?tekauer Pavol, Lieber Rochelle. Handbook of Word-Formation
Подождите немного. Документ загружается.
C
ONSTRAINTS ON PROD
U
CTIVITY
341
were located in a first stratum together with the corresponding phonological rules,
while the Germanic affixes were located in a second stratum, extrinsically ordered
after the first one. By locating, for example, the stress shifting rule at the first
s
tratum, this theor
y
was able to predict the correlation between affix order and stress
that we observed in the examples
g
iven above. Sie
g
el’s theor
y
became quite popula
r
and was inte
g
rated in a sli
g
htl
y
modified form into the hi
g
hl
y
influential theor
y
of
Lexical Phonology of Kiparsky (1982). Nevertheless it was also soon pointed out
that the predictions of the theory are not entirely correct. Strauss (1982) noted that
Germanic prefixes could regularly occur inside Latin suffixes (cf. [[un-gramatical
]
-
i
t
y
]
), while Aronoff & Sridhar (1983: 4ff.) showed that stress-neutral affixes may
o
ccasionally occur insid
e
stress-shifting ones (cf. X-able
vs.
X-abíl-it
y
,
X-
ise
vs.
X-
is
-
átion
,
X-ist
vs.
t
X-
íst
-
ic
). Finally, Fabb (1988) pointed out that level ordering also
failed to rule out man
y
impossible affix sequences inside the two strata. The theor
y
o
f level orderin
g
, in other words, turned out to be both too stron
g
and too weak. As a
c
onsequence of these serious empi
r
ical flaws, in the nineties
(
c
f
. Goldsmith 1990:
259-273, among many others) level ordering was either totally abandoned o
r
thoroughly revised (c
f
. Giegerich 1999). Today, in the light of the evidence
accumulated
,
it therefore see
m
s
p
referable to ado
p
t the
n
ull hypothesis that affix
o
rder in English is constrained by selecti
o
nal restrictions only, except for the general
processing constraints outlined in 2.
3.2 Affix-specific restrictions
B
efore turning to the question of what properties of bases or complex words may
be relevant in defining selectional restri
c
tions of affixes, it is worth dwelling on
s
ome general problems concerning the na
tu
r
eo
f
se
l
ec
ti
o
nal r
es
tri
c
ti
o
n
s.
T
h
e
fir
s
t
o
n
eco
n
ce
rn
s
th
e
l
ocus w
h
e
r
ese
l
ec
ti
o
nal r
es
tri
c
ti
o
n
s
ar
e
t
obee
n
coded.
I
n rule-based theories of word formation such as Aronoff
(
1976
)
, where a rule of
word formation is conceived of as a function that takes a base as its input and
transforms it into a more complex word as its output, restrictions are normall
y
defined over the input, thus narrowin
g
the domain of application of the rule. In
theories, such as Lieber (1981), where affixes are
g
iven lexical entries
j
ust like
words and stems, their combinatorial
p
otential is encoded in
s
ubcategorisation
frames
.
In order to express the generalisation that the suffix -
able
in
p
roductive use
m
ay only be attached to transitive verbs, one would say in the first kind of
framework that the application of the word formation rule introducing -
able
i
s
r
estricted to transitive verbal bases, while in a Lieber-style framework one woul
d
s
ay that the lexical entry -
able
carries a subcategorisation frame
V
tr
_. The empirical
r
r
esult is exactl
y
the same in this case. When we look at other patterns of word
formation, however, it turns out that the two wa
y
s of encodin
g
selectional
r
estrictions are not mere notational variants. S
p
anish, for exam
p
le, has a rule of
c
ompounding that forms exocentric possessive adjectives out of a noun and an
adjective, linked by the vowel -
i-
:
p
el
o
‘hair’
,
r
oj
o
‘r
ed
’
ĺ
pelirroj
o
‘redhaired
,
lit.
h
airred’, etc. The ad
j
ective semantically determines the noun, but it agrees with the
h
ead noun of the noun phrase and not with
t
he first constituent of the compound:
34
2
F
RANZ RAINER
una niña pelirroja
/*
pelirrojo
*
‘a redhaired girl’ (
pelo
(
(
is masculine
,
niña
feminine
,
-
o
is the agreement mark
e
r for the masculine
,
-
a
for the feminine
)
. This kind of
c
ompounding is productive, but severely restricted
(
c
f
. Rainer 1993: 289
)
: in fact,
t
he noun may only designate some part of the body, human or animal (c
f
.
*
puertirrojo
*
‘with a red door’
;
puer
t
a ‘door’), and must be bisyllabic (cf.
*
b
arriguigord
o
, beside the well-formed phrase
b
arriga gorda ‘fat belly’). Now, this
k
ind of restriction ma
y
be encoded neither in the lexical entries of the respective
n
ouns nor in those of the ad
j
ectives, because outside this t
y
pe of compound
r
o
jo
an
d
o
ther ad
j
ectives show no analo
g
ous restrictions as to semantic class or len
g
th of the
n
ouns they may determine. The locus where we have to state this restriction is
c
learly the pattern as such, and not one of its constituents. The same is also true fo
r
t
he requirement of placing a linking vowel -
i
- between noun and adjective. What this
and numerous analogous examples show is that restrictions on patterns of word
formation are best located at the level of the pattern (or rule, if one prefers) itself.
Theories of word formation with only lexical entries and general mechanisms of
c
ombinin
g
them, but without patterns or rules, do not seem to be viable.
The second
g
eneral problem to be address
e
d
i
sw
h
e
th
e
r r
es
tri
c
ti
o
n
s
ar
e
t
obe
d
efined on the input or on the output. Traditionall
y
, restric
t
i
o
n
s
in
wo
r
d
f
o
rmati
o
n
u
sed to be defined on the input, but since the nineties, partly in the wake of
O
ptimality Theory (cf. Raffelsiefen 1999, Plag 1999), output restrictions have
r
eceived more and more attention. In a patt
e
r
n-based theory of word formation, we
would expect both kinds of restrictions to occur, since both may easily be encoded
o
n a pattern. The prosodic restriction (bisyllabicity) on the nominal constituent of
o
ur Spanish compound t
y
pe, for example, has clearl
y
to be stated as an input
r
estriction, since the ad
j
ectival constituent shows no prosodic restriction, which
m
eans that the total number of s
y
llables of these compounds is variable. The
situation is less clear with res
p
ect to the semantic restriction to
p
arts of the human o
r
animal body: this could be encoded either as an input restriction on the nominal
c
onstituent, or alternatively as an output restriction on the pattern as a whole, stating
t
hat this kind of ad
j
ective may only refer to physical properties of persons o
r
animals. A clearer example of an output condition would be Plag’s (1999: 145-194)
analysis of English -
ise
,
which says essentially that this suffix in productive use
t
ends to favour outputs with a dact
y
lic contour (c
f
. rándomìs
e
,
d
ánd
y
ìs
e
,
etc.
)
. If the
base does not fit this prosodic schema, it is either avoided (
*
ká
r
stì
z
e
, etc.) or adapted
i
n a pertinent wa
y
(c
f
.
f
eminin
e
ĺ
f
éminìse
,
patina
ĺ
p
á
tin
ì
se, etc.
)
.
A
third general problem is whether restri
c
ti
o
n
ss
h
ou
l
dbee
n
coded o
n th
eb
a
se or
o
n the affix. Traditionally, restrictions have been encoded on the rule, the pattern, o
r
t
he affix itself, if this is
p
rovided with a separate lexical entry. Quite recently,
h
owever, Plag – following Giegerich (c
f
. Giegerich 1999) – has proposed to encode
r
estrictions in the base rather than in the affix. According to him, restrictions are
‘base-driven’
,
not ‘affix-driven’.
I
nstead of sa
y
in
g
that the suffix -
ation
se
l
ec
t
sve
r
bs
i
n -
ise
,
one should sa
y
that -
ise
se
l
ec
t
s
-
ation
.
Does that make an
y
difference? Pla
g
(1996: 777) ar
g
ues that it does. The advanta
g
e, accordin
g
to him, would be that we
need not state separately that verbs in -
ise
do
n
o
t tak
e
th
e
ri
v
al a
bs
tra
c
t
su
ffix
es
-ag
e
,
-al
,
-ance
,
-
ment
,
o
r
-
y
. Now, this result is only obtained if the
base
-
d
r
iven
rest
r
iction
s
tat
es
that -
ise
ONLY selects -
ation
. The – equally base-driven
–
C
ONSTRAINTS ON PROD
U
CTIVITY
343
s
tat
e
m
e
nt that
-ise
se
l
ec
t
s
-
ation
would not preclude the possibility that this suffix is
also compatible with other a
b
s
tra
c
t
su
ffix
es.
Th
e
r
e
f
o
r
ewe
c
an
see
that it i
s
n
o
t
b
a
se-
d
ri
ve
nn
ess
per s
e
which achieves the desired result
,
but the tacit introduction of two
k
inds of restrictions, one kind which simpl
y
states that a certain affix or base ma
y
s
elect a certain affix, and a stron
g
er variet
y
which states that a certain affix or base
m
a
y
onl
y
select a certain affix. Now
,
the same result could be achieved b
y
i
ntroducing two analogous types of affix-driven restrictions
:
-
ation
se
l
ec
t
s
-
ise
vs.
-
ation
is the only affix to select -
ise
.
The exclusive variant
,
both in its base-driven
and in its affix-driven wording, seems to contain a disguised paradigmatic statement,
s
ince it implicitly compares -
ation
with rival affixes. Such paradigmatic statements,
according to what we have said in 2.2.1, are the
j
ob of type blocking, the mechanis
m
Plag wanted to get rid of by the in
troduction of base-driven restrictions (
nn
cf. also Plag
2003: 66-67). But in realit
y
, as we can s
e
e, in Pla
g
’s account the paradi
g
matic
dimension simpl
y
creeps back throu
g
h the back door in the form of a new t
y
pe of
r
estriction with exclusive ri
g
hts: but a restriction with exclusive ri
g
hts is strictl
y
equivalent to a normal restriction + type blocking. Whether we do it the one way o
r
t
he other seems to be a matter of taste. The important thing is that there is no way of
avoiding explicit statements about the pa
r
adigmatic relationship between rival
patterns.
3
T
he fourth general problem is intimately tied to the last one. If the domain of one
general affix is curtailed by anot
her affix, as is the domain of
t
-
heit
by
t
-
ität
with
t
b
a
ses
in -
il
,
this fact must be stated twice
,
vi
z
.
as a
p
ositive restriction on
-ität
– it
t
attaches, amon
g
other ad
j
ectives, to ones in -
il
– and at the same time as a negative
l
res
tri
c
ti
o
n
o
n -
heit
– it attaches to any kind of adj
t
e
ctive with final stress exce
p
t those
in
-
il
(
this simplified statement will do for our purposes). This way of putting things
see
m
sso
m
ew
hat a
w
k
w
ar
dbec
a
use
it
does
n
ot state explicitly that the negative
r
estriction is somehow a consequence
o
f the positive one. Van Marle’s (1986)
D
omain Hypothesis was specifically designed to avoid this conceptual awkwardness
by postulating a general pri
n
c
iple according to which special
do
main
scu
rtail ri
v
al
g
eneral domains. Under this h
y
pothesis, th
e
ne
g
ative restriction would no lon
g
er be
n
ecessar
y
, but unfortunatel
y
, as we have alread
y
seen in 2.2.1, the Domain
H
y
pothesis is untenable. Whether a special domain curtails a
g
eneral domain or not,
i
s not predictable but must be stated case by case, either through a combination of
positive restriction + type bl
o
cking or through a restricti
o
n with exclusive rights,
which, as we have seen, amounts to the same. In this way we can avoid the explicit
s
tatement of negative restrictions, something that we would like to avoid fo
r
principled reasons. As is well-known (c
f
. Sokolov & Snow 1994 for a good review
o
f the literature), language learners can
o
nly rely on positive evidence: they first
l
earn sin
g
le instances of complex words and then proceed b
y
makin
g
–
g
enerall
y
v
er
y
conservative –
g
eneralisations on this
g
round (c
f
. Bowerman 1982: 324-328
,
Pinker 1989). We would thus expect to find no negative affix-specific restrictions in
3
In the case at hand
,
it is not clear whether we need
t
ype blocking or a restriction with exclusive rights at
all, since none of the rival suffixes is a general case (except -
i
ng
,
which may not really be
sy
non
y
mous). If one states the restrictions on the rival suffixes properl
y
, it ma
y
turn out that verbs in
-
ise
do not fall in their domains
,
so that no paradi
g
matic statement would be needed.
344
FRANZ RAINER
n
atural languages, except the ones attributa
b
l
e to type blocking or to some other
i
nde
p
endent cause.
U
p to now, the reader may have gained the impression that restrictions in word
formation are a question of all or nothing, i.e
.
that a
ce
rtain
b
a
se
i
se
ith
e
r in
c
l
uded
in
the domain of a particular pattern, or excluded. In reality, membership in a certain
domain is
j
ust as much of a gradual nature as membership in linguistic categories in
g
eneral (c
f
. Bauer 1983: 98-99). We often find a set of protot
y
pical members
s
urrounded b
y
more mar
g
inal ones, more acceptable to some speakers than to others.
T
his is the case, for example, of the Italian suffix -
issimo
(
c
f
. Rainer 2004
)
, which
expresses emphatic intensification and consequently is most compatible with
adjectival bases that may be used in highly emotional predications, such as
bello
‘beautiful’
,
ca
r
o
‘expensive’, etc., but less so with adjectives that involve a more
weighing attitude, such as
simile
‘similar’
,
s
ignificativ
o
‘significant’, etc. There is a
c
lear positive correlation, in this case, between degree of membership in the domain
of
-
issimo
o
n th
eo
n
e
han
d
an
d
th
e
n
u
m
be
r
o
f att
es
t
ed
f
o
rm
s
in -
issimo
an
d
acceptabilit
y
to native speakers on the other. In other cases
,
the relation between the
d
ifferent subdomains of a domain is better characterised b
y
the notion of famil
y
r
esemblance
,
i.e. the subdomains are
a
l
l somehow related, but no necessar
y
and
s
ufficient conditions can be formulated that would cover all the
p
ossible bases and
o
nly those. This situation is particularly frequent with less productive patterns,
where the forces of analogy generally a
r
e still very apparent. Good illustrations of
s
uch cases may be found in Adams’ (1973: Chapter 13) analyses of -
a
r
ian
,
-
ste
r
,
-
ee
r
,
-
nik
and many other patterns (
k
cf. also Rainer 2003
)
.
N
ow that we have discussed the main
g
eneral problems involved in the notion of
affix-specific restriction, we ma
y
turn our attention to the sin
g
le factors that ma
y
be
r
elevant for definin
g
the applicabilit
y
of a pattern of word formation.
3.2.1 Phonolo
gy
A
ll aspects of the phonolo
gy
of words ma
y
be relevant for definin
g
the
applicabilit
y
of a certain pattern of word-formation. The most common t
y
pe of
phonological restriction is th
e
sensitivity of a suffix to the presence of certain
phonemes in the base, especially at the right edge. So, for example, the suffix -
ee
r
(
c
f
. Adams 1973: 175-178) shows a preference for bases ending in [t]:
musketee
r
,
profiteer
,
racketee
r
,
etc. At the same time
,
a clea
r
preference for bisyllabic trocaic
r
bases may be observed, which would be an example of a prosodic restriction (c
f
.
cameleer
vs. *
r
giraffeer
,
profiteer
vs. *
r
gaineer
,
racketeer
vs. *
r
fraudeer
*
,
etc.
;
Raffelsiefen 1999: 231
)
. Note that this la
t
ter restriction could equall
y
well be
formulated as an output restriction (avoidance of stress clash). Man
y
other cases of
phonolo
g
ical output restrictions ma
y
be found in Raffelsiefen (1999) and Pla
g
(
1999
)
, both couched in the framewor
k
o
f Optimality Theory. An oft-quoted
example of a suffix sensitive to final stress of the base is the deverbal abstract suffix
-al
of
l
a
rr
ival
,
r
eferral
,
ret
r
ieval
,
su
r
vival
,
etc. According to Malicka-Kleparska’s
(
1992: 437) analysis of this pattern, however, this would be a spurious
g
eneralisation: final stress is said not to be relevant per s
e
,
but a mere side effect of
C
ONSTRAINTS ON PROD
U
CTIVITY
345
t
h
e
fa
c
t that -
al
preferably attaches to Latinate pr
l
e
fix-root verbs, which all happen to
h
a
ve
final
s
tr
ess.
Raffelsiefen, in her in-depth study of phonological restrictions in English word
formation, concludes that “phonolo
g
ical r
e
s
tri
c
ti
o
n
s
in
wo
r
d
-f
o
rmati
o
n ar
e
affix-
specific and are in principle independent of
the restrictions which characterize the
f
l
an
g
ua
g
e as a whole” (199: 235). An apparent exception to this statement seems to
be constituted by
h
aplolog
y
,
i.e. the avoidance of identical
p
honetic se
q
uences in
i
mmediate neighbourship at a morpheme boundary. Haplological effects in
derivation have first been noted by Dressler
(
1977
)
and have later been elaborated
o
n by Stemberger (1981), Menn & MacWhinn
e
y (1984) and, in more recent times,
by scholars working within the framework
of Optimality Theory, among them Plag
k
(
1998a) and Yip (1998). Haplological effects in morphology are found over and ove
r
a
g
ain in the lan
g
ua
g
es of the world, which seems to indicate that some ver
yg
eneral
c
o
g
nitive mechanism is at work here, but at the same time it has been shown that the
exact conditions under which it operates are
h
i
g
hl
y
lan
g
ua
g
e-specifi
c.
In hi
s
a
ccou
nt
o
f
co
ntra
s
t
ssuc
h a
s
s
trychniniz
e
/*
s
trychnize
,
classici
z
e
/*
classi
z
e
vs. *
femininize
*
/
feminize
,*
metathesisi
z
e
/
metathesi
z
e
,
for example, Plag (1998a: 202) concludes
that “the o
p
eration of OCP
(
onset
)
4
n
eeds
t
obe
r
es
tri
c
t
ed
t
o
th
ose c
a
ses
in
w
hi
c
h a
base with two unstressed syllables precedes -
i
z
e
.
” Independently of whether this is
t
he correct account or not, it serves the pu
r
pose of illustrating the bicephalous nature
r
r
o
f haplolo
gy
as a universal tendenc
y
with lan
g
ua
g
e-specific manifestations.
3
.2.2 Morphology
Patterns of word formation may also be sensitive to the presence of certain
m
orphemes in the base or more
g
enerall
y
to the morpholo
g
ical structure of the base.
This is no trivial matter
,
since several theories have denied rules of word formation
a
ccess
t
o
th
e
int
e
rnal
s
tr
uct
ure of the base, most radically the framework of
t
t
a
-
morphous morpholog
y
of Anderson (1992). Now, this position is certainly too
r
adical. The processing constraints presented in 2.2.2. and 2.2.3., for example, are
c
learly sensitive to the morphological stru
c
t
u
r
eo
f
wo
r
ds.
An
o
th
e
r
c
l
e
ar
c
a
se o
f
s
ensitivity to the internal make-up of the base which we have already seen is the
mu
t
u
al attra
c
ti
o
n
be
t
wee
n -
ation
an
db
a
ses
in -
ise
:
that it i
s
-
ise
q
ua morpheme
w
hi
c
h i
s
r
e
l
ev
ant h
e
r
e
i
ss
h
ow
n
be
th
e
fa
c
t that
ve
r
bs
in -
ise
w
h
e
r
e
thi
s
i
s
a m
e
r
e
endin
g
and not a morpheme do not automaticall
y
take -
ation
(
cf.
su
r
mise
/
*
su
r
misation
,
etc.
)
.
R
estrictions that limit the domain of a pattern to monomorphemic words are also
r
elevant here, since they oblige speakers to carry out a morphological analysis of the
base. Extensive use of this kind of restriction has been made by Fabb (1988: 532-
534), who claimed to have identified a whole range of English suffixes requiring
monomorphemic bases. More recently, Fa
bb’s observation has been taken up in
a
A
ronoff & Fuhrhop (2002: 469-485), where it is called the
M
onosuffix Constrain
t
and given the following content: “Suffixes that select Germanic bases select
u
nsuffixed bases” (2002: 473). The onl
y
excep
t
ion to this constraint, accordin
g
to
4
OCP = Obligatory Contour Principle, an anti-haplological output constraint.
346
F
RANZ RAINER
A
ronoff and Fuhrho
p
, would be -
ness
,
which is commonly attached to adjectives
ending in -
less
(
cf
f
f
home
-
less
-
ness
,
etc.). In reality, however, exceptions are far more
numerous, as Hay and Plag (in press) have pointed out, even among neologisms:
flattener
,
flattenee
,
childhoodless
,
princessl
y
,
etc. Hay and Plag therefore conclude
t
hat “[t]he monosuffix constraint is empirically inadequate and theoretically
superfluous”. The Monosuffix Constraint thus seems to have good prospects of
bein
g
the most short-lived constraint ever proposed in word formation.
B
ut the principle of opacit
y
of complex words is also challen
g
ed b
y
another t
y
pe
o
f morpholo
g
ical restriction,
vi
z
.
the supposed existence of so-called closin
g
m
or
p
hemes
.
This concept goes back to Nida (1949: 85), who pointed out that
“
[c]ertain mor
p
hemes ‘close’ the constructi
o
n t
o
f
u
rth
e
r f
o
rmati
o
n
.
” In m
o
r
e
r
ece
nt
t
imes it has been taken up again by van Marle (1985: 234-238) for Dutch and
A
ronoff and Fuhrhop (2002: 454-468) for German. In both works the closing
property is viewed as an idiosyncratic feature of certain morphemes that prohibits
further derivation. The effect of a closing morpheme would be,
j
ust as in the case of
ty
pe blockin
g
, the curtailin
g
of the domain of a more
g
eneral rival domain. One
notorious example from German is the puzzlin
g
unacceptabilit
y
of the suffix
s
e
qu
e
nc
e
X-lin
g
-in
.
Th
esu
ffix
-ling
forms personal nouns (
g
c
f
.
feig
‘cowardly’
g
ĺ
Feig-ling
‘coward’, etc.), and
g
-
in
is a suffix forming female personal nouns fro
m
m
asculine bases
(
cf. L
eh
r
er
‘teacher’
r
ĺ
L
eh
r
e
r
in
‘female teacher’, etc.
)
. Since the
do
main
o
f -
in
is normally described as very gener
a
l
and unrestricted
,
derivatives in
-ling
should be expected to be found among the possible bases, but speakers find
g
wo
r
ds suc
h a
s
Feig-ling-in somehow awkward. The explanation Aronoff and
F
uhrhop
g
ive for the unacceptabilit
y
o
f
suc
h
wo
r
ds
i
s
that
-ling
is a closing
g
m
orpheme, i.e. bears a diacritic that prohi
b
i
t
s
f
u
rth
e
r
de
ri
v
ati
o
n
.
Thi
s
l
oo
k
s
lik
e
a
b
r
u
t
e
f
o
r
ce so
l
u
ti
o
n that
o
n
ewou
l
d
lik
e
t
o
avoid as much as possible. First because
o
f our prejudice against negative restric
t
ions. Second, because closing morphemes
suc
h a
s
-ling
only prohibit further derivation, but not further inflection (
g
c
f
. Feigling
-
e
‘cowards’
)
: this would mean that all mor
p
hemes would have to bear a diacritic
[
+derivational
]
or
[
+inflec
t
ional], or presupposes some model of Split Morphology,
w
h
e
r
ede
ri
v
ati
o
n an
d
infl
ec
ti
o
n
wou
l
dben
eatly separated. Both consequences are
likely not to be acceptable to many mor
phologists. Appeal to
r
r
closing morphemes
should thus constitute onl
y
a last resort, when all alternative anal
y
ses fail. In ou
r
present case, for example, one mi
g
ht wonder whether the oddness of -lin
g
-in i
s
n
o
t a
c
onsequence of a semantic restriction on -
in
,
which seems to dislike dero
g
ator
y
n
ou
n
s.
Esel
‘ass’ has a female counterpart,
l
Esel
-
in
‘she-ass’
,
but when the word is
u
sed as an insult
,
it remains invariable: ev
e
n a
wo
man
wou
l
d
r
e
f
e
r t
o
h
e
r
se
lf a
s
an
Esel
,
and not as an
Eselin
.
Since words in
-ling
tend to have a negative connotation,
g
i
t might be worthwhile to investigate whether the oddness of -ling-in
cou
l
d
n
o
t
be
subsu
m
ed u
n
de
r thi
sse
manti
c
r
es
tri
c
ti
o
n
.
Another possibility that one might take
i
nt
oco
n
s
i
de
rati
o
n
wou
l
dbe
that th
edo
main
o
f -
in
after all is not so
g
eneral as is
t
hou
g
ht in the literature (
cf
. *
f
f
G
ästin ‘female
g
uest’, fro
m
G
as
t
‘guest’, etc.). But this
t
i
s not the place for a thorou
g
h reanal
y
sis
o
f the data presented in Aronoff and
F
uhrhop (2002). It may suffice here to have pointed out that analyses involving
c
losing morphemes are just as good as the analysis of the corresponding sector of
word formation that they crucially presuppose. Clearly more evidence is needed
C
ONSTRAINTS ON PROD
U
CTIVITY
34
7
before we can adopt the notion of closing morpheme as an indispensable tool of
m
orphological analysis.
Another feature, widely used in studies of English word formation, is
[
±Latinate], introduced b
y
Aronoff (1976: 51). Accordin
g
to Pla
g
(1996: 778), fo
r
e
xample, the followin
g
‘Latinate Constraint’ would be operative in En
g
lish word
formation: “Bases and affixes ma
y
combine onl
y
if their et
y
molo
g
ical features are
c
ompatible.” According to this constraint, a base marked [+Latinate] could only be
c
ombined with an affix marked [+Latinate
]
or [±Latinate
]
, a base marked
[
–Latinate] only with an affix marked [
–
L
atinate
]
or [±Lati
n
ate]. Since s
p
eakers
c
annot be supposed to know the etymology
o
f words
,
this would be a diacritic
feature attached to morphemes. Accordi
n
g to Plag, speakers have to learn this
feature of morphemes by observing their combinatorial behaviour. But if they have
t
o learn the combinatorial behaviour of affixes and bases an
y
wa
y
, is there an
y
further need for the Latinate Constraint? I know of no reall
y
convincin
g
demonstration of its indispensabilit
y
. On a priori
g
rounds, it would be desirable to
be able to dispense with this constraint, which is not a constraint on single affixes
–
o
nly the features would be located in concrete morphemes, not the constraint itself
–
but a supra-affixal constraint on English word formation as a whole. If it really
t
urned out to be necessary, it would c
ontradict the null hy
pothesis according to
y
which English word formation is constrained by affix-specific restrictions and
processin
g
constraints onl
y
.
3.2.3 Syntax
W
ord classes – or syntactic categories, as
t
hey are also called – have always been
c
onsidered of prime importance for the delimitation of the domain of patterns of
word formation. The frequent use of terms such as
denominal
,
deverbal
or
l
deadjectival
is sufficient to prove this point. In the
g
enerative literature, it is
l
c
ommon practice to use syntactic category features instead of syntactic categories,
which allows, for exam
p
le, to characterise as [+N
]
the domain of an affix that
attaches to both nouns and adjectives, and so, it is argued, to preserve a
g
eneralisation missed by the alternative characterisation relying only on word
c
lasses. This advantage, however, could b
e
more apparent than real. Normally, an
affix that attaches to both nouns and ad
j
ectives does not attach to all members of
those word classes indistinctl
y
, but onl
y
to some subset. Now, if the denominal and
dead
j
ectival subset ma
y
be characterised b
y
one com
m
o
n feature X
,
it is this
co
mm
o
n f
e
at
u
r
e
X
w
hi
c
h
w
ill
be
r
e
l
ev
ant f
o
r th
ec
hara
c
t
e
ri
s
ati
o
n
o
f th
edo
main
o
f
o
ur affix. A case in
p
oint would be the Italian abstract suffix
-
aggin
e
,
which, roughly
s
peaking, attaches to both adjectives (
cf
f
f
s
tu
p
id
o
‘stu
p
id’
ĺ
s
tupidaggin
e
‘stupidity’) and nouns (
cf
f
f
asino
‘a
ss
’
ĺ
asinaggin
e
‘stupidity’) referring to negative
h
uman qualities. The common denominator here is semantic – negative human
quality –, while the word class of the base does not seem to be directly relevant (c
f
.
Rainer 1989: 46-48
)
. If, on the other hand, no common denominator could be found
between the denominal and dead
j
ectival subsets, the feature [+N] would not be of
an
y
use either, since it could at most serve to
d
is
g
uise the fact that we have to do, at
l
east synchronically, with two separate patterns.
348
FRANZ RAINER
That the im
p
ortance of word classes in the delimitation of domains of
p
atterns of
word formation tends to be overestimated had already been pointed out by Plank
(1981), and also by Adams (1973), who, in her analysis of -
ste
r
,
arrived at the
c
onclusion “that the word-class of the stem and its paraphrasable relation to the affix
are perhaps less important in determining the way a group develops than those of
s
ound and meaning” (p. 175). More recently
,
in Plag (1998b), a study aiming at a
u
nitar
y
semantic anal
y
sis of the suffix -
ise
,
it has even been ven
t
ured “that with any
t
t
g
iven productive affix, the s
y
ntactic cate
g
or
y
of potential base words is onl
y
a b
y
-
product of the semantics of the process” (
p
.
237
)
. Future research will show to what
e
xtent the job of word classes may be taken over by semantics in the delimitation of
t
he domain of
p
atterns of word formation.
A similar situation may also be found with respect to other syntactic features
r
elevant in word formation. In many characterisations of domains of patterns o
f
word formation use is made, for example, of the feature [±transitive]. The suffix
-
able
,
among others, is commonly said to be s
e
ns
iti
ve
t
o
th
e
tran
s
i
tivity of the base:
visitable
vs. *
g
oabl
e
,
obse
r
vable
vs. *
lookable
,
etc. Just as with word classes
,
one
m
i
g
ht ar
g
ue here that in realit
y
transitivit
y
is a semantic, not a s
y
ntactic feature. The
t
y
polo
gy
of restrictions one arrives at is thus to a hi
g
h de
g
ree theor
y
-dependent,
especially on the borderline between syntax and semantics.
3.2.4 Ar
g
ument structure
T
he decision about what restrictions may count as referring to argument structure
i
s, of course, also highly theory-dependent: according to
t
he theory one subscribes
to, one and the same restriction may end up under the chapter Argument structure o
r
under Semantics. In this chapter we will adopt a view of argument structure as it is
c
ommon in studies of syntax, i.e. definable as the set of those highly abstract aspects
o
f semantic structure that are most relevant to syntax, including notions such as
external ar
g
ument, internal ar
g
ument, et
c
. It ma
y
suffice here to present one
anal
y
sis, Levin & Rappaport (1988), where crucial use is made in the delimitation of
th
edo
main
o
f th
esu
ffix
-er
of the notion of
r
external ar
g
umen
t
. Com
p
lex words
formed with this suffix are traditionally referred to as agent nouns, but Levin and
Rappaport argue that the correct definition of its domain should not refer to the
s
emantic role Agent but rather to the notion of external argument: “-
er
nominals are
r
o
nly derived from verbs that have external arguments, and they always refer to the
e
xternal argument” (1988: 1068). This genera
l
isation, according to the autors, allows
o
ne to account, among other things, for the fact that -
er
nouns may also denote
r
i
nstruments, but onl
y
if the instrument is able to occup
y
the position of the external
ar
g
ument of the base verb (c
f
. Dou
g
opened the can with the new
g
ad
g
e
t
/
The new
g
ad
g
et opened the can
/
the opener
vs.
r
Bill ate the meat with the fork
/ *
k
The
f
ork ate
the meat
/ *
t
the eater
‘fork’, etc.), or that only the unergative subclass of intransitive
r
v
erbs may take our suffix (cf. speaker
,
jumper
,
etc. vs.
*
d
isappearer
,*
existe
r
,
*
collapser, etc.), since only the verbs of this subclass have an external argument. On
t
he basis of their analysis of -
er
Levin and Rappaport even venture the more general
r
h
ypothesis “that productive morphological processes do not refer to semantic-role
l
abels but rather to notions defined over ar
g
ument structure” (1988: 1080). Whethe
r
C
ONSTRAINTS ON PROD
U
CTIVITY
349
s
yntactic argument structure is really relevant for the delimitation of the domains of
patterns of word formation, however, is a question that may not be considered as
being definitively settled. Ryder (1999), for example, questions the empirical
predictions of Levin and Rappaport (1988) and proposes an alternative, purel
y
se
manti
c
a
ccou
nt
.
3.2.5
S
emantics
The next anal
y
sis to be presented here, Barker’s (1998) stud
y
of the suffix -
ee
,
is
c
onsidered b
y
the author himself as involvin
g
ar
g
ument structure, in a more
s
emanticall
y
oriented conception, but most scholars would probabl
y
view the same
fa
c
t
s
a
sse
manti
c
tout cou
r
t
.
His analysis resembles Plag’s analysis of -
ise
that
we
alluded to in 3.2.3, in that it tries to give a unified account of what had hitherto been
c
onsidered a fragmented rule, and in its focus on the output rather than the input.
C
omplex words in
-ee
,
at first sight, seem to constitute a rather heterogeneous set,
since they may refer to the direct
o
b
j
ect associated with the verb (cf. employe
e
)
, to
t
he indirect ob
j
ect (cf. addresse
e
), to the ob
j
ect of a governed preposition (c
f
.
l
au
g
hee), to the sub
j
ect (c
f
. escapee), to a referent with no ar
g
ument relation to the
v
erb
(
c
f
. ampute
e
), and in a small
g
roup the base is even nominal (c
f
.
f
estschri
f
te
e
)
.
I
t i
s
har
d
t
osee
h
ow
th
ese v
ari
ous uses
cou
l
d be subsu
m
ed
u
n
de
r a
u
nifi
ed
a
ccou
nt
using syntactic argument structure. Barker th
e
refore shifts the analysis to semantics
and
p
ro
p
oses a unified account that im
p
oses
t
he following three semantic restrictions
o
n possible words in -
ee
:
th
e
r
e
f
e
r
e
nt
o
f th
e
-
ee
noun must be sentient
,
have
participated in an event
o
f the type corresponding to
t
he base
,
and lack volitional
co
ntr
o
l
ove
r th
eeve
nt
.
Semantic restrictions, however, ma
y
also relate to the input: a case in point
would be the Spanish relational suffix -
uno
,
which is attached almost exlusivel
y
to
n
ouns referrin
g
to animals (
cf
f
f
vaca
‘
cow
’
ĺ
vacuno
‘relatin
g
to cows’, etc.). It ma
y
c
ommonly be observed in word formation that words of the same semantic category
t
e
n
d
t
oc
h
oose
th
es
am
e
affix
.
3.2.6 Pra
g
matics and Sociolin
g
uistics
J
ust as single words, also patterns of word formation may be sub
j
ected to all
k
inds of pragmatic and sociolinguistic restrictions. Some patterns are confined to
i
nformal situations, others are neutral, still others stilted, etc. Competent speakers
k
now in which situations to use which patterns. Beside such output conditions, one
m
ay also observe, though less frequently, pragmatic or sociolinguistic restrictions on
the input. Learnèd affixes, for example, are often limited to learnèd bases, and when
the
y
are applied to ordinar
y
wo
r
ds,
j
ocular effects ma
y
arise.
4
. FINAL REMARK
S
C
onstraints on patterns of word formation constitute an intri
g
uin
g
field of
r
esearch where much has already been don
e
bu
t
s
till m
o
r
e
li
es
ah
e
a
do
f
us.
A
s
a
350
FRANZ RAINER
c
om
p
arison with the last state-of-the-art re
p
ort
(
Rainer 2000
)
5
shows
,
it is a field in
c
ontinuous movement which a
pp
ears to b
e
highly dependent on linguistic fashions
and innovations. Some of the
p
roblems that dominated the discussion in the
s
eventies and the eighties,
vi
z
.
the search for constraints located at
UG
and level
o
rdering, seem to have gone out of fash
i
on during the last decade. Other very
promising issues have come to the fore only recently, especially the investigation o
f
the influence of processin
g
, which is a consequence of the
g
reat pro
g
ress made in
ps
y
cholin
g
uistics. Theoretical innovations like the theor
y
of ar
g
ument structure in
the ei
g
hties or Optimalit
y
The
o
ry
in the nineties, could not fail to leave their traces
also in our field. The first of these has led to finer analyses of the constraints on
deverbal formations, while the latter has boosted research on phonological
c
onditioning. Since both of these theories a
r
e
characterised by a dynamic evolution,
analyses here are bound to evolve at the
p
ace of the general theoretical discussion.
Positive side effects on our field of study are also due to the renewed interest in
s
emantics (and cognition) in linguistics. And last but not least, our field of research
h
as also
g
reatl
y
benefited from technolo
g
ical pro
g
ress, which has made available
m
asses of data unima
g
inable even two decade
s
a
g
o, such as electronic dictionaries,
l
ar
g
e corpora of spoken and written lan
g
ua
g
e
o
r the Internet. No field of lin
g
uistics
c
an draw more profit from these new tools than the study of word formation.
Fr
an
z R
aine
r
Vienna University of
Economics and
B
usiness
A
dminist
r
ation
Vienna
Aust
r
ia
e-mail:
f
ranz.rainer
@
wu-wien.ac.a
t
REFEREN
C
E
S
Adams
,
Valerie. 1973. An Introduction to Modern English Word-Formation. London: Longman.
Anderson, Ste
p
hen R. 1992.
A
-Morphous Morpholog
y
.
Cambridge: Cambridge University Press.
Aronoff
,
Mark .1976. Word-Formation in
G
enerative
G
rammar. Cambridge, Mass.: MIT Press.
A
ronoff, Mark and Fuhrhop, Nanna. 2002. “Restricting
s
uffix combinations in German and English:
C
losing suffixes and the monosuffix constraint.” Natural Language & Linguistic Theory 20
,
451-490.
A
ronoff, Mark and Sridhar, S. N. 1983. “Morpholo
g
ical levels in En
g
lish and Kannada or Atarizin
g
Reagan.” In: Papers from the Parasession on the Int
e
r
play of Phonology, Morphology and Synta
x
.
C
hicago Linguistic Society, 3-16.
Baa
y
en, Harald and Schreuder, Robert. 2003.
M
orpholo
g
ical Structure in Lan
g
ua
g
e Processin
g
.
Berlin/New York: Mouton de Gru
y
ter.
Barker, Chris. 1998. “Episodic -ee in En
g
lish: A th
e
mati
c
r
o
l
eco
n
s
traint
o
n n
ew wo
r
d
f
o
rmati
o
n
.
”
Lan
g
ua
g
e
7
4
,
695-727.
Bauer
,
Laurie. 1983.
E
nglish Word-Formation
.
Cambridge: Cambridge University Press.
Botha
,
Rudolf P. 1988. Form and Meaning in Word Formation. A Study of Afrikaans Reduplication
.
C
ambridge: Cambridge University Press.
5
Which was actuall
y
written in 1992.