言語研究（
Gengo Kenkyu
）
135
: 49–84（2009）
49
Exocentricity in Compounding
S S A F F F
Univ. of Bologna Univ. of Tromsø/Bologna Univ. of Verona/Bologna
Abstract: e identifi cation of a compound as endocentric or exocentric depends
on the notion of head: if a compound has a head (or two), it is called endocentric;
if it has no head, it is called exocentric. Exocentricity, however, has been usually
assumed as a unitary notion, exactly because the notion of head has been gener-
ally interpreted as a unitary notion. In this paper we will fi rst provide typologi-
cally based data on the dimension and limits of exocentricity, and then we will
argue that the notion of head can be split into three diff erent subparts: categorial
head, semantic head and morphological head. Correspondingly, the notion of
exocentricity can be split into categorial exocentricity, semantic exocentricity and
morphological exocentricity. Our approach, based on features of the constituents
and not on constituents as a whole, will hopefully provide a new analysis of exo-
centricity in compounding.*
Key words: compounding, head, exocentricity, parameters, VN compounds
1. Identifi cation of the Head
ere are at least three points of view according to which the head can be identi-
fi ed. One point of view relies on semantics. Bloomfi eld (1933: 235), for example,
states that an endocentric compound denotes a hyponym of its head: a door knob
is a kind of knob and therefore the compound is endocentric. e same reasoning
is adopted by many other scholars, among which Zwicky.¹ Other scholars instead
claim that the head must be identifi ed only on formal grounds. Kageyama (2008,
2009), for example, states that “ e head should be defi ned as a category deter-
minant”. Finally, still other scholars, such as Namiki (2001), claim that the head
should be identifi ed on the basis of several parameters: semantic, morphological
and furthermore it should be taken into consideration also the claim that the
head is the morphosyntactic locus. is last position deserves some observations.
In the fi rst place, it has been shown (Guevara and Scalise 2009) that the head
is not always the locus infl ectionis: there are endocentric compounds that do not
* is research has been carried out also thanks to funds of the Italian Ministery of Univer-
sity and Research (PRIN 2005). Antonio Fabregas’ research has been fi nanced by postd-
doctoral grant EX2006-0968. We would like to thank two anonymous reviewers for com-
menting a previous version of this paper. Usual disclaimers apply.
¹ According to Zwicky (1985: 4), “We could take the head/modifi er distinction to be at root
semantic: in a combination X+Y, Y is the ‘semantic head’ if […] X+Y describes a kind of the
thing described by Y”.
50 S S, A F, F F
put any infl ection on the head (e.g. It. rosso mattone ‘red + brick = brick red’ is not
infl ected in the plural (due maglioni rosso mattone / *due maglioni rossi mattone ‘two
brick red sweaters’) and there are exocentric compounds that do have infl ection
(e.g. pickpockets). In the second place, while Namiki claims that the head should be
both formal and semantic, he does not attempt to establish any link between these
two notions. In the following pages, we will adopt Namiki’s position (without con-
sidering the head as locus infl ectionis) but we will try to set a relationship between
semantic head and the so-called formal head.
2. Dimensions of Exocentricity
Exocentricity is not a marginal fact (neither from a quantitative point of view,
nor as far as productivity is concerned). One could be tempted to think that exo-
centricity is a relic of the past and that productive contemporary word formation
is always endocentric. While this seems to be true for derivation it is not true for
compounding, as it will be shown below. Before facing the theoretical side of the
matter, it is interesting to have a general idea of the distribution of exocentric phe-
nomena in compounding across languages. In fact, the empirical observation leads
to some considerations, the fi rst one being that all languages exhibit some degree
of exocentricity but this degree can vary across languages.
ere are in fact languages (such as Turkana: Dimmendaal 1983) which are
reported to have almost only exocentric compounds. Furthermore, there are lan-
guages where the most productive pattern is endocentric but where exocentric
compounds are fully productive, such as V+N compounds in Italian (e.g. porta-
lettere ‘carry letters = mailman’).
As for the distribution of exocentricity in compounding in the languages of the
world, consider the following Table.
²
(1)³
Headedness
Mean %
³
Right headed 65.02
No heads 18.9
Left Headed 5.2
Two-heads 4.5
² e data (Guevara and Scalise 2008) are taken from Morbocomp, a database, developed at
the University of Bologna, that includes some 3000 compounds from 24 languages (Basque,
Bulgarian, Catalan, Chinese, Czeck, Dutch, English, Finnish, French, German, Greek,
Hungarian, Italian, Japanese, Korean, Latin, Norwegian, Polish, Portuguese, Russian, Serbo-
Croatian, Spanish, Swedish, Turkish). Each compound is analyzed in 18 searchable fi elds
(compound, script, internal structure, compound category, fi rst constituent, fi rst constituent
category, second constituent, second constituent category, third constituent, third constitu-
ent category, categorial head, semantic head, linking element 1, linking element 2, gender,
number, gloss and translation, remarks).
³ In this and in the following tables, the total sum of the percentages reported is not 100
Exocentricity in Compounding 51
We fi nd thus a clear hierarchy of the distribution of heads in compounding:
(2) Right > No Head > Left > Both
Overall, there are clearly more endocentric types than exocentric types and a
strong preference can be observed for right-headed types (barely two thirds of all
the types).
(3)
Headedness Mean % RO % GE % EA % SL %
Right headed 61.1 36.6 86.5 60.5 62.8
No heads 19.1 35.4 8.4 18.9 13.9
Left Headed 7.1 14.7 1.8 6.8 5.3
Two-headed 5.6 5.7 1.5 11.4
3.2
As it can be seen, exocentricity is attested in every considered language group of
the database.
As already said, the total number of compounds in the database is 3222. e
exocentric compounds are 617, representing a percentage of 19%, namely the fi fth
part of all compounds. is information alone supports a very simple consideration
we have already introduced, i.e.: exocentricity is not an idiosyncratic, possibly
ignorable, phenomenon in compounding.
e distribution across the compound classes identifi ed in Bisetto and Scalise
(2005) is the following:
(4)
ATTRIBUTIVE 5.1%
COORDINATE 4.6%
SUBORDINATE
8.5%
e attributive and the coordinate classes do not show a dramatically diff erent
behaviour as far as headedlessness is concerned. e subordinate class, on the con-
trary, has a higher incidence of exocentricity. is could refl ect a specifi c pattern, or
even a generalization, supporting the idea that one possible trigger of exocentricity
would be a complement-head relation between the constituents, probably—as it
will be shown ahead—depending on some typological properties of the licensing
of arguments.
Yet, the most pervasive compound type across the corpus is not a subordinate
but an attributive one: [A+N]
N
, such as the English pale face or [N+A]
N
, such as It.
viso pallido ‘face pale = pale face’ in left headed languages.⁴ e pale face type has
because spurious cases, accidental mistakes and problematic entries have deliberately been
left-out. Hopefully, the overall coherence of the results is not aff ected.
⁴ It is also interesting to notice a series of this type of compound with a slightly diff erent
fl avour, e.g. 奇麗所 kirei dokoro ‘beautiful + place = geisha’ in Japanese.
52 S S, A F, F F
been considered by some authors (e.g. Booji 2007) as endocentric on the grounds
that the semantic shift can be explained in terms of metonymy. However, it should
be noticed that semantics is not the only problem with this type of compounds
since, for example in Romance, they seem also to be exocentric in such formal
properties such as gender and number. For example the compound testa rasata
‘head + shaven = skin head’ can be masculine or feminine, singular or plural while
the constituent testa is obligatorily feminine singular.
is type of compound is found in all the 16 fusional languages of the corpus
(Bulgarian, Catalan, Dutch, German, Czech, English, French, Greek, Italian,
Latin, Norwegian, Polish, Portuguese, Russian, Serbo-Croatian, Spanish and
Swedish). In the agglutinative languages of the database this type is also well
represented, since it is present in Turkish, Finnish, Hungarian and Japanese. e
isolating language of the corpus, Chinese, also contains this type.
3. Limits of Exocentricity
It is not at all clear what are the limits of exocentricity, but we may safely assume
that there could be at least three dimensions along which the limits of exocentric-
ity could be investigated: typological, categorial and semantic.
3.1. Typological distribution
Let us now see the distribution of exocentricity among the main groups of our
database:
(5)⁵
Group %
Romance 35.4
Germanic 8.4
Slavic 13.9
East Asia
⁵
18.9
As it is possible to notice from the preceding table, Romance languages tend to
show a high rate of exocentricity. e most common exocentric types are [V+N]
N
,
[P+N]
N
, [V+V]
N
, and [A+N]
N
. Instead, Germanic languages are not signifi cantly
characterised by exocentric compounding. e most wide-spread headless struc-
ture is, as already mentioned, [A+N]
N
, such as pale face.
East Asian languages, as well as Slavic languages, contain an average amount
of exocentric types. ey share some types with other language groups, such as
[A+N]
N
or [V+V]
N
, but some types seem more language-specifi c, such as the
Chinese antonymic compound dàxiaor 大小 ‘large + small = size’ (although we will
challenge this view in the last section).
From a typological point of view, we may assume that diff erent types of lan-
⁵ Japanese, Korean and Chinese are grouped under the geographical label ‘East Asia lan-
guages’, which cross-cuts the typological and genealogical distinctions between them.
Exocentricity in Compounding 53
guages may exhibit diff erent kinds of exocentricity, a conclusion that is already
implicit in some data seen in (3). Romance languages, in fact, exhibit more exocen-
tricity than Germanic languages (35.8% vs. 8.9%). If we compare two languages,
for example Italian and English, we fi nd the following situation:
(6) Italian English
V+N]
N
—⁶
P+N]
N
—
N+A]
N
A+N]
N
V+V]
N
—
N+N]
N
N+N]
N
As expected, there are more exocentric patterns in Italian than in English. In both
languages the default category for exocentric compounds is N.
ese typological diff erences, as a matter of fact, must fi nd explanation in
some deeper reason, a reason concerning the very essence of compounding. As
we have seen in the preceding sections, exocentricity is spread among the world’s
languages.
Exocentric structures can be very diff erent from each other, all the lexical cat-
egories are involved, and there seems to be diff erent ways in which a compound
can be classifi ed as exocentric.
3.2. Limits of categorial combinations
We will now consider Chinese exocentric compound structures:
(7)
Chinese Script Example
V+N]
N
天葬
tiánfáng ‘(to) fi ll+room=second wife (of a widower)’
V+N]
A
缺德
quēdé ‘lack+morals=immoral’
A+N]
N
软卧
ruanwò ‘soft+(to) lie=(of a train) soft sleeper’
V+V]
N
裁缝
cáifeng ‘cut+sew=tailor’
N+N]
N
江湖
jiānghú ‘river+lake=vagabond’
A+A]
N
大小
dàxiaor ‘big+small=size’
Chinese seems to exhibit more exocentric compound types than both Italian and
English. Furthermore, in Chinese, the same compound structure (V+N) can give
rise to compounds with diff erent categories⁷—a very rare fact probably tied to the
existence of hybrid categories.
Both Chinese and Italian, but not English, have what we would like to call
‘Absolute Categorial Exocentricity (ACE)’, when the output is completely diff er-
⁶ In English there are some VN compounds such as pickpocket, killjoy, but this class is
not productive.
⁷ is fact has been studied and for [VN]
N
and [VN]
V
, cf. Ceccagno and Scalise (2005).