66 Nigel Fabb

3 Compounding
NIGEL FABB

1 Overview

A compound is a word which consists of two or more words. For example, the
Malay compound mata-hari ‘sun’ is a word which consists of two words: mata
‘eye’ and hari ‘day’. Compounds are subject to phonological and morpholo-
gical processes, which may be specific to compounds or may be shared with
other structures, whether derived words or phrases; we explore some of these,
and their implications, in this chapter. The words in a compound retain a
meaning similar to their meaning as isolated words, but with certain restric-
tions; for example, a noun in a compound will have a generic rather than a
referential function: as Downing (1977) puts it, not every man who takes out
the garbage is a garbage man.

1.1 Structure and interpretation

The meaning of a compound is usually to some extent compositional, though
it is often not predictable. For example, popcorn is a kind of corn which pops;
once you know the meaning, it is possible to see how the parts contribute to
the whole – but if you do not know the meaning of the whole, you are not
certain to guess it by looking at the meaning of the parts. This lack of predict-
ability arises mainly from two characteristics of compounds: (a) compounds
are subject to processes of semantic drift, which can include metonymy, so that
a redhead is a person who has red hair; (b) there are many possible semantic
relations between the parts in a compound, as between the parts in a sentence,
but unlike a sentence, in a compound, case, prepositions and structural posi-
tion are not available to clarify the semantic relation.

1.1.1 Endocentric and exocentric compounds Compounds which have a

head are called ‘endocentric compounds’. A head of a compound has similar
 Compounding 67

characteristics to the head of a phrase: it represents the core meaning of the

constituent, and it is of the same word class. For example, in sneak-thief, thief is
the head (a sneak-thief is a kind of thief; thief and sneak-thief are both nouns).
Compounds without a head are called ‘exocentric compounds’ or ‘bahuvrihi
compounds’ (the Sanskrit name). The distinction between endocentric and
exocentric compounds is sometimes a matter of interpretation, and is often of
little relevance; for example, whether you think greenhouse is an endocentric
or exocentric compound depends on whether you think it is a kind of house.
The major interest in the head of a compound relates to the fact that where
there is a clear head, its position seems to be constrained; endocentric com-
pounds tend to have heads in a language systematically on either the right
(e.g. English) or left (e.g. Vietnamese, French).

1.1.2 Co-ordinate compounds There is a third kind of compound, where

there is some reason to think of both words as equally sharing head-like char-
acteristics, as in student-prince (both a student and a prince); these are called
‘appositional’ or ‘co-ordinate’ or ‘dvandva’ (the Sanskrit name) compounds.
Co-ordinate compounds can be a combination of synonyms (example from
Haitian):

toro-bèf (bull-cow) ‘male cow’

a combination of antonyms (example from French):

aigre-doux (sour-sweet)

or a combination of parallel things (example from Malayalam):

acchanammamaaIf ( father-mother-pl.) ‘parents’

Co-ordinate compounds can have special characteristics in a language. For

example, in Mandarin, co-ordinate compounds behave differently in terms
of theta-assignment from (endocentric) resultative verb compounds. And in
Malayalam, co-ordinate compounds are not affected by gemination processes
which other compounds undergo (see section 5.2).

1.1.3 The semantic relations between the parts The semantic relations
between the parts of a compound can often be understood in terms of modifica-
tion; this is true even for some exocentric compounds like redhead. Modifier–
modifiee relations are often found in compounds which resemble equivalent
phrases; this is true, for example, of English AN%N1 compounds and many
Mandarin compounds (see Anderson 1985b). It is not always the case, though;
the French compound est-allemand (East German) corresponds to a phrase
allemand de l’est (German from the East). In addition, many compounds mani-
fest relations which can be interpreted as predicator–argument relations, as in
68 Nigel Fabb

sunrise or pull-chain. Note that in pull-chain, chain can be interpreted as an

argument of pull, and at the same time pull can be interpreted as a modifier
of chain.

1.1.4 Transparency: interpretive and formal The transparency and pre-

dictability of a compound are sometimes correlated with its structural trans-
parency. For example, in languages with two distinct types of compound where
one is more interpretively transparent than the other, the less interpretively
transparent type will often be subject to greater phonological or morphological
modification. A diachronic loss of transparency (both formal and interpretive)
can be seen in the process whereby a part of a compound becomes an affix,
as in the development of English -like as the second part of a compound to
become the derivational suffix -ly.

1.2 Types of compound

Accounts of compounds have divided them into classes. Some of these – such
as the exocentric, endocentric and appositional types, or the various interpretive
types (modifier–modifiee, complement–predicator, etc.) – are widespread across
languages. Then there are compound types which are language- or language-
family-specific, such as the Japanese postsyntactic compounds (4.3), Hebrew
construct state nominals (4.2), or Mandarin resultative verb compounds (4.1).
Other types of compounds are found intermittently; these include synthetic
compounds (section 1.2.1), incorporation compounds (section 1.2.2) and redu-
plication compounds (section 1.2.3).

1.2.1 Synthetic (verbal) compounds The synthetic compound (also called

‘verbal compound’) is characterized by a co-occurrence of particular formal
characteristics with particular restrictions on interpretation. Not all languages
have synthetic compounds (e.g. English does, but French does not). The formal
characteristic is that a synthetic compound has as its head a derived word con-
sisting of a verb plus one of a set of affixes (many writers on English restrict
this to agentive -er, nominal and adjectival -ing, and the passive adjectival -en).
Thus the following are formally characterized as synthetic compounds:

expert-test-ed
checker-play-ing (as an adjective: a checker-playing king)
window-clean-ing (as a noun)
meat-eat-er

(There is some disagreement about whether other affixes should also be

included, so that slum-clear-ance for example would be a synthetic compound.)
Compounds with this structure are subject to various restrictions (summarized
by Roeper and Siegel 1978), most prominent of which is that the left-hand
member must be interpreted as equivalent to a syntactic ‘first sister’ of the
 Compounding 69

right-hand member. We discuss this in section 3.2. There have been many
accounts of synthetic compounds, including Roeper and Siegel 1978, Selkirk
1982, Lieber 1983, Fabb 1984, Botha 1981 on Afrikaans, and Brousseau 1988 on
Fon; see also section 2.1.2. Many of the relevant arguments are summarized in
Spencer 1991.

1.2.2 Incorporation compounds In some languages, incorporation words

resemble compounds: for example, both a verb and an incorporated noun may
exist as independent words. Bybee (1985) comments on this, but suggests that
even where the two parts may be independently attested words, an incorp-
oration word may differ from a compound in certain ways. This includes
phonological or morphological differences between incorporated and free forms
of a word. (In fact, as we have seen, this is true also of some compound pro-
cesses; e.g. segment loss, which is found in Tiwi incorporation words, is found
also in ASL compounds.) Another difference which Bybee suggests may dis-
tinguish incorporation from compounding processes is that the incorporation of
a word may depend on its semantic class. For example, in Pawnee it is mainly
body part words which are incorporated, while various kinds of name are not
(such as personal names, kinship terms, names of particular species of tree,
etc.). It is possible that compounding is not restricted in this way; as we will
see, semantic restrictions on compounding tend to be in terms of the relation
between the parts rather than in terms of the individual meanings of the parts.

1.2.3 Repetition compounds Whole-word reduplication is sometimes

described as a compounding process, because each part of the resulting word
corresponds to an independently attested word. Steever (1988), for example,
describes as compounds Tamil words which are generated through reduplica-
tion: vantu ‘coming’ is reduplicated as vantu-vantu ‘coming time and again’. In
another type of Tamil reduplicated compound, the second word is slightly
modified: viyAparam ‘business’ becomes viyAparam-kiyAparam ‘business and such’.
English examples of this type of compound include words like higgledy-piggledy,
hotchpotch, and so on.

1.3 Compounds which contain ‘bound words’

In a prototypical compound, both parts are independently attested as words.
However, it is possible to find words which can be parsed into an independ-
ently attested word plus another morpheme which is not an independently
attested word but also does not appear to be an affix (see Aronoff 1976). Here
are some examples from English (unattested part italicized):

church-goer ironmonger television cranberry

The part which is not attested as an isolated word is sometimes found in other
words as well; in some cases it may become an attested word: for example, telly
70 Nigel Fabb

(= television). These parts fail to resemble affixes morphologically (they are relat-
ively unproductive compared to most affixes), and there is no good evidence
on phonological grounds for considering them to be affixes. They are also
unlike affixes semantically; judging by their contribution to the word’s mean-
ing, they have lexical rather than grammatical meanings. A similar example
from standard Tamil (Steever 1988) is the form nA[ ‘goodness’, which is not
independently attested, but can be found in compound nouns such as nalla-nA[
‘good-day’.

2 The structure of compounds

One thing that is reasonably clear about the structure of compounds is that
they contain two words, and the distinctness of these two components is vis-
ible to various rules. Other aspects of structure are not so obvious, however:
in the order of the parts regulated by rules 2.1, are their word classes visible
to any rules 2.2, and are three-word compounds hierarchically structured (2.3)?

2.1 Directionality
A compound can be ‘directional’ in two senses. One sense involves the posi-
tion of the head: whether on the right or the left. The other sense involves the
direction of the relation between the parts of the compound: the direction of
modification in a noun–noun compound (e.g. in log cabin modification is right-
wards) or the direction of complementation in a verb-based compound (e.g. in
push-bike complementation is rightwards). Notice that the two senses of direc-
tionality can be independent, because a compound can have internal modifica-
tion or complementation without having a head: killjoy has no head, but it does
have a predicator–complement order. This is an important descriptive issue;
some accounts assume that a modifier–modifee or predicator–argument rela-
tion inside a compound is itself evidence that part of the compound is a head.
To the extent that there are any useful claims about directionality of the head
to be made, it is probably best to focus on the narrowest definition of head
(which involves a semantic link between head and whole); see Brousseau 1988
for arguments to this effect.

2.1.1 The location of the head In English, the head of an endocentric word
is on the right. In French, the head is on the left (as in bal masqué, ‘masked
ball’). It has been argued (e.g. by DiSciullo and Williams 1987) that all true
endocentric compounds are right-headed, and that any left-headed compounds
should be considered as exceptions – for example, as phrases which have been
reanalysed as words. This is not a widely accepted account (it is based on the
controversial Right-Hand Head rule of Williams 1981b).
 Compounding 71

2.1.2 An argument about directionality and synthetic compounds

Brousseau suggests that the direction of relations inside a compound is respons-
ible for determining whether a language is able to have synthetic compounds.
She suggests that in a synthetic compound, the non-head must modify the head,
and be a complement of the head. This means that the direction of modification
must be opposite to the direction of complementation:

—modifies→
meat-eater
←takes as complement—

Because in English, the directions of complement taking and modification are

opposed, synthetic compounds are possible (the same is true of Fon). In French
and Haitian (a Fon–French creole), on the other hand, both modification and
complementation are leftward, and so the particular conditions which allow a
synthetic compound are absent – hence there are no synthetic compounds in
these languages.

2.2 The word classes of the component words

Another question which must be asked about the structure of compounds is

whether the class of the component words is relevant, or whether word class
is lost when the words are formed into a compound. ‘Relevant’ would mean,
for example, visibility of word class to a class-sensitive phonological or mor-
phological rule. Little attention has been focused on this question; clearly
some compound-internal affixation rules are class-sensitive (but this might be
because they are added before the compound is formed).
Most of the attention to word class in a compound has focused on the attested
word-class structures of compounds in a language. For example, in Punjabi
(Akhtar 1992) there are large numbers of compounds involving a combina-
tion NN, AN, AA, NV and VV, but none with a structure VA, and very few
with a structure VN. Selkirk (1982) suggests that these facts about a language
are best expressed by compound-specific rewriting rules analogous to phrase-
structure rules. This approach has been adopted by many people, and is useful
as a descriptive device. However, there are some fundamental differences
between structure-building rules for compounds and structure-building rules
for phrases:

(a) There is no true equivalent of X-bar theory as a constraint on

compound-building rules. Most obviously, compounds need not
have a head. More generally, it is hard to find structural generaliza-
tions across compound structures analogous to the generalizations
expressed by X-bar theory for phrases.
72 Nigel Fabb

(b) Compound-building rules would rarely be recursive. In English, for

example, the only clearly recursive type is the NN%N combination.
(c) There is a problem about productivity. Phrase-structure rules are
fully productive; each rule can underlie an infinite number of phrases
(partly because of recursion). But some rules for building compounds
are manifested by very few actual compounds. Selkirk recognizes
this, and distinguishes rule-built compounds from non-rule-built
compounds. Perhaps, though, there is an alternative way of explain-
ing the prevalence of certain compound types along functional rather
than formal lines. Thus, in English, the prevalence of NN%N and
AN%N types might be because of a functional need for compound
nouns before other word classes, and because these have a modifier–
modifee structure which is easily interpreted. This is a complex
problem which requires metatheoretical decisions about the place of
functional considerations and the meaning of productivity.

2.3 Subconstituency in three (or more) word

compounds
Where compounds consist of three or more words, the compound can some-
times be interpreted by breaking it down into subconstituents. This is true, for
example, of chicken-leg-dinner, which is interpreted by taking chicken-leg as a
subcompound within the larger compound. In some cases, ambiguity arises
from the possibility of two alternative groupings of words, as in American
history teacher (a history teacher who is American or a teacher of American
history). This fact about interpretation raises the question of whether three-or-
more-word compounds might perhaps have a subconstituent (hierarchical)
structure like a phrase (i.e. (b) or (c) rather than the flat structure (a)).

(a)

American history teacher

(b)

American history teacher

(c)

American history teacher

 Compounding 73

It is not obvious that the interpretive facts alone demonstrate the presence
of a complex structure. The interpretive rules might simply pick any pair of
adjacent units and make them into a unit, taking a ‘syntactic’ structure like (a)
and building a ‘semantic’ structure like (b) or (c). Compare hierarchical phrase
structure, which pre-exists any interpretive strategy, as can be shown by the
sensitivity of syntactic processes (such as binding theory) to constituent struc-
ture. But compounds are relatively inert compared to syntactic constituents
(no movement, anaphoric coindexing, etc.), so it is harder to find supporting
evidence for complex constituent structure.
As we will see (section 5.1.1), there is some evidence in English from the
stressing of four-word compounds that the compound-specific stress rules are
sensitive to a subconstituent structure, and hence that such structure exists
outside the interpretive component. Turkish also presents evidence for sub-
constituent structure in hierarchically interpreted compounds (Spencer 1991).
The ‘indefinite izafet compound’ has the structure Noun + Noun-poss. when
it consists of two words. The three-part compound has one of two types of
structure, each corresponding to a different interpretation of its hierarchical
structure (as bracketed):

[ Noun + Noun-poss. ] + Noun-poss.

Noun + [ Noun + Noun-poss. ]

Here the placement of the possessive affix is sensitive to hierarchical structure.

Most of the multiple-word compounds which are cited (particularly those
cited as hierarchical compounds) tend to be combinations of nouns. This sug-
gests that if there are structure-building rules for compounds, as discussed in
the previous section, they are not generally recursive.
Note that sometimes a compound can be interpreted as having an internal
subconstituency, without that subconstituent being an attested word. An ex-
ample from Fon (Brousseau 1988) is:

gàn-¡d-khn NNN%N money-stomach-cord = money belt

This compound is presumably interpreted as N[NN], but the [NN] sub-

constituent does not exist as an independent word. This is not necessarily a
strong argument against a hierarchical structure in this case, however, because
there are also many examples of two-word compounds where one ‘word’ is
not independently attested.
Note that some multiple-word compounds are not interpreted as having a
hierarchical structure; this is true particularly of dvandva compounds such as
the following from Tamil:

vDra-tDra-cakacaØ-kaŒ (courage-bravery-valour-pl.) ‘courage, bravery and valour’

An example of a compound which mixes non-hierarchical with hierarchical

interpretation is the following, from Mandarin:
74 Nigel Fabb

jı̄-yā-yú-ròu = ‘animal foodstuffs’ (i.e. chicken-duck-fish-meat)

Interpretively the structure is [[chicken + duck + fish] meat].

2.3.1 Hierarchies which include non-word components A three-member

compound need not contain three words. For example, a morpheme may appear
between two words in a compound or at one end of a compound. It has been
argued by a number of writers that in (English) synthetic compounds, the
suffix is a third constituent of the compound; synthetic compounds on this
analysis have a hierarchical structure such as [[meat-eat]-er].

3 The interpretation of compounds:

interpretive gaps

Extensive descriptive work has been undertaken on the semantic relations

holding between the components of English compounds (see e.g. Levi 1978,
Warren 1978). An interesting theme that arises from some of this work is the
possibility that there are certain gaps: semantic relations between the parts of
a word which are possible in principle, but are not attested in practice. In this
section we look at two such gaps in English compounds.

3.1 The missing goal

One of the commonest kinds of compound in English is the NN%N type.
Many different relations can be interpreted as holding between the two mem-
bers of such a compound. The interesting question in these cases is whether
any relationships are not attested; in surveys of NN%N compounds, both
Downing (1977) and Warren (1978) found that while source (something moved
away from) was attested, goal (something moved towards) was only mar-
ginally attested. Warren found only fourteen potential examples out of 3,994
compounds, and suggests that these may not even be true examples of ‘goal’
compounds. In her list of relations between the parts of nominal compounds,
Levi (1978) has ‘from’ (e.g. store-clothes) but not ‘to’.
The same gap can be seen in other compounds; for example, while we find
VN%N compounds like print-shop (a shop where printing takes place), there
are no compounds like go-place (meaning a place to which someone goes). This
gap also appears in synthetic compounds: heaven-sent can be interpreted only
as sent from heaven, not sent to heaven. Note that apparent goal compounds
like church-goer actually mean ‘someone who attends church’ (not someone
who moves towards church); similarly, sea-going means ‘going on the sea’ (not
to the sea).
 Compounding 75

3.2 Synthetic compounds in English: the absence

of ‘subject’
In a synthetic compound, the crucial interpretive restriction is that the left-
hand word (a noun, adverb or adjective) must be interpretable as a comple-
ment of the right-hand word (and must not be interpretable as an external
argument or subject). In effect, synthetic compounds with -ing or -er are like
reversed active verb phrases with equivalent components (play checkers >
checker-playing), while synthetic compounds with passive -en are like reversed
passive verb phrases (tested by experts > expert tested). Synthetic compounds
thus differ from other compounds (sometimes called ‘root compounds’); hence
while *bird-singing is excluded, there is a compound bird-song where the left-
hand member is interpretable as the subject of the right-hand member.
Synthetic compounds are interesting because the rules for their interpreta-
tion seem to be related to rules for building the meaning of sentences (e.g. the
assignment of thematic roles to particular positions in a sentence, depending
on the active or passive nature of the verb).

3.3 Ways of explaining interpretive gaps

Interpretive gaps in compounds could in principle be explained in one of three
ways:
(a) Constrain the compound-building rules to make them sensitive to
interpretation-relevant aspects such as thematic relations. Levi (1978) takes
this approach within a Generative Semantics framework, for compound nouns.
This is also the approach taken by Roeper and Siegel (1978) in an Extended
Standard Theory framework; they build synthetic compounds by a special
(transformational) rule which takes a combination of a verb and a subcat-
egorized complement as input. The gaps exist because there is no possibility
of building synthetic compounds which have a verb combined with a non-
subcategorized argument such as its subject. But Roeper and Siegel’s approach
(in particular) runs into a problem. Consider, for example, the compound
bird-singing. This cannot be built by the synthetic-compound-building rule,
because it combines the verb with a non-subcategorized argument (its subject).
But nothing stops it being built by an alternative rule – the root-compound-
building rule, which takes two nouns and combines them (bird + singing). This
rule is not subject to thematic constraints, as can be seen in subject–predicate
compounds like sunrise. So whatever rules out bird-singing as a root compound
is clearly not associated with the compound-building rules.
(b) Instead, it may be that the subject is ruled out by some filter which
looks at the compound, and if it has the structure of a synthetic compound,
applies certain constraints on interpretation to it. This would differentiate root
from synthetic compounds not in how they are built, but in their surface form.
76 Nigel Fabb

(c) A third possible approach would be to explain interpretive gaps in

compounds in terms of general constraints on the possible meanings of a word
(see Carter 1976). Such an approach does not necessarily require that compound-
internal thematic relations be specified, because these relations are not expli-
citly referred to. This approach might provide an explanation of the ‘goal’ gap.
Note that this gap is found also in non-compound word formation: Hale
and Keyser (1992) point out that while there is a verb shelve, meaning ‘put
on a shelf’, there is no verb church, meaning ‘go to church’. It may be that a
meaning of ‘movement towards’ is incompatible with some aspect of possible
word meaning. For example, Downing (1977) comments that ‘unambiguously
fortuitous or temporary relationships’ are ruled out in favour of generic or
habitual relationships. Perhaps ‘movement towards’ is ruled out in general
because it is not usually a generic or habitual relationship: in this light, it
is interesting to compare ‘source’ (movement from) with ‘goal’ (movement
towards). The source of something remains a stable and permanent pro-
perty of that thing; the goal of something is its goal only while it is travelling
towards it.

4 Compounds and syntax

In this section we look at some language (or language-family)-specific com-

pound types which have an internal structure open to syntactic manipula-
tion and visible to syntactic processes. Incorporation compounds are a clear
example, and synthetic compounds have also been argued to have such a
structure (e.g. by Fabb 1988). There are two complicating factors when con-
sidering the ‘syntactic’ aspects of compounds.
One is that compounds tend to have relatively fixed meanings, so that it is
difficult, for example, to modify them; the question of syntax vs morphology
may be irrelevant here. Thus, for example, the ASL compound ‘blue-spot’
(= bruise) cannot be morphologically modified to ‘*darkblue-spot’ (?= bad
bruise); and the French compound garde-malade (= nurse) cannot be syntactic-
ally modified to garde-bien-malade (?= good nurse).
The second problem relates to the possibility that some compounds are
the result of lexicalization of phrases. Thus, while in English it is generally
impossible to have the inside a compound, there is a word middle of the road
which looks like a compound but may best be analysed as a lexicalized phrase;
the same can be said for many French compounds such as pomme-de-terre
(= potato) or trompe-l’œil (= illusion), both of which contain typically syntactic
components, the preposition de or the article l’. Compounds containing and,
such as foot-and-mouth disease may perhaps be dealt with by claiming lexicaliza-
tion of a phrase; or it may simply be that co-ordination can involve parts of
words with no syntactic implications (see Sproat 1985 and Bates 1988 for dif-
ferent positions on this).
 Compounding 77

4.1 Mandarin resultative verb compounds

Y-F. Li (1990) discusses resultative verb compounds in Mandarin, which have
a VV%V structure:

Baoyu qu-lei-le ma
Baoyu ride-tired-asp horse
‘Baoyu rode a horse (and as a result it/Baoyu got) tired’

The resultative verb compound here is qu-lei. The first verb describes an action
‘ride’; the second verb characterizes the result of that action as a state, ‘tired’,
of the compound verb’s subject (Baoyu) or object (that horse). The compound
verb thus inherits theta-role-assigning properties from the component verbs.
This is one reason for thinking that the internal structure of the compound
may be visible in the syntax. Another reason comes from the fact that the first
verb can be duplicated and followed by the object, so that the above sentence
has this as an alternative form:

Baoyu qu ma qu-lei-le

A third reason for assuming syntactic visibility is that it is possible to modify

the compound by inserting de or bu between the two verbs; the particles add
the meaning ‘can’ or ‘cannot’ respectively (see Anderson 1985b).
Li suggests that the identification of the first verb as the head of the resultative
verb compound has important consequences. He compares the resultative verb
compound with a co-ordinate VV%V compound type. In the resultative verb
compound, only the first verb must assign its external argument to the subject
of the sentence, because only the first verb is the head of the compound word;
the second word is free to assign its external argument to the subject or the
object in the sentence, resulting in the ambiguity illustrated above. In a co-
ordinate compound, however, both verbs must assign their external argument
to the subject of the sentence, because they are both heads of the compound
word (hence the ambiguities found in the resultative verb compound are not
available to co-ordinate compounds).

4.2 Hebrew construct state nominals

Borer (1988) distinguishes two kinds of multi-word words in Hebrew, con-
struct state nominals:

ca9if ha-yaldá (scarf the-girl) ‘the girl’s scarf’

and compounds:

beyt xolím (house sicks) ‘hospital’

78 Nigel Fabb

The two kinds of word can be seen as two kinds of compound. Both kinds of
compound have word-like phonology (e.g. both have a single primary stress)
and morphology (e.g. both are pluralized in the same way). Neither com-
pound can be extracted from. The difference between them is that construct
state nominals are more transparent interpretively and syntactically than other
compounds; for example, definiteness and plurality on the components of
a construct state nominal are taken into account by sentence-interpretation
rules. Borer argues that the distinction should be achieved by having word-
formation rules in the lexicon (to build ordinary compounds) or in the syntax
(to build construct state nominals). A word built in the syntax will have its
components visible to syntactic processes.

4.3 Japanese post-syntactic compounds

Shibatani and Kageyama (1988) argue that Japanese has lexical and postsyntactic
compounds. A postsyntactic compound is ‘postsyntactic’ because it appears to
consist of the turning of syntactic material into a compound: it consists of a
noun phrase followed by a verbal noun which takes it as a complement (i.e.
the compound is subject to a kind of first-sister constraint, and so cannot take
a scrambled sentence as input). Here is an example of such a compound:

zikken-syuuryoo-go
experiment-finish-after = ‘after the experiment was completed’

Postsyntactic compounds are interpretively transparent. They are also trans-

parent to syntactic processes such as anaphoric co-reference, and they can
contain demonstratives and honorifics. They do not have the tonal pattern
associated with lexical words. Despite these syntactic characteristics, Shibatani
and Kageyama argue that the constituents as a whole are words (i.e. com-
pounds), because the case particle (following the NP) is omitted, there is no
tense on the verbal element, and the compound cannot be interrupted by an
adverb. Like Borer, they thus allow compounds to be built at different levels
(in this case, lexicon and PF).

5 Phonological and morphological processes and

compounds

Compounds resemble derived words in some ways, and resemble phrasal

combinations in other ways. This is true also of the phonological and morpho-
logical processes which apply to compounds. For example, some phonological
processes group compounds with derived words; others group compounds
 Compounding 79

with phrases; and others isolate compounds as a distinct class. Most recent
discussions of these groupings have been in a Lexical Morphology/Phonology
framework, and have been used to argue for a level-ordering of morphological
and phonological processes (see section 5.2 for an illustration).

5.1 Suprasegmental processes

5.1.1 Stress One of the most commonly cited compound-specific phono-

logical rules is stress assignment. Compounds may be subject to a rule which
places heavier stress on one word. In English, for example, the compound
big-foot (the name of a monster) is differentiated from the phrase big foot by
the heavier stressing of big relative to foot in the compound. (In English and
Danish, the first word is stressed; in Italian and Spanish, the second word.)
Even where there is not a compound-specific stress rule, compounds can still
show interesting stress properties. For example, Anderson (1985b) shows that
in Mandarin, contrastive stress ‘sees’ a compound as a single word, so there
is only one place in the compound where contrastive stress can be placed; by
comparison, a two-word phrase can have contrastive stress on either word.
The stress pattern of English compounds has been the source of extensive
analysis. It is relevant primarily in three areas:
(a) The stress pattern of compounds may indicate the presence of hier-
archical structure inside the compound. Since most versions of the compound
stress rule apply to two-word compounds, it is interesting to see what happens
to the stress in larger compounds. In these, there seems to be clear evidence
that stress assignment recognizes internal subcomponents; for example, in a
four-word compound like student essay récord book, the greatest stress is on
record because a rule recognizes the presence of a subconstituent record book
and places extra stress here. Note that a different hierarchical structure (cor-
responding to a different hierarchy of interpretation) involves a different stress
pattern, as in American hístory teacher association. This is one of the few pieces
of evidence that multi-word compounds in English have a complex internal
formal structure.
(b) Compounds must be handled by some system of rules for stress assign-
ment in English. Accounts vary in the extent to which they have rules specific
to compounds. Chomsky and Halle (1968) have a separate Compound Rule.
Breaking with this tradition, Liberman and Prince (1977) adapt the Lexical
Category Prominence Rule so that it applies to derived words and compounds;
similarly Halle and Vergnaud (1987) draw attention to the close similarity
between rules for compounds and rules for derived words. The question of
which rules apply to stress in compounds is relevant to questions of ordering
in the lexicon (e.g. level-ordering). This extends beyond English: Shaw (1985)
shows that the Dakota ‘syntactic compounds’ have two major stresses, while
the ‘lexical compounds’ have one major stress; she explains this by having the
80 Nigel Fabb

compounds formed at different levels in a level-ordered lexicon, with stress

rules also level-ordered.
(c) English compounds do not all have initial stress; examples of level-
and final-stressed compounds are knee-déep and apprentice-wélder respectively.
This opens up the possibility that there are structural (or interpretive) char-
acteristics of compounds, which stress rules are sensitive to. Bates (1988), for
example, develops an analysis in which the English compound stress rules
are sensitive to whether an initial adjective is predicative (initial stress) or not
(final stress). Analyses along these lines provide potential evidence for internal
structure in a compound (evidence which is otherwise sparse, see section 2.3).
Note that the initial/final stress issue is in some cases a matter of dialectal dif-
ference; compare British English ice-créam and hot dóg with American English
íce-cream and hót dog.

5.1.2 Other suprasegmental processes While stress is the most commonly

cited compound-specific suprasegmental process, other such processes may
differentiate compounds from derived words or phrasal combinations. In many
Mandarin Chinese compounds the second element loses stress because it first
loses its tone (Anderson 1985b). In Fon, nasalization occurs in a suffix, but not
in the second word in a compound (Brousseau 1988). And Klima and Bellugi
(1979) argue that in American Sign Language, temporal rhythm is analogous
to stress; a compound word and a simple word take more or less the same
amount of time to produce, because in a compound word, the first word is
made significantly more quickly. Hence there is a compound-specific temporal
reduction rule (cited in Liddell and Johnson 1986).

5.2 Phonological processes between the two words

Segmental phonological processes will tend to happen between two morphemes
– either a word and an affix, or two words. There are examples of such pro-
cesses which distinguish compounds from other combinations. In English,
some phonological rules apply within derived words but not within com-
pounds; M. Allen (1978) shows this for beauty-parlour vs beauti-ful. In Dakota
(Shaw 1985) a different alliance can be found, with compounds patterning
with derived words rather than phrases: ‘lexical compounds’ are subject to
an epenthesis rule adding a to the end of the first word, which then feeds a
continuant voicing rule; this rule is found also in derived words, but not in
‘syntactic compounds’.
Malayalam (Mohanan 1982) provides a particularly complex picture of the
visibility of compound structures to different phonological processes. Stem-
final nasal deletion and stem-final vowel lengthening apply just to compounds.
Other rules apply to both compounds and derived words (but not to sequences
of words in a phrase), such as a vowel sandhi rule where two adjacent vowels
are merged into one. A further complexity is that some rules apply only to a
 Compounding 81

subset of the compounds: stem-final and stem-initial gemination of obstruents

in Dravidian stems occurs in compounds with a modifier–modifiee structure
(‘subcompounds’) but not in co-ordinate compounds. Mohanan uses these
facts as evidence for a theory of level-ordered lexical phonology/morphology,
along the following lines:

stratum morphological vowel vowel nasal gemination

processes sandhi lengthening deletion
1 derivation yes
2 subcompounding yes yes yes yes
3 co-compounding yes yes yes
4 inflection

5.3 Segment loss

American Sign Language compounds tend to undergo more radical formal
restructuring than most spoken language compounds, with loss of segments
being a common feature of compound formation. For example, the compound
translated as ‘black-name’ (meaning bad reputation) involves loss of one of
the segments of the word for ‘name’ (which on its own consists of a sequence
of two identical segments). Liddell and Johnson (1986) comment that in ASL
‘the average compound has the same number of segments as the average
simple sign’.

5.4 Morphemes and compounds

5.4.1 Compound-specific morphemes In some languages, a morpheme
(with no independent meaning) may be inserted between the two words. This
morpheme may bear a historical relation to some affix, but is synchronically
found only in compounds. In German, for example, s or en may be inserted:

Schwan-en-gesang ‘swan song’

5.4.2 Inflectional morphology There have been two lines of research focus-
ing on whether an inflectional morpheme (e.g. marking plurality, case or tense)
can be found on a component word inside a compound.
One approach (e.g. Kiparsky 1982b) focusing on the question of level-
ordering of morphological processes, looks at the distinction between regular
and irregular inflectional morphology, and asks whether irregular morphology
is more likely to appear on a compound-internal word (because it precedes the
compounding process) than regular morphology (which comes later than the
compounding process). The evidence from English is not particularly clear. As
82 Nigel Fabb

predicted, regular inflection is lacking in some compounds like footprints (com-

pare *foots-prints), where irregular inflection is present in corresponding com-
pounds like teeth-marks. But, contra predictions, sometimes regular inflection is
present as in arms race where irregular inflection is missing as in out-putted.
The second line of enquiry focuses on the implications of having any inflec-
tion on a part of a compound, and whether this means that the internal struc-
ture of compounds is visible to the syntax (the answer to this depends on
one’s theory of inflection). For further discussion, see Anderson 1992: ch. 11.
Questions about the visibility of inflection relate also to questions about the
visibility of theta-assigning properties of subparts (as e.g. in Mandarin result-
ative compounds) and what this implies about the visibility of compound
structure in syntax.

5.4.3 Derivational processes and compounds Compounds on the whole

tend not to undergo derivational processes. However, there are some fairly
clear examples such as English bowler-hatted, which can only be interpreted
as bowler-hat + ed. It may be that English synthetic compounds are the result
either of a verb-final compound being affixed with -ing, -er, etc., or of a com-
bination of derivation and affixation (as in Roeper and Siegel’s (1978) com-
pound formation rule). There has also been discussion of compounds such
as transformational grammarian, which on interpretive grounds would seem to
involve suffixation of transformational grammar; there are, however, other ways
of dealing with these ‘bracketing paradoxes’ (see Spencer 1988b, e.g.). Liddell
and Johnson (1986) show that in ASL a reduplication process adds the mean-
ing of ‘regularly’ to a verb; this process reduplicates all of a compound verb
(though this may be an inflectional process).

5.4.4 Clitics and compounds Dakota postverbal clitics are attached in the
lexicon (Shaw 1985). They attach to the first member of syntactic compounds,
but not to the first member of lexical compounds.

6 Conclusion

Comparatively little theoretical work has been done on compounds. This is

because compounds tend to be less phonologically or morphologically active
than derived words, and less syntactically active than phrases: compounds are
relatively inert. My view is that the three most interesting questions about
compounds are as follows:

(a) If there are compound-building rules, what form do they take? Are
they recursive (i.e. building hierarchical structure), and do they take
word class into account? Are compound-building rules based on a
universally available rule type (i.e. is ‘compound’ a universal?).
 Compounding 83

(b) How do compound-building rules interact with other rules – or,

put slightly differently, what aspects of compound structure are
visible to what other processes, whether syntactic, morphological or
phonological? It is particularly important here to show that a certain
structural fact about a compound is visible to two distinct rules (e.g.
Bates (1988) argues that compound stress rules and interpretive rules
both ‘see’ that a particular modifier in a compound is predicative
rather than attributive). The question of the interaction of compound-
building rules with other rules has been discussed particularly in
the level-ordered systems of lexical phonology and morphology.
(c) Are there interpretive gaps? Are some sorts of meaning not creat-
able by putting compounds together – and if so, is this because of
compound-specific rules, or because of more general principles of
possible word meaning?

NOTE

1 XY%Z is to be interpreted as: [XY] is

a compound of word class Z.