Guevara, Forza and Scalise’s article
addresses a specific class of compounds, labelled as Adjectival Compounds.
The authors discuss the role of the adjective as a modifier in compounding and
analyse the relation between morphological and syntactic formations contain-
ing instances of adjectival modification.
Denis Delfitto and Chiara Melloni
4
COMPOUNDS DON’T COME EASY
Denis Delfitto Chiara Melloni∗
ABSTRACT:InthisarticlewepresentanoriginalanalysisofNNcompounds
in Germanic and Romance, proposing that their morpho-syntactic and inter-
pretive properties can be explained in compliance with narrow syntax con-
ditions on Merge and Projection, crucially related to Kayne’s Antisymme try
model. In particular, we contend that root compounding represents a specific
mode of syntactic computation (“Compound Phase”), whereby two struc-
turally identical syntactic objects – the compound members –getmerged
in a parallel fashion, hence yielding a symmetric configuration that prevents
label projection. Compound Phase computation can thus be seen as a “repair
strategy” allowing the derivation to get a label and convergeattheinterfaces.
On these theoretical grounds, the formal and interpretive contrasts between
Germanic and Romance, and, within each language, the differences between
compounds and prototypical syntactic constructions are essentially derived
from the syntax of Compound Phases, given the independent properties of
the lexical items involved in the computation.
KEYWORDS:rootcompounds,antisymmetry,compoundphases,interface
conditions, economy principles.
1. INTRODUCTION
One of the fields of grammar that traditionally has been taken to provide evi-
dence for the division of labor between morphology and syntax(Halle1973,
DiSciullo & Williams 1987) is the phenomenon of compounding.Compound-
ing can be roughly defined as the combination of two otherwise free morpho-
logical forms inside the same structure. This combination has been claimed
to illustrate a transformation in the lexicon because the structures that result
have properties which seem to fall into the scope of what is known as the
Lexical Integrity Hypothesis (Lapointe 1980), to the extentthattheydonot
allow for movement or other syntactic operations, contain specific morphemes
(called Linking Elements or Compound Markers, cf. Ralli 2008) and are sub-
ject to semantic and formal idiosyncrasy. In this paper we will concentrate on
compounds formed with two nouns (NN compounds) and we will argue that,
against what has been generally claimed in the morphologicaltradition,their
∗Earlier versions of this study have been presented at the 39th NELS Conference, Ithaca
(NY), 7-9 November 2008, and at the International Workshop “Words D on’ t Com e Easy’,
Verona, 1 7 - 1 8 N ovember 2008. We wou l d l i ke t o t h a n k t h e a u d i e nce of these conferences
for helpful comments and suggestions. The present article istheoutcomeofabroaderresearch
project on compounding also involving Antonio Fábregas, Gaetano Fiorin, Bianca Basciano,
whose contribution is gratefully acknowledged.
75
LINGUE E LINGUAGGIO VIII.1 (2009) 75–104
DENIS DELFITTO AND CHIARA MELLONI
properties are fully reducible to syntax and can be explainedbytheneedto
satisfy general conditions defined within the computationalsystemandclearly
related to Kayne’s LCA. Our study will compare NN compounds inGermanic
and Romance languages in order to show (i) that compounding corresponds
to a specific mode of syntactic computation (Compound Phase),wherebythe
two constituents that undergo “Merge” are too “parallel” foroneofthemtobe
able to project its label and (ii) that the differences between these two families
of languages, and, within each language, the differences between compounds
and prototypical syntactic constructions follow from the syntax of Compound
Phases plus independent properties of the lexical items involved in the com-
putation (the compound members). Conceptually, the assimilation of com-
pounding to “narrow syntax” is maintained by arguing that (i)theoperations
activated within the Compound Phase are exactly what is required in order
for the formal requirements on “Merge” to be satisfied when thetwoelements
that are merged are too “symmetrical”, and that (ii) CompoundPhasesrep-
resent a sort of repair strategy that applies to fix an excess ofsymmetrythat
would prevent the computation from proceeding. In this sense, compound-
ing constitutes a mode of computation whereby narrow syntax “corrects” a
protosyntactic configuration (i.e. a structure that does notcomplywithnarrow
syntax requirements), presupposing thus the existence of a non-syntactic (more
symmetric) type of computation (possibly related to protolanguage, much in
Jackendoff’s 2009 sense) that necessarily gets corrected when narrow syntax
supervenes to it.
The paper is divided as follows. First, we will describe the empirical con-
trasts detected with the formal and semantic properties of NNcompoundsin
Germanic and Romance. These empirical contrasts will lead ustodevelop
an analysis according to which the crucial property of NN compounds is that
they constitute the output of a Parallel Merge process resulting in a point of
symmetry, which has to be broken as soon as possible in the grammar of each
language. We will argue that in Germanic one of the two compound members
(arguably, the non-head) is syntactically activated by a formal feature associ-
ated to gender (section 2.1.), whilst in Romance this featurecannotbeacti-
vated, for principled reasons. As we will see below, this leads to the fact that
the operations performed inside the Compound Phase are categorially-driven
in Germanic and semantically-driven in Romance, deriving the presence of
strong interpretive constraints on compounds in Romance vs.theirvirtualab-
sence in Germanic (as recently emphasized by Jackendoff 2009forEnglish).
We wi ll also argue tha t the stud y o f compou nding con tributes in significant
ways to the definition of the Economy Principles that may be assumed to be
active within the language design.
76
COMPOUNDS DON’T COME EASY
2. EMPIRICAL BASIS OF NN COMPOUNDING
In this section we will concentrate on the formal and semanticcontrastsbe-
tween Germanic and Romance compounds.
2.1 Contrasts in formal properties
The first formal difference that catches the attention of someone describing
Germanic and Romance is the position of the head inside NN compounds.
While Germanic languages follow what Williams (1981) dubbedtheRight
Head Rule (1a), Romance NN compounds systematically have theheadtothe
left (1b, cf. Scalise 1984).
(1) a. En. peach man
b. It. uomo pesca
The fact that the head is the underlined constituent can be shown both by se-
mantic and formal considerations. The constituent which imposes its semantic
type to the whole compound is man in the first case – a peach man is a kind
of man – and uomo in the second case – a uomo pesca is a kind of uomo,
not a kind of pesca –(cf. Allen’s1978“ISA”rulewhichrelatessemantic
headedness with semantic hyperonymy). Also, the inflected element is man in
English – the plural of peach man is peach men,not*peaches man –anduomo
in Italian – plural uomini pesca,not*uomo pesche.
The second formal property that differentiates Germanic andRomanceis
that Germanic typically has specific morphemes between the two N’s that form
the compound; these morphemes are occasionally homophonic with the gen-
itive or nominative plural case markers and are familiar as Linking Elements
in the tradition. In (2) we present a list of languages which contain Linking
Elements (or Compound Markers). Notice that no modern Romance language
is listed in this series – although Latin did have this kind of markers, (2f) –,
while all major Germanic languages are (cf. 2a-e).
(2) a. (DE) Hund-e-futter dog+LE+food ‘dogfood’
b. (DU) boek-en-kast book+LE+case ‘bookcase’
c. (EN) doom-s man doom+LE+man ‘pessimist’
d. (NO) arbeid-s-dag work+LE+day ‘working-day’
e. (SW) jord-a-färd earth+LE+voyage ‘burial’
f. (LA) verb-i-velitatio word+LE+dispute ‘verbal dispute’
In contrast, Romance NN compounds do not contain linking elements (apart
from few exceptions, such as so-called neoclassical compounds, and a spe-
cific class of adjectival Spanish compounds). The closest syntactic object in
Romance has been claimed to be some prepositions which occur in so-called
“phrasal / prepositional compounds”, PCs (cf. Johnston & Busa 1999, Bassac
77
DENIS DELFITTO AND CHIARA MELLONI
&Bouillon2001). SeveralexamplesofPCsarelistedin(3). Wewillargue
later that PCs, in spite of this prima facie formal similaritywithLinkingEle-
ments, are the product of a distinct syntactic derivation, somehow related to a
specific type of Compound Phase (the Romance type).
(3) a. Fr. verre à vin glass + P + wine, ‘wine glass’
b. It. occhiali da sole glasses + P + sun, ‘sun glasses’
c. Sp. pantalones (de) campana trousers + P + bell, ‘bell trousers’
Although PCs do not have all the properties of familiar prepositionless com-
pounds, notice that they do not exhibit the properties of normal noun phrases
either. As shown by (4), PCs do not allow a full DP to be licensedinsidethe
complement of the preposition and the compound head to be modified (at its
least, the result of modification is often marked).
(4) a. NP raggi del sole raggi luminosi del sole
rays of the sun rays bright of the sun
‘bright rays of the sun’
b. PC occhiali da sole i. ??[occhiali] colorati da sole
glasses + P + sun glasses + coloured + P +sun
‘sun glasses’ ii. [occhiali da sole] colorati
glasses + P + sun + coloured
‘coloured sun glasses’
Another – often unnoticed or underestimated – difference between Ger-
manic and Romance is that, even though both languages allow for internal
inflectional markers, these are not interpreted in Germanic and are interpreted
in Romance. The German examples in (5) show that there is no correlation
between the number inflection of the non-head noun and its interpretation.
Notice, in particular, that even though a pram transports onechildatatime,
the compound in (5b) contains the plural noun Kinder ‘children’; conversely,
even though a circle is normally composed of more than one individual, the
compound in (5d) contains a noun in the singular.
(5) a. Sonn-en-shein ‘sunshine’ (from OHG gen.sg. -in)
b. Kind-er-wagen ‘pram for one child’
c. Schwein-e-braten ‘roast consisting of one pig or part of a pig’
d. Freund-e s-kreis ‘circle of more than one friend’ (gen. sg.)
However, when Romance compounds – in fact PCs, for reasons to be dis-
cussed below – contain internal inflection, the latter is always interpreted. In
(6a), the (Sp.) singular form sol ‘sun’ is in compliance with the fact that there
is only one sun in our solar system, and the plural form rayas ‘stripes’ reflects
the interpretation that in a striped T-shirt there is more than one stripe.
78
COMPOUNDS DON’T COME EASY
(6) a. gafas de sol ‘sunglasses’, lit. glasses of sun.sg
b. camisa a rayas ‘striped T-shirt’, lit. T-shirt with stripe.pl.
Another formal difference between Germanic and Romance thathasfre-
quently been observed (cf., a.o., Haider 2001) is that Germanic NN compounds
are recursive, while recursion is constrained in Romance. Take, for example,
the famous German compound in (7) or the English cases in (8).
(7) Donau
Danube
dampf
steam
schiff
ship
fahrt
journey
s
LE
gesell-schaft
journeyman-SUF
s
LE
kapitän
captain
s
LE
mütze
cap
‘Cap of the captain of the Danube steam ship company’
(8) a. Volume Feeding Management Success Formula Award
b. winter weather skin troubles
c. health management cost containment services
If we combine this observation with the taxonomy of compoundsfoundin
Bisetto & Scalise (2005), we notice that the compounds in (7) and (8) are sub-
ordinative (in the “extended” sense – we submit – that their interpretation in-
volves complex non-compositional patterns of predicate-argument structure).
The class of subordinative compounds involves some quite constrained forms
of recursion in Romance (here, subordination in the taxonomic sense will be
shown to correspond to an interpretive type whereby the non-head discharges
some position within default predicative structures associated with the lexical
representation of the head), of the kind illustrated in (9), featuring PCs. Recur-
sion in compounds is easily found, in Romance, with NN compounds of the
coordinative type, as illustrated in (10).
(9) a. [macchina
machine
DA
DA
cucire]
sewing
A
A
pedali
pedals
‘pedal sewing machine’
b. [ferro
iron
DA
DA
stiro]
ironing
A
A
vapore
steam
‘steam iron’
(10) bar ristorante pizzeria ‘pizzeria bar restaurant’
regista produttore attore ‘producer director actor’
2.2 Contrasts in semantic properties
Another striking – but curiously enough, largely underestimated – difference
between Germanic and Romance NN compounds refers to the rangeofpossi-
ble readings associated with the compound. The relationshipbetweenthehead
and the non-head in Germanic is purely arbitrary, connected to what Downing
79
DENIS DELFITTO AND CHIARA MELLONI
(1977) called an underspecified R function. The examples below (from 11 to
15) illustrate this interpretative freedom, whereby the semantics of compounds
only depends on context-related encyclopaedic information.
(11) tree man
‘man who is standing besides that tree / who is sitting on this tree /
that usually seats on trees’, ‘man who defends trees or forests’, ‘man
resembling a tree’, etc.
(12) apple-juice seat
i.e. ‘the seat in front of which an apple-juice had been placed’
(13) bike girl
i.e. ‘a girl who left her bike in the vestibule’ (Downing 1977)
(14) cradle song
i.e. ‘a song to lull a child in the cradle to sleep’
(15) Fischfrau ‘fish + woman’ (Heringer 1984)
a. ‘woman that sells fish’
b. ‘woman that has brought fish’
c. ‘woman standing close to fish’
d. ‘woman eating fish’
e. ‘woman looking like a fish’
f. ‘spouse of a fish’
g. ‘woman and fish at the same time’ (i.e. mermaid)
h. ‘woman having Pisces as zodiac’ (German Fisch)
i. ‘woman as cold as a fish’
This situation sharply contrasts with what is found in Romance, where the
semantic relationship between the two N’s inside the compound is very re-
stricted, in ways that will be discussed below and that correspond to the ac-
tivation of the default predicative structures made available by the “Qualia”
information encoded on the head. For prepositionless compounds, the seman-
tic relationship involved typically corresponds to the activation of the Formal
Quale on the head, whereby the non-head specifies one of the properties of the
head (form, dimension, size, etc.) made accessible by the Formal Quale. The
netto result is that Italian allows for only one of the readingvarietiesassociated
with the German compound in (15).
(16) uomo pesce ‘man + fish’
‘man resembling a fish, i.e. sharing some distinctive, formalproper-
ties of a fish’
To close th is section, table 1 collects all the described empirical contrasts be-
tween Romance and Germanic NN compounds. In the next section,wewill
provide an analysis that derives these differences from independently moti-
80
COMPOUNDS DON’T COME EASY
vated properties of lexical items in the two classes of languages, modulo the
computational principles that are inherent to the syntax of Compound Phases.
Germanic Romance
Head Right Left
Linking Elements Ye s ( ex c e p t En.) No
Light Prepositions No Yes
Recursion Yes Limited
Internal inflection Non interpreted Interpreted
Interpretation Unconstrained Constrained
TABLE 1: Germanic – Romance contrasts
3. THEORETICAL ASPECTS OF NN COMPOUNDING
In this section, we will concentrate on the theoretical aspects of NN com-
pounds. Our proposal is that root compounding in general, andNNcom-
pounding in particular, is a case of Parallel Merge, that is, acaseofMerge
between two elements sharing the same level of structural complexity or, be-
ing, anyway, ‘symmetric enough’ to create a Point of Symmetry(inthesense
of Kayne 1994, Moro 2000) when Merge applies to them. Abstracting away,
for the moment, from the precise categorial status of the elements to be com-
bined, we may simply state that NN compounding combines two elements
sharing the structure indicated in (17), proceeding then to combine them as
indicated in (18).
(17) XP XP
XYXY
(18) ?
XP XP
XYXY
As indicated by the question mark in (18), the combination of two symmetri-
cal structures by Parallel Merge produces a situation where none of them can
project its label to the dominating node. The reason is that, given the Antisym-
metry Model (Kayne 1994), projection takes place only when a head combines
with a non-head, the other structures (putting adjunction aside) being filtered
81
DENIS DELFITTO AND CHIARA MELLONI
out by the application of LCA (cf. 19), interpreted as a PF-oriented interface
principle relating hierarchy to linear order.
(19) LCA:
Given a phrase-marker M, be T the set of terminals in M, A the set
of ordered pairs of non-terminals in M such that the first element of
the pair asymmetrically c-commands the second, and d(A) the set of
terminals contained in A. Then d(A) must be a linear ordering of T.
Here, we consider Antisymmetry as a computational conditiononMerge,to
the effect that, when symmetric structures are merged, the label of the projec-
tion cannot be assigned computationally through External Merge (cf. Moro
2008 for a reinterpretation of LCA along these lines). In a nutshell, we assume
that whenever a Point of Symmetry is created, (label) projection is blocked
and the derivation crashes, unless a specific computational strategy is adopted
whose goal is the “resolution” of the Point of Symmetry. We propose that
Compound Phases constitute in fact the activation of such a computational
strategy. It’s not Internal Merge in general that is triggered by antisymmetric
requirements (as in Moro 2000), but the kind of Internal Mergethatisfound
in Compound Phases.
Let us consider the proposal in some more detail. If NN Compounding
is, by hypothesis, the result of Parallel Merge, then it yields a point of sym-
metry, which is barred by the Antisymmetry requirement on Label projection,
defined as an independent condition on External Merge. As a consequence,
the representations in (20) are not well-formed syntactic objects.
(20) a. PoS b. PoS (PoS=Point of Symmetry)
X° Y° XP YP
By independent principles, if follows that NN compounding will pose the same
problem for convergence both in Germanic and Romance: Merge of two iden-
tical nominal structures creates a point of symmetry. However, adopting but
partially reinterpreting the analysis put forward by Moro (2000), we propose
that the point of symmetry in (20) can be broken through movement. In partic-
ular, when one of the two compound members moves to a higher position in the
structure, the PoS disappears, to the effect that the derivation is allowed to con-
verge at the interfaces. Viewed from this perspective, Compound Phases are
syntactic domains in which symmetry is broken in order to allow the derivation
to proceed. They are constituted by the minimal number of syntactic opera-
tions (reducing to instances of External and Internal Merge)thatarenecessary
in order to break the PoS created by Parallel Merge. In particular, this entails
that movement (i.e. Internal Merge) has to apply before a compound can be
82
COMPOUNDS DON’T COME EASY
transferred to the interfaces. This leads to the strategy in (21), which we claim
is universally involved in compounding (thus both in Germanic and Romance).
(21) Insert a head able to attract one of the two elements.
Our proposal is that, for language-specific reasons inherenttothefeature-
constitution of the two compound members, the point in the derivation where
this attracting head is introduced is different in Germanic and in Romance. In
Germanic the attracting head (let’s call it the PROBE) targets the functional
head (let’s call it N) which directly selects the root, for theveryreasonthat
Ncontainsaformalfeaturethatcanparticipateinthesyntactic computation
(in particular, it can be attracted by the PROBE). It is one of our specific con-
tentions that N cannot be activated by a PROBE in Romance, for reasons that
will be investigated below. Therefore, the derivation has towaituntilother
features are introduced in the feature make-up of the two compound mem-
bers, to the effect that they can be attracted by a convenient PROBE. In other
words, the PoS created by merging symmetric elements cannot be broken, in
Romance, simply by merging parallel NPs (i.e. nominal expressions that con-
tain a minimum of functional structure besides the root); theelementstobe
merged have to be “bigger”, so to speak, in order for a PROBE to be able
to target a convenient formal feature within them. It followsthatGermanic
and Romance construct NN compounds with nominal structures of different
heights: NPs in Germanic and (as we will argue) projections involving whole
sets of phi features, which we label
ϕ
Ps, in Romance.
Before discussing the empirical basis for the proposed difference between
Romance and Germanic, let us briefly concentrate on the repairstrategythatis
adopted in order to break the PoS created by merging symmetricelements.As
we have seen, this repair strategy consists (i) in inserting aPROBE(formally:
afunctionalhead)thattargetsasuitableformalfeaturewithin one of the com-
pound members and (ii) in adjoining the targeted compound member to the
constituent obtained by merging the PROBE. In this way, all violations of the
interface conditions are repaired: the targeted compound member comes (via
Movement) to asymmetrically c-command the other (breaking the PoS) and
the resulting structure gets a label (as we will see below), asrequiredbythe
principles governing Merge. We will refer to the chunk of structure that results
from repairing the damage created by a PoS as a Compound Phase (22).
(22) Compound Phase (Cph):
The minimal domain of syntactic computation necessary to satisfy the
labelling requirement imposed by External Merge, after combining
two symmetric elements that create a PoS (Parallel Merge).
We contend that Cph’s are subj ect to a sor t of Earliness Principle, according to
which the repairing operations activated within a Cph can apply only to min-
83
DENIS DELFITTO AND CHIARA MELLONI
imal lexical units. This means that a PROBE within a Cph is not allowed to
target a phrase-marker A (corresponding to one of the compound members) if
there is a phrase-marker B (with B properly included in A) thatqualifiesas
atargetforthePROBE.ThereasonisthatBalreadycontainssyntactically
active grammatical features and can be conveniently lexicalized in the given
language, ensuring convergence at the interfaces. This Principle can be formu-
lated along the lines of (23).
(23) Earliness Principle:
Syntactic operations within a Cph are bound to the minimal struc-
tural chunks that satisfy the conditions defined on these operations
and qualify for interface-convergence.
The conceptual motivation for (23) is provided by the observation that Paral-
lel Merge (that is, a symmetric style of combining lexical units) constitutes a
marked syntactic operation, which got discarded when narrowsyntaxsuper-
vened to the protolinguistic style of linguistic computation. It seems reason-
able to assume that narrow syntax allows repair strategies toapplyinorder
to fix the damage (this is what Cph’s actually are) but only under the condi-
tion that these repair strategies apply as soon as possible. In layman language,
this entails that one cannot build up too large chunks of structure and then
pretend to combine them in the Parallel Mood: no ad-hoc fixing strategy is
allowed to apply at that point. Of course, how large portions of structure one
can build before activating the Cph depends on language-specific conditions
on interface-convergence and lexicalization, as expressedby(23). Thisisin
fact the ultimate reason for the kind of comparative variation that we find in
compounding, and that we purport to explain in this contribution.
In the next section we will provide empirical motivation for the Romance
/Germanicdifferencethatwillturnouttobecrucialinordertoexplainwhy
compounding looks so different in these two groups of languages.
4. A MINIMAL DIFFERENCE BETWEEN ROMANCE AND GER-
MANIC
In the morphological and typological tradition, a difference has traditionally
been made between noun classes and gender. While the first concept is nor-
mally used to refer to a declension class system used to arbitrarily distribute
nouns in several subcategories (Dixon 1968), the concept of gender refers to a
morphological property of nouns which is crucially involvedinsyntacticpro-
cesses such as agreement. Although some authors do not differentiate between
these two concepts and claim that gender is just one possible way of classify-
ing nouns (Corbett 1991), the case of Romance shows that thesetwonotions
are not identical. As noticed by Harris (1991), Romance declension classes
84
COMPOUNDS DON’T COME EASY
–therelevantnotionbeingthatof“wordmarker”–aresyntactically inactive
markers which are idiosyncratically selected by noun stems and do not de-
termine the gender of the noun. In fact, noun classes and gender classes in
Spanish do not show any correspondence between them, as showninTable2
(see Alexiadou 2004 for the same line of analysis).
Declension class Gender
carro ‘cart’ -o Masculine
mano ‘hand’ -o Feminine
problema ‘problem’ -a Masculine
mesa ‘table’ -a Feminine
puente ‘bridge’ -e Masculine
nave ‘ship’ -e Feminine
TABLE 2: Spanish declension classes and their lack of connection with gender
Similar points, or maybe even stronger points (see Acquaviva2008),canbe
made for Italian.
In contrast with Romance, where gender and declension classes do not
show any kind of correlation, German declension classes are clearly connected
to gender information. In other words, German allows the speaker to deduce
the gender of a noun by knowing the declension class to which itbelongs,
something which in impossible in Spanish or Italian. Declension classes and
gender are connected in this language, as shown in Table 3.
Masculine Feminine Neuter
Cl. Gen.Sg. Nom.Pl. Gen.Sg. Nom.Pl. Gen.Sg. Nom.Pl.
I. -s Ø Ø Ø(Umlaut) -s Ø
II. -s Ø(Umlaut) Ø-e -(e)s -e
III. -(e)s -e Ø-e (Umlaut) -(e)s -er (Umlaut)
IV. -(e)s -e (Umlaut) Ø-(e)n -(e)s -en
V. -(e)s -er(Umlaut) Ø Ø -ens -n
VI. -(e)n -en Ø-s -s -s
VII. -(e)s -en
VIII. -(e)s Ø
IX. -ns -n
X. -s -s
TABLE 3: German declension classes and their connectionwith gender
Similar observations can be made with respect to Norwegian Bokmål.
Let us see how this difference can be expressed syntactically. Notic e that in
German both declension classes and gender have independent morphological
exponents; in other Germanic languages, such as Dutch or English, one or both
of these distinctions is not marked with a dedicated morpheme. However, we
propose to tentatively extend the analysis motivated by the German paradigm
85
DENIS DELFITTO AND CHIARA MELLONI
to the whole Germanic family, as there is no counterevidence for it in any of
the other languages.
We follow Acq uaviva (2008) in the claim that m embership of a root to a
particular noun class is due to the fact that roots are licensed in the environment
of a head containing specific declension class information. Let us represent this
head as N, to capture the fact that declension class specification presupposes a
nominal status. The head N, then, both in Germanic and Romance, contains a
feature for Declension Class that ranges on a whole paradigm of morphological
exponents. However, the difference between Germanic and Romance is that in
the former case, Declension Class features are associated – or, alternatively,
have been reanalysed as – Gender features (24a). In Romance, this Gender
feature is not present (24b). This explains the (dis)connection between Gender
and Declension Class.
(24) a. Germanic NP b. Romance NP
NP NP
N
[Declension Class]
√N
[Declension Class]
√
[Gender]
Consider (24b). As Declension Class is a syntactically inactive feature that
does not intervene in agreement operations, it is by itself insufficient to trigger
the insertion of a PROBE within the Cph. If N is not syntactically active, then
it cannot take part in computational processes (cf. Chomsky’s 2004 Activation
Condition). However, External Merge (i.e. PROBE insertion)andInternal
Merge (adjunction of the targeted compound member to the PROBE projec-
tion) are, as we have seen above, necessary ingredients of therepairstrategy
that has to be activated within the Cph in order to break the PoSresultingfrom
Parallel Merge.
From these considerations, two consequences for the nature of NN com-
pounding follow, one universal and another related to language typology. The
universal consequence has to do with the nature of the root (√), as used in Dis-
tributed Morphology (DM) and other (related) frameworks. IfDMisrightin
claiming that a root is a category that lacks formal features,thenitfollowsthat
it will be universally impossible to solve the PoS within a Cphattherootlevel,
as the root would be inactive and consequently unable to countasatriggerfor
PROBE insertion and as a target for movement.
The second consequence is really crucial for the purposes of this contri-
bution, as it refers to the difference between Germanic and Romance that is
86
COMPOUNDS DON’T COME EASY
arguably responsible for the different surface status of root compounding in
these language groups. In Romance, as N contains only a Declension Class
feature, NP is a syntactically inactive category that cannotgiverisetoaCph;
in particular, Parallel Merge of two NPs cannot provide the basis of a success-
ful derivation, since the PoS will not be resolved. Thus, Romance is predicted
not to be able to build NN compounds with NP (or with roots, the option
which is universally unavailable); rather, it has to resort to some larger chunk
of structure that qualifies for Cph computation, in that it contains some suit-
able syntactically active feature. Conversely, Germanic has a Declension Class
feature in N that is associated with Gender, and qualifies thusasanappropriate
trigger for the Cph computation. Therefore, Germanic can break the PoS at the
NP level, as we will see in the next section.
5. NN COMPOUNDS IN GERMANIC AND ROMANCE
5.1 Compounds at the NP Level: Germanic
We have argued that Germanic NPs are syntac tically act ivatedbyaGender
feature associated with the Declension Class feature (DCl) expressed by N.
Therefore, Germanic can start a derivation by applying Parallel Merge and
break the resulting PoS already at the NP-level. In particular, as part of the
Cph computation, a dedicated head that spells out as the Linking Element is
introduced in the derivation, attracting one of the two NPs toitsedge.Werefer
to this operation as Categorially-Driven Compounding (CDC).
(25) Categorially Driven Compounding (CDC):
Merge a functional category F bearing a non interpretable butvalued
feature (Pesetsky & Torrego 2004), corresponding to an interpretable
but unvalued categorial feature on one of the two compound mem-
bers and driving adjunction of the selected compound member to the
structure obtained by applying External Merge of F.
The computational procedure in (25) can be conveniently described by saying
that F contains a valued DCl feature that acts as a PROBE for theunvalued
DCl feature on NP (26).
87
DENIS DELFITTO AND CHIARA MELLONI
(26) Hund-e-futter ‘dogfood’
N2P
N1P?
F
-e
PoS
[uDCl] N1P
Hund
N2P
Futter
[iDCl]
LEs attract an unvalued DCl feature on the NP in (26), triggering movement.
The obvious question that arises is which NP counts as the headwithintheCph,
that is, which NP provides, at the interpretive level, the predicative basis with
respect to which the semantic contribution of the other NP hastobecomputed,
and which NP determines the formal features of the resulting Cph (in strict
compliance with the traditional notion of “headedness”). Here is, in a nut-
shell, our principled answer to this question. The crucial observation is that F
does not activate, in Germanic, any conceptual-structure-related interpretable
feature. Nor is the activated Declension Class feature – though Gender-related
and thus visible for the syntactic computation – one of the grammatical features
that normally percolate to the top when a head projects. In these conditions,
we propose that grammar settles the issue of (semantic and formal) headedness
neither on interpretive nor on featural grounds, but on purely phrase-structural
grounds: the attracted NP (being an adjoined element) straightforwardly qual-
ifies as the non-head, whereas the in-situ element qualifies astheheadofthe
construction. In slightly different terms, we propose that at the point where
the PoS is broken in Germanic feature-endowment is not enoughtodecideon
matters of headedness. Under this solution, it is the adjoined NP that gets in-
terpreted as dependent on the predicative structure associated with the head.
Moreover, the Germanic Cph is (correctly) predicted to have awiderangeof
semantic interpretations: since no semantic feature is activated in the syntac-
tic derivation, we do not expect any grammatical constraint on interpretation
to apply. This line of analysis is consistent with Ramchand and Svenonius’
(2008) proposal that in the absence of syntactically pre-encoded semantic in-
formation the interpretative component produces a larger variety of readings,
with context and encyclopaedia acting as the only constraining factors. Let us
revise now some evidence that the head F is indeed the Linking Element and
that the trigger of the movement is attraction of a Gender-related DCl feature
on one of the two NPs.
88
COMPOUNDS DON’T COME EASY
5.1.1 The Distribution of LEs in German
The following table provides a list of the possible morphophonological real-
izations of the Linking Element in German.
Fugenelemente Examples
-e- Schwein-e-schmalz ‘lard’, Mäus-e-plage ‘plague of mice’
-er- Kind-er-garten,Ei-er-schale ‘egg shell’
-en- Frau- en- frage ‘woman’s issue’, Dorn-en-hecke ‘hedge of thorns’
-n- Schwester-n-liebe ‘love between sisters’, Seide-n-kleid ‘silk dress’
-es- Tag-es-ze i t ‘time of a day’, Kind-es-alter ‘childhood’
-s- Arbeit-s-lohn ‘wages’, Liebe-s-brief ‘love letter’
-Ø- Mutter-tag ‘mother’s day’
TABLE 4: LEs in German
It seems to be the case that normally each noun stem combines with only
one LE, although there are exceptions. According to Augst (1975: 134), only
10% of nominal lexemes have more than one linking form, as illustrated in
(27) for the word Kind,‘child’:
(27) Kind-er-wagen ‘pram’
Kind-es-alter ‘childhood’
Kind-s-kopf ‘person who behaves childishly’
Kind-taufe ‘child christening’
Leaving these exceptions aside, it is clear that the selection of the LE is not
random and some generalizations can be formulated. In fact, Montgomery
(2001) shows that declension classes determine the choice ofthelinkingele-
ment. In particular, a LE is ‘licensed’ for insertion after the adjoined NP if it
is that NP’s:
(28) - nominative sg. (always Ø)
-genitivesg.
-nominativepl.
Moreover, -s regularly appears when it is not licensed, that is, when it is not
found in the case paradigm associated with the declension class of the adjoined
NP. The relevant generalizations are summarized in (29):
(29) Generalizations
i. If the Nom. pl. of any NP is -s,thisnounwillnottake-s- as a
LE in a nominal compound.
ii. If the gen. sg. or the nom. pl are -(e)n then the LE will be
exclusively -(e)n-.Neitherthenominativesg.Ønorthedefault
-s- can apply.
89
DENIS DELFITTO AND CHIARA MELLONI
iii. Insert -(e)ns- only where it is licensed. Three lexical exceptions
are listed.
iv. Use the ‘default’ -s- if NP is morphologically complex (to the ef-
fect that the declension class of the stem is not directly accessible
by F)
v. Avoid m ultiple exp onence, in particular if it involves G en.Sg.
and Nom.Pl. This condition explains why:
-s- is disallowed if we have -s- in the Nom.Pl., and
-e- (schwa) is disallowed (it may be associated with the verbal
status of the first constituent)
Finally, it’s clear enough that number morphology on LEs needs not be inter-
preted, as shown by (5), restated here as (30):
(30) a. Sonn-en-shein ‘sunshine’ (from OHG gen.sg. -in)
b. Kind-er-wagen ‘pram for one child’
c. Schwein-e-braten ‘roast consisting of one pig or part of a pig’
d. Freund-e s-kreis ‘circle of more than one friend’ (gen. sg.)
On these empirical grounds, we are led to the hypothesis, already formu-
lated above, that the LE represents the realization of a valued DCl feature of
the adjoined NP. Crucially, this feature must be Gender-related, in order to
be visible and thus syntactically active. As we have seen, this is reasonable
enough for Germanic. The residual issue concerns morphological exponence:
how is it the case that a DCl feature is regularly made to correspond, mor-
phologically, to pieces of inflectional morphology (typically, as we have seen,
nom.pl. and gen.sg.)? The issue boils down to the following question: if LEs
are valued instances of DCl features, how can morphological specification take
place, given that DCl features do not have independent morphological expo-
nence? We take the answer to be along the following lines: in the absence of
independent morphological exponence, the PF-interface realizes valuation of a
given DCl feature by activating a choice function that explores the inflectional
paradigm associated with that class and selects a prototypical inflectional ex-
ponent of that class. This explains, in particular, why ambiguous exponents
(that is, forms that are characteristic of more than one declension class) are
systematically ruled out. It also explains the status of -sas default: when the
choice function cannot be activated because no declension class can be targeted
by LEs (due for instance to the morphologically complex structure of the ad-
joined NP, as is the case with noun-forming suffixes such as -heit,-keit,-ung,
etc.), valuation takes place by assigning default gender exponence (resulting
in -sat PF). Since valuation reduces to a searching procedure for paradigmatic
declension class exponence at PF, we do not exclude interference with other
PF-factors, integrating the content of the choice function and paving the way
for the additional role played, in certain conditions, by phonotactic factors in
90
COMPOUNDS DON’T COME EASY
the choice of LEs (a correct empirical consequence, it seems). Finally, the fact
that the inflections on LEs are often not interpreted, as exemplified in (30), sim-
ply follows from the fact that they count here as paradigmaticPF-exponentsof
acertaindeclensionclass,withoutmanifestingtheLF-oriented features that
are part, normally, of the featural make-up of inflections such as genitive and
number. In fact, we assume here that even when inflection on LEsappearsto
be interpreted, this interpretation is not grammatically-driven but depends on
the intervention of external interpretation systems (in thecaseofDutchboek-
en-kast ‘book shelves’, for instance, we assume that the PF-orientedchoiceof
the pl. morpheme -en accidentally overlaps with the encyclopaedic knowledge
that book shelves normally contain more than one book).
We conclud e th at there ar e solid emp irical g rounds to the hypothesis that
LEs represent the PF-realization of the valued counterpart of the unvalued DCl
feature realized on one of the two compound members. Valuation of the unval-
ued feature on the NP takes place through movement, and movement resolves
the PoS, enabling the Cph to converge.
5.2 NN compounds at the Phi level: Romance
The components of root (or primary) compounds in Romance are neither roots
nor NPs, because these two categories are syntactically inactive in this lan-
guage family, for the reasons explored above (basically, theformalindepen-
dence of Declension Class features from the Gender feature).NNcompounds
in Romance are constructed by Parallel Merge of two morpho-syntactically in-
flected nouns that contain phi-features triggering the insertion of a head F. As
proposed above, it is this head that resolves the PoS and ensures label projec-
tion, giving rise to a Cph (31).
(31) a.
ϕ
P[
√+N+alanguage-specificbundleof
ϕ
-features, including
Gender and Number]
b.
ϕ
P
ϕ
[Gender, Number]
...NP
N
[Declension Class]
√
Of course, the relevant issue is the nature of the F head that breaks the point
of symmetry at this structural level in Romance. As we have already seen,
the interpretation of Romance root compounds is quite more restricted than
in Germanic. In a nutshell, the head provides a default predicative structure
91
DENIS DELFITTO AND CHIARA MELLONI
of the kind P(x1...x
n), whereas the non-head provides an individual constant
borapredicativeconstantBthatgetscombinedwiththeheadby means of
constant-variable unification. Our particular contention is that the predicative
structures corresponding to the head do not depend on post-syntactic activa-
tion of conceptual structure as a whole, as is arguably the case in Germanic,
but are grammatically constrained: they strongly tend to coincide with the
default predicates made available by the Qualia Structure associated with the
nominal head in Pustejovsky’s GL, and in particular with the Formal Quale
(for the sake of clarity, we’ll assume here the GL classical model; see discus-
sion in Asher 2008). Below, we provide detailed empirical illustration. For
the time being, let’s assume that this view is essentially correct. The formal
model of Cph’s would be the same as in Germanic, with the different range of
available interpretations resulting from the nature of the predicative structures
introduced by the head: unconstrained predicates (and unconstrained types of
head / non-head composition in Germanic) and predicates strictly conform-
ing to the Qualia-format in Romance (with a single type of head/non-head
composition, i.e. constant/variable unification).
At this point, two questions arise: (i) how do we ensure that a specific
Quale (that is, a highly-specific predicative structure) is activated on the head;
and (ii) how do we derive left-headedness (that is, the fact that the relevant
Quale is activated on the
ϕ
Pthatgetsattracted)? Concerningthefirstissue,
we propose that
ϕ
Ps, being fully-inflected nouns, are semantically active, in
the very constrained sense that their Qualia Structure (specifically, the Formal
Quale) is made accessible for the purposes of the syntactic computation in the
form of a single unvalued Q-feature associated with the
ϕ
P. The i d e a is th a t F
represents, within the Cph, the valued counterpart of the selected compound
member, as was the case in Germanic. Specifically, as a matter of inner lexical
constitution, F is a FQ-oriented element and its role is to contribute to the
valuation of Q on
ϕ
Pbyattracting
ϕ
PtotheedgeofCph. Sinceinterpretive,
rather than purely categorial features are involved in this type of derivation,
we refer to the Romance way of constructing NN compounds as Semantically-
Driven Compounding (32).
(32) Semantically Driven Compounding (SDC):
Merge a functional category F with a valued Qualia-oriented feature
(FQ) that targets an unvalued Qualia-oriented feature on oneofthe
two compound members (
φ
Ps), driving adjunction of the selected
compound member to the structure obtained by applying External
Merge of F.
The attracted noun, which qualifies as the head on straightforward interpretive
grounds (i.e. it is the compound member whose interpretable features have
been activated within the Cph), ends up asymmetrically c-commanding the
92
COMPOUNDS DON’T COME EASY
non-head, deriving left-headedness and answering (ii). We further propose that
FinRomancemaybeinstantiatedbyaphoneticallyemptyhead(see below for
further qualifications). Given its FQ-oriented nature, F crucially contributes,
in Romance, to the compositional interpretation of the compound. Italian NN
compounds with an empty F are exemplified in (33).
(33) uomo ø lupo lit. ‘man + wolf’, ‘werewolf’
ϕ
1P
ϕ
1P?
F
/0
PoS
ϕ
1P
uomo
ϕ
2P
lupo
Interpretively, it turns out that F is strictly bound to Formal-Quale (FQ) acti-
vation (it values Q on
ϕ
PasFQ);itthusactivates(salient)second-orderprop-
erties (dimension, shape, color, position, etc.) on the headnoun,triggering
asemanticcompositionprocessinvolvingthepredicativeconstant introduced
by the non-head. This analysis directly accounts for the “restricted” inter-
pretation available for Romance NN compounds (so-called coordinative and
subordinative compounds of the type poeta-pittore ‘painter-poet’ and ufficio
acquisti ‘purchase office’ are excluded from the analysis provided here under
the hypothesis that they do not represent Cph’s, cf. Delfitto,Fábregasand
Melloni 2008). There is a last question to be answered: how do we account for
the fact that the semantic head also qualifies as the formal head in Romance?
Areasonableassumptionisthatactivationofasemanticfeature such as FQ
presupposes activation of formal features which are lower inthehierarchical
functional structure, such as (crucially) phi-features. Hence, not only semantic,
but also grammatical features are activated in SDC, ensuringthattheattracted
element qualifies as the head both at the interpretive and at the formal/featural
level.
Let us conclude by emphasizing that the different role playedbyFwithin
the Cph in Germanic and Romance (categorial vs. semantic feature activa-
tion) nicely matches the level of structural complexity at which the PoS is re-
solved: at the NP-level (root plus Declension Class specification) in Germanic
and at the
ϕ
P-level (fully inflected nouns) in Romance. At which level Cph
is computed depends on the interaction between the EarlinessConditionand
independent PF-properties of lexical items (lexicalization). All comparative
differences between Germanic and Romance directly follow.
93
DENIS DELFITTO AND CHIARA MELLONI
5.2.1 The Interpretation of Romance NN compounds
The proposal that F is restricted to FQ activation, rather than ranging over the
Qualia values of the noun that is displaced, explains why the interpretation
of most NN compounds in Romance is restricted to operations involving the
Formal Quale selected by F on
ϕ
P1 (s-selection as opposed to c-selection). For
illustration, consider the following examples of Italian nominal compounds,
where the non-head provides the predicative constant that specifies the value
of the predicative variable (ranging over colours, forms, etc.) introduced by the
FQ associated with the lexical representation of the head (in34b,forinstance,
spada ‘sword’ specifies the form of (a part of) the fish).
(34) a. pesce zebra →FORMAL Quale [YCOLOR (x)]
fish + /0 + zebra ‘zebra fish’
b. pesce spada →FORMAL Quale [YFORM (x)]
fish + /0 + sword ‘sword fish’
The empty F seems to play the same role as a phonetically realized light prepo-
sition, as shown by the fact that we can obtain interpretatively identical results
by inserting a preposition such as a.
(35) chiave a stella →FORMAL Quale [YFORM (x)]
key + A + star ‘cross wrench’
In (35), a comparable predicative structure is activated on the head noun (chi-
ave); syntactically, the question arises whether F selects objects of the same
structural complexity as light prepositions. In fact, we will argue below that
(34) and (35) are assigned different structures, depending on the different c-
selection properties of empty F with respect to light prepositions. We will
see that even this structural difference has a semantic reflex, in that empty F,
contrary to light prepositions, is bound to FQ-activation.
The hypothesis that we have put forward – that Romance Cph’s are se-
mantically driven – has noticeable consequences for a compositional analysis
of nominal compounds. In particular, notice that the logicalformsassociated
to each node of the syntactic phrase-marker for the compound structure pesce
spada (34b) turn out to be compatible with the satisfaction of severe compo-
sitionality requirements. The basic semantic operations tobeperformedare
constant-variable unification (the predicative constant introduced by the non-
head provides the value for the predicative variable introduced by the head)
and function composition at the top node. Clearly, syntax andsemanticsgo
hand in hand: the relevant predicative variable becomes accessible through F,
which activates FQ on the head pesce;moreover,functioncompositioncanbe
correctly performed at the top node (giving for pesce spada the semantic value
of ‘fish with the form of a sword’) because the head pesce has been displaced
(as a result of feature-attraction, as discussed above) to the left-edge of Cph
94
COMPOUNDS DON’T COME EASY
(under this analysis, the in-situ copy of pesce must be silent not only at PF
but also at LF). Of course, the fact that Romance root compounding turns out
to be (to a large extent) perfectly compositional under the analysis developed
here should be regarded not only as an appealing side-effect of the syntax of
Cph’s, but also possibly as an interesting kind of conceptualevidenceforthe
very existence of Cph’s.
(36) pesce spada ‘sword fish’ (Formal Quale: predicative variable)
ϕ
1P
λ
x.pesce(x) ∧SPADAFORM (x)
ϕ
1P
λ
x: pesce(x)
?
λ
x.SPADAFORM (x)
F(/0)
λ
x.YFORM(x)
PoS
ϕ
1P
pesce
ϕ
2P
spada
λ
y: spada(y)
At this point, two interesting issues arise. The first concerns the definition
of precise boundaries for Formal Quale-interpretation (seeAsher2008). Due
to space limitations, we have to leave this for a future occasion. The second
concerns one of the key comparative differences between Germanic and Ro-
mance: the constraints on recursion that are traditionally detected on Romance
root compounds. In fact, if we put aside the coordinative compounds in (9),
there is a clear-cut contrast between PCs such as (35) (where recursion is possi-
ble, witness 10) and prepositionless compounds such as (34) (where recursion
is rigidly excluded, witness the ungrammaticality of *pesce zebra spada). The
source of this contrast between prepositional and aprepositional forms cannot
be semantic: in fact, a *pesce zebra spada might well be a striped fish having
the form of a sword (maybe a biologically improbable object, but surely a se-
mantically sound one, as a result of double FQ-activation). Given the analysis
that we have developed, there is however a quite reasonable structural source
for a ban on recursion limited to prepositionless compounds.Supposethatthe
empty F involved in prepositionless compounds has specific c-selection prop-
erties (an hypothesis to be empirically substantiated below), to the effect that
it cannot select chunks of structure larger than
φ
Ps. Recursion would entail
that the second instance of F (corresponding, semantically,tothesecondFQ-
activation) necessarily selects a structure of the level of complexity of the Cph
created by merging two
φ
Ps. This is simply incompatible with the c-selection
properties of the empty F.
95
DENIS DELFITTO AND CHIARA MELLONI
With PCs the ban is lifted, since they are arguably allowed to select struc-
tures larger than
ϕ
Ps, as confirmed by their independent syntactic properties
and, specifically, by their lower degree of compliance with the surface effect
of the Lexical Integrity Hypothesis (as we will see in a moment). It seems thus
that the present analysis can offer a fine-grained explanation of the source of
the (partial) ban on recursion in Romance, achieving at the same time a high
level of empirical adequacy.
5.3 Number in PCs and other alleged compounds
The fact that Italian and Spanish NN constructions such as those in (38) may
contain plural features inside the non-head seems to count asacounterexample
to the Earliness Principle on Cph’s as defined in (23). According to (23), the
Cph computation is triggered as soon as the internal structure of the two paral-
lel constituents to be merged contains syntactically activefeatures. Thisleads
to the expectation that the Cph is activated as soon as Gender is introduced.
Formally, there is no reason to wait until a second phi feature(forinstance,
Number) is introduced: in fact, (23) predicts that we should not find Number
features realized on the non-head constituent of root compounds in Romance
(for the head, we may assume that number inflection intervenesafterthecom-
pound computation has been completed, capitalizing on the edge-position of
the head within the Cph). Of course, this argument is based on the hypothe-
sis that Gender and Number features are introduced by distinct heads and that
Number is higher than Gender in the functional hierarchy (Ritter 1991, Cinque
2005). The question is thus why this prediction is not fulfilled. Specifically,
plural number is allowed on the non-head of a Romance compoundintwo
cases: PCs and (for Italian) alleged NN compounds such as those in (37b).
(37) a. It. camicia a quadri ‘shirt + square+pl’, ‘T-shirt with squares’
b. It. ufficio professori ‘room + professor+pl’, ‘professor room’
It. treno merci lit. ‘train merchandise+pl’, ‘freight train’
For Italian, there is in fact a straightforward explanation for these data. As
shown by (37), in Italian Gender and Number are realized syncretically: the -i
ending in professor-ior quadr-i lexicalizes both Gender and (plural) Number.
This entails that there is no independent PF-realization forGender,totheeffect
that a Cph activating only the Gender feature could not be successfully trans-
ferred to the PF-interface (lexicalization would be impossible). For Spanish,
however, the situation is radically different, due to the presence of the sigmatic
plural in Iberic languages. Here, the -sending in mercancías (38b) represents
an uncontroversial violation of (23), since this sort of independent number ex-
ponence clearly implies that the Cph computation has been performed after
an additional functional layer has been added to the minimal structural level
required for feature-activation and lexicalization.
96
COMPOUNDS DON’T COME EASY
(38) a. Sp. camisa a cuadros ‘shirt with squares’
b. Sp. tren de mercancías lit. ‘train of merchandises’, ‘freight train’
Since the Earliness Condition is clearly irrelevant for these Spanish construc-
tions, the default hypothesis is that these structures should not be analysed as
Cph’s: there is no point of symmetry that must be broken to satisfy computa-
tional requirements. We argue instead that the alleged compounds in (37-38)
are the result of combining two nominal structures by means ofarelational
head that is syntactically manifested as a (light) Preposition (37a) or as a prepo-
sitionless genitive in Italian (37b). Before fleshing out ourproposalforPCs,
we would like to emphasize that that there is independent empirical motiva-
tion, beside the conceptual argument that we have just provided, to suggest
distinct structural analyses for the forms in (37-38).
First, these structures do not exhibit canonical compound behavior. A first
empirical difference between these constructions and actual NN compounds
concerns coordination; two non-heads can be easily coordinated with a single
head in PCs or in cases like Italian treno merci,buttheycannotincanonical
NN compounds (cf. 39).
(39) a. una camicia a quadri e a righe
ashirtAsquaresandstripes‘asquared-stripedshirt’
b. un ufficio professori e ricercatori
an office professors and researchers, ‘a professor and researcher
office’
un treno merci e passeggeri
atraingoodsandpassengers‘afreightandpassengertrain’
c. *un uomo lupo e rana
amanwolfandfrog,‘awerewolffrogmandiver’
Further, the non-head of the alleged compounds in (40a-b) canbemodified
independently from the head of the compound. The same operation is banned
in uncontroversial NN compounds (40c).
(40) a. una camicia a quadri rossi
ashirtAsquaresred‘ashirtwithred-squares’
b. un ufficio professori associati
an office professors associate, ‘an associate professor office’
c. *un uomo [lupo feroce]
amanwolfwild,‘abigbadwolfman’
Let us consider pronominalization of the non-head. In this case as well we
observe a non-canonical behaviour of the alleged compounds in a. and b.,
whose non-head can be anaphorically resumed outside the Cph,whilethisis
not an option for the non-head inside a NN compound (41c).
97
DENIS DELFITTO AND CHIARA MELLONI
(41) a. It. Ho comprato una camicia a [quadri]iche [ne]iha di diversi
colori e dimensioni.
‘I bought a shirt with squares which has them in different colours
and dimensions’
b. It. Hanno predisposto un ufficio [professori a contratto]iperché
[questi ultimi]inon avevano uno spazio per ricevere gli studenti.
‘They arranged an office for lecturers because the latter did not
have an office for meeting students.’
Sp. He comprado un tren de [mercancías]iporque transporto
muchasi.
IhaveboughtatrainofmerchandisesbecauseItransportmany.
c. It. *Odio gli uomini [lupo]iperché [pro]imi terrorizza.
Sp. *Odio a los hombres [lobo]iporque tengo miedo de él i.
IhatemenwolfbecauseIamafraidofit.
Significantly, there is a systematic correlation between thecontrastsillustrated
above and the fact that plural inflection inside PCs is allowed, whereas it is
banned inside NN compounds, as shown in (42) for Italian and Spanish.
(42) a. It. camicia a quadri ‘shirt with squares’
Sp. camisa a cuadros ‘T-shirt with squares’
b. It. ufficio professori ‘professor room’
Sp. tren de mercancías ‘train of merchandises’
c. It. *uomo lupi ‘man wolves’
It. *pesce spade ‘fish swords’, ‘swordfish’
Sp. *hombre lobos ‘man wolves’
Sp. *pez espadas lit. ‘fish swords’, ‘swordfish’
Let us proceed now to define the theoretical status of PCs in Romance. We
want to make a principled difference between the a. cases in (38-42), whose
behavior is uniform across Romance, and the b. cases in (38-42), which sur-
face as aprepositional expressions in Italian, while featuring a preposition in
Spanish and (to the best of our knowledge) in all other Romancelanguages.
For reasons of space, we will not discuss the b. cases here (fortheseweadopt
the syntactic analysis put forward in Delfitto and Paradisi 2009, to which the
interested reader is referred).
Concerning the a. cases, we take the presence of elements thatcategorially
qualify as prepositions as providing evidence in favor of a canonical mode of
syntactic computation, whereby a head (the ‘light’ preposition) combines with
amaximalprojection(oneofthetwocompoundsmember),infull compliance
with narrow syntax requirements (LCA and related conditions). Prepositions,
in fact, are dedicated syntactic heads that combine two nominal expressions,
while imposing categorial selection requirements on the selected chunk of
structure. In the case under scrutiny, light prepositions seem to require that the
98
COMPOUNDS DON’T COME EASY
constituent to be merged with them contain not only Gender butalsoNumber
features. When the syntactic process of combining two nominal constituents
is mediated by a preposition, Parallel Merge (as well as Cph-activation) is ex-
cluded. In particular, since no PoS has been created, there isnoneedtoresort
to a Cph as a repair strategy. On the contrary, it is standard antisymmetric
syntax that must be hold responsible for the derivation of these complex ex-
pressions, even though there are serious reasons to think that this class of light
prepositions is strictly tied to the semantics of Cph’s (since light prepositions
involve Qualia-related features) and should be carefully taken apart from regu-
lar DP-selecting prepositions. It is well-known that genitival prepositions (and
case-related prepositions quite generally) give rise to a rich set of interpretive
dependencies between the two arguments to which they apply (this severely
underspecified interpretive dependency is often dubbed ‘R-relation’ in the lit-
erature). For the sake of explicitness, we assume here that light prepositions
are dedicated elements that are constrained to a well-definedsubsetofthein-
terpretive possibilities made available by R-relations. Infact,theyselectthe
class of interpretations triggered by the activation of the Qualia Structure on
the head-noun.
For the purpose of this contribution, we conclude thus that the empirical
contrasts illustrated above are straightforwardly accounted for under the hy-
pothesis that PCs do not involve Cph’s. We propose instead that the two nouns
directly combine as the two arguments of a light preposition P, w h ich i m poses
categorial selection on the nominal constituent that it selects, forcing it to con-
tain at least the functional layer corresponding to the realization of Number
features and quite probably additional functional structure (cf. 43).
(43) PP
NumP
camicia
P
P
a
NumP
quadri
The phrasal nature of the structures in (43) has an interesting interpretive cor-
relate, in that they manifest more semantic complexity than Cph’s. We have
observed above that there are cases where light prepositionsseemtoplaythe
same role as F within compounds (cf. chiave a stella vs.pescespada)but
notice that the reverse situation does not hold, since F can only activate the
FQ-feature (cf. discussion above).
It has been independently noticed (cf. Johnston and Busa, 1999, Bassac
and Bouillon 2001) that in languages such as Italian and French, the selection
of one of the Qualia seems to be signalled by the choice of the appropriate
99
DENIS DELFITTO AND CHIARA MELLONI
light preposition. For example, the preposition ain Italian preferably selects
for the Constitutive Quale, hence it expresses a whole-part or possession re-
lation between the head’s (the possessor) and non-head’s referent (the posses-
sum)(44a). Onthecontrary,da typically selects for the Telic Quale, hence
indicating the purpose/function of the head (44b).
(44) a. bandiera a scacchi →CONST Quale [PART OF (y,)]
flag + A + chess(pl) ‘chequered flag’
b. bicchiere da vino →formal TELIC Quale [CONTAIN (x,y)]
wine + da + glass ‘wine glass’
Both da and acan activate the Telic role of the head noun in PCs. In some
cases, this gives rise to a series of subtle interpretive effects. As Pustejovsky
(1998) has shown, the event encoded in the telic role receivesamodalinter-
pretation and can be represented as an embedded structure made up of formal
and agentive, expressing respectively the Function Event and the Precondition
Event in the Telic of a noun. For instance, consider (45a-b): aknifecuts(re-
sult) something (the bread) if there is someone (the butcher)performingthe
action corresponding to the cutting event. Telic Quale activation is associated
with the presence of the preposition da in Italian (45a-b), ambiguously ex-
pressing the patient in the Formal Telic Quale and the agent intheAgentive
Telic Quale of the head noun. How ever, a dedicated light preposition ain Ital-
ian is systematically chosen to signal that the agent in the Agentive Telic Quale
is not human, but is instead an instrument or a causer (45bi), showing that the
choice of the appropriate preposition might depend on theta-related features of
the non-head argument (cf. Reinhart 2000, 2002). Notice, finally, that a modal
interpretation is regularly excluded for the Agentive Quale, which encodes an
existentially bound event signalling the “origin” of the entity expressed by the
head noun. The agent performing the default event in the Agentive is typically
expressed with di (45c).
(45) a. coltello da pane →Formal TELIC Quale [CUT_res (x,y)]
knife + DA + bread ‘bread knife’
b. coltello da macellaio →Agentive TELIC Quale
[USE (z,x) CUT ∧act (x,y)]
knife + DA + butcher ‘butcher’s knife’
bi.ferro a vapore →Agentive TELIC Quale
[LET (z,x) ∧WORK_act (x,y)]
iron + A + steam ‘steam iron’
c. foro di pallottola →AGENTIVE Quale [MAKE (y, x)]
hole + DI + bullet ‘bullet hole’
The matter is in fact quite intriguing and a thorough exploration of the lexi-
calization patterns of light Ps and Qualia-related featuresinItalianPCswould
100
COMPOUNDS DON’T COME EASY
clearly exceed the limits of the present contribution. We leave it thus for a
future occasion. The point that we want to make here is that, contrary to the
interpretive import of F in Cph’s, which is limited to FQ activation, light prepo-
sitions may activate larger portions of the Qualia Structureonthehead-noun.
The reason - we submit - is that they are endowed with a richer array of valued
Qualia-related features. On one side, the very existence of dedicated lexical el-
ements (light prepositions) that specifically activate Qualia-related predicative
structures in syntax provides extra support for the existence of a Qualia-related
empty head within Romance Cph’s. On the other side, the asymmetry between
the Formal Quale and the other Qualia that has emerged from a fine-grained
analysis of the syntax/semantics mapping seems to indicate that FQ enjoys
special prominence at the syntax/semantics interface. In fact, it has been in-
dependently noted that FQ plays a key role in determining the semantic type
of the relevant noun, and its modelling crucially involves establishing hyper-
onimic relations within the lattice structure of semantic types in which every
type is embedded (cf. Pustejovsky 1995, Asher 2008, Mery et al. 2007). Be
it as it may, the general point we wish to make is that matters that have been
traditionally ascribed to the domain of lexical semantics are better understood
as interface phenomena, requiring more attention for (hidden) syntactic pre-
encoding of interpretive relations.
6. CONCLUSIONS
The analysis presented above warrants a significant number oftheoreticalcon-
clusions.
The first conclusion is that there is no empirical reason to propose that
compounds are generated by a non-syntactic component of grammar, such as
morphology or the lexicon. The properties of compounds are infactbetterun-
derstood when they are linked to the properties of phases and the narrow syntax
conditions on Merge and Projection, as they are independently attested in other
syntactic domains. Secondly, we have shown that root compounding is a syn-
tactic process driven by narrow syntax requirements on parallel (symmetric)
structures; it is the application of these requirements thatfeedstheinterpretive
interface resulting in strong semantic constraints and strong compositionality
in Romance compounding. Thirdly, if our analysis is essentially correct, we
conclude that compounding involves more syntactic and semantic structure
than it appears at the surface. Furthermore, we have argued that a principled
account of linguistic diversity in compounding requires theadoptionofthe
tools for lexical decomposition of words recently developedinlexicalseman-
tics research. In particular, Pustejovsky’s Qualia Structure can be accessed by
the syntactic computation by means of unvalued interpretable features realized
on nouns (qualifying as formal syntactic features), and plays an important role
at the syntax/semantics interface. Finally, we have provided evidence that the
101
DENIS DELFITTO AND CHIARA MELLONI
source of variation in compounding cannot be identified in macro-parameters,
but essentially depends on the language-specific feature endowment of lexical
items, which determines which heads are syntactically active and may undergo
movement, complemented by a general Economy Requirement stating that the
point of symmetry arising in Parallel Merge must be resolved as soon as pos-
sible and cannot be postponed to a following stage in the derivation.
REFERENCES
Ackema, P. & Neeleman, A. (2004). Beyond Morphology.Oxford:OxfordUniversity
Press.
Acquaviva, P. (2008). Roots and Lexicality in Distributed Morphology. Paper given
at the Fifth York-Essex Morphology Meeting.
Availab le at http://ling.auf.net/lingBuzz/000654.
Alexiadou, A. (2004). Inflection Class, Gender and DP Internal Structure. In G.
Müller et al. (Eds.), Exploration in Nominal Inflection (pp. 21-50). Berlin: Mouton
de Gruyter.
Allen, M. (1978). Morphological Investigations.Unpublisheddoctoraldissertation.
University of Connecticut, Storrs.
Asher, N. (2008). A Web of Words: Lexical Meaning in Context. Unpublished
manuscript. Université Paul Sabatier, Toulouse.
Augst, G. (1975). Über das Fugenmorphem bei Zusammensetzungen. In G. Augst,
Untersuchungen zum Morpheminventar der deutschen Gegenwartssprache (pp. 71-
155). Tübingen: Narr.
Bassac, C. & Bouillon, P. (2001). The Telic Relationship in Compounds in French and
Turkish. In P. Bouillon & S. Kanzaki (Eds.), Proceedings of the 1st International
Worksh o p o n G e n e r a t i v e A p p r o a c h e s t o t h e L e x i c o n .Geneva.
Bauer, L. (1978). The Grammar of Nominal Compounding.Odense:OdenseUniver-
sity Press.
Bisetto, A. & Scalise, S. (2005). The classification of compounds. Lingua e Linguag-
gio, 4, 319–332.
Borer, H. (2005). In name only,Vol. 1oftheExoskeletal Trilogy.Oxford:Oxford
University Press.
Brody, M. (2000). Mirror Theory: Syntactic Representation in Perfect Syntax. Lin-
guistic Inquiry, 31, 29-56.
Chomsky, N. (2000). Minimalist Inquiries: The framework. InR.Martin,D.Michaels,
&J.Uriagereka(Eds.),Step by Step (pp. 89-156). Cambridge (MA): MIT Press.
Chomsky, N. (2001). Derivation by Phase. In M. Kensotwicz & K.Hale(Eds.), A
Life in Language (pp. 1–52). Cambridge (MA): MIT Press.
Chomsky, N. (2004). Beyond Explanatory Adequacy. In A. Belletti (Ed.), Structures
and Beyond. The Cartography of Syntactic Structures,vol. 3(pp. 104-131). Ox-
ford: Oxford University Press.
Cinque, G. (2005). Deriving Greenberg’s Universal 20 and ItsExceptions.Linguistic
Inquiry, 36, 315-332.
Citko, B. (2005). On the Nature of Merge: External Merge, Internal Merge, and
Parallel Merge. Linguistic Inquiry, 36, 475-496.
Corbett, G.G. (1991). Gender.Cambridge:CambridgeUniversityPress.
102
COMPOUNDS DON’T COME EASY
Delfitto, D. & Paradisi, P. (2009). Towards a Diachronic Theory of Genitive Assign-
ment in Romance. In P. Crisma & G. Longobardi (Eds.), Historical Syntax and
Linguistic Theory (pp. 292-310). Oxford: Oxford University Press.
Delfitto, D., Fábregas, A. & Melloni, C. (2008). Compounding at the Interfaces. Paper
presented at 39th NELS, Cornell University, Ithaca, November 2008.
Di Sciullo, A.M. & Williams, E. (1987). On the Definition of Word.Cambridge(MA):
MIT Press.
Dixon, R.M.W. (1968). Noun Classes. Lingua, 21, 104-125.
Donalies, E. (2005). Die Wortbildung im Deutschen [2nd edition]. Tübingen: Narr.
Downing, P. (1977). On the Creation and Use of English Compound Nouns. Lan-
guage, 53, 810-842.
Fábregas, A. (2005). The Definition of the Grammatical Category in a Syntactically
Oriented Morphology.Unpublisheddoctoraldissertation.InstitutoUniversitario
Ortega y Gasset.
Haider, H. (2001). Why Are There No Complex Head-initial Compounds? In Ch.
Schaner-Wolles, J. Rennison & F. Neubart (Eds.), Naturally! (pp. 165-174). Torino:
Rosenberg & Sellier.
Halle, M. (1973). Prolegomena to a Theory of Word Formation. Linguistic Inquiry,4
(1), 3-16.
Harris, J.W. (1991). The Exponence of Gender in Spanish. Linguistic Inquiry,22(1),
27-62.
Heringer, H.J. (1984). Wortbildung: Sinn aus dem Chaos. Deutsche Sprache,12,
1-13.
Jackendoff, R.S. (2009). Compounding in the Parallel Architecture and Conceptual
Semantics. In R. Lieber & P. Stekauer (Eds.), The Oxford Handbook of Compound-
ing.Oxford:OxfordUniversityPress.
Johnston, M. & Busa, F. (1999). Qualia Structure and the Compositional Interpreta-
tion of Compounds. In E. Viegas (Ed.), Breadth and Depth of Semantics Lexicons
(pp. 167-87). Dordrecht: Kluwer.
Kayne, R. (1994). The Antisimmetry of Syntax.Cambridge(MA):MITPress.
Lapointe, S. (1980). ATheoryofGrammaticalAgreement.Unpublisheddoctoral
dissertation. University of Boston, Amherst.
Mery, B., Bassac, C. & Retoré, C. (2007). A Montague-based Model of Generative
Lexical Semantics. In R. Muskens (Ed.), New Directions in Type Theoretic Gram-
mars (pp. 90-97). ESSLLI, Foundation of Logic, Language and Information. Avail-
able at http://let.uvt.nl/general/people/rmuskens/ndttg/ndttg2007.pdf
Montgomery, B.I. (2001). Studies in German and English with Special Reference to
‘Linking Elements’.Unpublisheddoctoraldissertation.UniversityofEdinburgh.
Moro, A. (2000). Dynamic Antisymmetry.Cambridge(MA):MITPress.
Moro, A. (2008). ANoteonLabellingandtheEPP.Unpublishedmanuscript.Avail-
able at http://ling.auf.net/lingBuzz/000635
Neijt, A.H. (2003). Linking schwa in Dutch compounds: a phonomorpheme? In J.
Koster & H. van Riemsdijk (Eds.), Germania et alia. A linguistic webschrift for
Hans den Besten.Availableathttp://o dur.let.rug.nl/∼koster/DenBesten/
Pesetsky, D. (1995). Zero Syntax.Cambridge(MA):MITPress.
Pesetsky, D. & E. Torrego (2007). The Syntax of Valuation and the Interpretability of
Features. In Phrasal and Clausal Architecture: Syntactic Derivation andInterpre-
tation.Amsterdam:JohnBenjamins.
103
DENIS DELFITTO AND CHIARA MELLONI
Pustejovsky, J. (1995). The Generative Lexicon.Cambridge(MA):MITPress.
Ralli, A. (2008). Compound Markers and Parametric Variation. Sprachtypologie und
Universalienforschung, 61, 19-38.
Ramchand, G. & Svenonius, P. (2008). Mapping a Parochial Lexicon onto a Universal
Semantics. In T. Biberauer (Ed.), The Limits of Syntactic Variation (pp. 219–245).
Amsterdam: Benjamins.
Reinhart, T. (2000). The Theta System: Syntactic Realization of Verbal Concepts.
OTS Working Papers in Lin guistics.Utrecht:UniversityofUtrecht.
Reinhart, T. (2002). The Theta System: An overview. Theoretical Linguistics,28,
229-290.
Ritter, E. (1991). Two Functional Categories in Noun Phrases: Evidence from Modern
Hebrew. In S. Rothstein (Ed.), Syntax and Semantics 25. Perspectives on Phrase
Structure: Heads and Licensing.SanDiego:AcademicPress.
Roeper, T. (1987). Implicit Arguments and the Head-complement Relation. Linguistic
Inquiry, 18 (2), 267-310.
Ryder, M.E. (1994). Ordered Chaos: The Interpretation of English Noun-Noun Com-
pounds.Berkeley/LosAngeles:UniversityofCaliforniaPress.
Williams, E. (1981). On the notions “Lexically related” and “Head of a word”. Lin-
guistic Inquiry, 12 (2), 245-274.
Denis Delfitto and Chiara Melloni
Università di Verona
viale dell’Università 4,
37129 Verona Italy
e-mail: denis.delfitto@univr.it, chiara.melloni@univr.it
104
