Variation in Finnish Vowel Harmony: An OT Account
Author(s): Catherine O. Ringen and Orvokki Heinämäki
Source: Natural Language & Linguistic Theory, Vol. 17, No. 2 (May, 1999), pp. 303-337
Published by: Springer
Stable URL: http://www.jstor.org/stable/4047991
Accessed: 18/05/2010 16:12
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=springer.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
Springer is collaborating with JSTOR to digitize, preserve and extend access to Natural Language & Linguistic
Theory.
http://www.jstor.org
CATHERINE
VARIATION
O. RINGEN AND ORVOKKI HEINAMAKI
IN FINNISH VOWEL HARMONY:
AN OT ACCOUNT*
This paperpresentsdata on vowel harmonywith disharmonicroots in Finnishwhich
show that when the last harmonicvowel in a disharmonicroot is back, in almost all
cases the only possible harmonicsuffix vowel is back, but when the last harmonic
vowel is front, there is usually variation in suffix vowel choice that seems to be
influencedby severalfactors,includingsonorityand stress.These data, whichcannot
easily be accountedfor in rule-basedtheories, can be accountedfor in Optimality
Theory.A highlyrankedalignmentconstraintaccountsfor harmonywith nativeroots
and loans in whichthe last harmonicvowel is back. Unrankedconstraints,whichtie
suffix vowel choice to stress and sonority, as well as alignment requirements,
determinesuffixvowel qualityfor the remainderof forms. Variationis seen to be a
functionof the relativefrequencywith which a particularsuffixvowel is designated
as optimalby the differentpossible rankingsof the unrankedconstraints.
0. INTRODUCTION
Finnish vowel harmony has received considerable attention in the
literatureof generative phonology because of its relevance for a variety
of issues in phonologicaltheory. Different analysesof Finnishvowel harmony, many of which include discussions of disharmonic roots, are
presented in Campbell (1980, 1981), Halle and Vergnaud (1981), Kiparsky (1973, 1981), Goldsmith (1985), Ringen (1988b), Steriade (1987),
Valimaa-Blum(1987), and Vago (1988), among others.
This paper reportson the resultsof our empiricalinvestigationof vowel
harmonywith Finnishdisharmonicroots.l Our studies were conductedin
* Portions of this paper were presented at the 1988 and 1998 LSA meetings, at Nordic
ProsodyV, 1989,in Turku,Finland;at the 81st SASS AnnualMeeting,Amherst,Massachusetts, 1991; at Kielitieteen Paivat 1990 in Helsinki, Finland; at MCWOP 1996 at the
Illinois;at the DepartmentalColloquium,March
Universityof Illinois, Urbana-Champaign,
1997, Universityof Iowa; and the Universityof Warsaw,Universityof Stuttgart,University
of Salzburg,and Eotvos L6randUniversity,Budapest,in October 1997. Thanksto Gregg
Oden for help with the statisticalanalysis, to RosemaryPlapp for assistancein tabulating
the data, and to Arto Anttila, Jill Beckman,Ellen Broselow,ChrisCuly, GaryDell, Gregg
Oden, Aimo Hakanen. Bob Harms, RichardHurtig, Anneli Pajunen, Jon Ringen, Jerzy
Rubach, Szilard Szentgyorgyi,Cheryl Zoll, and two anonymous reviewers for helpful
commentsand/ordiscussions.This, of course, does not mean that any of these individuals
bears any responsibilityfor any errorsor that they necessarilyagree with all of our claims
or interpretations.
1 Practicallyall disharmonicroots are loanwords,althoughnot all loanwordsare disharmonic. The statusof these wordsas loanwordsper se is not important.Our claimsare about
disharmonicroots, regardlessof whether or not they are loanwordsor whether speakers
recognizethem as loans.
NaturalLanguageand LinguisticTheory17: 303-337, 1999.
(g) 1999 KluwerAcademicPublishers.Printedin the Netherlands.
304
CATHERINE
0.
RINGEN AND ORVOKKI HEINAMAKI
Helsinkiwith native speakersof Finnishat the Universityof Helsinki. We
found that vowel harmonywith Finnish disharmonicroots has not been
accuratelyportrayed in the literature.2Although suffix vowel choice is
categoricallyfront or back with some disharmonicforms, there is variation
in suffix vowel choice in other cases. In almost all cases when the last
harmonicvowel in a disharmonicroot is back, the only possible harmonic
suffixvowel is back. When the last harmonicroot vowel is front, however,
most forms exhibit variationthat seems to be influencedby the qualityof
the vowel with primarystress, the quality of the vowel with secondary
stress, the quality of the most sonorous vowel, as well as the quality of
the last vowel in the root. These data, which cannot easily be accounted
for in rule-based theories, can be accounted for in Optimality Theory
(McCarthyand Prince 1993a, 1995; Prince and Smolensky 1993) if it is
assumedthat constraintsare partiallyranked. A highly ranked constraint
requiringthat the feature [+back] be aligned with the right edge of the
word accounts for harmony with native roots and disharmonicroots in
whichthe last harmonicvowel is back. Unrankedconstraintswhichrequire
that suffix vowels agree with the backness of the vowel which is most
sonorous, with the vowel with primarystress, with the vowel with secondarystress, and with the last root vowel determinesuffixvowel qualityfor
the remainderof forms. For native roots, any rankingof these unranked
constraints will give the correct result because all the vowels will be
front. For disharmonicroots, however, differentresults are possible. The
variation that occurs is seen to be a function of the relative frequency
with which a particularsuffixvowel is designatedas optimalby the different possible rankingsof the unrankedconstraints.
1.
FACTS
OF VOWEL
HARMONY
The essential characteristicsof Finnish vowel harmony are well-known
and uncontroversial.There are eight surface vowels which are listed in
(1):
(1)
2
neutral
front
harmonic
front
back
i [i]
y [y]
u [u]
e [e]
o[0]
a[a]
o [o]
a [a]
See Kontra and Ringen (1986, 1987) and Ringen and Kontra (1989) for discussionsof
empiricalinvestigationsof Hungarianvowel harmonywhich show that certainwidely cited
data are, in fact, not accurate.
VARIATION
IN FINNISH
VOWEL
HARMONY
305
In native Finnish noncompoundwords, front and back harmonicvowels
do not co-occur. Neutral or transparentvowels, on the other hand, are
found in words with either front vowels (harmonicor neutral) or with
back vowels. Some examplesare given in (2) :3
(2)
a. poiytai
'table'
b. pouta
'fine weather'
c. hamara
'dusk'
d. kasi
'hand'
e. koti
'home'
f. kesy
'tame'
g. veli
'brother'
h. vero
'tax'
i. tie
'road'
Harmonic suffix vowels usually agree in backness with harmonic root
vowels, as illustratedin (3):
(3)
a. poyta-na
'table'
essive
b. pouta-na
'fine weather'
essive
c. hamara-na
'dusk'
essive
d. kade-lla
'hand'
adess
e. koti-na
'home'
essive
f. kesy-lla
'tame'
adess
g. vero-lla
'tax'
adess
If all root vowels are neutral, harmonicsuffix vowels are (usually) front,
as illustratedby the examplesin (4):
3 We give here orthographic representations: long vowels and consonants are represented
by double letters. Primary and secondary stress are marked throughout.
306
(4)
CATHERINE
0.
AND
RINGEN
ORVOKKI
a. velje-lla
'brother'
adess.
b. tie-lla
'road'
adess.
HEINAMAKI
But loanwords often violate harmony restrictions,as illustratedby the
disharmonicforms in (5):
(5)
disharmonicloans
a. vuilgaari
'vulgar'
b. tyranni
'tyrant'
c. afaari
'affair'
d. analy'ysi
'analysis'
It is sometimes claimed that suffix vowels following such disharmonic
roots agree with the last harmonicroot vowel, e.g., Kiparsky(1973). The
situation is not this simple, however. While it is generally agreed that a
disharmonicroot in which the last harmonicvowel is back requiresback
harmonicsuffixvowels, there is variationwith disharmonicroots in which
the last harmonicvowel is front. Moreover, what the facts are has not
been clear. Different Finnish (normativeand descriptive)grammarsdo
not agree with each other or with generativephonologistswho have made
claims about which suffix vowels occur with such disharmonicroots. The
LanguageBoard of the Finnish LiteratureSociety recommendedin 1945
(Sadeniemi 1946, pp. 79-80) that in additionto i and e, y should also be
treated as neutral in loanwordswhereas o and a should alwaysbe considered harmonic.Saarimaa'snormativegrammar(1971, p. 17) presents y
as basically neutral, while o and d are said to be harmonic, especially
when they are long. Penttila's grammar(1963, p. 17) also considersy to
be neutral, but o and d are said to be alwaysharmonic.Ikola (1986, pp.
134-135) considersall the frontvowels y, o and d to be basicallyharmonic,
but states that y may also be neutral when unstressed. Campbell (1980,
p. 251), on the other hand, claims that in certain disharmonicloanwords
both front roundedvowels y and o may be treated as neutral,so that both
alternatesof forms such as those in (6) are possible:
(6)
a. y/o - neutral
b. y/o - harmonic
hieroglrfi-a
hieroglrfi-a
'hieroglyph'part.sg.
sutenoon-a
suten6bri-d
pimp' part. sg.
amatoori-a
amatoori-a
'amateur'part. sg.
VARIATION
IN FINNISH
VOWEL
HARMONY
307
In addition, Campbell, referring to Ikola (1971) and Saarimaa(1971),
claims that the forms in (6a), with a back suffix vowel, occur more frequentlyand are consideredto be better, more prestigious,while the forms
in (6b), with a front suffix vowel, are more colloquial. However, both
Ikola and Saarimaagive normative rules and say nothing about the frequencies of these forms in actual use. Kiparsky,as reported in Steriade
(1987), claims that all the front vowels y, o, d are optionally treated as
neutral in disharmonicloanwords. Thus, accordingto Steriade, Kiparsky
claims that both the forms in (7) are possible:
(7)
hydrosfaari-a
2.
hy'drosfaari-a 'hydrosphere' part.sg.
EMPIRICAL
INVESTIGATIONS
There is very little researchon what suffixvowels are found with disharmonic loanwords.4In order to find out what kind of variationoccurswith
disharmonicloanwords such as those in (6) and (7) we administereda
series of questionnairesto native speakersof Finnish in Helsinki. In our
firsttest,5the subjectswere 50 studentsin four differentcoursesin summer
school at the University of Helsinki. The courses were mathematics,
statistics, physics, and Swedish for students of math and the natural
sciences. Ages rangedfrom 17 to 49. The subjectswere all native speakers
of Finnish,and half were male (=25) and half female. Fortyof the subjects
residedin Helsinkior the surroundingarea;the other 10 were fromvarious
places in Finland.6 Subjects were presented with 29 written sentences
containingblanks and asked to supply an appropriateform or forms of a
word given in its nominative singular.No suffix identical to that appropriate for the test word occurred in the sentence frames since Kontra,
Ringen, and Stemberger (1991) found that a preceding identical suffix
can influence harmonicsuffix vowel choice.7 Subjectswere instructedto
4
One study is Levomaki (1972), who tested acceptability of front vs. back vowel suffixes
using students in the humanities as subjects. He found a great deal of variation and concluded
that, contrary to claims in Finnish grammars, the behavior of y is not the same as the neutral
vowels i and e.
S A copy of this questionnaire is given in Appendix A.
6 There was no significant difference between the responses of subjects who were from the
Helsinki area and those from other places in Finland.
7 In a study of Hungarian vowel harmony, Kontra, Ringen, and Stemberger (1991) found
that a preceding morphologically identical suffix (within a sentence) influenced speakers'
choice of suffix vowel: a preceding (morphologically identical) front vowel suffix increased
the number of front suffixes that Hungarian speakers used with loanwords, whereas a
preceding (morphologically identical) back vowel suffix increased the number of back vowel
suffixes they used. To check whether there was a similar priming effect in Finnish, we gave
another group of 50 students a test where the target words occurred in three environments:
308
CATHERINE
0.
RINGEN
AND
ORVOKKI
HEINAMAKI
preserve the singularsin their answersand to make decisions only on the
basis of what sounded more natural, not what they thought the rules of
grammarmight say.
All the test words were loanwords,but ten were fillerscontainingonly
back vowels. All subjects used back vowel suffixes for all fillers. Some
examples are listed in (8):
(8)
All subjectsused back suffixvowels for all fillers
part. sg.
'Tabasco'
a. thbasko-a
b. juirtta-a
'yurt'
part. sg.
c. rotund'a-sta
'rotunda'
elat. sg.
d. sailuun'a-sta
'saloon'
elat. sg.
e. tornado-a
'tornado'
part. sg.
There were two types of disharmonicloanwords. In one, the order of
vowels was front harmonicvowel then back vowel. The results of this
group are given in (9). We see virtuallyno variation;practicallyall suffix
vowels are back.8
a. sy'ntaksi
% Front
0.0
% Back
100.0
% Both
0.0
b. symptomi
2.0
96.0
2.0
'symptom'
c. tyranni
0.0
98.0
2.0
'tyrant'
d. foljetongi
0.0
100.0
0.0
'serial'
e. k'sta
0.0
100.0
0.0
'cyst'
(9)
'syntax'
The other type of loanwordhad a back vowel followedby a frontharmonic
vowel. Suffixvowels with these words exhibit variation.There were nine
of these words, three for each front vowel height. The results are given
in (10)
in two of them, there was a primer,i.e., a precedingidenticalsuffixwith a front vowel or
a back vowel. The third environmentdid not containany suffixidenticalto the test word.
No primingeffect wasfound.The responsesin the threeconditionsdid not differsignificantly.
8 The situationis more complexwhen a frontharmonicvowel is followedby a backharmonic
vowel and then more than one neutralvowel as in dynamiitti'dynamite'where variationis
found for some forms. Such examplesare discussedbelow.
VARIATION
IN FINNISH
VOWEL
309
HARMONY
% Front
80.0
% Back
12.0
% Both
8.0
50.0
36.0
14.0
'analysis'
marttyyri
42.0
42.0
16.0
'martyr'
b. su'tenoori
94.0
2.0
4.0
'pimp'
jongloori
86.0
6.0
8.0
'juggler'
dmatoori
82.0
10.0
8.0
'amateur'
100.0
0.0
0.0
'millionaire'
96.0
80.0
2.0
16.0
2.0
4.0
'hydrosphere'
'affair'
(10)
a. hieroglyfi
analyysi
c. mljonauri
hydrosfaari
aifaari
Gloss
'hieroglyph'
The numbersindicate the percentages of subjectsthat respondedwith a
suffix containing a front vowel, a back vowel, or two alternatives, one
with a back vowel and one with a front vowel. There is some variation
for all but one word, mfljonaari. However, the majorityof subjects responded with a front suffixvowel for all but one of the words, marttyyri.9
Some have suggestedthat the variationis relatedto prestige;one variant
has greater prestige than the other. It will be recalled that Campbell
suggeststhat the back suffixvowel "soundsbetter"and is "morefrequent"
with these disharmonicroots. However, when asked directly which one
sounded better, the majorityof our subjects respondedwith front suffix
vowels, not back suffix vowels. Given the formalityof the situation, we
would not have expected subjectsto have used the more colloquial, lower
prestige forms.
To see if we could get subjectsto respond with more back vowels, we
administeredanother questionnaireto a differentgroup of 30 studentsat
the University of Helsinki, There were 17 females and 13 males, ranging
in age from 19 to 39. This questionnairewas identical to the first except
that the subjectswere asked to use the correctforms, the ones they would
use in a formaltext. The differentdirectionsdid not cause subjectsto give
more backvowel responses.In fact, in six of the nine cases, the percentage
9 One reviewerhas suggestedthat we may have found back suffixvowels with these disharmonic roots which end in a front vowel because the fillersentencesall had test wordswith
back vowels and thereforethere was a primingeffect. Whilewe cannotestablishthat there
is no primingeffect, we do not think that there was one. First, we checked for priming
effects withinsentences (see fn. 7) and found none. Second, we find that in some cases all
suffixvowelswere front(e.g., mfljonaari) even thoughthe precedingtest wordhad only back
vowels (dnafora) and this is one of the last test wordson the questionnaire.Third,when we
have tested these samewordson other questionnaireswithno fillers,resultsare not different.
310
CATHERINE
0.
RINGEN AND ORVOKKI HEINAMAKI
of front responses was higher, not lower than on the first questionnaire.
We concludethat althoughthere is clearlyvariationin suffixvowel choice,
our data do not supportCampbell'sclaim that Finnishspeakersuse back
suffixes more frequentlythan front suffix vowels with loanwordssuch as
in (10) or his claim that the back suffixvowels are viewed as more learned
and more prestigious.
Likewise, our results do not supportthose who suggest that different
styles accountfor the variation.These includeValimaa-Blum(1987), who
claims that different rules apply in differentstyles, Kiparsky(1981) who
says that different vowels are opaque in different styles, and Steriade
(1987), who says that vowel harmony applies at different points in the
derivation in different styles. We found that the same subjects in the
same situation on the same day sometimes used back vowel suffixesand
sometimes used front vowel suffixeswith the forms in (10).10
Another explanation for the variation that has been offered in the
literature(e.g., by Halle and Vergnaud1981)is that there are two dialects,
one that uses front vowels and another that uses back vowels in suffixes
of disharmonicloans. Those who vary in their use of suffixvowel choice
are mixing dialects. But this account is not plausible either. First, there
is no evidence that we know of, nor that Halle and Vergnaudprovide,
for the existence of either dialect other than the existence of variation.
Second, even if we accept the claim that there are two dialects, the
majorityof our subjects(68%) on the firsttest sometimesused back suffix
vowels and sometimes used front suffixvowels with disharmonicroots in
which the last harmonicvowel is front. These are the subjectswho mix
dialects accordingto Halle and Vergnaud.We are still left with the question of why so many more of these subjectsuse front vowels with some
forms (e.g., hierogl'fi) than with others (e.g., marttyyri).If this is simply
a matter of dialect mixing, we would expect that hieroglyfiand marttyyri
should be equally likely to take back vowel suffixes. Finally, Halle and
Vergnaud predict that those who mix dialects will exhibit variation in
suffixvowel choice with formslike miljonaddriand hydrosfddri,but we find
essentiallyno variationin suffix choice with these forms.
Yet another explanationthat has been offered in the literaturefor the
suffixvowel variationin loanwordsis that some of the words are analyzed
by speakers as compounds. With compounds, suffix vowel quality is
determinedby the harmonicqualityof the vowels in the last member of
10 We have not, of course, shown that there are no differences in usage in different styles,
but we have shown that there is substantial variation that does not seem to be a function of
style.
VARIATION
IN FINNISH
VOWEL HARMONY
311
the compound. For example, Valimaa-Blum (1987) cites Sadeniemi
(1949), who suggeststhat subjectsanalyzeloanwordssuch as those in (11)
as compoundsbecause their stress patterns are like those of compounds,
which have secondarystress on the first syllable of the second element in
the compound (e.g., kirja#kauppa'bookstore').
(11)
appelsiini
'orange'
karamelli
'candy'
dynamiitti
'dynamite'
The results with some of the disharmonicloanwords in (10) might be
explained as cases that speakersanalyzeas compounds.This is illustrated
in (12).
hiero#glgfi
% Front
80%
ana#lyysi
50%
sute#noori
94%
ama# toori
82%
mfljo#nadri
100%
(12)
96%
hydros#faari(or hydro#sfairi)
The suggestion is that speakers who use front vowels are analyzing,for
example, dnalyysias a compound, ana#lyysi. Since the second member
of the 'compound'containsonly front vowels, the suffixvowels are front.
In contrast,those speakerswho use backsuffixvowels are usinga prescriptive rule which treatsall front vowels as neutralin disharmonicloanwords.
Hence, for these speakers a disharmonicloanword such as analyysi will
require back suffixvowels because there is a back vowel in the word.'1
There are two problemswith this account. First, we still cannot explain
the difference between the forms. Why, for example, do we have 100%
front suffix vowels with miljonadari
but 50% front suffixvowels with analyysi? Second, three of the words we tested are not plausiblecompounds
and hence are incorrectlypredictedto allow only back suffixvowels. These
are given in (13a) below. Thus, even if we accept the claim that the back
suffixes for some of the disharmonicloans can be explainedby assuming
that some speakers treat all front vowels in disharmonicloanwords as
" This position is similar, but
not identical to that of Valimaa-Blum (1987).
312
CATHERINE
0.
RINGEN
AND
ORVOKKI
HEINAMAKI
neutraland others analyzethe same words as compounds,there are other
disharmonicloans that cannot be so explained. In addition to the three
listed in (13a), we tested other words that cannot be analyzed as compounds on a second questionnaire.The results for these words are given
in (13b):
% Front
80.0
% Back
16.0
% Both
4.0
jongloori
86.0
6.0
8.0
marttyyri
42.0
42.0
16.0
64.0
32.0
4.0
58.3
29.2
12.5
(13)
a. afaari
b. kasoori
vulgaari
Finnish (non-derived)nouns usually consist of two syllables. If three syllable words such as marttyyri are analyzedas compounds,then one of their
component words must be monosyllabic. Although Finnish does have
some monosyllabicnouns, these contain long vowels or diphthongs,e.g.,
maa 'earth',yo 'right'.12 Neither the firstnor the last syllableof the three
syllablewordsin (13) fit this pattern.Finnishalso has restrictionson wordfinal consonants:no word-finalconsonantclustersare permitted,and only
dentals, i.e., t, s, n, r, and 1, are permitted word-finally - and even these
are relatively rare in the nominative singularform. Hence, four-syllable
words where the second syllable ends in a vowel are the best candidates
for a compoundanalysis.For example, there is no way to analyzej6ngloori
as a compound. While loonriis a possible Finnishnoun, jong is not. Nor is
there any other way to divide this word into two possible Finnishnouns.
In short, the 'compound'analysis hypothesis might account for some of
the variation, but it cannot account for all of it.
We did, however, find some evidence in supportof the idea that speakers do sometimes treat disharmonicloanwords as compounds. Neutral
vowels are not supposed to have any effect on the suffix vowel choice.
Yet it has been reported(e.g., by Campbell1980;Levomaki1972;Sadeniemi 1946) that there are loanwordswith back vowels followed by several
neutralvowels with whichpeople tend to use frontharmonicsuffixvowels.
We included words with back vowels followed by several neutral vowels
on questionnaires1 and 2 in order to see if this is true. These are given
in (14). As can be seen, not all of these formsshow any significantnumber
of front suffix vowels. Indeed, the only forms for which many subjects
12 yo is
a diphthong.
VARIATION
IN FINNISH
VOWEL
HARMONY
313
used front suffix vowels are the first four, those that are most plausibly
analyzedas compounds.
(14)
arkkitehti
bolshevikki
karamelli
d9naml'ltti
katrilli
fakulte'etti
a'djektl'lvi
artikkeli
fakiiri
matrikkeli
partikkeli
ateisti13
% Front
48.0
26.0
16.0
8.0
6.0
4.8
4.0
4.0
0.0
0.0
0.0
0.0
% Back
34.0
70.0
78.0
70.8
94.0
95.2
92.0
96.0
100.0
96.0
96.0
96.0
% Both
18.0
4.0
6.0
20.8
0.0
0.0
4.0
0.0
0.0
4.0
4.0
4.0
Gloss
'architect'
'Bolshevik'
'candy'
'dynamite'
'quadrille'
'faculty'
'adjective'
'article'
'fakir'
'register'
'particle'
'atheist'
It seems to us that these resultsdo not supportthe claim that roots ending
with more than one neutralvowel tend to have more front suffix vowels
than otherwisewould be predictedby the regularoperationof harmony.14
Rather, these data seem to supportthe claim that the suffixvariationthat
is foundis due to (some) speakers'analysisof certainformsas compounds.
That is, the reason that some speakers use front suffix vowels with the
first four forms is that they are treating these words as if they were
compounds.If this is correct,then when a compoundanalysisis plausible,
variationcan be expected in words with a back vowel followed by several
neutral vowels. Specifically, if subjects were treating all the forms in
(14) as noncompoundforms, we would expect only back suffix vowels,
essentially as we see in the last eight forms. If, however, some of the
subjectstreat some of the forms as compounds,we would expect to see
front suffix vowels, as we do in the first four forms. For example, if
arkkitehtiis analyzedas drkki#tMhti,
then by the regularrules of harmony,
we would expect front suffix vowels. If all speakers analyze this form as
a compound,we would expect only front suffixvowels, which is not what
we find. It appears that, on average, about 25% of the subjects treated
those forms for which a compoundanalysisis plausible as compounds.
13 Valimaai-Blum (1987) mentions dteisti as a form which could be analyzed as a compound,
but few of our subjects seem to have treated it this way.
14 Neutral vowels do affect suffix harmony in Hungarian. Subjects use almost exclusively
front harmonic suffix vowels in words with back vowel(s) followed by more than one neutral
vowel (bronchitisz 'bronchitis' bronchitisznek) (Ringen 1988a; Kontra and Ringen 1986,
1987).
314
CATHERINE
0.
RINGEN AND ORVOKKI HEINAMAKI
Summarizingat this point, we find that in disharmonicloanwordsthere
is no variation in suffix vowel choice when a front harmonic vowel is
followed by a back vowel, but there is variationwhen a back vowel is
followed by a front harmonicvowel. Contraryto what Campbell(1980)
has claimed, however, with disharmonicroots in which the last harmonic
vowel is front, front harmonicsuffixvowels - and not back vowels - are
more frequent, sound more natural,and are thoughtto be correctby the
majorityof our subjects.Our resultssuggestthat the variationof the suffix
vowels cannot be accountedfor by saying that there are two dialects and
that individualvariation results from dialect mixing (as Halle and Vergnaud 1981 have claimed). Likewise, our results do not supportthose who
suggest that different styles accountfor the variation.Finally, our results
show that the 'compound analysis hypothesis' cannot alone account for
all of the data although they do seem to support the claim that some
loanwordsare analyzedas compounds.
The measure of the variationin suffixvowel choice that we used for an
individualis the percentageof subjectswho respondedwith two alternatives, one with a back suffixvowel and anotherwith a front suffixvowel.
Subjectswere not asked whether both forms would be appropriate.We
thought that actual individualvariation might be greater than our tests
suggested. As can be seen in (10), the percentage of people who gave
more than one form of the test words is very small: it was under 10% for
all but two disharmonicwords, analyysiwith 14% and marttyyriwith 16%
both responses. In order to see whether individuals'variationis greater
than our tests suggested, the same questionnairewas given to a group of
students twice, with an interval of one month between tests. The group
consistedof 50 studentsof foreignlanguages,38 female and 12 male. The
ages ranged from 18 to 34. Thirtyof the studentsresided in the Helsinki
area, while the other 20 were from variousplaces in Finland. The results
are given in (15). % Stable Front and % Stable Back show the percentage
of subjects who were consistentin their choice of front or back suffix
vowels on both tests. % Variationshows the percentageof subjectswho
either gave both alternativeson one or both of the tests, or who gave
differentresponses on the two tests, e.g., a front suffixvowel on one and
a back suffix vowel on the other test.
VARIATION
IN FINNISH
VOWEL
315
HARMONY
(15) Results of Tests 3 & 4
% Stable
Front
Back
6.0
68.0
40.0
12.0
24.0
44.0
a. hierogl'yfi
analy'si
marttyyri
% Variation
26.0
48.0
32.0
b. su'tenoori
amatoori
jongloori
90.0
80.0
72.0
0.0
2.0
2.0
10.0
18.0
26.0
c. hydrosfaari
mfljonaari
afaari
96.0
92.0
68.7
0.0
0.0
6.3
4.0
8.0
25.0
When we compare the figures in (15) to those in (10), we see that the
variation on an individuallevel is clearly greater than suggested by the
percentage of subjectswho supplied two forms. The greater variationin
(15) is not due to differencesbetween the groupsthemselves. The reader
is referred to Appendix B to see that the subjects' both responses are
very similarto those of the group whose resultswere given in (10).
In (1Sa), where a back vowel is followed by y, roughly one fourth to
one half of the subjectsshowed variationin their suffixvowel choice. In
(15b), where a back vowel is followed by o, the variationrangesbetween
10% and 26%. Least variationwas found in (15c), where a back vowel is
followed by d: even though dfddrishows 25% of variationin suffixvowel
choice, hy'drosfddriand miljonddri remain under 10% in variation. Taken
together, the variationin suffix vowel choice is greatest in words where
the front harmonicvowel y follows a back vowel.
3. AN OT ACCOUNT
We have seen that earlierattemptsto explainvariationare unsatisfactory.
Moreover,it is not clearthat any rule-basedaccountwithinany framework
can accountfor the variationwe found. The challengethese data present,
when consideredin the context of native roots, is that suffixvowel choice
for native roots shows no variation- it is categorical:whetherfront vowel
suffixesor back vowel suffixesoccur is a functionof the qualityof vowels
in the root. Similarly, when the last harmonic vowel in a disharmonic
loanword is back, the suffix vowels are categoricallyback. But there is
significantvariation when the final harmonicvowel is front in a disharmonic loanword.We suggestthat OptimalityTheory (OT) providesa way
316
CATHERINE
0.
RINGEN AND ORVOKKI HEINAMAKI
not only to account for the this kind of asymmetry- there is variation
with disharmonicroots in which the finalroot vowel is front but not when
the final root vowel is back or with harmonicroots - but also to predict
how much variation occurs. It is not clear how any rule-based account
could do the former. It is clear that no rule-basedaccount has ever been
proposed that is capable of doing the latter.
We will argue that the choice of suffix vowel is influenced by stress
and openness of root vowels. Initial vowels are stressed in Finnish with
secondary stress on subsequent odd vowels except that vowels in final
syllables are not stressed. Stress seems to play some role in suffix vowel
choice as claimed by Anderson (1980), among others, who suggests that
the harmonicstrengthof vowels is weaker when they are unstressed.For
example, subjectsreport more back suffixvowels with analyysiwhere the
vowel with primarystress (the first vowel) is back than with hie'roglyfi
where the vowel with primary stress is neutral. And they report more
back suffix vowels with marttyyrithan hieroglyfi,both of which have y
preceded by a back vowel, but the formerhas a stressedback vowel and
an unstressedfront harmonicvowel whereas the latter has an unstressed
back harmonicvowel and a front harmonicvowel with (secondary)stress.
Similarly, subjects report substantiallymore front suffix vowels with
hydrosfddrithan with dfddri, even though both end with a back vowel
followed by (long) d and then i. Here the back vowel is unstressedand
the front harmonicvowel is stressedin the former, but the back vowel is
stressed and the front harmonic vowel is unstressed in the latter. In
addition, the data in (10) and (15) supportthe suggestionthat in disharmonic loans harmonicstrengthof front vowels correlateswith openness:
a is most stronglyfront harmonicand y is weakest.
Many recent analyses of vowel harmonyin the frameworkof OT have
assumedthat harmonyresults from alignmentconstraints(McCarthyand
Prince 1993b) which require that two linguisticconstituentsbe aligned at
some edge as suggestedby Kirchner(1993).15 Zoll (1996), however, points
out that alignmentconstraintsdo not, as commonlyassumed, differentiate
between constituentsthat miss alignmentat some edge by one segment
or by two or several segments. Either the constituents are aligned or
they are not; more misalignmentdoes not result in multiple violations.
Following Zoll (1996), we adopt the formulationof alignmentsuggested
in Ellison (1995), NO-INTERvENING, for which there is a straightforward
way to compute multipleviolations:
15
This analysis will work equally well if we assume a constraint MULTIPLE
LINK[+back]
which requiresthat the feature [+back] be multiplylinked.
VARIATION
(16)
IN FINNISH
VOWEL
HARMONY
317
No INTERVENING[+ back]
No vowel intervenes between [+back] and the right edge of
the word.
One violation is assessed for each vowel that intervenes between the
feature [+back] and the right edge of the word.
In Finnish, root vowels determine suffix vowel quality, never the reverse. We assume a positionalfaithfulnessconstraintthat mandatesfaithfulness to input specificationfor backness in harmonicroot vowels. Segments in prominentpositions such as roots, stressed syllables and initial
syllables are often subject to more stringentfaithfulnessconstraintsthan
are other segments, as has recentlybeen well documented(Beckman1997,
1998; McCarthy and Prince 1995).16
(17)
IDENTJIOHARPMVROOT
Correspondentharmonic root vowels have identical specifications for backnessin the input and output (where harmonic
vowels are vowels that are either low or round).
This means that an input harmonicroot vowel must have the same specification for backness as does its correspondentvowel in the output and an
output harmonicroot vowel must have the same backnessspecificationas
does its correspondentinput vowel.
In addition, we assumean inventoryconstraintwhich disallowsnon-low
back unroundedvowels, which do not occur in Finnish (Kiparsky1981,
Ringen and Vago 1995, 1998; Smolensky1993):
(18)
*iA
If a vowel is non-low and unrounded,it must be front.
These constraintswill give us the desiredresultsfor native roots with back
vowels and for disharmonicloanwordsin which the final harmonicvowel
is back. In particular,the predictionis that these formswill alwaysrequire
back suffix vowels.
In the case of a native back harmonicroot koti, these constraintswill
correctlychoose an output candidate with a back suffix vowel over one
with a front vowel (3d), as illustrated in (19). Vowels unspecified for
backnessare indicatedby capitalletters. Note that the optimaloutputhas
a vowel unspecifiedfor backness. Unroundednon-low vowels unspecified
for backnessmust be interpretedas front.
16
Beckman (1998) attributes the original idea to Selkirk (1994) who uses positional PARSE(F)
constraints.
318
(19)
CATHERINE
0.
RINGEN
AND
ORVOKKI
HEINAMAKI
input koti + nq
I
II
+b-b -b
Candidates
ID-IlC TiA ooT
*iA
No-INT + B
a. ko't-na
+b
...........
b. k6ti-na
w
~ ~~ ~ ~..,. ~ ~ ~ ~ .....
c.k +bb
n.
.........
..:.:.
^
b jS~~~~~~~~~~~~~~~~~~~~~~...
.....
C. ko'ti-na
These constraintsalso give the correct results for the disharmonicloans
such as those in (9), Ithat
V is, loanwords in which the final harmonic vowel
is back. The tableau in (20) shows this for syntaksi (19a):
(20)
~~~~~~~~~~..........
.....
.'''.''..
. .. .
...........
~"I'l, ....................
.................................
input syntaksi-na
11 I 1
..~~~~~......
.
.. ..:..........
..>
-b+b
\
,
a. sSrntaksl-na
\
I1
-b +b
1~.-
L
*!
/
*
cd.s)ntasl-na.........
+b ........-...
............
..........
..... ........ ...
.........~~~~~~~~~~.........
.
~~~~~~~~~~~~~~~~~~~~~~..
......
!
;;.
su We assume arbitrainly,that the input suffixvowel is specifiedas [lback] Regardlessof
ak
this
whether this sui vowel is specifieda
unspecifiedfor backness
the same output candidatewill be optimal.
VARIATION
IN FINNISH
VOWEL
319
HARMONY
The unsuffixedform of syntaksi in (21) shows that *iA must be ranked
higherthan NO-INTERVENING
[+BACK].
(21)
input: syntaksi
-b+b -b
Candidates
a. syntaksi
*iA
ID-IOARm/RooT No-INT + B
-b +b -b
S
b. s'ntaks*
.............................
....sy
}...........................................
b ~~~~~~~~~~~~.......................
b
b
*!
-_ +b,
.........
~~~~~...........
. . ....
.-
.
..........
.
.... . .
... . . . .
No-INT+
*jA
These contraintsdo not predictthe correctform for suffixvowels following
native roots with only front vowels or for disharmonicroots in which the
~ ~ ~In
~ ~(22)
~~~the
~~~native
~~.
final harmonicvowel is front.
root hamara (2c, 3c) is
shown. Since the alignmentconstraintrefers only to [+back], it will not
............
X1...~~~~~%-X
align a [-back] feature.
=
(22)
.
|
=. A.
.
....._..
%..: .........
.......
.....
input:hanmara
+ nai
W/I
-b
-b
Candidates
*jA
ID-IOHARM/RooT
No-INT
+
B
b. hama.ra-na
-b
c. hamaira-na
-b
+b
|...i.X
Notice that for a native root with only front vowels such as haimara'dusk'
several constraints would work. A constraint requiring that no vowel
intervene between the feature [-back] and the rght edge of the word
would give the correct output, as would a constraintrequiringthat all
vowels have the same specificationfor backness as does the vowel with
320
CATHERINE
0.
RINGEN
AND
ORVOKKI
HEINAMAKI
primarystress (or the root-initialvowel), the vowel with secondarystress,
or a constraint requiringthat all vowels have the same specificationas
does the most sonorousroot vowel. That any of these constraintswill give
the same and the correct output for front harmonicroots is the key to
explainingthe distributionof suffixvowels with disharmonicloans in which
the last harmonicvowel is front. Supposewe assumethat there are several
constraints,all of which would give the same result for native front harmonic roots, and that they are unranked below the three constraints
and NO-INTERVENING[+ BACK].
assumedso far: *iA, IDENT-IOHARM,ROOT,
Since they all make the same prediction,this lack of rankingmakes no
difference in the case of native roots. Whicheverrankingis utilized will
have the same result.'8But in the case of disharmonicloanwordsin which
the last harmonicvowel is front, these constraintsare capableof predicting
differentoutcomes, dependingon the other vowels in the word. Here we
follow Kiparsky(1993) and Anttila (1997) who have used such unranked
constraintsto account for variation.Given that such constraintsall make
the same predictions for the vast majority of Finnish words, it is not
unreasonable to assume that they are not ranked because there is no
evidence in Finnish as to an appropriateranking.
We suggestthat there are two constraintsthat are concernedwith stress,
one right alignmentconstraint,and one that designatesthe most sonorous
vowel as the triggerof harmony:
(23)
NO-INTERVENING[-
BACK]
No vowel intervenes between a [-back] feature and the right
edge of the word.
(24)
PRIMARY
STRESS
All vowels have the same specificationfor [+back] as does the
vowel with primarystress.
(25)
SECONDARY
STRESS
All vowels have the same specificationfor [+back] as does the
vowel with secondarystress.
(26)
SONORITY
All vowels have the same specification for [+back] as does the
most sonorous vowel of the root.
18
Note that when a native root such as poytd is combined with a suffix such as -1a/lid, there
will be no secondary stress and the decision will be passed to the next constraint in the
ranking.
VARIATION
IN FINNISH
VOWEL
HARMONY
321
According to the sonority hierarchy,Finnish vowels would be ranked as
in (27):
(27)
SonorityScale19
a, a > 6, o, e > i, y, u (low vowels are more sonorousthan mid
vowels, mid vowels are more sonorous than high vowels).
Because Finnish has both long and short vowels, we need to know how
length figuresinto this scale. Assuming that length enhances sonority, we
suggest that all long vowels are more sonorous than their short counterparts. This will give the following rankings:
(28)
SonorityScale
aa,aa>a, a,>i, y, u
ee, oo, 66 > 6, o, e > i, y, u
ii, yy, uu > i, y, u
Let us suppose that all long vowels EXCEPT yy and ii are more sonorous
than any short vowel. yy and ii are then only more sonorousthan y, i and
U.
The resultingscale, in terms of sonority and length, will be as in (29):20
(29)
Sonority Scale
aa, aa > 66, oo, ee>uu> a, a>o,
e > ii, yy> i, y, u
o,
There are 24 different possible rankings (4!) of the four constraintsin
(23)-(26). In six of these rankings,PRIMARY STRESS will be highestranked
and hence will determine the optimal output. If we take the predicted
variation to be a function of the number of times a particularcandidate
is designated as optimal in the 24 rankings,then it is possible to predict
differentpercentagesof back suffixvowels for differentdisharmonicloanwords. On this view, the grammarconsists of some ranked constraints
and 4 which are not ranked.To obtain an output, however, it is necessary
to have some arbitraryrankingof these unrankedconstraints.There are
24 suchpossible rankingsand the grammarwill then yield differentoutputs
with different rankings.With a native root like hdmara,with front harmonic vowels, all rankingswill produce the same optimal output - one
with a front harmonicsuffix vowel. But for disharmonicroots, different
results are possible. This is illustratedfor hieroglyfi(6a, 10a) which can
take either a back vowel or a front vowel suffix.
19 Anttila (1997) suggestthat sonorityplays a role in the selection of variantsof the GENPL in Finnish.
20
The data we have are consistentwith other scales. For example, Harms'(1982) tonality
scale for front vowels or Bladon and Lindblom's(1981) perceptualscale will also work.
322
CATHERINE
O. RINGEN AND ORVOKKI HEINAMAKI
(30) input: hierloglC3-1 Crankingsyieldfi
4+b-b b
-
Candidats
a.
*A
Looar
v'rogltR
Jo +b 4
b. hierloglpl-
-b +b -b +b
c. hierogldl-a
/
1
-b fb
*!
Io-IA>RMsoar > No-ltr + B
If PRIMARY
is ranked immediatelybelow No-INrERvENING[+sAcK]
then
the candidate with the front suffix vowel will be optimal. There are 6
rankingsin which PRIMARY
STRESS
is ranked over SECONDARY
STRESS,
SONORITY,
and NO-INTERVENINGE-BACK]
(because there are 3I rankings
of the other constraints)and hence in 6 out of the 24 rankingsthe optimal
output will have a front harmonic suffix vowel. These 6 rankings are
illustratedin (31):
(31)
PRIMARYSECONDARY SONORITY
NO-INTt-]
BACK]
PRIMARYSECONDARY NO-INT[-BS
XCK]SONORI
NO-INTE-BACK]
PRIMARYSONORITY
SECONDARY
PRIMARYSONORI
NO-INT[-BSXCK] SECONDARY
PRIMARYNO-INTE-BACK]
SONORITY
SECONDARY
PRIMARYNO-INTE-BACK]
SECONDARY
SONORITY
r
Slnce the top-ranked constraintswill never be the deciding constraints
when a front harmonicvowel is final in a disharmonicroot, these constraintsare not includedin tableauxwhich follow.
If SECONDARY
STRESS
is highestranked,as in (32), there are 5iXrankings
which will also yield a front harmonicsuffixvowel.2l
=1 It might appearthat SECONDARY
would not have an effect with three-syllableroots like
afaari because there is no secondarystress on the final root vowel since no stress falls on
word-finalsyllables.But when a suffixis added, e.g., afaarl-ssa, there will be stress on the
thirdvowel becauseit is no longer in the final syllable.
VARIATION
(32)
IN FINNISH
-a
input hierol
323
VOWEL HARMONY
6 rankings yield a
I
VlI
-b +b -b +b
hieroglyEl-a~~~~~~
SEC SON No-NT - B PRIM
Candidates
a
...-...
.........
. . . .8
.....
..."#*...... ...
. .. .. ....
.*..**8*..
. .. .....
ir:
VW * _~~~~~~~~~~.
^,. ......
.~
v~
~ ~~ ~ ~ ~ ~ ~ ~~~~....
,,
........
......
/
V
/
*.-.a .
V
.
s*
.....,,*4
,
...
,.-'.-'-'
-.-'2---
...........
.......... . .
b. hieroglyfi-a
Vt VI
**!
4*
a* a*. t.+
*.
.
b.
-b +b -b .+...
In contrast, if SONORITYis ranked highest,
then
the
back
.......
.... ....I.
............
...~- alternant
...
... vowel
will be optimal as illustratedin (33):
................
.
....
11 . .....
,~~~~~~~.
. '.'.. ' . . '.x.
....:.
. :.:.... >.>.:.:.:.:.:.
(33)
input hieroglE-Y
b
v
4
4 ""~.'.x.:.''."..'...... . .
+b-b-b+b+b
-b+bb
CandidateS
. ,...
.
>'
. .......
..
....
SONZ No-INT -B
...
......
i.w
....*sa..
PRIM SEC
................................
a. hierogly'fi-:
*****i
b. hiWroglyfi-a
VI
:
..~~~~~~.....................
4
V/IV/
: .
....x.........
.. a
6* rankings
yield
I
Vt 1 1/1*
. :.
-
-*.#8:.::
:':..:::.,,:8:::
.........
**
#
**....................
.......
Finally, in six of the rankings No JNTERVENING[ BACK] Wi1lbe ranked
highest and the optimal candidate will have a front harmonic suffix vowel.
This is illustrated in (34):
2
We assumethat o is more sonorousthan the diphthong
ie.
/D.*
*/
324
(34)
-b
W ngerfglyV-a
**** +b
+ 4.. -b
CATHERINE
*I J
1s /
rr
/::::::
...
:::
4,.,.-#E,.....
.. ..
::::.: ..
::::::::.:::::::::::.
..
,-.,,
*.....
:,@
..
,.,
*:
2 .-.8. *-....t*^
a.*.,
t
^
. ::::.
;
.
.....
:.::.:.::::
.
.
<
E
t.
O. RINGEN AND ORVOKKI HEINAMAKI
1
input: hveroglyv-
6 rankingsyield a
-b +b -b +b
No-IN r-B
Candidates
a. hiemglyE-a
PEM SEC SON
.,,.,.,,,.,.,
.,,.,...,,.,..,,,
,,,
,,.,
,...,,.,
x.
,,,,,
,.,
,,,,
...,;
Thus 18 of the 24 possible rankingsof these constraints(that is, 75% of
the possible rankings)will choose the front vowel suffix as optimal. fIowever, as noted above, eYidence suggests that those fbrms that can be
are indeed so analyzedapproxianalyzedas compoundssuch as hieroglyfi
mately 25%of the time. If we assume that 25%of Finnish speakers
asa compound, then the harmonyconstraintswe have
analyze hierog0fi
assumedwill requirethat suffixvowels be front, no matterwhich ranking
there are only front vowels. As we have just seen
is used, because in gly>fi
is not treated as a
of the speakers for whom hieroghyfi
for the 75°MP
compound, the likelihood of a rankingthat yields a front suffix vowel is
of all
75°/O.Overall the grammarpredicts front suffix vowels in 81.3%
cases (0.75 x 0.75 + 0.25 x 1.00 = 813%).
OUFtests did not involve a forced choice and hence we do not know
what percentage of our subjectswould have reported front suffix vowels
if they had been forced to give only one response. We do
with hieroglyfi
know that 40 subjects responded with a front suffix vowel and 6 with a
back suffix vowel while 4 gave two brms, one with^afront vowel and
one with a back vowel. In computingthe observed frequenciesfor comparison with the predictionsof our grammar,we have eliminated those
subjects who gave both responses. This means that 40 of 46? or 87%,of
the subjectsrespondedwith a forrnwith a front suffixvowel for haeroglyfi.
Assumingthat plausiblecompoundsare assumedto be compounds25%
of the time, the predictedvs. observed frequenciesof front suffixvowels
and
kasoori,jonglooriX
are given below. Since marttyyri,
are not plausiblecompounds,we assumethese forms shouldnever receive
the compound analysis and, hence, the predicted frequencies for these
forms are those predicted by the twenty-fourdifferent rankingsof the
constraints.Wherewe tested two differentwordsfor whichthe constraints
make identicalpredictions we have combinedresponses.
affaari
vulga'ari
VARIATION
(35)
IN FINNISH
325
VOWEL HARMONY
% FRONT SUFFIXES OBSERVED COMPARED WITH
PREDICTED VALUES
observed% predicted%
testword
hieroglyfi
87.0
81.3
analy'ysi
58.0
62.5
mdirttyyri
50.0
50.0
sutenoon
98.0
81.3
jongl6Ori/kiso6ori
84.0
75.0
amatoo5nr
89.0
81.3
hydrosfaairi
100.0
100.0
miljoniaairi
98.0
100.0
ri/vlguaan
78.0
75.0
tdbasko/jirttafr6tunda/t6rnado
00.0
00.0
syntaksi/symptomi/tyranni
00.5
00.0
foljetongi
00.0
00.0
The correlationr (Pearson product moment correlationcoefficient), between the predicted and observed results is +0.9903 indicating a very
strong linear relationship23between the predictedand observedresults.
4. CONCLUSION
We have shown how OptimalityTheorycan accountfor both the categorical suffixvowel choice which occurswith most roots and the variationthat
occurswith other roots - not only where the variationwill occur, but also
how much variationwill occur. Our analysisshows that variationdata can
be modeled by a grammarwith partiallyrankedconstraints.
There are several questions raised by our discussionthat deserve some
comment. The first has to do with underspecification.It is assumedthat
there are no constraintson inputs in OT, which means that the crucial
use of input underspecification,as has been commonin rule-basedanalyses
of transparentvowels, is inconsistent with the basic tenets of OT. The
23
In fact there is evidence that the relationship between the predicted and obtained results
is the identity relationship; the predictions directly account for 97.97% of the variance in the
data.
326
CATHERINE
0.
RINGEN
AND
ORVOKKI
HEINAMAKI
analysiswe have assumeddoes not make crucialuse of inputunderspecification, but some outputs are underspecified. The status of output under-
specificationin OT is less clear. On the one hand, the analysis of Ito,
Mester,and Padgett(1995) workswhetheror not inputsare fully specified,
but some of their outputs do involve underspecifiedforms. On the other
hand, Kirchner(1997;pp. 97-89) claimsthat Smolensky(1993) has shown
that "phonologicalinactivityof predictablefeaturesmay be attributedto
rankingsof a particularclass of constraints,ratherthan to the absence of
such featuresfrom the representationat some stage of a derivation".The
example he uses is that of Finnish. While it is beyond the scope of this
paper to consider the entire question of output underspecificationhere,
Kirchner'sargumentdoes not show underspecificationis unnecessaryin
an accountof Finnishvowel harmonybecause the analysishe proposes is
problematic.24
Kirchner'saccount involves the constraintsALIGN(+bk-R),which requiresthat a [+backj specificationbe linked to a segment at the rightedge
of the word,25 *EMBED,which prohibitsembeddingof a [-back] domain
inside a [+back] domain, *[-low, -round, +bk] (whichis the same as our
*iAconstraint),PARsE(bk),whichis a faithfulnessconstraintrequiringthat
input and output be minimallydifferentwith respect to specific features,
and *[-low, +rd, -bk], which prohibits ii and o. Kirchner gives three
hypotheticalforms and concludesthat there is no need for feature underspecificationat underlyingor intermediatestages of the derivation.
24
Unlike Kirchner, Smolensky (1993) assumes that all suffix vowels are unspecifiedfor
backness in the input. If we make this assumption, it may be possible to account for Finnish
vowel harmony without underspecification of transparent vowels as Smolensky suggests, but,
of course, such an analysis does involve input underspecification and hence cannot be used
to argue that no input underspecification is necessary.
25 Kirchner's constraint is violated only if there is no [+back] feature aligned with the right
edge of the word (if there is a [+back] feature somewhere in the word), whereas our
alignment constraint is violated for every [+back] feature that is not aligned with the right
edge of the word.
VARIATION
IN FINNISH
VOWEL
327
HARMONY
(36)
-rd, +bkJ AUGN(+bk, -R)
*-4ow,
_____
*EMBED
PARSE(bk) *[-40W, +rd, --bk]
a. lu-I-UI
U
i
U
_
__
X
X
I
|
m X
..:
.
. ......................
~~~~~~~~~~~~~~~~~~~............
e
u-i-u
u-i-u
~~~~~~~~~~~~~~~~~~~~~..
....
.,. . .. .
.'.,
.__
..........................................
'.'.
a1i.......
=*_____
~~~~~~~~~~~~~~~~~~~~~~~~.....
u-i-u
~~~~~~~~~~.
U-1-U
. . , , ,,,, .. ...,........
*
. ..
.
s...... ...-.,,.,.,,.-:
.._ .... ....
.......,,,,,
b. l-I-U!
.. . .
U-i-U
R
:
_ u-i-U
*!.
*..
. l.......i-u!......
~.u-i-u
.
.
... -
U-i-U**
U i
*.1
*. :t
- >
^;
....
, . .........
_~~~~~.
......
..
..........
.
.-- 8
..
_=
...
**
;
...
........
.........
.. .
...
~~~~~~~..........
....... . .....
~~~~~~~~~~~~~~..
....... . ...
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~......
-.....
*!
!
u-X-u
=
...
=
U l-0
ju-i-UI
C.
., .....
-^:
_
--- : ?? . ?.?SX-:-
..
__
U
.
____
. ...........
___._
**
/u-i-u
.
.. . . . . ...
.. . . . . .
w=_;
.;.:
Although
this set of constraints.......
seems to handleneutral
vowel
tranparency
..
;;;.
.....................;
.
.
..............
.. ..............
....U
without
this
constraint
set
does
not
work
with
the-.-.rest
underspecification,
W~~~~~~~~~~~~~~~~~~~~~~~~~~~
... .
......
of the language. To take just one simple case, consider a root which
.+X...;..
requiresfrontharmonicsuffix
vowels
;h> .. >.. . suchaspoytai:
.~......... ....
,
.
_ .
_
.
.
.
.
poyta-nai.
......
. .
If thisroot
is followed by a suffix with an underlyingback vowel, the wrong output
~~~~~~~~~~~~~~~~~~~...
.
oana
a.a.
would result:
*!...........
S......
(37)
_____*[.0w,
a
-1-U
_
-a+a
-rd, +1* ALIGN(+bk,-R)
.
PARSE(bk) *[.4W, -srd, k]
.. . . .
*EM4BED
*!
The correct surface form would be one with a front harmonic vowel
(po'yta-na), yet this constraintsystem chooses one with a back harmonic
vowel (*poyta-na). Nor is it clear how this constraintset could be modified
a+a
cFa+a
328
.......
*
CATHERINE
*!
.......
..................
.:2:<.
;N
O. RINGEN AND ORVOKKI HEINAMAKI
to handle the forms in (36) as well as front vowel roots. Adding a constraint, Align(-b-R) does not help, as illustratedin (38):
(38)
*[4ow, -rd, +b]|
ALIGN(+bk,_R)
| ALIGN(_bk,
-R) |PARSE(bk)
| *[-10W,
+rd, -bk]| *EM]3ED
a. Ia-a/
@a+a
*!
<.^8:.S.-
We conclude that it has not been demonstratedthat it is possible to
provide an analysis of a significantbody of data in a vowel harmony
languagesuch as Finnishwithout some type of underspecification,output
underspecificationas assumedhere, or inputunderspecificationas assumed
in Smolensky(1993).26
Our aim in this paper is not to argue that output underspecificationis
or is not necessaryto account for transparentvowels in Finnish. Indeed,
the issue we want to address,namely how to accountfor the variation,is
independentof the issue of underspecificationin the sense that our account
of vanation does not cruciallydepend on assumingthat output forms are
underspecified.That is, when the final harmonicroot vowel is back, there
is no variation. Some high-rankingconstraintor constraintsmust account
for this. We have adopted an account that involves output underspecification. It is possible, however, that there is an interestingaccountof the
forms with no variation (that is those in which the final harlllonicroot
vowel is back) which does not involve output underspecification.Our
account of the vanation that occurs when the last harmonic vowel is
front would not, in principle,be incompatiblewith such a (fully-specifiedoutput) account of forms in which the last harmonicvowel is back.
Second, it might seem that we are not justified in assumingthat the
variation in suffix vowel choice is a result of grammarswith unranked
constraintsbecause it is conceivablethat the variationcould be accounted
for by different grammarswith differentfixed rankings.This is a version
of the 'differentdialect'hypothesis.Suchan alternativeis not supportedby
our data. For example,two of the subjectswhose responsesare reportedin
(10) reported back suffix vowels with amatoori, analyysi, and marttyyri,
The facts of Hungarianvowel harmonyare also incompatiblewith an analysisalong the
lines that Kirchnersuggests,as well as with that of Smolensky(1993), becausesuffixvowels
in Hungariancannot all be unspecifiedfor backness. See Ringen and Vago (1998) for
discussion.
26
VARIATION
IN FINNISH
VOWEL
HARMONY
329
and front suffixvowels for the other seven words in (10). There is no fixed
ranking of the four constraints PRIMARY,SECONDARY,
SONORITY,
and NO-
rNT[- BACK] that will predict this distribution.A ranking of PRIMARY
highest will choose an output with a back suffix vowel as optimal for
dmatoo ri, dnalyysi, and marttyyri (if we assumethat amatoo ri and dnalyysi
are not analyzed as compounds), but this ranking will also incorrectly
choose as optimal a candidate with a back suffix vowel for sutenoo0iri,
jongloori,
and dfddri, which is not consistent with these subjects' re-
sponses. Even if we assume that these subjects analyze sutenoo05ri as a
compound, thereby explainingtheir choice of front suffix vowels for this
form, we are still left with the wrong predictionsfor jongloori and dfddri,
whichcannotbe analyzedas compounds,as we have seen above. Similarly,
two other subjects reported back suffix vowels for dfddri, analyysi, marttyyri, and hieroglyfiand front suffixvowels for the other five wordsin (10).
There is no fixed ranking of the four constraintsPRIMARY, SECONDARY,
SONORITY, and NO-INT[-back]that will predict this distributionof responses. The responsesof subjectswho gave front responsesfor all forms
in (10) can be accounted for by assuming a fixed ranking of NO-INT[-back]
highest. But aside from these, the reports of the rest of the subjectsare
not consistentwith any fixed rankingof these constraints.This, of course,
does not show that there is no set of constraintsfor which differentfixed
rankingscan account for the variation data in (10). But the burden of
proof is on those who would suggest that variationin suffixvowel choice
can be accounted for by assuming different fixed rankings of a set of
constraintsto propose a set of constraintsthat can account for variation
in suffixvowel choice for (some) disharmonicforms. We have shown that
these data can be accountedfor if we assumeone grammarin which some
constraints are unranked and may occur in different rankings for one
speakerin differentproductionsof a given form.
Aside from the fact that there is no apparentway to account for the
variationdata with a fixed ranking,there is anotherreason to believe that
the account with unrankedconstraintsis correct. This analysis, but not
the fixed rankingsalternative(the 'differentdialect' account)predictsthat
there is intrasubjectvariation. Our tests with these same words with
another group of subjects on two separate occasions show that there is
considerableintrasubjectvariation(see (15) above). These resultsare not
consistentwith the idea that the variationin (10) can be accountedfor if
we assume that there are different grammarsof Finnish with different
fixed rankingsof the same constraintset.
Third, we may well wonder about the status of the constraintswe have
proposedin accountingfor the variationwe find with Finnishdisharmonic
330
CATHERINE
0.
RINGEN AND ORVOKKI HEINAMAKI
loanwords. The constraintsNo-INTERVENING[-BACK] and NO-INTERVENING[+BACK] are not particularlycontroversialand have been employed in
accounts of other vowel harmony systems (Kirchner 1993, Ringen and
Vago 1995, Smolensky 1993). However, the other constraints(PRIMARY,
SECONDARY, and SONORITY) do not have this status. There is some evidence that stressdoes play some role in vowel harmonyin other languages.
For example, Ultan (1973) notes that unstressedvowels are more likely
to harmonize than are stressed vowels and that stressed vowels often
determinethe qualityof unstressedvowels, and Beckman (1998) discusses
cases where the stressedvowel is the triggerfor assimilation.But if these
constraints are indeed universal, we make certain unsupported crosslinguisticpredictions.For example,it is difficultto imaginethat a language
exists in which root faithfulnessis ranked above PRIMARY STRESS, which
in turn is ranked above NO-INTERVENING[ + BACK] and No-INTERVENING
[- BACK], resultingin non-adjacentassimilationof all suffixvowels to the
backness of vowels with primarystress. Hence, it may be that certain
rankingsof these constraintsare universallyprohibitedor, as one reviewer
has suggested, it may be that some of these constraintsrepresentavailable
strategies for dealing with loanwords. There is some evidence that this
may be the case. As Holden (1972) notes, when Kazakh(a languagewith
front-backharmony)borrowsfrom Russian,it is the qualityof the stressed
Russian vowel that determinesthe harmonicquality of all the vowels in
the word in Kazakh.
Fourth, it may be importantto clarifywhat is being claimed here with
respect to languageproduction.When a linguistelicits data from a single
consultant,we take the consultantto be providingdata about the language
that, in turn, must be accounted for by a grammarof that language. If
the consultantreportsthat there are acceptablealternativepronunciations
or alternativeways of expressingsome idea, we expect an analysis(grammar) to provide an account of this. We take the results of elicitation to
be data about the languagethat our grammarsshould model.
In this study we have consulted a large number of consultants and
recordedtheirjudgmentsabout the appropriateformsto determinewhere
and how much variationoccurs. We have demonstratedthat an OT grammar with some unrankedconstraintscan model the variation, if we take
the variationpredictedby the grammarto be a functionof the numberof
times a particularcandidateis designated as optimal by the 24 possible
rankingsof the unrankedconstraintsin the grammar.The grammarwe
constructedhas unrankedconstraints,but to relate any given input to a
single output, we must impose some arbitraryranking on the unranked
constraints. By computing the frequency with which a particularinput
VARIATION
IN FINNISH
VOWEL
HARMONY
331
resultsin variousoutputs, when all possible permutationsof the unranked
constraintsare considered, it is possible to derive statements like "this
grammarpredictsthat both dfddrillii and dfddrilla are possible in Finnish,
but dfddrilld is three times more likely than dfddrilla".
Such an accountis consistentwith the traditionalview that grammarsare
about linguisticcompetence,ratherthan production.But it is legitimateto
ask why our consultantsexhibited the patternof responsesthat they did.
Our account would appear to be consistent with differentconceptionsof
the role of grammarin production.That is, our accountis consistentwith
the view that speakersactuallyconsult the constraintswhen they produce
a form so that the actual form that is produced on a given occasion is
determinedby the particularrankingof the unrankedconstraintsthat they
'happento use' on a given occasion. This accountis undoubtedlyconsistent
with connectionist models where the (unranked) constraints might be
viewed as equallyweightedunits in an interactiveactivationmodel. Thus,
the activationof any particularconstraintmight reduce the probabilityof
activationof the others. Full discussionof this importantissue is beyond
the scope of this paper.
Finally, we have tested a relatively small number of forms. Clearly,
more data need to be gathered to see if similar results are found when
more forms are tested. However, our preliminaryresults are promising,
suggesting that the variation in suffix vowel quality found with Finnish
loanwordscan be accountedfor with the same constraintset that accounts
for the categoricalsuffixharmonyof native roots, if we adopt the proposal
of Kiparsky(1993) and Anttila (1997) that some constraintsare not ranked
in the grammar.Not only are we able to predict where variation will
occur, but such a theory is capable of modeling frequency of various
alternantswith considerableaccuracy.
332
CATHERINE
0.
RINGEN
AND
APPENDIX
Ika:
vuotta
age
years
Merkitse rastilla oikea vaihtoehto:
ORVOKKI
A:
TEST
HEINAMAKI
1
nainen
markthe correctalternativewith an X: female
Kotipaikka: Helsinki
;
mies
male
muu (mika?)
other (where?)
home town
Aidinkieli:
;
suomi
mothertongue: Finnish
;
muu mikai?
other (which?)
1. Haluaisin maistaa
2. En ole kiinnostunut
(tabasko)
(afaari)
3. En ole koskaan rakentanut
(jurtta)
4. Pyysimmepiirustukset
(arkkitehti)
5. En ole viela nahnyt
(fakiiri)
6. Oireet johtuvat
7. Tamapiirreesiintyy
(kysta)
(katakana)
8. Nyt keskustellaan
(analyysi)
9. Poliisi etsii
(sutenoori)
10. Sheriffi tuli ulos
(saluuna)
11. Haluaisin opiskella
(syntaksi)
12. Opas kertoi meille
(marttyyri)
13. Meidan ei tarvitsepelata
14. Laakarikiinnostui
(trakooma)
(symptomi)
(katrilli)
(hieroglyfi)
(kontrabasso)
15. Koko illan he tanssivat
16. Liisa otti kuvan
17. Haluaisitko soittaa
18. Han kuulostaamelkein
_
(tyranni)
19. Mina katselin
(jonglbori)
20.
21.
22.
23.
Kaikki lapset pitavat
Nayttelysiirrettiinpois
Tuo mies nayttaaihan
Monet pelkaavat
(karamelli)
(rotunda)
(amatoori)
(tornado)
24.
25.
26.
27.
28.
Muutos on jo havaittu
En halua lukea
Liisa piti esitelman
Sain rahat
Han vaikuttaa ihan
(hydrosfaari)
(foljetongi)
(anafora)
(miljonaari)
(bolshevikki)
29. Kaikkiolivatmukana
(karonkka)
VARIATION
IN FINNISH
VOWEL
333
HARMONY
DIRECTIONS: The following directions (in Finnish) were given to the
subjects in writing and also read to the subjects before they began the
task.
On the following page there are 29 sentences. Each of them has an empty
space marked with a line. Your task is to fill in the space by writing
an appropriateform of the word that is given in parentheses after the
sentence.
Example:
1. Tamanneuvon sain
This piece of advice I received
(juristi)
(lawyer)
The appropriateform would be juristilta, so that your answer is as
follows:
2. Tamanneuvon sain
This piece of advice I received
juristilta
from a lawyer
(juristi)
(lawyer)
If in your opinion more than one form is possible, write them on the
line (one on top of the other or one after the other) and
markthem with a numberin the orderof theirnaturalness; the most
naturalalternativeis 1. Example:
3. Taiteilijatyoskenteli
(ateljee)
(The artist worked
(workshop)
If you think that the most appropriateform would be ateljeessa, but
you think the form ateljeessa is also possible, answer as follows:
4. Taiteilijatyoskenteli
(The artistworked
1 ateljeessa
2 ateljeessa
in the workshop
(ateljee)
(workshop)
If on the other hand you think that the different forms are
equallynatural, write them on the line and put braces around the
words. Example:
5. Taiteilijaviimeisteli
{krusifiksiakrusifiksia}
(The sculptorwas finishing {thecrucifixthe crucifix}
(krusifiksi)
(crucifix)
Make your choice of forms only on the basis of what you feel is more
natural.Do not try to rememberany rules!
All the words given in parentheses are in the singular. Keep the
singularin your answers. Try to make your handwritingas clear and
legible as possible!
334
CATHERINE
APPENDIX
B:
0.
RINGEN AND ORVOKKI HEINAMAKI
STABILITY
OF VS.
VOWEL
analyysi
marttyyri
sutenoori
amatoori
jongloori
hydrosfaari
miljonaari
afiaari
27
Test 327
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
Stable
Varying
Test 3
Test 4
IN SUFFIX
% Front
76.0
78.0
68.0
suffix vowel
% Back
8.0
14.0
6.0
52.0
54.0
40.0
22.0
28.0
12.0
30.0
36.0
24.0
58.0
56.0
44.0
96.0
94.0
90.0
0.0
2.0
0.0
test word
hieroglyfi
VARIATION
CHOICE
88.0
90.0
80.0
4.0
8.0
2.0
84.0
80.0
72.0
2.0
14.0
2.0
98.0
96.0
96.0
0.0
2.0
0.0
100.0
92.0
92.0
0.0
4.0
0.0
77.6
79.2
8.2
16.6
Tests 3 and 4 were identical to Test 1 given in Appendix A.
% Both
16.0
8.0
26.0
26.0
18.0
48.0
12.0
8.0
32.0
4.0
4.0
10.0
8.0
2.0
18.0
14.0
6.0
21.0
2.0
2.0
4.0
0.0
4.0
8.0
14.2
4.2
VARIATION
IN FINNISH
68.7
Stable
Varying
Test 3:
Test 4:
Stable:
Varying:
VOWEL
335
HARMONY
6.3
25.0
Results of the group when tested the first time.
Results when the same group was tested the second time,
one month later.
The percentage of subjectswho chose a front suffixvowel
or a back suffix vowel on both tests.
The percentage of subjectswho chose both suffix alternatives on at least one of the tests, or who chose a front suffix
in one and a back vowel suffixon the other.
REFERENCES
Anderson, Lloyd: 1980, 'Using Asymmetricaland GradientData in the Study of Vowel
Harmony',in RobertM. Vago (ed.), 1980, pp. 271-340.
Anttila, Arto: 1997, 'DerivingVariationfrom Grammar',in FransHinskens, Roeland van
Hout and Leo Wetzels (eds.), Variation, Change and Phonological Theory, John Benjam-
ins, Amsterdam,pp. 35-68.
Beckman, Jill: 1997, 'PositionalFaithfulness,PositionalNeutralization,and Shona Vowel
Harmony',Phonology14(1), 1-46.
Beckman, Jill: 1998, PositionalFaithfulness,unpublishedPh.D. dissertation,Universityof
Massachusetts.
Beckman,Jill, LauraWalshDickey, and SuzanneUrbanczyk(eds.): 1995, UMOP18:Papers
in Optimality Theory, GLSA, Amherst, Massachusetts.
Bladon, R. A. W. and Bjorn Lindblom:1981, 'Modelingthe Judgmentof Vowel Quality
Differences',JASA 69, 1414-1422.
Campbell,Lyle: 1980, 'The Psychologicaland SociologicalReality of FinnishVowel Harmony', in R. Vago (ed.): 1980, pp. 245-269.
Campbell,Lyle: 1981, 'GenerativePhonologyvs. FinnishPhonology:Retrospectand Prospect', in D. L. Goyvaerts(ed.): 1981, pp. 147-182.
Ellison, T. Mark:1995, 'PhonologicalDerivationin OptimalityTheory',unpublishedmanuscript,Universityof Edinburgh.
Fujimura, Osamu (ed.): 1973, Three Dimensions of Linguistic Theory, TEC Corp., Tokyo.
Goldsmith,John:1985,'VowelHarmonyin KhalkhaMongolian,Yaka, Finnish,andHungarian', Phonology Yearbook 2, 253-275.
Goyvaerts,D. L. (ed.): 1981, Phonology in the 1980s,Story-Scientia,Ghent.
Halle, Morrisand Jean-RogerVergnaud:1981, 'HarmonyProcesses',in W. Klein and W.
Levelt (eds.), Crossing the Boundaries in Linguistics, Reidel, Dordrecht, pp. 1-22.
Harms, Robert: 1982, 'What Helmholtz Knew about Neutral Vowels', Texas Linguistic
Forum19, 67-88.
Holden, Kyril: 1972, Loan-words and Phonological Systems, unpublished Ph.D. dissertation,
Universityof Texas.
Ikola, Osmo 1971, Nykysuomenkdsikirja(A Handbookof ContemporaryFinnish),second
edition, Weilin and Goos, Helsinki.
Ikola, Osmo (ed.) 1986, Nykysuomenkdsikirja(A Handbook of ContemporaryFinnish)
second, revisededition, Weilin and G66s, Espoo.
336
CATHERINE 0. RINGEN AND ORVOKKI HEINAMAKI
It6, Junko and Armin Mester: 1995, 'Core and PeripheryStructureof the Lexicon and
Constraintson Reranking',in Beckmanet al. (eds.): 1995, pp. 181-209.
It6, Junko, Armin Mester, and Jaye Padgett: 1995, 'Licensingand Underspecificationin
OptimalityTheory', Linguistic Inquiry 26, 571-613.
Kiparsky,Paul: 1973, 'PhonologicalRepresentations',in 0. Fujimura(ed.), pp. 3-136.
Kiparsky,Paul: 1981, 'Vowel Harmony',unpublishedmanuscript,MIT.
Kiparsky,Paul: 1993, 'VariableRules', paper presentedat The First Rutgers Optimality
Workshop(ROW#1), October, RutgersUniversity,New Brunswick,New Jersey.
Kirchner,R.: 1993, 'TurkishVowel Disharmonyin OptimalityTheory',paperpresentedat
The First Rutgers OptimalityWorkshop(ROW#1), October, RutgersUniversity,New
Brunswick,New Jersey.
Kirchner,Robert: 1997, 'Contrastivenessand Faithfulness',Phonology 14(1), 83-111.
Kontra, Mikl6s and Catherine 0. Ringen: 1986, 'Vowel Harmony:The Evidence from
Loanwords',Ural-Altaic Yearbook, 1-14.
Kontra, Mikl6s and Catherine0. Ringen: 1987, 'Stressand Harmonyin HungarianLoanwords', in KarolyR6dei (ed.), Studien zur Phonologie und Morphonologie der uralischen
Sprachen, Verbandder WissenschaflicheGesellschafteOsterreichs,Vienna, pp. 81-96.
Kontra, Mikl6s, Catherine0. Ringen, and Joseph P. Stemberger:1991, 'The Effect of
Context on Suffix Yowel Choice in HungarianVowel Harmony', in Werner Bahner,
JoachimSchildtand Dieter Viehweger(eds.), Proceedings of the Fourteenth International
Congress of Linguists, vol. 1, Akademie-verlag,Berlin, pp. 450-453.
Levomaiki,Mauri:1972, 'Vierasperaistensanojensuffiksaalivokaalisointu',(Suffixalvowel
harmonyin words of foreignorigin), Virittajd76, 254-62.
McCarthy,John and Alan Prince:1993a,Prosodic Morphology I: Constraint Interaction and
Satisfaction, unpublishedmanuscript,Universityof Massachusettsand RutgersUniversity.
McCarthy,John and Alan Prince:1993b,'GeneralizedAlignment',in Geert Booij and Jaap
van Marle (eds.), Yearbook of Morphology, pp. 79-154.
McCarthy, John and Alan Prince: 1995, 'Faithfulnessand Reduplicative Identity', in
Beckmanet al., 1995, pp. 249-384.
Penttila,Aarni:1963, Suomen kielioppi (Finnishgrammar),second, revisededition,WSOY,
Porvoo, Finland.
Prince, Alan and Paul Smolensky:1993, Optimality Theory: Constraint Interaction in Generative Grammar, Technical Report #2 of the Rutgers Center for Cognitive Science,
RutgersUniversity,New Brunswick,New Jersey.
Ringen, CatherineO.: 1988a, 'Transparencyin HungarianVowel Harmony',Phonology 5,
327-342.
Ringen, Catherine O.: 1988b, Vowel Harmony: Theoretical Implications, Ph.D. dissertation,
IndianaUniversity,Bloomington,Indiana,1975. Publishedby Garland,New York, 1988.
Ringen, Catherine0. and MiklUsKontra:1989, 'HungarianNeutral Vowels', Lingua 78,
181-191.
Ringen, Catherine0. and Robert M. Vago: 1995, 'A ConstraintBased Analysisof HungarianVowel Harmony',in IstvdnKenesei (ed.), Approachesto Hungarian,vol. 5.
Ringen, Catherine0. andRobertM. Vago: 1998,'HungarianVowel Harmonyin Optimality
Theory', Phonology 15(3).
Saarimaa,E. A.: 1971, Kielenopas (Languageguide), eighthed. WSOY, Helsinki.
Sadeniemi,Matti:1946, 'Marttyyreja
vai marttyyreja?(Marttyyrejaor marttyyreja?)Virittdja
50, 79-80.
Sadeniemi,Matti: 1949, Metriikkamme perusteet (Fundamentalsof Finnishmetrics),Otava,
Helsinki.
Selkirk, Elisabeth: 1994, 'OptimalityTheory and Featural Phenomena', lecture notes,
Linguistics730, Universityof Massachusetts,Amherst.
Smolensky,Paul: 1993, 'Harmony,Markedness,and PhonologicalActivity',paperpresented
at The FirstRutgersOptimalityWorkshop(ROW#1), October,RutgersUniversity,New
Brunswick,New Jersey (ROA 87-0000,ROA 37-000).
VARIATION
IN FINNISH VOWEL HARMONY
337
Steriade,Donca: 1987, 'RedundantValues',in Anna Bosch, BarbaraNeed, and Eric Schiller
(eds.), CLS 23, ChicagoLinguisticsSociety, Chicago,pp. 339-362.
Ultan, Russell: 1973, 'SomeReflectionson Vowel Harmony',WorkingPaperson Language
Universals 12, 37-67.
Vago, Robert M. (ed.): 1980, Issues in VowelHarmony,John Benjamins,Amsterdam.
Vago Robert M.: 1988, 'Vowel Harmonyin FinnishWord Games', in Harryvan der Hulst
and Norval Smith (eds.), Features, Segmental Structure and Harmony Processes, Part II,
Foris, Dordrecht,pp. 185-205.
Valimaa-Blum,Ritta:1987,'FinnishVowel Harmonyas a PrescriptiveandDescriptiveRule:
An AutosegmentalAccount', in A. Miller and Joyce Powers (eds.), 4th ESCOL 1987,
The Ohio State University,Columbus,pp. 511-522.
Zoll, Cheryl: 1996, Parsing Below the Segment in a Constraint Based Framework, unpublished
Ph.D. dissertation,U.C. Berkeley.
Received 26 November1997
Revised 28 September1998
(Ringen)
Departmentof Linguistics
Universityof Iowa
Iowa City, IA 52242
[email protected]
(Heinamaki)
Departmentof Linguistics
Universityof Helsinki
PL 4 HelsinginVliopisto
FIN-00014Finland
[email protected]