Northeastern University

524
JOllrnal of 1'1\'('/wlill):lIi,Hi(' Researc/r, ~'ol, 9, No, 6, 191W
-ity: A Study in Wor(1 Fornlalion
Reslri(~lions
Janet H. Randalll
A('('('pll'ci
All/I'
}.J, 1979
am\,
SlIhj"('/,I' jllcig('(l l/r,' a('{'('plahilil), 01' P,I"lItiO lI'ortiJ formed from a Jlem, an oplional
(/11" -ity, Til<' "',1'/1/1.1' 1"011' l/ral I"O,\(' 1I"IfIl.\ ill II'/ric/r -ity II'I/S ,atijacel/I 10 a If/III/til"
1/10r,,/l<'me "'1'1'1' mort' a('('(,,,lahlt, I"al/ l/ro.l'" iI/ ... /rich il lI'as nOl, Ol/rer .fewlo", III"",
illfll/"IIn'til"!' (/l'{'('rllahilily /'alillgl 11,'/,,' jel/lliliarily all" /",mogel/cily "I'I/">rpl/{',",, "'" ,
/" Ih(';r ('I'lI/IIalil'n,\ 01 ,h(, pon·,."/,' 11",,.41\. \11";('('1,\' WI'fl' ,H"I.\';I;I'(' 10 hi.\/"f"I( Ii/ I r;,
C/i\li"clio",\ j)/ ,\/f'"'' tl1ld df!i\(·\
INTRODUCTION
One feature of human language which distinguishes it from other kinds llf
animal communication systems is what has been referred to as its
"creative aspect" (Chomsky. 1965, 1975). As tluent speakers. we ha\c
the ability to make up and understand new expressions. In particular. \\ e
can form new words. by combining the sets of already existing stems and
affi.xcs of the language in novel ways.
When we construct these forms. however. we do not do so imliscri­
minately. We prefer certain combinations to others. For instance.
Aronoff (1976) claims that of the two possible words "perceptivcne<;~"
and "perceptivity." native speakers will mm;t probahly prefer "rL'l'l'l'I'
IDepartment of Lingtlistic~.
Massachuselts 11100.',
University of Massachusetts
at
i\mhc"t.
\""1<'1.1
52-'
tl4~')(lt,I~I~ ~,. llli(\(I'\'I·n ~no/n
I
)11)0(0 1'1<'1111111 Pllpll ... hl!l\~ (,,,\,,It ""
"
Randall
tivcncss" (p. 37). Thesc intuitions, which speakers share about word
prcfercnces, correspond to the "productivity" of the allixes of the
languagc. That is. onc will bc more apt to coin a ncw word by using an
affix which appcars commonly in existing words than by using one which
IS rarcr.
A rcccnt experimcnt by Aronoff and Schvanevcldt (197M) supports
this view. Subjects indicated which of two pseudo words "sounded
bcttcr," Each was composed or a stem and one of two competing atlixes.
-1/(',\,\ or -ity. Subjects prcfcrrcd thc words in -1/('.1'.1'. illustrating that
speakcr<;' judgcmcnt<; of productivity do indecd corrcspond to thc actual
rclativc productivity of these allixes in the language (Aronoff, 1976. ch,
31.
One question to be answercd is how speakers provide these intui­
tions: in clTect. how allixes become more or less productive. We can
hegin to explore this question by noticing that our judgements of
morpheme combinations are sensitive to a number of features. including
the historical origin of the stem.
Certain allixes combine more readily with stems belonging to one
historical class than to another. For example. -hood is more often
attached to native. Anglu--Saxon stems. as in "motherhood" and "neigh­
horhood." While -ity sounds more natural when combined with a Latin
hase. as in "identity" and "urbanity" (Aronoff, 1976. ch. 3). We can
ohserve this preference in novel forms as well: "crocinity" seems bcttcr
than "toothity," despite our unfamiliarity with the stem crocin-.
Scholars have differed on just how subtle our distinctions are.
Jespersen (1954) discusses a discrimination among Greek. Latin. French.
etc .. whereas Chomsky and Halle (1968). Aronoff (1976), Siegel (1978),
and others discuss a division into two major classes: a classical or
"latinate" group composed of Greek and Romance forms. and a "nativc"
group. We will adllpt the former position as a working hypothesis.
following Jespersen. because it makes the finest difierentiations. How­
ever, as will be discussed later, the results of this study suggest that the
latter division may be adequate.
Distinctions of this kind are not limited to English morphology.
Selkirk and Dell (1978) argue for the existence of a morphological feature
I ± learned) which serves to divide the suffixes and roots of French into
two classes. A phonological rule of vowel backing operates with respect
526
iIy: Word Formation Restrictions
Randall
52~
I\n:THOI>
to this,flvision; the learned class undergoing the rule, the non-learned
class l1u1 suhject to it~ The two classes of morphemes cannot "e
distinguished on purely phonological grounds.
The purpose of this study is to investigate the degree to which
English speakers distinguish among hases of different origin in adding
aflixes to form pseudo-words. Further. the stlldy deterlTlincs whether
'1llhjects arc sensitive to conditions 0\1 word formation rules "a'it'd upon
thi, distinction proposed in Sil'gl'l C197H).
In Sit'gel's framework, there i, a (onditilll1 permitting -ity to alladl
only to latinate hases. The condition states that in order for -ity to attach.
the immediately preceding transformational cycle (Chomsky and Halle.
196X) mllst uniquely contain a latinate form. In more informal terms, a
word becomes availahle for -ity attachment if and only if the last
morpheme attached (on either side of the stem) is latinate. If nothing ha ..
heen attached. the stem itself must be latinate.
This restriction would make the following claims. A configllration
.. uch as cOIl-centr-ity would be allowed, since here, -ity has been
attached in the cycle immediately following the attachment of ('011-. a
latinate form. Although -ity is not contiguous (or "physically adjacent")
to a latinate morpheme. it is cyclically adjacent to one. In contrast. a
contiguration such as poly-saner-ity would be disallowed. Although the
morpheme contiguous to -ity is latinate. poly-. attached in the immeJi­
ately preceding cycle, is not.
In addition to testing the adjacency condition and its superiority to a
condition on physical adjacency, this study will examine other aspects of
our ahility to recognize and lise etymological distinctions in \\ Illd
formation. One question to he answered is on the effect of cla:-;s
homogeneity. Will a word which is composed entirely of 1ll0rphl.'OIC"
"l'iol1ging to the same historical dass he nHlre ac(cptahlc than ()ne \\ IW"l'
morphemes derive from different classes'?
Another issue which will be examined is the eftect of stem familiarit\
on acceptability jlldgements. Will words formed out of stems that we h:1\ t'
heard before he more acceptahle than those which contain lInfamiliar
morphemes?
Finally. we will examine the suhtlety of speakers' intuitions: do the\
hOllor the bipartite distinction (classical vs. native) of Chomsky and Halle
and others. or the additional subdivisions suggested hy Jespersen?
Materials
A list of pseudo words of the 10 types shown in Table I was
constructed. Examples appear in the Appendix.' Each word was formed
from a hase and the suffix -ity. The hases were of three types: simple (a
and h), stem + sutlix ({'-.t), and prefix + stem (.l?-i). Each was varied with
respect to whether its morphemes were latinate.
The latinate stems were of two Iypes: LF (Latin stems whid1 arc
familiar and have heen incorporated into English) and LU (Latin stems
which have not been borrowed and are unfamiliar to most English
speakers). The non-latinate stems were either Greek (G) or Anglo-Saxon
(A). and were drawn fl'(nn the corpus of English morphemes.
The latinate prefixes were ('011- land its variants com- and col-) and
t'1/- Um-. i/-, ir-). The non-latinate prefixes were syn- (sym-, syl-) and poly-.
The suflixes -al. -i<' (latinate). and -osis and ~fil/ (non-latinate) were used.
Procedure
Six list s were dist ribllted each of I} students, a total of 54 subjects.
Each of the 4() word types (the 4 stem types, G, A, LF, and LU, combined
with the 10 atlixation possibilities) appeared once on each list. No stem
Was used more than once on a list. and 14 real words which corresponded
to the phonological shapes of the test words were scattered among the
items.
The suhjects were asked to listen to the words spoken hy a reader (to
assnre uniform slress .Issignmcnt), and give each a rating of I. 2. 3, or 4: I
being perfectlv acceptable. 4 being unacceptable.
It was explained that a \\ord \\;lS to he judged "acceptahle" if it
"sounded" like a possible (if morphologically complex) word of English.
Examples of the following kind were given: "kindness," "luctality."
"dentosis ... The students were reminded nol to judge the w(lrd-; on their
meanings. bllt on their phonol~lgical :Icn'plahility. Sample "unaccept­
ahk" words were given as well: "walkness," "cheekosisness,"
Examples of words which might rate a :; or a 3: "wonderfulness,"
"knowledgeahleness" (2): "refinedness," "rapedness" (3).
ITwo additional types, composed of a stem and an affix without the putative latinate
re,triction were also tested. The data are nOI reported here.
Randall
528
527
ily: Word Formation Re<;trktiOl1.~
These comparisons appear in tables associated with each of the following
sections. Every difference claimed met the 0.05 level of significance.
•
Table I. t\cccrtahili1y Ratings"
-------;~~-.-------------------.. structure
1\ I'C
Aflhc'
Morpheme
fJatinale .] .
l<;tem
• ,/."
(l
J'
non-Ialinate
[ stem
;,
/(\.
.
\;lli~ate
Patin;llc ,]
-01
[
-I('
L'tC[![
,un IX rnon-tatinal':
[ lstcm
,I
-.fid
~(}si.\'
"
il\' lalim.ltc
sum,,;
J.
1n(ln~latinate 'J
ill'
sufhx
lrrellx
{,,,I\' -
:!.OR
0, III
:!.4R
11.10
~.Ih
\Ill'
For words composed of stem + it)' (types a and h), those with latinate
stems were preferred over those with non-latinate stems, as comparison
I in Table" indicates. Comparison 2 shows that the a(LF) words (type (J
words with LF stems) are significantly more acceptable than the a(LU)
words (J .85 vs. 2.3 t). There was thus a tendency for items with "familiar"
stems to be judged more acceptable. This point will be discussed later.
Comparisons 3 and 4 determine whether "latinacy" of the a(LU)
words affected the subjects' preferences. The Anglo-Saxon (A) stems.
when comoincd with -it.\', were less preferred than the LU stems (2.64 vs.
:!. ~ I) but the difference did not reach significance (0.10 < P < 0.05). The
!.~h
stem
lUl~
, .!Il
O.\\',
'lib
\1\)":'
.
Table II. Preferences Among Stem + ity Types
Compari,on
J
J
palir~alc [+ non-latinall~l
h
Latinate Stems
1/"
patinate [. latinate].] ity
\prefix
stem
('f)ll­
il/­
1
rl.alinalc .] non-Iatinat.: .] , ,ullix
/IV [ lstcm
rn.on-latinate
[ lstcm
l'
.J
S.l
Mcan
in
:.~q
a
!.7:'
'.110
tl
:'!
J
tl~
II (IX
6
7
8
IYli­
rmm~latinale [+ non-latinate1,]
Lrrehx
stem
J
it."
~.57
(1 n~
9'
10
II
1'.,11.," inc uf-.le". \\"HI!\ pc .. j!".:n I" il'':: in the Ie'!. the ,tati'li" I,'" ~,'[h
(An~,,>-~"ax(}n) -".~i\ \(;reeK) are Ji~l'laye.l: in the fotltno tes the -".f;.f statislic' are !!i' en
"f kl c. and in the
lui
alone. _ .
OJU
I'
r--
J
[latin",,' , non-latinate
,tem
,ullh
fJ '. ,\\ \
If"
Mean difference h
0.0­
'~*
rnon-Ialinate [. latinate].] it."
Lrrelix
stem
Word types
('
b
aILU)
hiM
b(GI
e
g
d
h
i+d
d+h
h
I~
t'
IJ
14
K
r~
1 11
a(LF)
a(LU)
o(LU)
d
c-Ii
I
j
e+h
i4 e
O.4(Y'
··0.46"
0.:13
I).
no
-1.lW'
-0.43/1
-0.7oa
0.16
-0.2811
-0.56<1
-O.~711
Ii
h
-\l.:!O"
t'
. n.MII
d-e
-(l.f)'"
() 15
I
"I' "
(Ul~.
"A negative lIitlcrence means Ihill fhe lirs' "<1rd Iype "as preferred over the second.
'Type e (only -".,i.f)
9
RESllLTS
The acceptahility ratin!!s of the word types displayed ill Table I Ullld
s
in some cases, averages of the ratings of several word lype ) wert"
\:ompared lIsing. I tests. Lower ntlmhers reflect ~\ higher acceptahilil\ .
10
!4
15
if" "I'".
II
d. I,
d
I' .,)
('
d-e
(l.IQ<'
(lAo"
--O.o~tI
--0.44"
ity: Word Formation Restrictions
~2q
Cireek-haseli stems, h(G). had the same score as the lI(LU) stems C. '\ I)
",uggestllig another effect on acceptahility-that of "classicality." II
appears that the Latin and Greek stems. or "fancy" stems as oppo,ed til
"plain" mach. 1978). do form a super set, consistent with the claim,
of Aronoff (1976). Chomsky and Halle (1968). and Siegel (1978).
Cyclic Adjacency vs. Contiguity
According to Siegel's cyclic adjacency theory, word types in which a
latinate morpheme occurs in the cycle adjacent to -ity should be more
acceptable than those in which .a non-latinate morpheme Occurs in the
adjacent cycle. In comparisons I and 5-7 (Table Il), this is indeed the
case. (Comparison 8 shows a very small mean difference, but this re"lIt j,
nonsignificant.) It is important to note. however, that 5 and (, Illay not ~l'
clear tests of the cyclic adjacency theory. In these compari,ons. a
homogeneous sequence (latinate stem + affix + ity, i.e .. whele ;"1
clements arc latinate) is hcing contrasted with a non-holllogclIcl1l1'
concatenation. The effect of homogeneity will he discussed in thc nnl
... ection.
A set of more fruitful comparisons may he 9-11, in which hoth of the
compared word types are made lip of a mixture of morpheme clas...e.... In
comparisons 9 and 10. combinations of word types have been examined to
discern the effects of physical (or contiguous) and cyclic adjacency. In
comparison 9. a dilTerence favoring i and d would indicate an etTect of
physical adjacency. A difTerence favoring d and II in comparison 10 would
indicate an effect of cyclic adjacency. Both comparisons were significant.
indicating that both factors were etTective. Comparison II compares the
magnitudes of the two effects. It contrasts words in which -ity is attached
cyclically adjacent to a latinate morpheme with words in which it i...
attached physically adjacent to a latinate morpheme. (Two words pf thi,
type would he 1<'0" [+tootll[ + I ity and [JYfI [+f1ahl] + I ity). In this
comparison, the cyclically adjacent strings (type II) were preferred over
the physically adjacent ones (type i). These three results taken tl'gl'tht',
Indicate the primacv of cyclic adjacl'Ill'Y o\'er physical adjaccnn
Homogeneity
A factor which seelllcli to infll1ence acceptahility slighth \\a~ the
hOlllogeneity of the classes constituting each word type. Comparisons 12
and 13 illustrate this effect. The types which contained non-latinate stem,
~JO
Randall
(ti and II) were somewhat less acceptahle than their non-homogeneous
counterparts Ie and R). Both of these comparisons are controlled for
adjacency: all four word types have a latinate morpheme cyclically
adjacent to -ity.
It is important to note that the effect of homogeneity in both of these
cases is small, in fact it is smaller than any of the significant differences due
to adjacency. This fact is consistent with the word formation preferences
which we have heen examining. If. in fact, word formation is a process
dependent on the presence or absence of restrictions on each affix.
dictating its attachment possibilities. then another process of scanning
words for homogeneity would be redundant and possibly contradictory.
There is another fact. illustrated hy comparisons 12, 14. and 15,
which corrobordtes this hierarchical relationship of adjacency over
homogeneity. The ditTerence between word types (' and d (0.20) is
significantly smaller than the difference hetween d and e (0.84). as is clear
from comparison 15. In the first pair. d differs from (' in that d is
non-homogeneous while (' is homogeneous. The adjacency facts in each
case are the same. In the second pair. hoth word types arc non­
hOlllogeneous. hilt d has a latinate morpheme adjacent to -ity (in this case
it is a suffix. and is therefore hoth physically and cyclically adjacent).
Adjacency. then. is re\ ealed as a more significant factor than homo­
gencitv.
LF and LU
The category "latinate" to which we have appealed comprises two
kinds of latinate word types. The LF (Latin familiar) words are those
composed of latinate morphemes which have been incorporated into
English. Examples are recto /Iloll. and plt'll. LU (Latin unfamiliar) words
are made up of Latin stem'> which English has not borrowed. for example.
Rldhi". crocill. and llit.
A plausihle hypothesis might state that LU words should not diller
from LF words in their ahility to undergo -ity attachment. if in deciding on
a word's latinacy we consider only the phonological features and stress
assignment of words. Table III illustrates that this is not the case.
In all hut one of the cases in which the scores ditTered. the LU words
were given a score higher than the LF words. Apparently we do not
evaluate a morpheme for latinacy based purely on phonological and stress
criteria. Since these are identical in the I.F and LU words. there mllst he
other factors causing our jlldgements to favor words in the LF class.
531
!Ol
ity: Word Formation Rt'!Itrictlons
-l
Affix
Type
II
f' '
Mean
DitTercm.:e
2.31
-0.46"
IGAI + LFI+ lit)'
2,91
2,1)3
IGAI + I ,tIl +-lity
2.91
3.54
. (). \) I II
con·
ILl + LFI + lit)'
2,31
2.114
11.[ +LUI+ lit)'
3.19
-O.R7"
·mis
:/111
111.1' +-IGA + lity
JJlI
3, II
IlLU + IGA + lity
2.96
3,67
..
II LF +-IL + lit)'
1.81
2,(}7
II LU + IL+ lity
2.28
2.46
.;..
-til
TOlal..
Composilion
(LU + lit)'
H'II-
,.
Mean
I.R~
Iiolv-
('
Composition
These results point to an efTeet based on "classicality." It is unclear
at this point how one would assign a word or morpheme to a classical
category; children. when learning a new word, are not explicitely taught
this distinction. We may make a distinction between the more common,
everyday words and those that are rare or specialized in their usc, but
there may be phonological determinants as well. Native words tend to be
monosyllabic, and if polysyllabic. receive initial stress. In addition. we
may recognize certain spelling patterns as belonging characteristically to
one group or another. In general. ung and initial yare non-latinate while
words containing qu sequences are latinate. kn is an orthographic
sequence particular to Anglo-Saxon words. and words beginning with psy
are of Greek origin.
It is not possible. from this study. to draw firm conclusions ahout the
influence and psychological reality of etymological-phonological factors
on word formation; but the results here do suggest a native speaker's
ability to differentiate in some way among these classes. Determining the
criteria that we use is one problem for future research.
LU
ILF+lity
{l
"
.. Cla!'i!'iicalily"
.Table III. Acccptability Ratings of Latinate-Stemmcd Forms
LF
2.42
2,04
2.90
Randall
().oo
(Ul()a
(1.()4"
(1.~6"
. 0,47"
0.0<;,
DISCUSSION
The results show that people arc sensitive 10 distinctions in historical
origin among the morphemes of English. Of all the parameters innuencing
word formation preferences that we have examined. "classicality" is the
most salient. although familiarity has an dIec! as well. As it has been
estahlished that native speakers can distinguish between those forms
which are classical and those which are not. we can now use their
intuitions to test the claim that there exist "free" unrestlicted atlixes and
to discern whether restrictions based on factors besides latinacy operate
in word formation rules.
We may further conclude that adjacency of latinate morphemes
dearly int1uences acceplahility. and as Siegel proposes, cyclic adjacency
is more important than physical adjacency. Homogeneity. although a
recognizable factor. was less innuenlial than either classicality or adja­
cency.
It is not clear to what extent phonological form contrihutes 10 the
morphological divisions we make. But the tendency for LU words to be
preferred over the familiar-stemmed Anglo-Saxon neologisms suggests
that this is it question w(lrth examining.
I-'amiliarity may be important. The LF stems were familiar. whcrea~
the LU stems were nol. But this effect. while it may have inlluem:ed thl'
ratings of the LF and LU stems. cannot uniformly account for the results.
average normative Kucera and Francis (1967) frequencies were O. ~:"
..\9, and 50 for LU. Greek. Anglo-Saxon. and LF. respectively. while the
mean acceptahility ratings were :'.60. :'.~4. :'.7:'. and :'.0:' for these f(1l1r
da""es.* The failure of familiarity to correlate with these ratings indicalc~
that there must be still other factor'i operating. The generally higher
acceptability ratings given to Greek and latinate stems than to Anglo ..
Sa xon stems suggests one possihle faclor. While the familiar Greek slem~
\\ere 'iignilicantly preferred over the unfamiliar LU stems (2.34 vs. 2.60."
.- 0.(5) they were also preferred to the even more familiar Anglo-..<iaxon
'-Iellls (:'.~.t VS. :'.72. p < 0.05). Furthermore. the LU class tended 10 he
mol',' acceplahlc than the Anglo-Saxon stems (260 vs. :'.7:', . J() < f1 < (UI" I.
: I he'e fllI:an, relleclth", nVl'rall acceplanility of "orJ~ wilh eilh.:r a latinate aflh
·ill. Iype, 0. h. c. d. f,!, anJ fr.
"I III>
"IIi,
~
".
5.'-'
ity: Word Fnrm:dion Restrictions
REFERENCES
The matter of what factors condition word formation has emerged ii'
a C(lIllr'~" onc. This study has cstahlished what some of these factors arc.
and hopeflllly, has clarified some areas for flllllre research in the licltl
APPENDIX
")pe_
11.(7)
II
(LU)
h
1Ci)
(A)
,.
(I.F)
!I.U)
d
(G)
(A)
t'
(LF)
(LU)
Randall
5.U
i\wnoll. M, (11176), Wonl formal inn in g.:n.:raliv.: grammar. ril/glli... ti,' It''luir), U"''''l!rtll,h
No. I. M IT Press. Camhridj!l'. Mass.
AnllwtT. M.. and Schvaneveldl. R. (1117H). li·stillR Morl'lwlolliclIl Produt'liritv. Manuscripl
suhmil1ed for publication.
Bach. F. (19711). Personal Commllni<:alion.
Chomsky. N. (1%51. A.lp,·cts
(I,,' 111<,ory "f Srllftlx. MIT Press, Camhridgc.
Chom~ky. N. (1973). Conditions on Iransformations. In Anderson. S. R .. and Kirparsky. P.
(cds.l. ,1 1.'t'.Hscilri/i }"r Morri.! flallt'. Holl. Rinehart and Winston. New York.
Chomsky. N. (19751. Rl'/I,'c,iOlI.1 till I.t/II.I:IIIII:C·. Random HOlls,:. Nl'W York.
Choll"ky. N .. and lIalle. M. (1%11). lilt' SOIllIt! l'tlUr·rllofEnl'Ii.l'Ir. Harper and Row. Nt'w
York.
JC'l.pcrsen. O. (19.'14). A Modcm El1gli.l'ir (irallllllt/rol1l1i.l'lOri('(l1 Principles. Allen & Unwin.
London.
KII~cra. II. and Fral1cis. W. N. (1')t,7). COlllflll(tlfiolltll AII(lIY.I'is '~f l'r,·sc'lIt·O".I' AIIl('fict/11
f.'II1'li.I/I, Brown Universily Press. Providence, R.1.
Selkirk, E. 0 .. anJ Dell. F. (1978). On morphologically governed vowel al1ernations in
French. In Recent transformalional studies in European languages. l.illl'lIi.l'(i,· III</Iliry
MOllogrllph No.3. MIT Press, Camhridge, Mass.
SIegel. D. (197HI. The adjacency conslrainl and the theory of morphology. In Stein. M. L
(cd.), "wcCI,dilllls ,~r tlrt' Eif.llrtlr Anllltal Mt't'tillll of the North Eastern l.illlluistics
.'10";"/1', University of Massachusetls Departmenl of Linguislics. Amhersl. Mass.
or
Type
Test items
obliquity
fortity
pervility
derosity
f
cyanity
archeity
stoutity
shrillity
!?
lacticity
orbitality
scelicity
I'll rfalit y
(G)
(A)
hypnosisity
phonifulity
heartosisity
youthflliity
conllndit~
inopportllnit)
(LU)
concrocinit~
insubidity
II
(G)
(A)
(LF)
chloricity
pentagonalit y
earthicity
moonality
labiosisity
plentifulity
umerosisity
balaniflliity
(LF)
Test items
concentrity
intheosity
condoomity
inripity
polyglutinit~
synsanctity
(LU)
polygalbinit~
synnablity
.i
(G)
(A)
polydermity
synchromity
polythornity
syntoothity
\CKNOWLEDGMENTS
I thank Charles E. Clifton. Jr. for his patient and continued discu,­
,ions. and for his critical experimental and editorial assistance. I am also
!!ratcful to Thomas Roeper and Alan S. Prince for helpful comments on an
callier version of this paper.