Coordination

Abstract

Introduction The term coordination refers to syntactic constructions in which two or more units of the same type are combined into a larger unit and still have the same semantic relations with other surrounding elements. The units may be words (e.g. verbs (1a)), phrases (e.g. noun phrases (1b)), subordinate clauses (e.g. (1c)) or full sentences (e.g. (1d)). All languages appear to possess coordination constructions (or coordinate constructions) of some kind, but there is a lot of cross-linguistic variation. Individual languages may possess a wealth of different coordinate constructions that relate to each other in complex ways. It is the purpose of this chapter to introduce and discuss a wide range of conceptual distinctions that are useful for describing the cross-linguistic and language-internal variation. This entails the use of a large number of technical terms (printed in boldface on first occurrence), each of which is explained and illustrated as it is introduced. Terminological issues are discussed further in an appendix. © Cambridge University Press 2007 and Cambridge University Press, 2010.

Full text

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
1 Coordination
Martin Haspelmath
0 Introduction
The term coordination refers to syntactic constructions in which two or more
units of the same type are combined into a larger unit and still have the same
semantic relations with other surrounding elements. The units may be words
(e.g. verbs (1a)), phrases (e.g. noun phrases (1b)), subordinate clauses (e.g.
(1c)) or full sentences (e.g. (1d)).
(1) a. My husband supports and adores Juventus Turin
b. My uncle or your in-laws or the neighbours will come to visit us
c. I realize that you were right and that I was mistaken
d. The Pope dissolved the Jesuit order, and all the Indian missions
were abandoned
All languages appear to possess coordination constructions (or coordinate con-
structions) of some kind, but there is a lot of cross-linguistic variation. Individual
languages may possess a wealth of different coordinate constructions that relate
to each other in complex ways. It is the purpose of this chapter to introduce and
discuss a wide range of conceptual distinctions that are useful for describing
the cross-linguistic and language-internal variation. This entails the use of a
large number of technical terms (printed in boldface on first occurrence), each
of which is explained and illustrated as it is introduced. Terminological issues
are discussed further in an appendix.
The particle or affix that serves to link the units of a coordinate construction is
called the coordinator. In (1) and in the other numbered examples in this chap-
ter, the coordinator is printed in boldface. By far the most frequently occurring
coordinator is ‘and’ (i.e. English and and its equivalents in other languages),
but coordinate constructions can also involve various other semantic types of
linkers, such as ‘or’, ‘but’ and ‘for’. ‘And’-coordination is also called conjunc-
tive coordination (or conjunction), ‘or’-coordination is also called disjunctive
I am indebted to Tim Shopen, Orin Gensler, and especially Edith Moravcsik for detailed helpful
comments on an earlier version of this chapter.
1

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
2Martin Haspelmath
coordination (or disjunction), ‘but’-coordination is called adversative coor-
dination, and ‘for’-coordination is called causal coordination. Examples of
each of these four types are given in (2).
(2) a. (conjunction) Snow White ate and drank
b. (disjunction) She was a countess or a princess
c. (adversative coordination) The dwarfs were ugly but kind
d. (causal coordination) She died, for the apple was poisoned
The units combined in a conjunctive coordination are called conjuncts, and,
more generally, the units of any coordination will be called coordinands here.
Adversative coordination is always binary, i.e. it must consist of two coor-
dinands. Ternary or other multiple coordinations are impossible here. This is
illustrated in (3).
(3) a. *The queen tried to kill Snow White but Snow White escaped
but she went through much hardship
b. *The mountain climbers were tired but happy but bankrupt
By contrast, conjunctions and disjunctions can consist of an indefinite number
of coordinands. The examples in (4) show six coordinands each.
(4) a. You can vote for Baranov or Wagner or Lef`evre or McGarrigle
or Ram´ırez or Abdurrasul
b. Cameroon, Nigeria, Niger, Libya, Sudan and the Central
African Republic have a common border with Chad
Languages differ with respect to the number and the position of the coordi-
nators used in coordinate constructions. For instance, while English generally
shows the pattern A co-B (where co stands for coordinator), Kannada (a Dra-
vidian language of southern India) shows the pattern A-co B-co:
(5) Narahariy-u: So:maˇse:kharan-u: pe:t
.e-ge ho:-d-aru
Narahari-and Somashekhara-and market-dat go-past-3pl
‘Narahari and Somashekhara went to the market’
(Sridhar 1990:106)
The patterns of coordinator placement and the types of linkers are discussed
further in Section 1.
Many languages have several alternative patterns for a given semantic type of
coordination, as illustrated in the English examples (6a,b). Coordination with
the two-part coordinator both . . . and describes the coordinands as contrasting
in some way: (6a) is appropriate, for instance, if the hearer expects only one of
them to make the trip. This construction will be called emphatic coordination
in this chapter.

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 3
(6) a. Both Franz and Sisi will travel to Trieste
b. Franz and Sisi will travel to Trieste
Moreover, many languages have special coordinators for negative contexts, as
in the English example (7a). This sentence is roughly equivalent semantically
with (7b), but again it has a more emphatic flavour. The construction in (7a)
will be called emphatic negative coordination.
(7) a. Neither Brahms nor Bruckner reached Beethoven’s fame
b. Brahms and Bruckner did not reach Beethoven’s fame
Emphatic and negative coordinate constructions are discussed further in
Section 2.
We saw in (1) above that a coordinate construction can consist of different
types of coordinands: words, phrases, clauses or sentences. But as the defini-
tion of coordination says, each coordinand must be of the same type within a
coordinate construction. Thus, (8b) and (9b) are ungrammatical, because the
coordinands are syntactically different (np vs pp in 8b) or at least semantically
different (manner vs comitative in 9b).
(8) a. Guglielmo wrote to his bishop and to the Pope
b. *Guglielmo wrote a letter of protest and to the Pope
(9) a. Guglielmo spoke with the abbot and with the cardinal
b. *Guglielmo spoke with eloquence and with the cardinal
Different languages may require different coordinators depending on the syn-
tactic type of the coordinands. For example, Yapese (an Austronesian language
of Micronesia) has ngea ‘and’ for np conjunction (10a), but ma ‘and’ for sen-
tential conjunction (10b) (Jensen (1977:311–12)):
(10) a. Tamag ngea Tinag ea nga raanow
Tamag and Tinag conn incep go.du
‘Tamag and Tinag will go’
b. Gu raa yaen nga Donguch, ma Tamag ea raa
Ifut go to Donguch and Tamag conn fut
yaen nga Nimgil
go to Nimgil
‘I will go to Donguch, and Tamag will go to Nimgil’
Types of coordinands and their relevance for the structure of coordination are
discussed further in Section 3.
In addition to the major semantic distinctions that we saw in (2), numer-
ous more fine-grained distinctions can be made. For example, many languages
distinguish between two types of disjunction: interrogative disjunction and

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
4Martin Haspelmath
standard disjunction. Mandarin Chinese uses two different coordinators for
these two cases, h´
aishi and hu`
ozhe (both translate as ‘or’) (Li and Thompson
(1981:654)):
(11) a. Nˇıy`ao wˇob¯ang nˇı h´aishi y`ao z`ıjˇızu`o
you want I help you or want self do
‘Do you want me to help you, or do you want to do it yourself?’
b. W ˇomen z`ai zh`eli ch¯ı hu `ozhe ch¯ıf`andi`an d ¯ou x´ıng
we at here eat or eat restaurant all OK
‘We can either eat here or eat out’
More fine-grained semantic distinctions such as these are discussed further in
Section 4.
Next, I discuss some special types of conjunction. Since conjunction is the
most frequent kind of coordination, it exhibits the greatest formal diversity, and
some of these patterns are examined in Section 5. The most prominent ‘special
type’ of conjunction involves the use of a comitative marker (i.e. a marker
expressing accompaniment), as in Hausa, where da means both ‘with’ (12a)
and ‘and’ (12b) (Schwartz (1989:32, 36)):
(12) a. Na je kasuwa da Audu
I.pfv go market with Audu
‘I went to the market with Audu’
b. Dauda da Audu sun je kasuwa
Dauda and Audu they.pfv go market
‘Dauda and Audu went to the market’
In addition to coordinations in which each coordinand is a regular syntac-
tic constituent (e.g. an np,avp, or a clause), many languages allow non-
constituent coordination, as illustrated in (13). For the sake of clarity, the
coordinands are enclosed in square brackets in these examples.
(13) a. [Robert cooked the first course] and [Maria the dessert]
b. Ahmed [sent a letter to Zaynab] or [a postcard to Fatima]
c. [Martin adores], but [Tom hates Hollywood movies]
In (13a) and (13b), the first coordinand is an ordinary constituent (a sen-
tence and a vp, respectively), but the second coordinand is not. In (13c), only
the second coordinand is an ordinary constituent. In order to assimilate non-
constituent coordinations to patterns found elsewhere in the grammar, linguists
have often described them in terms of ellipsis (or coordination reduction).
That is, abstract underlying structures such as those in (14a–c) are posited,

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 5
which show ordinary constituent coordination. In a second step, a rule of ellip-
sis of identical elements deletes the words underlined in (14), resulting in the
surface patterns in (13).
(14) a. Robert cooked the first course and Maria cooked the dessert
b. Ahmed sent a letter to Zaynab or sent a postcard to Fatima
c. Martin adores Hollywood movies, but Tom hates Hollywood
movies
Non-constituent coordination and ellipsis are discussed further in Section 6.
Finally, in Section 7 I discuss ways of distinguishing coordination from less
grammaticalized constructions and, perhaps most importantly, from subordi-
nation and dependency. The latter two notions will be discussed briefly here.
The primary contrast is that between coordination and dependency. In a coor-
dination structure of the type A(-link-)B,Aand Bare structurally symmetrical
in some sense, whereas in a dependency structure of the type X(-link-)Y,Xand
Yare not symmetrical, but either Xor Yis the head and the other element is a
dependent. When the dependent element is a clause, it is called subordinate
clause.
Although the distinction between coordination and dependency is, of course,
fundamental, it is sometimes not evident whether a construction exhibits a coor-
dination relation or a dependency relation. The best-known distinctive property
of coordinate structures is that they obey the coordinate structure constraint
(J. R. Ross (1986)), which prohibits the application of certain rules, such as
extraction of interrogative words from coordinate structures. This is illustrated
in (15–16), where the (i) sentences show the basic structure, and the (ii) sen-
tences show fronting of who. As the examples make clear, only the dependency
structures allow extraction (15aii and 16aii).1
(15) a. dependency (subordination)
(i) (basic sentence) You talked to someone before Joan arrived
(ii) (who extraction) Who did you talk to before Joan arrived?
b. coordination
(i) (basic sentence) You talked to someone and then Joan arrived
(ii) (who extraction) *Who did you talk to and then Joan arrived?
1Not all dependency/subordination structures allow extraction. For instance, extraction from rel-
ative clauses is blocked in many languages (see (a)), while extraction from complement clauses
is typically possible (see (b)).
(a) You think Joan saw someone./ Who do you think Joan saw ?
(b) You knowa woman who admires someone. /*Whom do you know a woman who admires
?
Thus, the possibility of extraction is only a sufficient, not a necessary, condition for a dependency
relation.

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
6Martin Haspelmath
(16) a. dependency
(i) (basic sentence) You saw Marvin with someone
(ii) (who extraction) Who did you see Marvin with ?
b. coordination
(i) (basic sentence) You saw Marvin and someone
(ii) (who extraction) *Who did you see Marvin and ?
Obeying the coordinate structure constraint is a formal property of constructions
that is sometimes taken as the decisive criterion for coordinate status. In this
chapter, by contrast, I will work with a primarily semantic definition of coor-
dination, as given at the beginning of this section. The reason for this is that
only semantically based notions can be applied cross-linguistically – formal
criteria are generally too language-particular (for instance, not all languages
have extraction constructions that would show the effect of the coordinate struc-
ture constraint).
1 Types and positions of coordinators
Coordinate constructions may lack an overt coordinator (asyndetic coordina-
tion) or have some overt linking device (syndetic coordination). So far in
this chapter, all examples have shown syndetic coordination. If we restrict our-
selves for the moment to binary coordinations, syndetic coordinations may have
either a single coordinator (monosyndetic) or two coordinators (bisyndetic).
Monosyndetic coordination is illustrated by Franz and Sisi (cf. (6b)), and bisyn-
detic coordination is illustrated by both Franz and Sisi (cf. (6a)). Coordinators
may be prepositive (preceding the coordinand) or postpositive (following the
coordinand). In English, all coordinators are prepositive, but we saw an example
of the postpositive coordinator -u: in Kannada earlier (example (5)).
The logical possibilities for binary coordination are shown schematically in
(17) (the two coordinands are represented as Aand B, and the coordinator is
represented as co).
(17) a. (asyndetic) AB
b. (monosyndetic) A co-B (prepositive, on second coordinand)
A-co B (postpositive, on first coordinand)
A B-co (postpositive, on second coordinand)
co-A B (prepositive, on first coordinand)
c. (bisyndetic) co-A co-B (prepositive)
A-co B-co (postpositive)
A-co co-B (mixed)
co-A B-co (mixed)

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 7
As we will see below, with one exception (co-A B), all these possibilities occur
in languages. However, not all of them are equally common.
1.1 Asyndetic coordination
Coordination without an overt linker occurs widely in the world’s languages,
and, although in European languages monosyndesis of the type A co-B is the
norm, asyndesis (also called juxtaposition) also occurs commonly, especially
with the meaning of conjunction:
(18) a. (English) Slowly, stealthily, she crept towards her victim
b. (German) ein elegantes, ger¨aumiges Foyer
‘an elegant, spacious entrance hall’
c. (French) Dans quel philtre, dans quel vin, dans quelle tisane
noierons-nous ce vieil ennemi?
(Baudelaire in Grevisse (1986:§253))
‘In which love potion, in which wine, in which herbal tea
shall we drown this old enemy?’
In European languages, asyndesis occurs mostly with modifying phrases such
as adverbials and adjectives, or with clauses. Asyndetic coordination of npsis
more restricted and quite impossible in many cases (cf. ??I met Niko, Sandra
‘I met Niko and Sandra’). Many non-European languages have no such restric-
tions, and asyndetic coordination is very wide-spread in the world’s languages.
The following examples are from Sarcee (an Athapaskan language of Alberta,
Canada), Maricopa (a Yuman language of Arizona), and Kayardild (a Tangkic
language of northern Australia).
(19) a. `ıstl´ıg´uts`ıs d ´o´on´ı`ıc¯ıctc `ud, g¯ın´ı
horse scalp gun I.capture they.say
‘ “I captured horses, scalps, and guns”, they say’
(Cook (1984:87))
b. John Bill ˜ni-ʔ-yuu-k
John(acc) Bill(acc)pl.obj-1-see.sg-realis
‘I saw John and Bill’ (Gil (1991:99))
c. wumburu-nurru wangal-nurru bi-l-d
spear-having boomerang-having they-pl-nom
‘They have spears and boomerangs with them’
(Evans (1995:250))
In asyndesis, intonation is the only means by which the coordinated structure
can be indicated, and it is probably not an accident that languages with a long
written tradition tend to have a strong preference for syndesis: intonation is not
visible in writing (see Mithun (1988)). Languages that lack writing (or lacked

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
8Martin Haspelmath
it until recently) often lack indigenous coordinators and now use coordinators
borrowed from prestige languages such as Spanish, English, Arabic and Rus-
sian. Asyndesis is often preferred in natural conjunction, i.e. when the two
conjuncts habitually go together and form some kind of conceptual unit (see
Section 4.1 below).
1.2 Monosyndetic coordination
There are three occurring patterns of monosyndetic coordination: A co-B,A-co
B, and A B-co, which are illustrated in (20–22). The logically possible type
co-A B is unattested (this fact will be explained below).
(20) A co-B (Lango, a Nilotic language of Uganda; Noonan (1992:163))
`
Ok´el `o`om `at`ocˆay k`ed`e c`ak
Okelo 3sg.drink.pfv tea and milk
‘Okelo drank tea and milk’
(21) A-co B (Classical Tibetan; Beyer (1992:240))
Blama-s bgegs-daŋndre btul
lama-erg demon-and spirit tamed
‘The lama tamed demons and spirits’
(22) A B-co (Latin)
senatus populus-que
‘the senate and the people’
The two types A co-B (medial prepositive) and A-co B (medial postposi-
tive) can be distinguished on the basis of evidence for different constituency
divisions: [A] [co B] vs [A co] [B]. Relevant constituency tests include:
(i) Intonation: in certain cases, English and forms an intonation group with
the following phrase, not with the preceding phrase (Joan, and Marvin, and
their baby, not *Joan and, Marvin and, their baby; here commas represent
intonation breaks). Of course, this test does not apply in the simplest
cases: a construction such as Joan and Marvin forms a single intonation
group.
(ii) Pauses: in English, it is much more natural to pause before and (Joan . . .
and Marvin) than after and (??Joanand...Marvin).
(iii) Discontinuous order: in special circumstances, the coordinands may
be separated by other material, as when a coordinand is added as an
afterthought. In English, the coordinator must be next to the second coor-
dinand (e.g. My uncle will come tomorrow, or my aunt, not *My uncle or
will come tomorrow, my aunt).
(iv) (Morpho)phonological alternations: when the coordinator or one of
the coordinands undergoes (morpho)phonological alternations in the

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 9
construction, this is evidence that they form a constituent together. For
instance, in Biblical Hebrew the coordinator w
ə
‘and’ has the alternant ¯
u
when the first syllable of the following phrase has a schwa vowel (e.g.
w
ə
-
ð
æræx ‘and (a) way’, ¯
u-
ðə
r˚
ax-¯
ım ‘and ways’). In Latin, the element -
que and the preceding conjunct form a single domain for stress assignment
(e.g. p´
opulus ‘the people’, popul´
us-que ‘and the people’).
In principle, one could imagine cases in which none of these criteria yields
a clear asymmetry, so that one would have a symmetrical pattern A-co-B in
addition to prepositive A co-B and postpositive A-co B. But no case of a lan-
guage that requires such an analysis has come to my attention. Monosyndetic
coordination seems to be universally asymmetric.
When the coordinator is linked by phonological processes to its coordi-
nand (see (iv) above), it is generally regarded as a clitic or affix rather than
an independent word. (Criteria for clitic or affix status are largely language-
particular and cannot be discussed further here.)2Due to the universal pref-
erence for suffixation over prefixation, postpositive coordinators are typically
suffixed and thus written as one word with the coordinand to which they are
attached. Prepositive coordinators, by contrast, are rarely prefixed and written
together with the coordinand. Thus, when a language has a coordinate con-
struction of the form AcoB, where co is not an affix on Aor B, it is likely
that constituency tests will show co to be a prepositive coordinator, like English
and.
Postpositive coordinators may follow the complete phrase, or they may enc-
litically follow the first word of the coordinand. The latter is illustrated by
Turkish postpositive de in (23).
(23) Hasan ıstakoz-u pisir-di, Ali de balıˇgı
Hasan lobster-acc cook-past(3sg) Ali and fish-acc
‘Hasan cooked the lobster, and Ali (cooked) the fish’
(Kornfilt (1997:120))
As is noted in Stassen (2000), the order of the coordinator correlates with
other word order patterns of the language, in particular verb–argument order:
languages with a postpositive coordinator (such as Latin and Classical Tibetan)
tend to have verb-final word order, whereas verb-initial languages tend to have
a prepositive coordinator. However, Stassen’s generalizations are based exclu-
sively on conjunctive coordinators. Disjunctive coordinators may conform to
different ordering patterns. For instance, Kanuri (a verb-final Nilo-Saharan
language of northern Nigeria) has (bisyndetic) postpositive conjunctive coor-
dinators (-a...-a, see (24a)), but a (monosyndetic) prepositive disjunctive
2But note that coordinators apparently never show suppletion, i.e. totally different shapes depend-
ing on the lexical class or the phonological shape of their host. In this sense, they are universally
closer to clitics than to affixes. (Such suppletion is not uncommon with affixes.)

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
10 Martin Haspelmath
coordinator rˆ
a((see 24b)). A similar asymmetry is found, for instance, in Lez-
gian (a verb-final Daghestanian language; Haspelmath (1993:327, 331)).
(24) a. kˆam ´adə-ak´am´ut´ud´u-a
man this-and woman that-and
‘this man and that woman’ (Cyffer (1991:70))
b. kit´awu ´adərˆat´udu ra ˆam
book this or that you.like
‘Do you like this book or that one?’
As I noted above, the pattern co-A B is unattested and seems to be non-
existent, at least for conjunction (see Stassen (2000), who examined a sample
of 260 languages). This generalization can be explained diachronically if the
two main diachronic sources of conjunction constructions are (i) a comitative
modifying construction of the type ‘A with B’ (see Section 5.1), and (ii) a
construction with an additive focus particle of the type ‘A, also B’. An example
of a comitative-derived construction is Lango cˆ
ay k`
ed`
ec
`
ak ‘tea and milk’ (cf.
(20)), which comes from a dependency construction in which k`
ed`
ec
`
ak is a
modifier meaning ‘with milk’ (in fact, the phrase can still have this meaning;
Noonan (1992:163)). Since languages with modifier–noun order tend to have
postpositions and languages with noun–modifier tend to have prepositions (cf.
Greenberg (1963); Dryer (1992)), the patterns A-co B and A co-B are the
most expected ones from the comitative source. The focus-particle source of
conjunction always has the marker on the second conjunct: ‘A, also B’, or ‘A,
B too’. When the focus particle is postpositive (like too), this yields A B-co,
and when the focus particle is prepositive (like also), this yields A co-B. There
is thus apparently no common diachronic source for the pattern co-A B, whose
non-existence or extreme rarity is thereby explained.
1.3 Bisyndetic coordination
When there are two coordinators in the binary coordination, there are again
four logically possible patterns, but in this case, all four patterns are attested
(see (25–28)). However, the mixed patterns (27–28) seem to be extremely rare.
In the non-mixed patterns (co-A co-B,A-co B-co), both coordinators generally
have the same shape, whereas this is not the case in the mixed patterns.
(25) co-A co-B (Yoruba, a Kwa language of Nigeria; Rowlands
(1969:201ff.))
`ati `emi `ati K´e
.h`ınd´e
and I and Kehinde
‘both I and Kehinde’

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 11
(26) A-co B-co (Martuthunira, a Pama-Nyungan language of
W. Australia)
puliyanyja-ngara-thurti jantira-ngara-thurti
old.man-pl-and old.woman-pl-and
‘old men and old women’
(cf. also examples (5) and (24a)) (Dench 1995:98)
(27) A-co co-B (Homeric Greek, cf. Dik (1968:44))
Atre´ıd¯es te ka`ı Akhille´us
Atreus’s.son and and Achilles
‘Atreus’s son and Achilles’
(28) co-A B-co (Latin, cf. Dik (1968:44))
et singulis universis-que
‘both for individuals and for all together’
Stassen (2000) finds that, for conjunctive coordination, postpositive bisyn-
desis (A-co B-co) is fairly widely attested, especially in the Caucasus, north-
eastern Africa, Australia, New Guinea and southern India. By contrast, prepos-
itive bisyndesis (co-A co-B) is only found as an emphatic variant of prepositive
monosyndesis. Thus, besides (25), Yoruba also has the non-emphatic monosyn-
detic pattern `
emi `
ati K´
e
.h`
ınd´
e‘I and Kehinde’, and several European languages
have similar patterns (e.g. French (et) Jean et Marie ‘(both) Jean and Marie’,
Russian (i) Nina i Miˇ
sa ‘Nina and Misha’) (see further Section 2.1).
1.4 Multiple coordinands
So far we have only examined binary coordinations, but for conjunction and
disjunction, all languages seem to allow an indefinite number of coordinands,
i.e. multiple or n-ary coordination. This will be symbolized by a sequence of
letters A, B, C, . . . M, N, where A, B, C stand for the initial coordinands, and
M, N for the final coordinands.
The question now arises how the basic pattern that is used in binary coor-
dination is applied to multiple coordination. In the bisyndetic types, this is
straightforward: the type A-co B-co becomes A-co B-co C-co..., and the type
co-A co-B becomes co-A co-B co-C . . . , i.e. each coordinand is associated
with a single coordinator:
(29) A-co B-co C-co...(Nivkh, an isolate of Sakhalin; Panfilov
(1962:169))
˜
Ni jozo-γomeut¸u-γopos-ko γe-d¸.
I lock-and rifle-and cloth-and buy-finite
‘I bought a lock, a rifle and cloth’

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
12 Martin Haspelmath
(30) co-A co-B co-C...(French)
Le congr`es sera tenu ou `a Paris ou `a Rome ou `a Varsovie
‘The congress will be held either in Paris or in Rome or in Warsaw’
When the monosyndetic type occurs with multiple coordinands, there are two
possibilities: a full pattern and a pattern with coordinator omission. In the full
pattern, only one coordinand lacks its own coordinator, the same that lacks the
coordinator in the binary construction. Thus, A co-B becomes A co-B co-C . . .
(the first coordinand lacks a coordinator), A-co B becomes A-co B-co . . . N
(the last coordinand lacks a coordinator), and A B-co becomes A B-co C-co . . .
(again the first coordinand lacks a coordinator).
(31) A co-B co-C...(Polish)
Tomek iJurek iMaciek przyjechali do Londynu
‘Tomek and Jurek and Maciek went to London’
(32) A-co B-co...N(Lezgian; Haspelmath (1993:327))
K’¨ud warz-ni,k’¨ud juˇg-ni,k’¨ud deq’iq’a alat-na.
nine month-and nine day-and nine minute pass-past
‘Nine months, nine days and nine minutes passed’
(33) A B-co C-co...(West Greenlandic; Fortescue (1984:124))
ini igavvil=lu qalia-ni=lu sinittarvi-it marluk
room kitchen-and loft-loc-and bedroom-pl two
‘a living room and a kitchen and two bedrooms in the loft’
But, in many languages, coordinations with multiple coordinands allow (or
even require) coordinator omission, by which, most commonly, all but the last
coordinator are eliminated. Thus, English can reduce A and B and C to A, B and
C, and French can reduce AouBouCto A,BouC. In fact, coordinator omission
is strongly favoured in English and other European languages. Keeping the
coordinators on all coordinands has an emphatic value and is appropriate only
under special circumstances.
Coordinator omission is found quite similarly in languages with postpositive
coordinators:
(34) West Greenlandic (basic pattern: A B-co): AB...N-co
tulu-it qallunaa-t kalaall-il=lu
Englishman-pl Dane-pl Greenlander-pl-and
‘Englishmen, Danes and Greenlanders’ (Fortescue (1984:127))
(35) Amharic (basic pattern: A-co B): AB...M-co N
ˇc
.¨aw b¨arb¨are-nna qəbe am¨at
.t
.awh
salt pepper-and butter I.brought
‘I brought salt, pepper and butter’ (Leslau (1995:725))

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 13
Table 1.1 Correspondences among coordination patterns
multiple coordination
basic/binary full pattern with coordinator omission
A co-B A co-B co-C . . . AB...coN
A-co B A-co B-co ...N AB...M-co N
A-co B-co A-co B-co C-co . . . AB...N-co
co-A co-B co-A co-B co-C . . . AB...co-N
A B-co A B-co C-co . . . AB...N-co
The correspondences among the major patterns of binary coordination, mul-
tiple coordination and coordinator omission (with omission of all but the last
coordinator) are shown in Table 1.1.
However, these are not the only possibilities of coordinator omission. For
instance, in Classical Tibetan and Amharic (both of which have a basic A-co B
pattern), coordinator omission eliminates all but the first coordinator (A-co B
C...):
(36) a. Classical Tibetan (Beyer (1992:241))
sa-daŋtˇshu me rluŋ
earth-and fire water air
‘earth, fire, water and air’
b. Amharic (Leslau (1995:725))
ˇc
.¨aw-ənna b¨arb¨arre qəbe am¨at
.t
.awh
salt-and pepper butter I.brought
‘I brought salt, pepper and butter’
Some languages can be even more radical in applying coordinator omission
to multiple coordination: they can completely omit coordinators from multiple
coordinands, even though a coordinator would be required or preferred for
binary coordinations. This is reported, for instance, for Classical Tibetan (Beyer
(1992:241)), Cantonese (Matthews and Yip (1994:289)) and Nkore-Kiga (a
Bantu language of Uganda; Taylor (1985:57)).
However, not all languages allow coordinator omission. For example, in Pon-
apean (an Austronesian language of Micronesia; Rehg (1981:333)) the coordi-
nation in (37) cannot be reduced by deleting all but the last coordinator.
(37)
Soulik oh Ewalt oh Casiano oh Damian pahn doadoahk lakapw
Soulik and Ewalt and Casiano and Damian fut work tomorrow
‘Soulik, Ewalt, Casiano and Damian will work tomorrow’

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
14 Martin Haspelmath
For Yoruba, Rowlands (1969:36) cites examples with coordinator deletion such
as epo, e
.ran, ata `ati `
al`
ub´
o
.s`
a‘palm-oil, meat, pepper and onions’ (coordinator
`
ati, as in (25) above), but he notes that this is possibly a case of imitation of
English usage.
1.5 The scope of coordinators
In English, the coordinators can be either within the scope of prepositions (e.g.
(38a)), or outside their scope (e.g. (38b)). There is perhaps a slight semantic
difference here: in (38a), it seems more likely that we are dealing with a joint
present for a couple, whereas (38b) is preferred if two different presents for
unrelated people are referred to:
(38) a. I bought a present for [Joan and Marvin]
b. I bought a present [for Joan] and [for Marvin]
The more strongly an adposition is grammaticalized, the more likely it is to
be repeated in coordination (i.e. the more likely it is that the coordinator has
scope over the adposition). For example, in French the preposition `
acan take
the coordinator et ‘and’ in its scope if it has a spatial (allative) meaning (e.g.
(39a)), but it must be inside the scope of the coordinator if it has the more
grammaticalized ‘dative’ meaning ((39b), cf. Melis (1996:67)):
(39) a. Je vais `a [Turin et Venise]
‘I’m going to Turin and Venice’
b. J’ai emprunt´ecelivre[`a Jean] et [`a Marie] (*...`a Jean et Marie)
‘I borrowed this book from Jean and Marie’
Case affixes, which are even more grammaticalized, have a strong tendency
to occur inside the scope of coordinators; and this can be the case with either
monosyndetic or bisyndetic coordination.
(40) a. Lezgian (a Daghestanian language of the eastern Caucasus)
[Ali-din]-ni [Weli-din] buba
Ali-gen-and Weli-gen father
‘Ali’s and Weli’s father’ (Haspelmath (1993:326))
b. Kunuz Nubian (a Nilo-Saharan language of Egypt)
[it-todon]-go:n [e:n-godon]-go:n
man-com-and woman-com-and
‘with the man and with the woman’ (Abdel-Hafiz (1988:277))
However, in some languages, even case affixes can be outside the scope of the
coordinator. In Classical Tibetan (e.g. (41a)) and in Turkish (e.g. (41b)), the
wider scope of the case suffix can be seen in the absence of the case suffix on

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 15
the first coordinand (cf. Johannessen (1998:9–24) for further examples).
(41) a. [ri-daŋluŋpa]-la
mountain-and valley-all
‘to mountain and valley’ (Beyer (1992:240))
b. [ev-le sokaˇg]-a
house-and street-dat
‘to the house and the street’ (Underhill (1976:83))
In these languages, the coordinator (-da
ŋ
,-le) comes from a former case-marker,
so it is perhaps not so surprising that it should not co-occur with a case-marker
in the same word. However, there are even some languages in which a case
suffix follows the suffixed coordinator in the same word:
(42)
a. Djabugay (a Pama-Nyungan language of Australia; Patz (1991:292))
yaba-nggu nyumbu-djada-nggu
brother-erg father-and-erg
‘my brother and father’
b. Tauya (Trans-New Guinea; MacDonald (1990:138))
awa ya-pi-sou afe ya-pi-sou-ni me watamu ya-tu-i-ʔa
father I-gen-and mother I-gen-and-erg this thing me-give-3pl-ind
‘My father and my mother gave me this thing’
2 Emphatic coordination
2.1 Conjunction and disjunction
Many languages distinguish between normal coordination such as A and B,Xor
Y, and what might be called emphatic coordination:both A and B,either X or
Y. The semantic difference is that in emphatic coordination it is emphasized that
each coordinand belongs to the coordination, and each of them is considered
separately. Thus, (43) is felicitous only if there was some doubt over one of
the conjuncts, and (44) is impossible with emphatic coordination, because two
things cannot be separately similar.
(43) Both Guatemala and Belize are in Central America
(44) (*Both) Spanish and Portuguese are similar
Likewise, either X or Y emphasizes the contrast between both coordinands and
requires that they be considered separately.
In European languages, this distinction is well known, but it is far less often
described for non-European languages. As a rule, European languages have
monosyndetic A co-B for normal coordination and bisyndetic co-A co-B for

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
16 Martin Haspelmath
emphatic coordination. The two coordinators (both . . . and,either . . . or) are
often called correlative coordinators in such emphatic constructions, because
at least one of them does not occur without the other. (Note, however, that there
are also languages where bisyndetic coordination is the normal, non-emphatic
construction; in these languages, the coordinators are apparently always post-
positive, and they always have the same shape – see Section 1.3.)
In emphatic coordination, it is not uncommon for both coordinators to have
the same shape and to be identical to the single coordinator (45a; for the moment
we again restrict ourselves to binary coordination). In other cases, only the
second coordinator is identical to the single coordinator (e.g. (45b)), and, more
rarely, the two coordinators are identical to each other, but not identical to the
single coordinator (e.g. (45c)):
correlative single
coordinators coordinator
(45) a. conjunction:
Russian i...i i
Italian e...e e
Modern Greek ke...ke ke
Albanian edhe...edhe edhe
disjunction:
Polish albo...albo albo
Dutch of...of of
Basque ala...ala ala
Somali ama...ama ama
b. conjunction:
English both...and and
Irish idir...agus agus
disjunction:
English either...or or
German entweder...oder oder
Finnish joko...tai tai
c. conjunction:
Hungarian mind...mind ´
es
Korean -to...-to -hako
disjunction:
Lezgian ja...ja waja
d. conjunction:
German sowohl...alsauch und
Polish jak...tak(i) i
Finnish sek¨
a...ett
¨
aja
Indonesian baik...maupun dan

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 17
A final possibility, at least for conjunction, is that both correlative coordi-
nators are different from the single coordinator and are not identical in shape
either (e.g. (45d)). This latter case typically derives from a circumlocution of
the semantic type ‘A as well as B’. For instance, Polish jak A tak (i) B literally
means ‘as A, so (also) B’.
2.2 Emphatic negative coordination
Many languages also have special correlative coordinators that are restricted
to the position in the scope of negation, such as English neither...nor.
Again, such negative coordinators have mostly been described for European
languages (cf. Bernini & Ramat (1996:100–6)), and it is unclear whether they
are indeed a peculiarity of Europe or are simply insufficiently described for other
languages.
Negative coordination of the type We met neither Marvin nor Joan could be
described either as conjunction (because a possible paraphrase is ‘We didn’t
meet Marvin, and we didn’t meet Joan either’), or as disjunction (because
another possible paraphrase is ‘We didn’t meet either Marvin or Joan’).3This
is related to the well-known logical equivalence of disjunction with wide-scope
negation and conjunction with narrow-scope negation (in the notation of sym-
bolic logic: ¬(p ∨q) ≡¬p&¬q).4Accordingly, some languages have
emphatic negative coordinators that are related to disjunctive coordinators (e.g.
(46a)), whereas other languages have emphatic negative coordinators that are
related to conjunctive coordinators (e.g. (46b)). A third group of languages
have negative correlatives that are not formally related at all to semantically
related expressions (e.g. (46c)). In (46), the emphatic negative coordinators are
shown in the left-hand column, and related elements are shown in the right-hand
column.
(46) a. English neither...nor either...or
German weder...noch entweder...oder ‘either...or’
Swedish varken...eller antingen...eller ‘either...or’
b. Latin ne-que...ne-que -que ‘and’
c. Italian n´
e...n
´
ee‘and’, o‘or’, non ‘not’
Dutch noch...noch en‘and’, of ‘or’, niet ‘not’
Maltese la...ulanqas u ‘and’, jew ‘or’, ma ‘not’
In quite a few languages (47), the emphatic negative coordinators are also used
as scalar focus particles of the type ‘not even’ or ‘neither’, as in Polish (e.g.
(48a,b))
3J. R. Payne (1985) uses the term rejection, implying that negative coordination is neither a type
of conjunction nor a type of disjunction.
4In this notation, ¬means ‘not’, ∨means ‘or’, & means ‘and’, and ≡means ‘is equivalent to’.

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
18 Martin Haspelmath
(47) Polish ani...ani, Russian ni...ni, Hungarian sem...sem,
Modern Greek ´
ute...´
ute, Albanian as...as, Romanian nici . . . nici
(48) a. Ani mnie, ani jemu sienie udalo
neither i.dat nor he.dat refl not succeeded
‘Neither I, nor he succeeded’
b. Karliczek ani sl´owka mi nie powiedzial
Karliczek not.even word me.dat not said
‘Karliczek didn’t even say a word to me’
Languages without special negative coordinators can use their emphatic con-
junctive coordinators (49a) or their emphatic disjunctive coordinators (49b) to
express the same content.
(49)
a. Indonesian (Sneddon (1996:348); baik A maupun B ‘both A and B’)
Baik kepandaian maupun kecantikan tidak berguna
both ability and beauty not useful
untuk mencapai kebahagiaan
for achieve happiness
‘Neither ability nor beauty is useful for achieving happiness’
b. Lezgian (Haspelmath (1993:334); jaAjaB‘either A or B’)
I k’walaxda-l ja aburu-n ruˇs, ja gada razi tuˇs-ir
this job-obl either they-gen girl or boy satisfied be.neg-past
‘Neither their girl nor the boy was satisfied with this job’
Less wide-spread than correlative negative coordinators are special coordi-
nators which do not occur in correlative pairs, but are restricted to positions in
the scope of negation. This is the type (not) A nor B, which can be paraphrased
by ‘not A or B’ or by ‘not A and not B’.
(50) a. His father wouldn’t give the money nor would he lend it
b. (Italian) Giovanni non parla n´esimuove
‘Giovanni does not talk nor move’
(Bernini and Ramat (1996:100))
In English, (50a) also has an emphatic counterpart with correlative coordinators
(His father would neither give nor lend the money), but Italian (50b) has no
such counterpart. When clauses are coordinated, Italian cannot use n´
e...n
´
e,
its negative correlative pair for nps (cf. (46c)). Irish completely lacks emphatic
negative coordinators and only has the single word n´
a‘nor’. The closest Irish
equivalent to ‘He has neither a son nor a daughter’ is (51), where the first
negative word is the ordinary sentence negation.

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 19
(51) N´ıl mac n ´ain´ıon aige
neg.is son nor daughter at.him
‘He doesn’t have a son nor a daughter’
3 Types of coordinands
The definition of coordination at the beginning of this chapter contains the
phrase ‘two or more units of the same type’. This can be seen as an automatic
consequence of the required identity of semantic roles of the coordinands: if
two expressions have different semantic roles (e.g. patient and location), it will
not be possible to coordinate them (e.g. *We want to eat pizza or in a Thai
restaurant). It is sometimes said that the coordinands must belong to the same
phrasal category; for instance, *[pizza]
np
or [in a Thai restaurant]
pp
is said to be
ungrammatical because it consists of an np and a pp. However, coordination of
different phrasal categories is often possible when both have the same semantic
role:
(52) a. Mr Hasegawa is [a legal wizard]np but [expensive to hire]ap
b. She felt [quite happy]ap and [at ease]pp in her new office
c. There will be typology conferences [in August]pp and [next
April]np
d. [His kindness]np and [that he was willing to write letters to me]s
amazed me
(53) a. Maltese (Borg and Azzopardi-Alexander (1997:81))
−
Harbu [malajr]adv, [bil-mobi]pp u[malli setgu]s
escape.pf.3pl quickly with-stealth and when can.pf.3pl
‘They escaped quickly, stealthily and as soon as they could’
b. Italian (Scorretti (1988:246))
La situazione [meteorologica]ap e[del traffico]pp `e buona
the situation meteorological and of.the traffic is good
‘The weather and traffic condition is good’
Conversely, if two expressions belong to the same phrasal category but have a
different semantic role, coordination is generally not felicitous (see also (9b)).
(Ill-formed structures such as (54) are often called zeugma.)
(54) a. *Ms Poejosoedarmo bought a book [in Penang]pp and
[in the spring]pp
b. *I still smoked [last year]np and [cigarettes]np
c. *[Go home!]sand [are you hungry?]s
The examples in (52–54) seem to suggest that semantic factors alone deter-
mine whether two expressions can be coordinated. But there are also cases in

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
20 Martin Haspelmath
which two syntactically dissimilar phrases have the same semantic role but do
not coordinate felicitously:
(55) *[Waterskiing]np and [to climb mountains]vp can be fun
(Grover 1994:764)
There is also some cross-linguistic variation. For instance, Italian allows coor-
dinations like (56), whose direct counterparts are impossible in English.
(56) Evitate gli accordi [poco chiari]ap,o[che potrebbero danneggiarci
gravemente]s(Scorretti (1988:246))
‘*Avoid insufficiently clear agreements, or which could hurt us
seriously.’
In many languages, the semantic–syntactic type of the coordinands is rel-
evant for the choice of the coordinators. The most widespread contrast for
conjunction is that between np conjunction and event conjunction (i.e. vp or
clause conjunction). For instance, Korean has the suffix -(k)wa for np con-
junction (57a), but event coordination is expressed by a suffix -ko on the
verb (57b). The Turkish contrast between -la and -ıp (58a,b) is completely
analogous.
(57) a. yenphil-kwa congi
pencil-and paper
‘pencil and paper’ (Martin and Lee (1986:51))
b. Achim mek-ko hakkyo ka-ss-eyyo
breakfast eat-and school go-past-ind
‘I ate breakfast and went to school’
(58) a. Hasan-la Amine
Hasan-and Amine
‘Hasan and Amine’
b. C¸ ocuk bir ka¸sık ¸corba al-ıp i¸c-er
child one spoon soup take-and eat
‘The child takes a spoon of soup and eats’
In such cases, there is often some doubt over whether the event coordination
really constitutes coordination, or perhaps rather some kind of subordination
(or ‘cosubordination’, cf. Van Valin and LaPolla (1997:454)). Verb forms suf-
fixed with such quasi-coordinating markers as Korean -ko and Turkish -ıp are
commonly called converbs (see Haspelmath and K¨onig (1995)), and the closest
syntactic analogue of (57b) is perhaps the English participial construction Hav-
ing eaten breakfast, I went to school. The issue of the coordinate or subordinate
status of these constructions is discussed further in Section 7.1.

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
Coordination 21
While the binary contrast between np coordination and event coordination is
certainly the most wide-spread in languages, we also find languages in which
there are more contrasts. For example, Yoruba has `
ati for nps (59a), t´
ıfor relative
clauses (59b) and s`
ıfor main clauses (59c) (Rowlands (1969:201–3)):
(59) a. `emi `ati K´e
.h`ınd´e
I and Kehinde
‘Kehinde and I’
b. epo ni mo ´n-r`a t´ımo t´un ´n-t`a
palm.-oil foc Iprog-buy and I repeat prog-sell
‘It is palm-oil that I buy and in turn sell’
c. ´om´umil’´ara d´a, `emi k`ıyi´os`ıgb`agb´e
he cause me in body well I neg fut and forget
‘He caused me to get better, and I shall not forget’
Somali has iyo ‘and’ for nps (60a), oo ‘and’ for vps (60b) and the suffix -na
‘and’ for clauses (60c) (Berchem (1991:324–7)):
(60) a. rooti iyo khudrat
bread and fruit
‘bread and fruit’ (p. 324)
b. Suuqa tag oo soo iibi rooti
market go and andat buy bread
‘Go to the market and buy bread!’ (p. 325)
c. Carrur-tu waxay joogaan dugsi-ga waxay-na
children-art 3pl.foc be school-art 3pl.foc-and
bartaan Af-Soomaali
learn language-Somali
‘The children are in school, and they learn Somali’ (p. 327)
The use of different formal means for expressing np conjunction and event
conjunction is probably the majority pattern in the world’s languages. Welmers
(1973:305) says that he is not aware of any African language that expresses
np conjunction and sentence conjunction in the same way. This is in striking
contrast to European languages, where the ‘and’ word is always used for both
purposes. But the twofold use of ‘and’, both for np conjunction and for event
conjunction, is also found often outside of Europe, e.g. in Chukchi (Chukotka,
eastern Siberia), Chalcatongo Mixtec (Mexico) and Samoan.
While conjunctive coordinators are thus often selective with respect to the
syntactic–semantic type of the coordinands, this is much less true of disjunctive
coordinators. Quite a few languages have different coordinators for np and event
conjunction, but one and the same coordinator for np and event disjunction, e.g.
the languages in (61):

P1: RPU/... P2: RPU
9780521581578c01a.xml CUUK171B-Shopen V. 2 December 24, 2006 14:11
22 Martin Haspelmath
(61) np event np & event
conjunction conjunction disjunction
Maori (Polynesian) me aa raanei
Chamorro (Austronesian) yan ya pat
Yapese (Micronesian) ngea ma faa
Supyire (Gur, Mali) n´
ak
`
a/m`
al
`
aa
J. R. Payne (1985:5) proposes an implicational sequence that constrains the
possible ranges of coordinators: S – vp –AP–pp –np.5The prediction that
this makes is that individual coordinators are restricted to cover contiguous
categories, e.g. S and vp, or AP, pp and np. There can be no coordinators,
according to this hypothesis, that only link sentences and APs, but not vps, or
vps and nps, but not APs and pps, and so on.
Sometimes languages are also selective with respect to which coordinand
types they even allow to be coordinated. For instance, Koromfe (a Gur lan-
guage of Burkina Faso) only allows event disjunction, and no np disjunction,
so that a sentence like ‘Do you want coffee or tea?’ must be rephrased as
‘Do you want coffee, or do you want tea?’ (Rennison (1997:93)). Somali
does not allow the conjunction of predicative adjectives, so that a sentence
like ‘That house was new and big’ must be rephrased as ‘That house was
new and it was big’ (Berchem 1991:327)). Arabic does not permit conjunc-
tion of two verbs, so that ‘Ahmed ate and drank’ must be rephrased as
‘Ahmed ate and he drank’ (Harries-Delisle (1978:527)). Finally, Tinrin (an
Austronesian language of New Caledonia) allows sentence coordination and
np coordination with mˆ
e‘and’, but not vp coordination (Osumi (1995:258–
9)) (note that this seems to contradict Payne’s implicational sequence). These
are just a few random examples. Clear cross-linguistic patterns have yet to be
discovered.
4 Semantic subtypes of coordination
The three main semantic types of coordination are conjunction, disjunction and
adversative coordination. But languages can make more fine-grained semantic
distinctions. We already saw the important difference between non-emphatic
and emphatic coordination in Section 3. Some further semantic subtypes are
discussed in this section (see also Section 5 below for other special kinds of
conjunction).
5Payne refers to this as an implicational hierarchy, but it is not a hierarchy in the usual sense
(in which, for instance, the noun phrase accessibility hierarchy is a hierarchy). Rather, it is
a special (one-dimensional) case of an implicational map. See Haspelmath (2003) for gen-
eral discussion of implicational (or semantic) maps and the difference between maps and
hierarchies.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 23
4.1 Semantic subtypes of conjunction
The most important distinction in conjunction is the difference between nat-
ural conjunction and accidental conjunction (W¨alchli (2003)). In natural
conjunction, the conjuncts ‘habitually go together and can be said to form
some conventionalized whole or “conceptual unit”’ (Mithun (1988:332)). Typ-
ical examples of natural conjunction are ‘mother and father’, ‘husband and
wife’, ‘boys and girls’, ‘bow and arrows’, ‘needle and thread’, ‘house and
garden’. Natural conjunction generally consists of only two conjuncts (hence
the term binomial; cf. Malkiel (1959); Lambrecht (1984)). When a language
makes a formal distinction between natural and accidental conjunction, this
often involves the lack of an overt coordinator or of an intonation break in
natural conjunction (Mithun (1988); Stassen (2000)). Examples from Erzya
Mordvin (a Finno-Ugrian language of Russia; W¨alchli (2003)) and Burushaski
(an isolate of northern Pakistan; Lorimer (1935:105, 381)) are given in (62–63):
(62) a. t’et’at-avat b. t’ikˇse´n di sivel’e´n
father.pl-mother.pl grass and meat
‘father and mother’ =‘parents’ ‘grass and meat’
(63) a. mu: mu:mi b. jεkεu:ŋ
father mother I and you
‘father and mother’ ‘I and you’
The conjuncts in natural conjunction may be so tightly linked that the con-
struction can be regarded as a single compound word, i.e. a coordinative com-
pound (cf. W¨alchli (2003)). The spelling with a hyphen in (62a) points in this
direction. The explanation for the contrast between zero-marking and overt
marking must be sought in economy: since the conjuncts in natural conjunction
occur together very frequently, the relation between them is quite predictable
and overt marking is redundant.
The distinction between natural and accidental conjunction also plays a role
in the scope of elements that apply equally to both conjuncts, such as articles.
In English, one definite article for two conjuncts is sufficient for natural con-
junction (the house and garden), but not in accidental conjunction (*the house
and stamp collection). In Bulgarian, the subjunctive particle da is not repeated
in natural conjunction of verbs: see (64b), contrasting with (64a) (from W¨alchli
(2003)):
(64)
a. Ivan veˇce moˇz-eˇse da ˇcet-e ida pluva
Ivan already can-past.3sg sjnct read-3sg and sjnct swim(3sg)
‘Ivan could already read and swim’ (accidental conjunction)

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
24 Martin Haspelmath
b. Ivan veˇce moˇz-eˇse da ˇcet-e ipiˇs-e
Ivan already can-past.3sg sjnct read-3sg and write-3sg
‘Ivan could already read and write’ (natural conjunction)
In German (and to some extent also in English), the definite article may be
omitted from both conjuncts, e.g. Messer und Gabel ‘knife and fork’, Bleistift
und Papier ‘pencil and paper’ (Lambrecht (1984)). Mparntwe Arrernte (Pama-
Nyungan, Australia) has a special ‘binary-and’ construction (A uthene B uthene)
that is used when the two conjuncts ‘are commonly thought of as occurring
naturally together’ (Wilkins (1989:369)), e.g. alkere uthene angkwelye uthene
‘sky and clouds’, but not ??pwerte uthene angkwelye uthene ‘rocks and clouds’.
Yoruba has the special pattern t-A-t-B for natural conjunction, e.g. t-o
.ko
.-t-aya
‘husband and wife’, t-`
o
.san-t-`
oru ‘night and day’ (Rowlands (1969:202)). There
are probably quite a few further types of formal differences between natural
and accidental conjunction in languages, but grammars rarely describe them
in detail, perhaps because this conceptual and terminological distinction is not
widely known among linguists. Often just a few examples of conjunction are
given, and these are, of course, often examples of natural conjunction, because
natural conjunction is so frequent.
A special type of conjunction can be called representative conjunction.
In this construction, the conjuncts are taken as representative examples of a
potentially larger class. In Koasati (a Muskogean language of Louisiana), the
suffix -o:t is used to connect a number of categorically similar nouns (Kimball
(1991:413)):
(65) akk´ammi-t ow-i:s´a-hci hahci-f-´o:t oktaspi-f-´o:t k´ammi-fa
be.so-conn loc-dwell.pl-prog river-in-ex swamp-in-ex be.so-in
‘So they live in rivers and in swamps and in suchlike places’
This suffix is not only a linker, because it can also be used on a single noun which
is intended as a representative of a larger set of nouns (Kimball (1991:414)):6
(66) as´a:l-o:t talibo:li-t sco:pa-t
basket-ex make-conn sell-conn
‘She made and sold things like baskets’
Japanese also has representative coordinators. According to Kuno (1973:114,
121), the linker ya or yara is used for giving examples:7
6In this use, the suffix -o:t comes close to a plural marker of the type that is often called associative
plural (e.g. Corbett and Mithun (1996); Corbett (2000)).
7Kuno adds that ‘yara seems to be suitable only when the speaker is annoyed (or affected) by
actions or states enumerated by the constructions’ (1973:121). This highly specific semantic
feature of the construction shows how difficult it is to put constraints on possible coordinator
meanings in languages.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 25
(67) John yara Mary (yara) ga yattekita
John ex Mary ex nom came
‘John and Mary (among others) came’
Another special type of conjunction involves the combination of several
identical elements to express intensity of an action or a high degree of a property,
as in She ran and ran, The city grew bigger and bigger, or in (68) from Syrian
Arabic:
(68) L-əmnaaqaˇse stamarret saaʕ-aat u-saaʕ-aat
the-argument continued hour-pl and-hour-pl
‘The argument went on for hours and hours’ (Cowell (1964:394))
This type of conjunction can be called augmentative conjunction. Although it
is semantically very distinctive, I am not aware of a language that uses a special
kind of coding for augmentative conjunction.
4.2 Semantic subtypes of disjunction
The most important distinction in disjunction is the difference between inter-
rogative disjunction and standard disjunction that was already illustrated in (11)
above. Another example comes from Finnish (coordinators tai and vai):
(69) a. Anna-n sinu-lle kirja-n tai albumi-n
give-1sg you-all book-acc or album-acc
‘I’ll give you a book or an album’
b. Mene-t-k¨o teatteri-in vai lepo-puisto-on
go-2sg-q theater-ill or rest-garden-ill
‘Are you going to a theatre or to a park?’
The distinction between standard and interrogative disjunction cannot be
reduced to the occurrence in declarative vs interrogative clauses, because stan-
dard disjunction may occur in questions as well. This is illustrated by (70a,b)
from Basque (Saltarelli (1988:84)).
(70) a. Te-a ala kafe-a nahi duzu
tea-art or coffee-art want you.it
‘Do you want tea, or coffee?’ (=‘Do you want tea or do you
want coffee?’)
b. Te-a edo kafe-a nahi duzu?
tea-art or coffee-art want you.it
‘Do you want tea or coffee?’ (=‘Do you want either tea
or coffee?’)

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
26 Martin Haspelmath
Interrogative disjunction occurs in an alternative (or disjunctive)question,
i.e. a question by which the addressee is asked to specify one of the alternatives
in her answer. This is the case in (70a), where the answer must be either ‘tea’ or
‘coffee’. Example (70b), by contrast, shows standard disjunction which happens
to occur in a question. This is not an alternative question, however, but a polar
question that requires ‘yes’ or ‘no’ as its answer.8The distinction between
interrogative disjunction and standard disjunction is made in many of the world’s
languages. Even English can be said to make the distinction in some way: the
emphatic disjunctive markers either . . or only express standard disjunction (Do
you want either tea or coffee? cannot be an alternative question). According to
Moravcsik (1971:28), this is a more general property of emphatic disjunctive
coordinators.
A semantic distinction that is well known (especially from the work of logi-
cians) is that between exclusive and inclusive disjunction. These notions are
defined in terms of truth values: an exclusive disjunction is true if only one but
not both of the disjoined propositions are true, while an inclusive disjunction
is true if either one or both disjoined propositions are true. Examples might be
(71a–b):
(71) a. exclusive disjunction
Marvin died on Tuesday or Wednesday
b. inclusive disjunction
Mike is a psychologist or a linguist
It is often said that languages may distinguish between these two semantic
types by using different disjunctive coordinators. Typically Latin aut ‘or (excl.)’
and vel ‘or (incl.)’ are cited as illustrating this distinction. However, this view
seems to be erroneous (Dik (1968:274–6)). First, the logical distinction between
exclusive and inclusive disjunction cannot be applied well to natural languages
because many sentences with disjunction have no truth value (e.g. questions and
commands). But, more importantly, the Latin distinction between aut and vel
is evidently of a different nature,9and no other good case of a language mak-
ing precisely the exclusive/inclusive distinction is known. Modern technical
and bureaucratic writing sometimes uses the artificial compound coordinator
8An interesting formal description of the semantic difference between (70a) and (70b) is given in
Dik (1997:206).
9According to K¨uhner and Stegmann (1914:108), the difference between the two is that with vel
the speaker does not decide between the two coordinands and leaves the choice between them
open. Similarly, Dik (1968:275) proposes that with vel, the choice between the two coordinands
‘is left to the interpreter, or is immaterial to the argument.’ However, it is not clear whether
one would want to say that in an aut disjunction, the speaker makes a choice between the two
coordinands. Perhaps the difference is not so much a semantic one as a stylistic one: K¨uhner and
Stegmann (1914:107) observe that vel is very rare in the classical language, and becomes much
more common in late Latin.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 27
and/or which can be said to express inclusive disjunction, but such coordina-
tors are not found in ordinary speech. Ordinary disjunction always presents an
alternative between A and B, and whether or not ‘A and B’ is compatible with
the situation as well depends on the context. In (71a), the pragmatic context vir-
tually excludes the inclusive reading, but if we change the verb (Marvin left on
Tuesday or Wednesday) the two coordinands are no longer necessarily mutually
exclusive. Marvin may of course have left on both days, and in this case the
proposition would not be false. (See McCawley (1993:315–17) for arguments
that no exclusive ‘or’ need be assumed, and that the exclusive sense arises from
the pragmatic context.)
Another type of disjunction can be called metalinguistic disjunction,
because here the alternative is merely between two names for the same thing.
For instance, earlier in this subsection I used the expression alternative (or dis-
junctive) question. In many languages, the ordinary ‘or’ word can be used in this
way. Italian has a special coordinator (ovvero) that is restricted to metalinguistic
disjunction, and while the ordinary ‘or’ word (o) can also be used in this way,
there is a stronger form of this (oppure) that cannot be used metalinguistically
(Scorretti (1988:254)):
(72)
a. l’ Irlanda o/ovvero/*oppure l’ isola verde
the Ireland or the island green
‘Ireland, or the green island’
b. Voglio comprare un dizionario o/oppure/*ovvero una grammatica
I.want buy a dictionary or a grammar
‘I want to buy a dictionary or a grammar’
Finally, a type of disjunction-like coordination that is widely attested is tem-
poral alternation. In this construction, several events are said to occur alternat-
ingly at different times. In the example in (73), special correlative coordinators
are used to express this relation (cf. also literary English now...now).
(73) a. Zaza (an Iranian language of Turkey; Selcan (1998:667))
Na ˙rozu gˆehewro, gˆepakao
these days now cloudy now clear
‘These days it is sometimes cloudy, sometimes the skies are clear’
b. Russian
Xolodnyj doˇzdik to usilivalsja, to oslabeval
cold rain now strengthened now weakened
‘The cold rain became now stronger, now weaker’

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
28 Martin Haspelmath
4.3 Semantic subtypes of adversative coordination
Adversative coordination is signalled by English but and its counterparts in other
languages. While it is fairly common for languages to have a ‘but’ coordinator,
other languages express the same idea exclusively by means of a concessive
subordinate clause. In English, too, concessive clauses with although are often
roughly equivalent to ‘but’ coordinations:
(74) a. It is raining, but we are going for a walk
b. Although it is raining, we are going for a walk
Here, but expresses the denial of an expectation: the fact that it is raining would
lead one to expect that we would stay inside, and but cancels this expectation.
English but can also express a contrast between a negative and a positive
expression, where the positive expression substitutes for the negative one. This
could be called substitutive adversative coordination. In some languages,
there is a special substitutive coordinator, e.g. German sondern (which contrasts
with ordinary adversative aber ‘but’), shown in (75b):
(75) a. I did not go to Mindanao, but (rather) to Cebu
b. Ich bin nicht nach Mindanao gereist,
sondern nach Cebu/*aber nach Cebu
Some languages have a special oppositive coordinator that is used when there
is a contrast between the two coordinands, but no conflicting expectations. In
many cases, English would translate such a coordinator as ‘and’. For instance,
Ponapean (an Austronesian language of Micronesia; Rehg (1981:331–2)) has
a contrast between ordinary conjunctive oh ‘and’ (76a) and oppositive ah ‘and,
but’ (76b).
(76) a. Soulik pahn mwenge oh e pahn meir
Soulik fut eat and he fut sleep
‘Soulik will eat and he will sleep’
b. I laid, ah e meir
I fish but he sleep
‘I fished, and/but he slept’
A similar contrast between a concessive and an oppositive type of ‘but’ is well
known from Polish (ale/a) and Russian (no/a).
5 Some special strategies of conjunction
As the most frequent type of coordination, conjunction exhibits the greatest
diversity of formal patterns and has also been studied the most thoroughly. In
this section, we look more closely at conjunction patterns in which the marker is

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 29
identical to the comitative marker (Section 5.1), as well as two other strategies
which deviate to some extent from the standard pattern (Section 5.2–5.3).
5.1 Comitative conjunction
In many of the world’s languages, the conjunctive coordinator for nps is identical
in shape with the marker for accompaniment, i.e. the comitative adposition or
case-marker. Here I will call such cases comitative conjunction, exemplified
in (77–8). The (a) example illustrates the comitative use of the marker, and the
(b) example shows the use in conjunction.
(77) Samoan (a Polynesian language; Mosel and Hovdhaugen (1992:148))
a. Ia, alu atu Sina ma le ili-tea
well go.sg dir Sina with art fan-white
‘Well, Sina went there with the white fan’
b. ‘Ua ¯o atu Sina ma Tigilau
perf go.pl dir Sina and Tigilau
‘Sina and Tigilau left’
(78) Retuar˜a (a Tucanoan language of Colombia; Strom (1992:64–5))
a. J˜u˜a-re turi-koʔo paki-ka
Juan-core travel-past father-com
‘Juan travelled with his father’
b. Anita-ka Gloria-re wiʔi-˜er˜a baa-yu
Anita-and Gloria-core wet-purp do-pres
‘Anita and Gloria are going to get wet’
The extension of a comitative marker to express a conjunctive relationship
is of course very natural: the meaning of (77a) is not very different from the
comitative ‘Sina left with Tigilau.’ In fact, one might be tempted to argue
that comitative conjunction does not constitute coordination at all: languages
with supposed comitative ‘conjunction’ might simply lack a formal means of
conjunction, and speakers might substitute the ordinary comitative construction
when asked to translate a coordinate phrase such as ‘Sina and Tigilau’. This
may indeed be true in some cases, but for the majority of languages with (what
I call here) comitative conjunction, there is evidence of various kinds that the
construction is really a type of conjunction, a construction separate from the
comitative construction.
One kind of evidence is semantic. While conjunction and accompaniment are
often similar and difficult to distinguish, there are also cases where they clearly
have different entailments. For instance, the sentence Joan and Marvin ate
entails for both Joan and Marvin that they ate, while in Joan ate with Marvin it
is possible (though perhaps unlikely) that Marvin did not eat. In many languages

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
30 Martin Haspelmath
with comitative conjunction, the meaning clearly shows that we are dealing with
a conjunction construction, not just a comitative construction that happens to be
the closest translation equivalent of English and. (With verbs of motion, as in
(77–8), this test does not work well, because it is hardly possible to accompany
a moving person without moving oneself.)
The morphosyntactic evidence for a special construction of comitative con-
junction is often less subtle. Most strikingly, comitative-conjoined nps often
trigger plural agreement on the verb, as in (77b) and in (79) from Russian:
(79) Maˇsa sKostej priˇs-l-i pozdno
Masha with/and Kostya come-past-pl late
‘Masha and Kostya came late’
Comitative-conjoined nps may also obey the coordinate structure constraint.
Thus, in Russian, a comitative conjunct with s‘with’ cannot be questioned
(80c), just like an ordinary conjunct with i‘and’ (80b), contrasting with non-
conjunctive comitative phrases (80a) (Yakov Testelec (p.c.)):
(80)
a. (comitative) Maˇsa priˇsla sKostej / Kto priˇsel sKostej?
‘Masha came with Kostya / Who came with Kostya?’
b. (i-conjunction) Maˇsa iKostja priˇsli / *Kto iKostja priˇsli?
‘Masha and Kostya came’ / (lit.) ‘Who and Kostya came?’
c. (s-conjunction) Maˇsa sKostej priˇsli / *Kto sKostej priˇsli?
(lit.) ‘Masha with Kostya came’ / Who with Kostya came?’
That comitative-coinjoined nps are truly coordinate can also be seen when a
modifier has scope over both conjuncts, as in (81) from Amele (a language of
Papua New Guinea; Roberts (1987:109)):
(81) ija na sigin sapol ca
iof knife axe and/with
‘my knife and axe’ (i.e. ‘my knife and my axe’)
If (81) still meant ‘knife with axe’, one would not expect the possessive modifier
ija na ‘my’ to have scope over both elements.
Another indication comes from word order. Thus, in Retuar˜a, comitative
phrases typically follow the verb as in (78a), so that the comitative-marked
np Anita-ka in initial position, adjacent to Gloria-re (78b), must be the first
conjunct of a coordinate construction. Similarly, Russian has a different word
order in (80a) and (80c). But interestingly, the agreement criterion and the word
order criterion need not coincide: languages may show plural agreement on the
verb even if the two comitative conjuncts are not adjacent.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 31
(82) Krongo (a Kadugli language of Sudan; Reh (1985:278))
nk-´aa b´ar´ak ´oor`a´os ´un´ı´
υυd`ak´ubl´e y´a-`ıtt ´oŋ
pl-be jackal inf.share meat down com-rabbit
‘The jackal and the rabbit share the meat’
(lit. ‘The jackal share(pl) the meat and (‘with’) the rabbit’)
Such patterns are synchronically unexpected, but they can be understood
diachronically as erstwhile comitative constructions in which only the agree-
ment pattern, but not the word order pattern, has been adapted to the new
conjunctive sense.
Comitative-marked conjunctions may also involve more than two phrases,
like other conjunctions, but unlike ordinary accompaniment constructions (cf.
the strangeness of Joan with Marvin with Esther):
(83) Krongo (Reh (1985:278))
m-´aa ´ad `
ukw`at`ımy´aar´e y´a-t `
υnk ´υl´υb´aŋy´a-s`ar´ı
f-be inf.take log with-knife with-basket
‘And she takes the log, the knife and the basket’
Given the comitative origin of the construction, we would not necessarily
expect that coordinator omission could occur in constructions with multiple
conjuncts. Thus, Loniu (an Austronesian language of New Guinea) does not
allow this in its comitative-derived conjunction pattern: AmaBmaCcan-
not be reduced to A,BmaC(‘A, B and C – ma ‘with, and’), although the
synonymous coordinator
ε
‘and’ normally shows the pattern A, B
ε
C(‘A, B
and C’) (Hamel (1994:102)). Here, ma still seems to behave in accordance
with its original comitative function. However, there are also languages which
do exhibit (optional) coordinator omission in multiple comitative conjunction,
e.g. Ndyuka (an English-based creole language of Surinam) (Huttar and Huttar
(1994:237)):
(84)
baana, bakuba, angooki, kumukomu anga ala den soutu sani de
plantain banana gherkin cucumber and all these sort thing there
‘plantains, bananas, gherkins, cucumbers, and all these kinds of things’
Clearly, such behaviour is only expected if the comitative marker has already
become a coordinator.
Equally strikingly, there are many languages in which the original comitative
marker occurs not just with one of the coordinands, but bisyndetically with
each of them. For instance, in Tauya the comitative suffix -sou follows both
conjuncts when it means ‘and’, so that we get the contrast in (85) (MacDonald
(1990:137)):

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
32 Martin Haspelmath
(85) a. Ya-ra Towe-sou yate-e-ʔa
i-top Towe-com go-1(sg)-ind
‘I went with Towe’
b. Ya-sou Towe-sou yate-ene-ʔa
I-and Towe-and go-1pl-ind
‘Towe and I went’
Similarly, alongside the pattern AdaB(cf. 12b), Hausa also allows the prepos-
itive bisyndetic pattern daAdaB.
That the ‘ex-comitative’ coordinator no longer behaves like a true comi-
tative is also clear when it takes a case-marked np in its scope, as in (86),
where -wan (otherwise a comitative case suffix) occurs outside the genitive case
suffix -pa.
(86) Huallaga Quechua (Weber (1989:350))
Kampu-pa alwasir-nin-pa-wan ka-n mas huk-pis
marshal-gen alguacil-3sg-gen-and be-3 more other-even
kustumri-n rura-na-n-paq.
custom-3 do-sub-3pl-purp
‘The marshal and his alguacil have another custom to do’
However, even when a comitative construction shows clear signs of marking
conjunction, it may retain clear traces of its comitative origin. Thus, Russian
comitative conjunction (79) is restricted to animate conjuncts, and the two
conjuncts are typically thought of as participating in the situation together (see
McNally (1993) and Dalrymple, Hayrapetian, and King (1998) for detailed
discussion of the meaning of this construction).
In all cases where we have some diachronic evidence, we see that comitative–
conjunctive polysemy of particles and affixes goes back to a diachronic exten-
sion of the original comitative marker, which acquires the additional sense of
coordinator and with it different syntactic properties. Theoretically, one could
imagine the reverse diachronic process, from coordinator to comitative, also
giving rise to the same synchronic polysemy, but this apparently never hap-
pens. The change from comitative to conjunctive coordinator is a commonly
found path of grammaticalization (Stassen (2000)), and, like other grammati-
calization processes, it is unidirectional (C. Lehmann (1995)). Stassen (2000),
who looked at np conjunction in a large sample of 260 languages worldwide,
finds that languages with comitative conjunction (‘with-languages’) are partic-
ularly found in sub-Saharan Africa, East Asia, Southeast Asia and the Pacific
Islands, as well as in northern North America and lowland South America.
By contrast, languages lacking the comitative strategy (‘and-languages’) are

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 33
concentrated in northern and western Eurasia (including all of Europe), India,
northern Africa, New Guinea, Australia and Meso-America.
5.2 Inclusory conjunction
A semantically peculiar type of conjunction is what I call here inclusory con-
junction. In the usual case, a conjunction of two set-denoting nps refers to the
union of the two sets. Schematically, we can say that ‘{A, B}and {C, D}’
yields the set {A, B, C, D}. However, there also exist conjunction constructions
in which the result of the conjunction is not the union, but the unification of the
sets. That is, if some members of the second conjunct set are already included
in the first conjunct set, they are not added to the resulting set. Schematically,
we can say that ‘{A, B, C}and {B}’ yields the set {A, B, C}. Some examples
of inclusory conjunction are given in (87):
(87)
a. Russian
my stoboj
we with you.sg ‘you and I’
b. Chamorro (an Austronesian language of Guam; Topping (1973))
ham yan si Pedro
we with art Pedro ‘I and Pedro’
c. Yapese (an Austronesian language of Micronesia; Jensen (1977:185))
gimeew Wag
you.pl Wag ‘you(sg) and Wag’
d. Tzotzil (a Mayan language of Mexico; Aissen (1989:524))
voʔoxuk xchiʔuk i jtzebe
you.pl with def my.daughter ‘you(sg) and my daughter’
e. Maori (a Polynesian language of New Zealand; Bauer (1993:374))
maaua ko te rata
we.two.excl spec the doctor ‘the doctor and I’
f. Tagalog (Philippines; P. Schachter and Otanes (1972:116))
sila ni Juan
they gen.art Juan ‘he/they and Juan’
g. Mparntwe Arrernte (central Australia; Wilkins (1989:409))
Margaret anwerne-ke
Margaret we.pl-dat ‘to Margaret and us’
As the examples show, the inclusory conjunct (i.e., the one that denotes the
total set) is generally a non-singular personal pronoun. The included conjunct

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
34 Martin Haspelmath
is often linked by means of a comitative marker (Russian, Chamorro, Tzotzil),
but the marker may also be of a different kind (Maori, Tagalog), or the two
conjuncts may simply be juxtaposed (Yapese, Mparntwe Arrernte). Inclusory
conjunction is found widely throughout the Austronesian language family, but
is also attested elsewhere in the world, e.g. in many dialects of northwestern
France (nous deux Jean ‘Jean and I; Tesni`ere (1951)). In most cases, the inclu-
sory pronoun precedes the included conjunct, but (87g) shows that it may also
follow it.
When the inclusory pronoun is plural, as in the Russian example, this con-
struction can be translated into English in two ways: my s toboj can be ‘you and
I’ (in this case the unification sets are {you, I}and {you}), or ‘we and you’ (in
this case the unification sets are {you, I, X,...}and {you}). When the inclu-
sory pronoun is dual, as in (87e) from Maori, there is only one translation into
English: ‘the doctor and I’. When the language has both dual and plural pro-
nouns, like Mparntwe Arrernte, again only one translation is possible. Example
(87g) can only mean ‘to Margaret and us’ (‘to Margaret and me’ would require
the dual pronoun).
Inclusory conjunction as in (87) is impossible when the non-inclusive con-
junct outranks the inclusive conjunct on the person hierarchy (1 <2<3), so
*you(
pl
) with me (‘you(sg) and I’) and *they with me (‘he and I’) are excluded
(Schwartz (1988b)). This follows straightforwardly from the fact that second
person pronouns cannot include the speaker, and third person pronouns cannot
include the speaker or hearer. For reasons that are not clear, there seems to be a
general preference for first and second person pronouns over third person pro-
nouns in inclusory conjunction. And so far I have found only a single language
in which the inclusory word is not a non-singular pronoun, but a non-singular
full noun: in Margi, a Chadic language of Nigeria, the construction in (88) is
attested (Hoffmann (1963:57)).
(88) K`ambə`r`aw ´azh´a-’y`ar `ag´a m`al`ag
ə´nd`a
Kamburawazha-ass.pl with wife of.him
‘Kamburawazha and his wife’
The inclusory noun in (88) is in the associative plural form (cf. B`
ashir-’y`
ar
‘Bashir and his family’ – see note 6). The construction in (88) differs in no
way from Margi’s more typical inclusory construction with an inclusory pro-
noun (e.g. n`
a’y `ag´a M´
ad`
u

(we with Madu) ‘Madu and I’; Hoffmann (1963:
238)).
In addition to the construction in (87–88), where the inclusory conjunct
and the included conjunct occur contiguously and form a phrasal inclusory
conjunction, many languages also have a construction in which the inclusory
pronominal element is a clitic pronoun or a coreference marker on the verb

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 35
(89a–d) or on the possessed noun (89e). This is called split inclusory con-
struction because the inclusory conjunct and the included conjunct do not
form a phrase (Lichtenberk (2000)).
(89) a. Nkore-Kiga (a Bantu language of Uganda; Taylor (1985:99))
tw-a-gyenda na Mugasho/
1pl-past-go with Mugasho
‘I went with Mugasho / Mugasho and I went’
b. Yapese (Jensen (1977:187))
Ku gu waarow Tamag
perf 1excl go.du Tamag
‘Tamag and I went’
c. Turkish (Kornfilt (1997:298))
Ahmet-le d¨un sinema-ya git-ti-k
Ahmet-com yesterday movies-dat go-past-1pl
‘Yesterday Ahmet and I went to the movies’
d. Hausa (Schwartz (1989:30))
Audu yaa gan m`u jiya da Binta
Audu 3sg.m.perf see us yesterday with Binta
‘Audu saw Binta and me yesterday’
e. Toqabaqita (an Oceanic language of Vanuatu; Lichtenberk
(2000:22))
nuu-maroqa tha Uluta
picture-2du.poss art Uluta
‘the picture of you and Uluta’
Some languages (apparently especially in Polynesia) use pronominal inclu-
sory conjunction also for conjoining two nps. The first conjunct precedes the
inclusory pronoun, which is then followed by the other included conjunct(s) in
the usual way.
(90) a. Samoan (Mosel and Hovdhaugen (1992:680))
Peni laua ma Ruta
Peni they.du with Ruta
‘Peni and Ruta’
b. Maori (Bauer (1993:128))
Tuu raatou ko Hine, ko Pau
Tuu they.pl spec Hine spec Pau
‘Tuu, Hine and Pau’
Inclusory conjunction is discussed from a theoretical point of view in
Schwartz (1988a, 1988b), Aissen (1989) and Lichtenberk (2000).

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
36 Martin Haspelmath
5.3 Summary conjunction
Summary conjunction is the term adopted here for a construction in which
conjunction is signalled not by an element that links the conjuncts together in
some way, but by a final numeral or quantifier that sums up the set of conjuncts
and thereby indicates that they belong together and that the list is complete.
Examples with numerals come from Mongolian (91a), Classical Tibetan (91b)
and Huallaga Quechua (91c) (Weber (1989:351) calls this construction list-
and-count conjunction).
(91) a. bagˇs, Gombo xojor
teacher Gombo two
‘the teacher and Gombo’ (Vietze (1988:41))
b. lus ŋag yid gsum
body speech mind three
‘body, speech and mind’ (Beyer (1992:241))
c. Pusha-ra-n Pedru-ta Jacobo-ta Hwan-ta kimsa-n-ta
lead-past-3 Peter-acc James-acc John-acc three-3-acc
‘He led off Peter, James and John’ (Weber (1989:351))
The final quantifier may also be the word ‘all’, as in the following example
from Cantonese (Matthews and Yip (1994:289)):
(92) Yanfa seui, leuhts¯ı fai, g¯ıngg ´ei y ´ung d¯ou yiu b ´ei ge
stamp duty lawyer fee agent commission all need pay prt
‘You have to pay stamp duty, legal fees and commission’
More intriguingly, summary conjunction of two conjuncts may also be
expressed by a dual affix on the second conjunct, which refers to the num-
ber of the whole construction (dual conjunction). Example (93a) comes from
Wardaman (a Yangmanic language of northern Australia; Merlan (1994:90))
and (93b) from Khanty (a Finno-Ugrian language of western Siberia; Nikolaeva
(1999:45)):
(93) a. yibiyan yingawuyu-wuya yawud-janga-n
man(abs) wife-du(abs) 3nonsg-come-pres
‘The man and his wife are coming’
b. a:´si jik-ŋə n
father son-du
‘father and son’
Khanty also allows the dual on both conjuncts (94a), and a similar kind of
conjunction construction is attested in Vedic Sanskrit (94b). This construction

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 37
may be called double-dual conjunction (see Corbett (2000:228–31) for some
discussion).10
(94) a. a:´se:-ŋə njik-ŋən
father-du son-du
‘father and son’
b. Mitr-¯a Varun
.-¯a
Mitra-du Varuna-du
‘Mitra and Varuna’
Mparntwe Arrernte (Australia) uses its numeral ‘two’ (therre) in this
construction: Sandy therre Wendy therre ‘Sandy and Wendy’ (Wilkins
(1989:371)). Dual conjunction seems to be restricted to natural conjunction
wherever it occurs.
6 Ellipsis in coordination
In many languages there are some ellipsis phenomena that are specific to coor-
dination constructions. This can be illustrated by the contrast between (95) and
(96). The two ellipsis processes in (95) are possible both in coordinate (i) and in
subordinate (ii) constructions, while the two ellipsis processes in (96) are pos-
sible only in coordinate (i) constructions, and ungrammatical in subordinate
(ii) constructions. In the examples here and below, an ellipsis site is indicated
by ‘[ ]’.
(95) a. VP ellipsis (<write a novel>)
(i) Joan wrote a novel, and Marvin did [ ], too
(ii) Joan wrote a novel after Marvin did [ ]
b. N ellipsis (<poems>)
(i) Zhangsan admires Lisi’s poems, but Lisi despises Zhangsan’s
[]
(ii) Zhangsan admires Lisi’s poems, though Lisi despises
Zhangsan’s [ ]
(96) a. V ellipsis (<cooked>)
(i) Robert cooked the first course, and Marie [ ] the dessert
(ii) *Robert cooked the first course, while Marie [ ] the dessert
10 A related construction is the representative dual (or associative dual,orelliptic dual), where
just one of the conjuncts is used and the other one is inferred, e.g. Khanty a:´
se:-
ŋə
n(father-du)
‘father and son’, Vedic Mitr-¯
a(Mitra-du) ‘Mitra and Varuna’, Classical Arabic al-qamar-aani
(the-moon-du) ‘sun and moon’, Mparntwe Arrernte Romeo therre (Romeo two) ‘Romeo and
Juliet’. It is unclear what the exact relation between the representative dual and double-dual
conjunction is: perhaps the former has been expanded into the latter (Delbr¨uck (1893:138)),
or perhaps the former is a reduction of the latter. The representative dual is reminiscent of
representative plural (or associative plural) markers as exemplified in (66) above.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
38 Martin Haspelmath
b. NP ellipsis (<Hollywood movies>)
(i) Martin adores [ ], but Tom hates, Hollywood movies
(ii) *Martin adores [ ], because Tom hates, Hollywood movies
The functional motivation of the ellipsis is the same in both cases: identical
material need not be repeated, for reasons of economy. A much more difficult
question is why certain ellipsis processes are restricted to coordination, and an
answer to this question is far beyond the scope of this overview article. Here
I will limit myself to describing and illustrating the types of ellipsis that are
particularly associated with coordination.
There is no agreement among linguists concerning the extent to which ellip-
sis should be assumed in coordinate constructions. An extreme view (which
perhaps no contemporary linguist holds) is that all phrasal (i.e. non-sentential)
coordination involves ellipsis, and that the corresponding non-elliptical (‘under-
lying’) structures all involve sentential coordination. Thus, (97a,b) are said
to be derived from (97ab) by an ellipsis process (often called coordination
reduction) that eliminates identical elements and turns the underlying biclausal
structure into the monoclausal surface structure.
(97) a. Joan and Marvin got a pay raise
a.Joan got a pay raise and Marvin got a pay raise
b. I’ll ring you today or tomorrow
b.I’ll ring you today or I’ll ring you tomorrow
The main motivation for this derivation is the desire to see coordination as
uniformly sentential at the underlying level, following a long tradition of philo-
sophical logic in which only conjunction and disjunction of propositions is
assumed, but no conjunction or disjunction of terms.
However, not all cases of phrasal coordination can be derived from sentential
coordination in this way. Consider the examples in (98), in which the predicate
denotes a joint action or a situation that interrelates the two coordinands in some
way:
(98)
a. Joan and Max met (*Joan met and Max met)
b. Bob and Marie are similar (*Bob is similar and Marie is similar)
c. Mix the soy sauce and the vinegar (*Mix the soy sauce and mix the vinegar)
These sentences cannot be derived from the corresponding clausal structures
because these are syntactically incomplete. Another class of sentences with
coordinate structures does have well-formed biclausal counterparts, but these
do not have the same meaning:

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 39
(99)
a. Poland’s national flag is white and red
(= Poland’s national flag is white and Poland’s national flag is red)
b. Many people believe in God and do not go to church
(= Many people believe in God and many people do not go to church)
c. Did you play football or go for a walk? (yes / no or alternative question)
(= Did you play football or did you go for a walk? (only alternative
question))
Again, semantic considerations rule out a straightforward derivation from
biclausal underlying structures. Sentences like (98) and (99) seem to require that
languages (perhaps in contrast to logic) also have phrasal coordination, not only
sentential coordination. Now, if this is the case, then the motivation for assum-
ing coordination reduction in sentences like (97) disappears.11 While coordi-
nation reduction was widely assumed by transformationalists a few decades
ago, most linguists today would describe the sentences in (97–99) as phrasal
coordination.
But ellipsis rules cannot easily be eliminated entirely from the domain
of coordination, because some coordinate structures involve coordinands that
are not constituents (non-constituent coordination). For instance, in (96ai)
(Robert cooked the first course, and Marie the dessert), the second conjunct
Marie the dessert cannot be described as an ordinary constituent, and it differs
from the first conjunct (Robert cooked the first course) in that it lacks a verb.
This situation is most conveniently described by a rule of ellipsis (or, in other
words, deletion).
Ellipsis in coordination can be either forward ellipsis (or analipsis) (i.e. the
ellipsis site is in the second coordinand), or backward ellipsis (or catalipsis)
(i.e. the ellipsis site is in the first coordinand). The two types are exemplified
in (100–101). Again, an ellipsis site is indicated by ‘[ ]’, and the identical
material in the other coordinand (the antecedent of the ellipsis) is enclosed in
brackets.
(100) Analipsis (=forward ellipsis)
a. Hanif [loves] Khadija and Khadija [ ] Hanif
b. Mr Sing [wrote] his father a letter and [ ] his grandmother a
postcard
c. Bergamo [is beautiful], and Lucca [ ], too
d. Bill’s [story] about Sue and Kathy’s [ ] about Max
11 Another argument against coordination reduction is that many languages have different coor-
dinators in sentential and phrasal coordination (cf. Section 3). In these languages, one would
have to assume a rule that changes the form of the coordinator in addition to reducing the
coordinands.

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
40 Martin Haspelmath
(101) Catalipsis (=backward ellipsis)
a. Birds eat [ ], and flies avoid, [long-legged spiders]
b. I think that Joan [ ], and you think that Marvin, [will finish first]
c. Joan sells [ ], and Fred knows a man who repairs, [washing
machines]
Analipsis and catalipsis have not been studied in great detail for many languages
(but see Sanders (1977), Harries-Delisle (1978), Mallinson and Blake (1981)
for cross-linguistic surveys). In English and similar European languages, the
most common type of analipsis consists in the ellipsis of the verb, as in (100a).
Since it generally leaves a gap between the remaining preverbal and postverbal
constituents (and Khadija [ ] Hanif), it is called gapping (J. R. Ross (1979);
Neijt (1979). By contrast, the most common type of catalipsis consists in the
ellipsis of elements at the right periphery of the first coordinand (e.g. the direct
object in (101a)), and is called right periphery ellipsis (H¨ohle (1991)), or (in
obsolete transformational terms) right node raising (Postal (1974)).
Gapping is illustrated in (100) by cases in which a single verb is ellipted, but
in fact more elements can be omitted together with the verb, such as adverbs
(102a), objects and subjects (102b,c), and additional higher verbs (102d).
(102) a. Simon [quietly dropped] the gold and Jack [ ] the diamonds
b. Fred [sent the president] a nasty letter, and Bernice[]abomb
c. In China [they drive] on the right and in Japan[]ontheleft
d. John’s father [managed to get him to read] the Bible and his
mother [ ] the Communist manifesto
In English, gapping requires that exactly two remnant constituents are left after
ellipsis, so that (103a) is impossible. In German, however, there is no such
restriction (cf. 103b).
(103) a. *Mr. Singh [sent] his father a postcard and Ms. Bannerjee[]her
grandmother a fax
b. Herr Singh [schickte] seinem Vater eine Postkarte,
und Frau Bannerjee [ ] ihrer Großmutter ein Fax
There also exist gapping-like types of analipsis in which the two remnant con-
stituents are both postverbal (100b), or in which the ellipted element is a noun
(100d).
Gapping and right periphery ellipsis differ not only in that the former affects a
medial constituent, and the latter a final element (Hudson (1976)). Some further
differences are: first, in gapping only major phrasal categories (such as np,pp,
AdvP) are left as remnants, but in right periphery ellipsis (RPE) other elements
may stay behind as remnants. Thus, in (104a), the remnant the white is not a
major phrasal category, but it is allowed in (104b).

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
Coordination 41
(104)
a (gapping) ?The black [horse] won and the white [ ] lost
b. (RPE) Dirk chose the white [ ] and Bernd wanted the red [Volvo]
Second, in gapping the ellipted element need not be strictly identical inflec-
tionally. The non-elliptical version of (105a) would have the verb form like
rather than likes in the second conjunct. Such agreement differences cannot be
ignored in right periphery ellipsis (cf. 105b). In the non-elliptical version, the
first conjunct would have the object np herself, so, because the two elements
are not strictly identical, (105b) is ungrammatical.
(105) a. (gapping) Julia [likes] Mendelssohn, and her parents[]the
Rolling Stones
b. (RPE) *Joan greatly admires [ ], and Marvin constantly
criticizes, [himself]
Third, gapping primarily affects coordinate clauses, but right periphery ellipsis
is quite productive at the noun phrase and pp level as well:
(106) right periphery ellipsis
a. both in front of the blue [ ] and behind the white [house]
b. I read Dik’s book [ ] and Ross’s article [about coordination]
Finally, in English, many cases of right periphery ellipsis exhibit an intonation
break (represented by a comma in writing) in front of the antecedent in the
second coordinand (e.g. 96bi, 101, 105b). Thus, gapping and right periphery
ellipsis are specific rules with their unique characteristics and cannot be reduced
to medial analipsis and final catalipsis, respectively. Interestingly, gapping and
right periphery ellipsis can occur together in the same coordination. In (107),
antecedents and ellipsis sites are matched by subscripts.
(107) Joan [visited]iher youngest [ ]jand Marvin [ ]ihis oldest [brother]j
Equivalents of both gapping and right periphery ellipsis are attested in many
European languages (see Wesche (1995), Wilder (1997) for extensivediscussion
of German compared to English). A few examples are:
(108) a. French (gapping; Grevisse (1986:§260))
Philippe [revient] des champs, et son fils [ ]
Philippe returns from.the fields and his son
du chemin de fer
from.the way of iron
‘Philippe comes back from the fields, and his son from the
railway’

P1: RPU/... P2: RPU
9780521581578c01b.xml CUUK171B-Shopen V. 2 December 24, 2006 14:12
42 Martin Haspelmath
b. Latvian (right periphery ellipsis; Mallinson and Blake (1981:223))
Puika redz¯eja[], un meitene dzird¯eja [suni]
boy saw and girl heard dog
‘The boy saw and the girl heard the dog’
c. Welsh (right periphery ellipsis; Mallinson and Blake (1981:256))
Gwelodd Gwen [ ], arhybuddiodd Ifor, [y dyn].
saw Gwen and warned Ifor the man
‘Gwen saw, and Ifor warned, the man’
However, by no means all languages with SVO basic order admit gapping
of the verb. Gapping is impossible in Thai and Mandarin Chinese (Mallinson
and Blake (1981:218)), and even in the southern European language Maltese
(closely related to Arabic), the same verb occurring with a different subject and
object is normally repeated.
(109) Maltese (Borg and Azzopardi-Alexander (1997:82))
Jien adt kaf`e uhu a luminata
I took.1sg coffee and he took.3sg.m lemonade
‘I had coffee, and he (had) lemonade’
In languages with verb-final word order, catalipsis usually affects the verb, and
we get examples like those in (110).
(110) a. Basque (McCawley (1998:286))
Linda-k ardau [ ] eta Ander-ek esnea [edaten dabez]
Linda-erg wine(abs) and Ander-erg milk(abs) drink they.it
‘Linda will drink wine and Ander milk’
b. Lezgian (northeastern Caucasus; Haspelmath (1993:339))
ˇ
Caqal-di sa werˇc[], ˇzanawur-di sa lapag [ ˇga-na]
jackal-erg one chicken wolf-erg one sheep bring-past
‘The jackal brought a chicken, and the wolf a sheep’
c. Marathi (Indo-Aryan; Pandharipande (1997:176))
Sudha Mumba¯ı-l¯a[] ¯an
.im¯ı Triwendram-l¯a [dzaın]
Sudha Mumbai-all and I Trivendram-all went
‘Sudha went to Mumbai, and I to Trivendram’
If we want to apply the terminology that has become usual for English to these
languages, we could either say that (110) shows right periphery ellipsis of the
verb, or that it shows backward gapping. Thus, it is not clear how the terms
right periphery ellipsis and gapping should be applied to languages with a basic
word order other than svo. Here I will use the more neutral terms analipsis and
catalipsis instead.

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
Coordination 43
Besides the final catalipsis pattern of (110) (so[]+so[v]), some sov lan-
guages such as Basque (111) and German (112) also allow final analipsis (so[v]
+so[ ]) (cf. also example (23) above from Turkish):
(111) Basque (McCawley (1998:286); cf. 110a)
Linda-k ardau [edaten du], eta Ander-ek esnea [ ]
Linda-erg wine(abs) drink he.will and Ander-erg milk(abs)
‘Linda will drink wine and Ander milk’
(112) German
a. . . . dass Georg Wein [ ] und Barbara Bier [trinkt]
b. . . . dass Georg Wein [trinkt] und Barbara Bier [ ]
‘...that Georg drinks wine and Barbara beer’
But verb-final sov languages often also allow medial analipsis (s[o]v+
s[]v):
(113) a. Turkish (Kornfilt (1997:120))
Hasan [istakoz-u] pis
ir-di, Ali de [ ] ye-di
Hasan lobster-acc cook-past(3sg) Ali and eat-past(3sg)
‘Hasan cooked the lobster, and Ali ate it’
b. Korean (Mallinson and Blake (1981:224))
Sonyen-i [swuley-lul] kul-ko sonye-ka [ ] mile-ss-ta
boy-nom cart-acc pull-and girl-nom push-past-decl
‘The boy pulled, and the girl pushed the cart’
So far we have considered only medial and final ellipsis. Initial ellipsis cannot
be illustrated well from svo languages like English, because a sentence like Joan
arrived and began immediately would not be analysed as involving ellipsis
([Joan] arrived and[]beganimmediately), but rather as showing simple vp
coordination (Joan [[arrived] and [began immediately]]VP). But German has
ovs patterns which allow initial analipsis ([o]vs +[]vs) (Zifonun, Hoffmann,
and Strecker (1997:574))
(114) [Das Buch] kaufte mein Vater und [ ] las meine Mutter
the book bought my father and read my mother
‘The book was bought by my father and read by my mother’
In a verb-final language with relatively free order of subject and object, we may
get the pattern [o]sv +[]sv, as in Malayalam (a Dravidian language; Asher
and Kumari (1997:151)):
(115) [Pustakam] Raamu vaaŋŋ ipaks
.e[] Kr
.s
.n
.an vaayiccu
book Ramu bought but Krishnan read
‘Ramu bought but Krishnan read the book’

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
44 Martin Haspelmath
Table 1.2 The coordination ellipsis site in relation to clausal word order
patterns
svo sov vso osv/ovs
analipsis
(=forward
ellipsis)
medial: s[v]o+s[]o
(=gapping,
100a)
s[o]v+s[]v
(113,b)
final: so[v]+so[]
(111, 112b)
initial: [v]so +[]so
(116)
[o]vs +
[]vs /[o]sv +
[]sv (114,
115)
catalipsis
(=
backward
ellipsis)
final: sv[]+sv[o]
(=right
periphery ellipsis,
101a)
so[]+so[v]
(110a–c)
vs[]+vs[o]
(108c)
In German vso sentences, initial analipsis is possible, too ([v]so +[]so):
(116) [Liebt] Julia Romeo und [ ] Kleopatra C ¨asar
loves Juliet Romeo and Cleopatra Caesar
‘Does Juliet love Romeo, and Cleopatra Caesar?’
The patterns that are more widely attested are summarized in table 1.2.
So far we have only looked at ellipsis patterns as they concern the major clause
constituents subject, verb and object. Sanders (1977) presents an ambitious
typology of ellipsis constructions, and he argues that what counts is not the
grammatical function of the constituent in the ellipsis site, but only its position.
Starting out from an abstract pattern ‘ABC & DEF’, there are thus six logically
possible types of ellipsis (Sanders (1977:255)):
(117) [ ]BC & DEF A-ellipsis initial catalipsis
A[ ]C & DEF B-ellipsis medial catalipsis
AB[ ] & DEF C-ellipsis final catalipsis
ABC & [ ]EF D-ellipsis initial analipsis
ABC & D[ ]F E-ellipsis medial analipsis
ABC & DE[ ] F-ellipsis final analipsis
Table 1.2 already suggests that analipsis is generally more common than catal-
ipsis, and that, of the three catalipsis types, final catalipsis is the most com-
mon one. Now Sanders examines the available evidence for a wide variety
of languages and asks which ellipsis types are possible in each language.

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
Coordination 45
For instance, English allows C-ellipsis (=final catalipsis, or right periphery
ellipsis), D-ellipsis (e.g. [Yesterday] Joan left and [ ] Marvin arrived), and
E-ellipsis (=gapping, or medial analipsis), but not the other three types.
It turns out that no ellipsis type is universally impossible, but there are strong
restrictions on which combinations of ellipsis types a language can have. Out of
sixty-four logically possible combinations, only six are in fact attested, accord-
ing to Sanders (1977:255–6). In (118), the permitted ellipsis sites are underlined.
(118) Chinese A B C D EF
English, Japanese A B C D E F
Quechua A B C D E F
Russian A B C D E F
Hindi, Zapotec A B C D E F
Tojalabal A B C D E F
This pattern is clearly not random and can be reformulated in the implicational
hierarchy in (119):
(119) Accessibility hierarchy for ellipsis types
A>B>C
F>E>D
This hierarchy should be read as follows: if a language allows any ellipsis
type (i.e. if a position is accessible to ellipsis), then all types to the right on
the hierarchy are also possible. Sanders argues that this state of affairs has
a straightforward functional explanation: the less accessible ellipsis types are
more difficult to decode. Decoding difficulty of an ellipsis construction depends
on two factors. First, a purely temporal factor: catalipsis is more difficult than
analipsis because the antecedent of the ellipsis has not been processed at the
time when the ellipsis site is encountered. Second, Sanders argues that decod-
ing difficulty depends on the ‘memory prominence’ of the antecedent. Memory
prominence is known to be determined by the ‘serial position effect’: begin-
nings and ends are learned faster and recollected more accurately than middles.
Thus, A and F should be the best antecedents, and C and D should be the worst
antecedents of ellipsis. This means that D and C should be the most favoured
ellipsis sites, and A and F should be the least favoured ellipsis sites. The com-
bination of the temporal factor and the prominence factor yields exactly the
pattern in (118–119).
7 Delimiting coordination
In this final section, I will discuss ways in which coordinate constructions
can be delimited against related constructions, in particular, dependency/

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
46 Martin Haspelmath
subordination constructions and less grammaticalized constructions. Finally,
I ask whether coordination constructions are universal.
7.1 Coordination versus dependency/subordination
The formal symmetry of the terms coordination and subordination does not
correspond to a similar conceptual symmetry. First of all, while coordination
is applied to the combination of both phrases and clauses, subordination is
generally restricted to clauses. For instance, in the sentence If you see Pat, tell
me immediately, we would say that the clause if you see Pat is subordinate
(to the main clause), but not that the direct objects Pat and me or the adverb
immediately are subordinate (to the verb). Instead, the term dependency is used
as a general term for both phrases and clauses.12
As I noted in Section 0, an important difference between coordination and
dependency is that two coordinate elements A–B are symmetrical, whereas
two elements X–Y in a dependency relation are asymmetrical, with X being
the head and Y being the dependent (or vice versa). This is often thought of
as a difference in the syntactic/structural relations of the elements: in head–
dependent relations, we find asymmetrical formal phenomena such as person–
number agreement of the head with the dependent (e.g. verbs agreeing with
their arguments), or case–number agreement of the dependent with the head
(e.g. adjectives agreeing with the nouns they modify), or government of the
dependent properties by the head (e.g. verbs governing the case of their argu-
ments). Such asymmetries are often absent from coordinate structures, and the
coordinands are often structurally more on a par, thus mirroring their identical
semantic roles.
But coordinate constructions may also show a fair amount of structural asym-
metry, especially when they have their origin in comitative structures (which
are, of course, dependency structures – see Section 5.1). Structural asymme-
tries are attested in non-comitative coordination as well, e.g. Norwegian han
og meg (he.nom and I.acc) ‘he and I’ (Johannessen 1998:1). And, conversely,
head–dependent relations are not always reflected in formal asymmetries, e.g.
in languages that lack agreement and case-marking. Thus, it seems best to define
both coordination and dependency in semantic terms,13 and to take as criterial
the identity vs non-identity of the semantic roles that the connected elements
play. Formal tests for subordination vs dependency, such as the coordinate struc-
ture constraint (Section 0), will largely yield the same results as the semantic
12 Thus, a subordinate clause is more or less the same as a dependent clause (though Haspelmath
(1995:26) makes a subtle distinction between them), and subordination is now more or less
equivalent to clausal dependency.
13 See also Croft (1996, 2001) for a semantic definition of heads and dependents.

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
Coordination 47
criterion, but, as we saw in Section 5.1, mismatches between the semantic cri-
teria and the formal criteria, as well as among different formal criteria, are not
uncommon.
When the coordination or dependency status of a sequence of two clauses is
in question, i.e. when we are unsure whether we are dealing with coordination
or subordination, the semantic criterion is often difficult to apply (see Cristofaro
(2003) for some discussion). For instance, with converb constructions of the
type illustrated above in Section 3, it is often unclear whether we should describe
them as subordinate or coordinate. Example (57b) from Korean is repeated here
for convenience:
(57) b. Achim mek-ko hakkyo ka-ss-eyyo
breakfast eat-and school go-past-ind
‘I ate breakfast and went to school /
After eating breakfast, I went to school’
Both of the English translations given seem appropriate here, so one wonders
whether there are formal criteria that would be of help in deciding the issue.
In Haspelmath (1995), I noted that, across languages, subordination struc-
tures generally have the following properties:
(i) only subordinate clauses can be in internal position (i.e. with the subordi-
nate clause inside the main clause): At eight o’clock, after eating breakfast,
I went to school.
(ii) only subordination constructions allow extraction of wh-pronouns
(because of the coordinate structure constraint, Section 0): Where did you
go after eating breakfast?
(iii) only subordinate clauses can be focussed: It was after eating breakfast that
I went to school.
(iv) only subordinate clauses allow backwards anaphora: After meeting heri
again, I admired Joanieven more.
But again, as in the case of comitative conjunction, mismatches occur. When
we try to apply the criteria to the case of the Korean -ko converb used in (57b),
the evidence is mixed (see Rudnitskaya (1998) for detailed discussion). When
the verb shows tense (e.g. the past-tense suffix -ass), the converb clause cannot
be in internal position, but must precede the finite clause (120a). When the verb
lacks tense, it can be inside the finite clause (120b).
(120) a. Swunmi-nun cakiaphathu-lul phal(-ass)-ko
Sunmi-top self’s. apartment-acc sell(-past)-conv
cohun cip-ul sa-ss-ta
good house-acc buy-past-decl
‘Sunmi sold her apartment and bought a good house’

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
48 Martin Haspelmath
b. Swunmi-nun cohun cip-ul [caki aphathu-lul phal-ko]
Sunmi-top good house-acc self’s apartment-acc sell-conv
sa-ss-ta.
buy-past-decl
‘Sunmi bought a good house, having sold her apartment’
As Rudnitskaya (1998) shows, there are a number of diverse factors that deter-
mine whether subordination tests are positive or negative. Thus, it is often not
straightforward whether a verbal (converb) marker signals subordination or
coordination.
Similarly, Culicover and Jackendoff (1997) show that there is a class
of English clause-combining constructions that show mixed subordinate–
coordinate behaviour, as illustrated in (121).
(121) You drink another can of beer and I’m leaving
(=If you drink another can of beer, I’m leaving)
This construction is semantically subordinate, but the syntactic evidence is
mixed. Most strikingly, the linker and does not look like a clause-final subordi-
nator, but much more like a medial coordinator. But Culicover and Jackendoff
(1997:206) show that this construction does not obey the coordinate structure
constraint, and behaves as subordinate also with respect to backwards anaphora.
Thus, structural tests show no more than a tendency to correlate with semantic
criteria, not a strict one-to-one correspondence. But the investigation of the
attested types of mismatches and constraints on mismatches is a rich area for
future discoveries.
7.2 Degrees of grammaticalization
The patterns and coordinators discussed in this chapter are primarily those
that show the highest degree of structural integration or grammaticalization.I
have not said much about further semantic types of coordination such as causal
coordination, consecutive coordination (e.g. French Je pense donc je suis ‘I
think, therefore I am’) or explicative coordination (e.g. The film is open only
to adults, i.e. people over 18). These coordination types are marginal, and the
linkers used in them are not always clear cases of coordinators. In conjunction
and adversative coordination, too, there are some linkers (e.g. then,moreover,
yet,however) that are not generally recognized as coordinators, but are typically
treated as linking adverbs (or conjunctional adverbs). The criteria for treating
them as adverbs rather than coordinators are typically formal, not semantic. For
instance, it is commonly said that coordinators are always in initial position and
that they do not co-occur with other coordinators. The first criterion excludes
however, the second excludes yet and then (cf. She was unhappy about it,

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
Coordination 49
and yet / and then she did as she was told), and both criteria qualify for as a
coordinator. In German, with its verb-second word order, a formal criterion is
that adverbs, but not coordinators, occupy the preverbal slot and force the subject
into postverbal position (e.g. contrast und Lisa kam / *und kam Lisa ‘and Lisa
came’ with dann kam Lisa / *dann Lisa kam ‘then Lisa came’). But, as is so often
the case, these various formal criteria do not always yield consistent results.
For instance, then and yet behave like coordinators in that they can link not just
sentences, but also vps (e.g. The car turned suddenly, then screeched to a halt),
but they are unlike coordinators in that they can co-occur with and. And German
doch ‘however’ allows either word order pattern (doch Lisa kam sp¨
at / doch
kam Lisa sp¨
at ‘however, Lisa came late’). Thus, the category of coordinators
does not have sharp boundaries, and, in a cross-linguistic perspective, it seems
best to focus on the most grammaticalized members of the category.
7.3 Is coordination universal?
The degree of grammaticalization is also relevant for another important ques-
tion: whether coordination is a universal that is found in all languages, or
whether some languages lack coordinate patterns. Gil (1991) argues that Mari-
copa (a Yuman language of Arizona) has no coordinate structures, though he
defines coordination formally, starting out from English-like patterns. At the
same time, Gil notes that Maricopa speakers have a variety of ways of express-
ing ‘A and B’, e.g. simple juxtaposition (cf. (19b) above), or a form of the verb
u
ð
aav ‘accompany’ (so that ‘John and Bill will come’ is literally ‘John, accom-
panying Bill, will come’). If Gil’s analysis is right, Maricopa is a language that
has no specific grammatical constructions dedicated to expressing coordination,
although it can express the same concepts by using its lexical resources, or by
leaving them implicit. That may of course be the case, and it would constitute
an important finding.
However, another possibility is that Maricopa coordinate structures simply
exhibit a fairly low degree of grammaticalization, so that they are easily mis-
taken as completely non-grammaticalized. We saw in Section 5.1 that many lan-
guages with comitative conjunction at first blush appear to show no dedicated
conjunction pattern, simply replacing ‘A and B’ by ‘A with B’. But when these
comitative-conjoined patterns are examined more closely, it is often found that
they are grammatically and semantically distinct from their comitative source
constructions, even though they still show the same overt marker. It may well
be that the Maricopa patterns are also on their way toward grammaticalization,
and that a closer look would reveal evidence for this. But, whatever the right
description of Maricopa turns out to be, it is clear that there is a universal ten-
dency for languages to grammaticalize coordination markers from a variety of
sources, and eventually the formal features of these coordination patterns seem

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
50 Martin Haspelmath
to converge. Different languages, or different constructions, exhibit different
degrees of grammaticalization and of similarity with the source pattern, but the
cross-linguistic similarities are quite striking as well.
8 Appendix: terminological issues
As in other domains of grammar, the terminology for coordination and related
phenomena is often disparate and sometimes confusing. The following remarks
point out synonyms and homonyms of the terms chosen in the main body of
the chapter.
(1) Coordinator: This is a non-traditional term for what has more often been
called coordinating conjunction. The term conjunction in this traditional
sense comprises both coordinators and markers of subordination (subordi-
nators). I have avoided this term in this chapter because I want to reserve
conjunction to denote a special type of coordination (‘and’-coordination,
or conjunctive coordination).
(2) Conjunction and disjunction: An older term for conjunctive coordina-
tion (=conjunction) that is now rarely used is copulative coordination.
(However, the term copulative compound (=coordinative compound)is
still fairly common.) Besides disjunctive coordination (=disjunction), one
also finds alternative coordination. Since conjunction is by far the most fre-
quent type of coordination, the term conjunction is sometimes (erroneously
or carelessly) used as a synonym of coordination.
(3) Coordinand: This term is introduced in the present chapter for the units that
are combined in a coordinate construction (cf. Dixon (1988:161), where
I have found this term used in the same sense). There is no traditional
term for this concept. Dik (1968) uses the term term (of a coordination).
Sometimes the term conjunct is used as a synonym of coordinand (just as
conjunction is sometimes used as a synonym of coordination), but this is
confusing and should be avoided.
(4) Emphatic coordination: This term is used in the present chapter for coor-
dinations such as both A and B,oreither X or Y. There is no traditional term
for such constructions. J. R. Payne (1985) uses the feature ‘[±separate]’,
so my emphatic coordination corresponds to Payne’s ‘[+separate] coordi-
nation’. (Payne uses the feature ‘[±emphatic]” to distinguish between A,
B and C and A and B and C, so my use of this term is very different from
Payne’s.)
(5) Inclusory conjunction: This term is inspired by Lichtenberk (2000), who
introduces the term inclusory pronominal construction. There is no tra-
ditional term for this concept. For phrasal inclusory conjunction as in
(87), Schwartz (1988a, 1988b) and Aissen (1989:523) use the term plu-
ral pronoun construction, and in Australianist circles the term inclusive

P1: RPU/... P2: RPU
9780521581578c01c.xml CUUK171B-Shopen V. 2 December 24, 2006 14:13
Coordination 51
construction seems to have been used (Wilkins (1989:407); but see now
Singer (2001)). For the split construction as in (89), Schwartz (1988a,
1988b) uses the term verb-coded coordination. I prefer Lichtenberk’s term
inclusory, because it allows one to capture the similarities between the two
constructions. It is better than inclusive, because the neologism inclusory
makes it very clear that a special kind of construction is referred to.
(6) Analipsis (=forward ellipsis) and catalipsis (=backward ellipsis): These
are inspired by Zifonun et al.’s (1997 1:571) Analepse/Katalepse, which
have antecedents in the late nineteenth century. The prefixes ana- and cata-
are used in the same sense here as in anaphoric and cataphoric.
9 Suggestions for further reading
General overview articles on coordination are J. R. Payne (1985) (with emphasis
on cross-linguistic diversity), van Oirsouw (1993), and Grover (1994) (with
emphasis on ellipsis phenomena in English).
The best general book-length study of coordination is still Dik (1968),
although much of the discussion of early transformational grammar is primarily
of historical interest now.
Much of the literature on coordination from a formal syntactic point of view
has been concerned with ellipsis in coordination in English and similar European
languages, for instance the book-length studies by van Oirsouw (1987), Neijt
(1979) and Wesche (1995). A formal semantic approach is adopted in Lang
(1984).
The typological literature on coordination is rather scarce. For ellipsis, the two
most important references are Harries-Delisle (1978) and, especially, Sanders
(1977). For coordination in general, three important references are Moravcsik
(1971), Mithun (1988), and Stassen (2000). A collection of papers describing
coordinating constructions in various languages is Haspelmath (2004).