Vowel harmony in Turkish and Hungarian∗

Harry van der Hulst

University of Connecticut

1 I NTRODUCTION

In this paper I apply the approach to vowel harmony developed in van der Hulst (2011b, to appear,
ms a.) to Turkish and Hungarian. In section 2 I briefly introduce the model, while sections 3 and 4
deal with Turkish and Hungarian, respectively. In section 5 I offer some conclusions.

2 T HE R CV P M ODEL

2.1 G ESTURES , E LEMENTS AND H EADEDNESS

In RcvP1 each vowel segment has a tripartite structure consisting of the Laryngeal, Manner and
Place gesture, as in (1). Within each gesture, we find precisely two elements.

∗
This paper reflects the result of work in progress in which I explore the application of a model for vowel harmony that
is in development to two languages, Turkish and Hungarian. I regard the analyses presented here as first explorations
of some of the complexities that these languages offer. I gladly contribute this paper to this collection which honors my
colleague and friend Glyne Piggott. I am very lucky for having had the opportunity to directly work with Glyne and
profit from his sharp and encouraging style of thinking. I look forward to many more of his insightful contributions to
the field of phonology and beyond.
1
RcvP stands for ‘Radical CV Phonology’, a dependency-based approach that is developed in van der Hulst (1995, 2005,
2011b; mss. a, b, c). This section presents a simplified version of the model and is almost identical to section 3 in van
der Hulst (2011b), where the model is applied to vowel harmony in Yoruba. I need to include this brief introductory
section here in order to give the reader the necessary background for assessing the analyses offered in this paper.

McGill Working Papers in Linguistics, Volume 22.1, Winter 2012

2012
c by Harry van der Hulst
VOWEL HARMONY IN T URKISH AND H UNGARIAN 2

(1)

Laryngeal (tone)

{L, H} Manner (aperture) Place (color)

{A, ∀} {U, I}
The six elements in (1) can in fact be replaced by just two, viz. |C| and |V|, as in (2) (hence the name
Radical cv Phonology).

(2)
Element names RcvP coding
Manner A ∀ V C
Place U I V C
Laryngeal L H V C

This reduction is possible because each gesture contains exactly two elements. This allows us to say
that the element labels |A|, |U| and |L|, because they occur under different gestures, are paradigmat-
ically speaking in complementary distribution, and so can be reduced to one and the same element,
viz. |V|. The same holds for |∀|, |I| and |H|, which can be reduced to |C|. Complementary distribution
is a familiar criterion used to reduce allophones to phonemes (where allophones are in complemen-
tary distribution in a syntagmatic sense). However, the same criterion can be applied to elements,
provided that the elements that we reduce to |C| or |V| have something in common. In RcvP, the
claim is that in each gesture, |A|, |U| and |L| represent vowel- or rhyme-oriented choices, and so
reduce to |V|, while |∀|, |I| and |H| represent consonant- or onset-oriented choices, and so reduce to
|C|. However, for practical purposes, I will normally use the element labels |A, U, L,∀, I, H| so as to
avoid cumbersome expressions such as ‘|Place: V|’(instead of ‘|A|’).
Gestures may contain a single element or a combination of elements. In the latter case, the
elements enter into a head-dependency relation, such that one element is the head and the other the
dependent. Headedness will be graphically indicated by underlining.
Following Kaye, Lowenstamm and Vergnaud (1985), I assume that an element in dependent
position corresponds to a single phonetic attribute while an element in head position (without an
accompanying dependent) denotes a complete segment – an idea that is also implicit in Dependency
Phonology (Anderson and Ewen 1987). For example, in isolation the head element |U| denotes back
rounded /u/, while dependent |U| denotes labiality (i.e., rounding), as in /y/2 . If the element |U|
occurs with a dependent |I|, the result is a ‘fronted /u/’, i.e. a kind of ‘central’rounded vowel. I will
return to the interpretation of such structures shortly.

2
For ease of exposition I refer to phonemic entities (hence the use of slant lines), although it is important to bear in
mind that IPA symbols represent approximate phonetic values. A notation such as /u/ is shorthand for the correspond-
ing elemental structure, which represents a cognitive phonological entity that plays a contrastive role in a particular
language.
3 H ARRY VAN DER H ULST

The idea of having different (but related) phonetic interpretations for elements is referred to as
the ‘dual interpretation’of elements (cf. van der Hulst 1988a). Consider for example (3):

(3) The dual interpretation of A and ∀3

Head Dependent
A /a/ ‘low (or retracted)’
∀ /i/ ‘high (or rather ATR; cf. below)’
Combinations of |A| and |∀| denote intermediate aperture degrees and correspond to intermediate
central vowels, as in (4):

(4) A A∀ ∀A ∀
/a/ /3/ /@/ /1/ (central)
/2/ /7/ /W/ (back unrounded)
In the case of a three-way distinction in aperture, there is a choice of IPA symbols provided we
make the (not uncommon) assumption that languages never employ a phonological contrast between
central and back unrounded vowels.
All vowels considered so far are ‘manner-only’or ‘colorless’vowels. Let us therefore now con-
sider the ‘color’elements, viz. |U| and |I|. These have the dual interpretations in (5):

(5) Interpretation of U and I

Head Dependent
U /u/ ‘round’
I /i/ ‘front’
Adopting (for convenience) the ‘autosegmental line-notation’that was proposed in Kaye, Lowen-
stamm and Vergnaud (1985), in which headedness is indicated by underlining, yields the following
representations for the four front unrounded vowels:

(6) /æ/ /E/ /e/ /i/

∀ ∀ ∀
A A A
I I I I4
The next question that must be addressed concerns the interpretation of combinations of |U| and |I|.
Consider (7), where |U| and |I| are combined with the aperture element |∀|:

(7) /u/ /0/ /y/ /i/

∀ ∀ ∀ ∀
U U U
I I I

3
Here I use articulatory labels for the phonetic interpretation of elements. Elements are also associated with an acoustic
interpretation.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 4

Both Anderson and Ewen (1987) and Kaye, Lowenstamm and Vergnaud (1985) stipulate that there
is no difference between |UI| and |UI|, which both yield a rounded front vowel. However, I assume
that these combinations in fact denote two distinct vowels, which are sometimes referred to as ‘out-
rounded’ /y/, i.e. a rounded front vowel, and ‘in-rounded’/0/, a fronted back-round vowel. These
vowels are sometimes contrastive, for example in Swedish.
Cross-classifying aperture and color, and allowing for both colorless and mannerless vowels,
yields 25 different monophthong vowels, given in (8):

(8)
I IU Colorless UI U
∀ i y 1∼W 0 u
Mannerless I Y @ (∼ schwa) U
∀A e ø @∼7∼5 8 o
A∀ E œ 3∼2 Æ O
A æ Œ a∼A 6
Each language selects a subset of these vowels by imposing combinatorial constraints, which will
be exemplified below for the Yoruba system.

2.2 T HE VOWEL HARMONY RELATION

In a unary system, it is impossible to say that vowel harmony is the result of ‘needy’, i.e. under-
specified, vowels (cf. Nevins 2010), since there is no underspecification to express this ‘neediness’.
Nevertheless, the idea of ‘neediness’ is similar in spirit to the Government Phonology approach,
where ‘emptiness’, or more specifically empty nuclei, require licensing (cf. Kaye, Lowenstamm and
Vergnaud 1990). In van der Hulst (2011b), I suggest that a vowel that undergoes vowel harmony
possesses the harmonizing element as a variable element, i.e. an element which can be present only
if it is licensed by an adjacent non-variable, i.e. licensed, occurrence of that same element. Thus,
alternating vowels contain the harmonic element as a ‘variable’; variable elements will be repre-
sented between parentheses5 . Segmental structures that contain a variable element will be called
morphophonemes. For example in Finnish, all vowels that are front (in stems) or can be front (in
suffixes) are represented as morphophonemes:

(9) /u/ ∼ /y/ {U (I) }

/o/ ∼ /ø/ {AU (I)}
/a/ ∼ /æ/ {A (I) }
Variable elements represent the neutralization of the contrast between presence and absence of an
element. In a binary feature system, absence of contrast would be expressed by saying that the
harmonic feature is not specified, but in the current model it is expressed in terms of a variable
element ‘(X)’which means ‘X or nothing’. It must be borne in mind that the so-called zero that is
used in underspecification also refers to a disjunction namely ‘+ or –’.

5
I will refine this notion in our discussion of ATR-harmony in Yoruba.
5 H ARRY VAN DER H ULST

While a complete theory of licensing remains to be developed (see van der Hulst ms.b), van der
Hulst (2011b) proposes two types of licensing6 :

(10) a. Positional licensing

An element X is licensed in position P (where P is the first/last syllable in domain D,
where D is a Word or Stem/Root)
b. Lateral licensing
A variable element (X) is licensed by an adjacent occurrence of X

The crucial aspect of positional licensing is that position P licenses the presence of element E which
implies that in this position we can find a contrast between the presence of E and its absence. In all
other positions, no contrast is allowed which means that in those positions we always find (E).
If positional licensing rules alone, we find that a potential contrast is neutralized in the non-
privileged position. If lateral licensing is active, we get the effect that the non-privileged positions
agree with the privileged position for the element E. In other words, the fact of vowel harmony is
expressed in terms of lateral licensing.
It might be asked why I use the notion of variable element and lateral licensing. The motivation
for using the variable notation, rather than saying that vowels in non-privileged positions simply lack
the element E, comes from disharmonic roots and non-alternating affix vowels. With the variable
notation we can distinguish vowels that alternate from vowels that do not alternate either by always
having E or never having E. Thus we can make a three-way distinction:

(11) a. (E) b. E c.
X X X
a = alternating vowel element must be licensed to get interpreted
b = invariant E
c = invariant non-E

A disharmonic root with (for example) a front vowel preceding a back vowel (e.g., ü - o) would
have the positionally licensed I-element associated to the front vowel and there would be no vari-
able element on the back vowel. This can be illustrated with the Hungarian disharmonic root büro
‘desk’(Hungarian has palatal harmony, i.e. harmony for the element |I|):

6
In van der Hulst (2011b) positional licensing refers to ‘a variable element (X)’. Here I simplify that to ‘an element
X’. Also, here I assume that the directionality of lateral licensing does not need to be stated. Any variable element is
licensed by either a preceding or following licensed instance of the same element; see van der Hulst (ms. b) for further
discussion of this point.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 6

(12) a. A
I
U U
b ü r o

b. A
I
U U
k o sz t ü m

The second example illustrates an invariable occurrence of the element |I| in non-initial position.
We could say that in this case the element |I| is ‘lexically licensed’, which essentially means that its
occurrence is irregular, just like büro is irregular in lacking the variable (I) in non-initial position.
Affixes can also have disharmonic vowels. A suffix that is invariably back would have option
(11c), whereas a suffix that is invariably front would have option (11b). Regular alternating suffixes
would have option (11a). If we were to adopt a spreading (or copy) model, using unary elements,
we cannot distinguish between (a) and (c).
Now, whereas vowels in harmonic affixes de facto alternate (which is adequately expressed by
supplying its vowels with the variable element, i.e. the neutralization option), vowels in stems do
not (in so-called stem or root-controlled systems). It might be said that we could therefore assume
that all vowels are marked with this element as invariable. However, there is interesting evidence
(provided in Harrison and Kaun 2001) that non-initial vowels in, for example, Turkish and Finnish
are not invariably specified with the harmonic value. This evidence is based on certain language
games. I refer to van der Hulst (ms. b) for discussion.
Another question that arises is whether vowels that contain the harmonic element predictably
(such as /i/ and /e/ in Finnish which always contain the element |I|, or /i/ and /u/ in Yoruba which
contain the ATR-element predictably) must have this element lexically (invariably in the privileged
position and variably in other positions) or rather acquire it by a redundancy rule. In line with my
discussion in van der Hulst (2011b), but contrary to van der Hulst (to appear) where I consider
the alternative of full specification, I will suggest here that predictable elements remain unspecified
except when their presence is licensed. This licensing happens in two circumstances. Firstly, as
shown in (13b), the neutral vowel /i/ in Hungarian will be followed by a front suffix when preceded
by a front vowel that is when its ‘latent’element |I| is laterally licensed, but not when preceded by a
back vowel (see 13a):
7 H ARRY VAN DER H ULST

(13) a. ∀ ∀
A
(I)
p a p i: r – b o: l ‘of paper’

b. ∀ ∀ ∀
A
I > I > (I)
ü v e g – b ö l ‘of glass’
Secondly, when a root only contains neutral vowels, the neutral vowel in the first syllable will be
positionally licensed and thus be present7 :

(14) “front /i/”

∀
A
I > (I)
v i: z – n e k
Thus, neutral vowels act as licensors when preceded by a front vowel or when they are the only
vowel(s) in the stem. This raises the question how this element is interpreted if it is not laterally
or positionally licensed, for example in the (second syllable in the) form /papi:r/. At the phonemic
level every Hungarian vowel that has the element |∀| must have either |I| or |U|. I will argue in section
4 that predictable elements are added in order to satisfy the constraints that hold for the phonemic
level, the output of the lexical phonology, i.e. at the phonemic level where all segmental structure
must be wellformed, i.e. in accordance with the constraints. This implies that lexical representations
can deviate from phonemic structures, but only, as I will argue, in terms of forming substructures of
phonemic structures.

3 T URKISH

In this section I will first offer an analysis of vowel harmony and disharmony in Turkic languages. I
start the discussion with a detailed analysis of standard Turkish. The most recent extensive treatment
is offered in Clements and Sezer (1982) and Kabak (2011), from which this section takes much of
its examples. The formal description that I offer, however, assumes a rather different conception
regarding the nature of phonological primes than these authors do.

7
Hungarian has about 60 stems which are exceptional in taking a back vowel in the suffix; see section 5 for an analysis
of Hungarian.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 8

3.1 T HE BASIC FACTS

It is not difficult to formulate the general vowel harmony requirements that hold for Turkish. The
statement appears again and again in elementary textbooks and numerous studies. Turkish has 8
contrastive vowels:

(15) [front] [back]

[non-round] [round] [non-round] [round]
[high] i y 1 u
[low] e ø a o
Any of the eight Turkish short vowels may appear in the first syllable of a word. Any following
vowel must agree with the preceding vowel in frontness. As a result, all vowels in a regular stem
agree in frontness. This is traditionally called palatal harmony. For example:

(16) Examples of palatal harmony

hyviyet ‘identity’ k1m1lt1 ‘movement’
kysyly ‘annoyed’ oyuncak ‘plaything’
netice ‘result’ sogukça ‘ice cold’
In addition, a non-initial high vowel agrees with the preceding high or non-high vowel in roundness.
This is apparent in the form oyuncak ‘plaything’: the high /u/ is rounded just like the initial /o/, but
the /a/ is not rounded as it is low. This process is called labial harmony.
Because the roundness of initial vowels does not extend to low vowels in subsequent syllables,
the vowels /o/ and /ø/ may not occur as a result of vowel harmony in any syllable except the first.
I shall refer to this constraint on as the Ònon-initial /o-ö/ prohibitionÓ. Below I will discuss the
nature of this constraint. I summarize the three harmony requirements of Turkish in (17):

(17) Generalizations
a. All vowels in a regular stem agree in frontness
b. Any high vowel agrees with the immediately preceding vowel in roundness
c. The vowels /o/ and /ø/ can only occur in stem-initial syllables.

Following Haiman (1972), I would like to discuss an explanation as to why constraint (17c) should
be part of the grammar of Turkish. Suppose for a moment that Turkish only had palatal harmony,
i.e. (17a). In that case, in non-initial syllables a four-way contrast would exist8 :

(18) a. [-back] [+back]

[-round] [+round] [-round] [+round]
[+high] i y 1 u
[-high] e ø a o

8
When I use traditional binary features I do this to refer to ‘phonetic properties’. These features have no status in Element
Theory.
9 H ARRY VAN DER H ULST

b. non-initial syllables (palatal harmony, phase I)

[-round] [+round]
[+high] I (i/1) U (y/u)
[-high] A (e/a) U, A (ø/o)

c. non-initial syllables (palatal harmony, phase II)

[-round] [+round]
[+high] I (i/1) U (y/u)
[-high] A (e/a)
Constraint: |U, A| is blocked from non-initial syllables
d. non-initial syllables (after adding labial harmony)
[-round] [+round]
[+high] I (i/1)
[-high] A (e/a)
Constraint: |U, A| is blocked from non-initial syllables

In (18b), the capital letters represent elements. If we now add constraint (17c) to the grammar, this
set is reduced to that of I, U and A, which can be identified as the typical minimal, unmarked vowel
system. As Haiman (1972) shows, in languages like Russian and Greek we see that in unstressed
syllables only the vowels /a/, /i/ and /u/ occur, whereas in stressed syllables a richer set occurs. By
eliminating |U,A| from non-initial syllables, Turkish achieves an unmarked three-way distinction in
non-initial syllables, as shown in (18c).
It would seem then that Russian, Greek and Turkish are united in limiting the position in which
vowels are maximally contrastive to one: the stressed position in Russian and Greek, the initial
position in Turkish, both ‘privileged’positions. Palatal harmony causes there to be two variants for
the three unmarked vowels, as shown in (18c).
Now, once rounding harmony is present in the grammar, the contrast in non-initial syllables is
reduced to A and I (18d), but due to the constraint introduced earlier in (18c), rounding harmony
cannot cause low round vowels. With labial harmony present, the motivation for (17c) is no longer
transparent, but apparently this did not immediately cause this constraint to disappear. As a result,
from the view point of Turkish-cum-labial harmony, the constraint is an arbitrary statement. I infer
from this that the constraint must have been added to the grammar before labial harmony because
the presence of (17c) can be understood if this condition became part of the Turkish grammar before
rounding harmony.
Thus we see that a restriction on labial harmony finds motivation in the original palatal system
when labial harmony was not yet around. As a result we do not find non-initial /o/ and /P/ (in roots
or suffixes) even where these vowel ‘could’result from labial harmony.
There are other possible explanations for the fact that low vowels refuse to harmonize in round-
ing, such as the idea that rounding is more marked for low vowels than for high vowels (see Kaun
2004) but I find Haiman’s account very insightful.
Unrestricted or fully symmetrical harmony systems can be described in a number of ways, and
it would be difficult to choose between different descriptions if one is not otherwise predisposed
VOWEL HARMONY IN T URKISH AND H UNGARIAN 10

toward a particular framework. It is often the case that restrictions on or exceptions to a particular
harmonic pattern, causing various patterns of asymmetry, shed more light on its nature than the
regular instances. In the next section, I will show how it is possible to integrate the disharmonic
stems into an analysis of synchronic harmony9 .To achieve this goal, I first review which stem vowel
combinations are regular, which are disharmonic but do occur, and which are disharmonic and cate-
gorically ruled out, i.e. are not attested. Then, I will offer an account within the present framework.
As it will turn out, the statement as to which vowels can occur in disharmonic roots and which
cannot, can be reduced to a rather simple formula. I will take this as evidence that our account deals
with vowel harmony in a promising way.

3.2 ROOT DISHARMONY

All descriptions of palatal and labial harmony in Turkish acknowledge that there is a host of excep-
tions to both processes, as well as to the non-initial /o-ø/ prohibition. I shall illustrate these below.
As Clements and Sezer (1982) (henceforth: CS) point out in great detail, within stems many excep-
tional patterns to vowel harmony arise. In particular, vowels from the set /i e a o u/ may combine
quite freely. However, patterns which include the vowels /y ø1/ are much more limited. CS argue
that they are absent except for the occurrence of a number of stems combining /i/ and /y/ (in vi-
olation of labial harmony). Kabak (2011) disagrees with CS and points to disharmonic roots that
involve /y ø1/. Having noted his observations, I still assume that the relative rarity of such vowels
in disharmonic roots has a systematic basis. CS do not offer an explanation for the rarity of these
vowels in disharmonic roots, but decide on the basis of the exceptions that harmony is no longer
active in roots. I shall argue against this view and suggest that the disjunction in the vowel set can
be understood if I assume that vowel harmony is governed by unary elements that either regularly
‘extend’over the word domain, or, irregularly, are linked to specific vowel positions.
First, I summarize the co-occurrence patterns of vowels. In the table below, an empty box
indicates that the pattern /...V1...V2.../ is regular (and attested). A mark in a box indicates that the
pattern is disharmonic with respect to either palatal (P) and/or labial (L) harmony. The symbol C is
given when the pattern would violate the constraint against non-initial low round vowels. However,
as was pointed out above, disharmonic roots involving the vowels /i e a o u/ may still occur. Irregular
pattern that thus occur are marked with bold. The row of vowels stands for the first syllable, while
the column represents non-initial syllables.

9
See van der Hulst and van de Weijer (1997) on which this discussion is based.
11 H ARRY VAN DER H ULST

(19)
σ1\σ2 i y 1 u e ø a o
i L P PL C P PC
1 L PL P C P PC
1 P PL L P PC C
u PL P L P PC C
e L P PL C P PC
ø L PL P C P PC
a P PL L P PC C
o PL P L P PC C

L = disharmonic due to labial harmony

P = disharmonic due to palatal harmony
C = illegal due to the “non-initial /o-ö/ prohibition"
Bold = attested as an exception

On the basis of the exceptions, CS conclude that within stems neither palatal nor labial harmony
holds in Turkish. I fail to see that we should draw that conclusion. Synchronic harmony on suffix
vowels is independently needed. The stems which conform to the harmonic pattern can therefore
simply get a ‘free ride’on the harmony rules. This simplifies their underlying representation consid-
erably10 .

3.2.1 F ORMAL PRELIMINARIES

Under the approach presented in section 2 the vowel inventory of Turkish is represented as follows:

(20) /i/ /e/ /y/ /ø/ /1/ /a/ /u/ /o/

∀ ∀ ∀ ∀
A A A A
I I (I) (I) U U
U U
The vowel /1/ is analyzed as merely |∀|, although below I will suggest that this vowel seems to
behave as a |IU| combination (i.e. /W/) in stems (see the table in 8).
A simple situation of harmony can be represented as follows:

(21) ∀ - A A - ∀ A - A ∀ - ∀
I > (I) I > (I) I > (I) I > (I)
U > (U) U > (U) U > (U) U > (U)
The elements |I| and |U| are licensed in the first syllable and variable elsewhere. Recall that a variable
element is licensed if preceded by a licensed instance of that element.

10
Within stems epenthetic vowels also harmonize although this may have to be analyzed as an independent harmony
system (see Kabak 2011).
VOWEL HARMONY IN T URKISH AND H UNGARIAN 12

3.2.2 H ARMONY AND DISHARMONY

If we take bisyllabic stems as representative, we need to consider four combinations:

(22) ∀ - A A - ∀ A - A ∀ - ∀

We can cross-classify these stem types with the four possible combinations of harmonic elements:

(23) a. No harmonic element

b. I harmonic element only
c. U harmonic element only
d. I and U harmonic elements

I shall discuss the combinations of intrinsic properties and harmonic properties in turn.
No harmonic element
If no harmonic element is present the regular patterns surface as follows:

(24) ∀ - A A - ∀ A - A ∀ - ∀
/1 a/ /a 1/ /a a/ /1 1/
g1rtlak ‘throat’ alt1 ‘six’ kara ‘black’ k1s1m ‘part’
h1tta ‘province’ yal1 ‘villa’ tavşan ‘rabbit’ s1n1r ‘border’
k1yak ‘excellent’ kad1 ‘judge’ hasta ‘sick’ s1g1r ‘sick’

I harmonic element only

We now move to the stems which contain one harmonic element. First, let us consider regular cases
with the I-harmonic element. This harmonic element associates to all vowels in the stem:

(25) ∀ - A A - ∀ A - A ∀ - ∀
I > (I) I > (I) I > (I) I > (I)
/i e/ /e i/ /e e/ /i i/
ince ‘thin’ degis ‘change’ kere ‘time’ kişi ‘person’
igne ‘needle’ yedi ‘seven’ gebe ‘pregnant’ gibi ‘like’
diçer ‘other’ aski ‘cold’ tepe ‘hill’ inci ‘pearl’

Stems with the I-harmonic element can be disharmonic in two ways. A non-initial syllable either
misses the variable element (26) or it has the element ‘lexically licensed’(27):

(26) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
/i a/ */e 1/ /e a/ */i 1/
siyah ‘black’ — elma ‘apple’ —
inan ‘believe’ beyan ‘declaration’
idrak ‘perception’ mezat ‘auction’
13 H ARRY VAN DER H ULST

(27) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
*/i e/ /a i/ /a e/ */1 i/
— tatil ‘vacation’ haber ‘news’ —
dani ‘also’ kardes ‘brother’
hangi ‘which’ anne ‘mother’

We see that all patterns are possible exceptions, except those which would produce a bare |∀| in a
root that contains the element |I|. It would seem that bare |∀| acts as a ‘black hole’that must attract
the element |I|, if present.
U harmonic element only
Let us turn to stems containing the U-harmonic element:

(28) ∀ - A A - ∀ A - A ∀ - ∀
U > (*U) U > (U) U > (*U) U > (U)
/u a/ /o u/ /o a/ /u u/
tuhaf ‘strange’ soguk ‘cold’ boga ‘bull’ kuru ‘dry’
bugday ‘wheat’ yorgun ‘tired’ oda ‘room’ ugur ‘fortune’
muhak ‘new moon’ oku ‘read’ dogar ‘is born’ nutuk ‘speech’

Non-initial |U| cannot be licensed by a preceding licensed |U| due to constraint (17c). Thus, in
the two cases in the box, even though the U-element is licensed it is nonetheless prevented by a
sequential constraint. However, it must be noted that the pattern /o-o/ does occur, which would
require a lexical association of U to both V-positions. /u-o/ is not reported in CS:

(29) A - A
U U
/o o/
horoz ‘rooster’
moloz ‘debris’
lodos ‘southwest wind’

In addition, exceptional patterns can arise due to lexical association of |U| to either of the two V-
positions:

(30) ∀ - A A - ∀ A - A ∀ - ∀
U U U U
/u a/ */o 1/ /o a/ */u 1/

Examples of lexically associated |U| in the first syllable with |A| in the next syllable produce the
same result as regular cases because |U| does not associate to |A| due to (17c).
VOWEL HARMONY IN T URKISH AND H UNGARIAN 14

(31) ∀ - A A -∀ A - A ∀ - ∀
U U U U
/1 a/ /a u/ ?/a o/ */1 u/
— marzul ‘lettuce’ (see 17c) —
arzu ‘desire’
yakut ‘emerald’

As in the case of the |I| element, we note that disharmonic patterns which would result in a bare ∀
position are ill-formed. This, however, does not account for the absence of /a – o/, which again is
due to (17c). Since irregular non-initial /o/ occurs with other vowels preceding it (i.e. /e o/), one
would expect that exceptions involving /a o/ should not be ‘too bad’.
We conclude the following: that in the presence of |I| and/or |U|, bare ∀ is not permitted in a
root.
I and U harmonic elements
With respect to the combined presence of both harmonic elements, we also start with the regular
pattern:

(32) ∀ - A A - ∀ A - A ∀ - ∀
I > (I) I > (I) I > (I) I > (I)
U > (*U) U > (U) U > (*U) U > (U)
/y e/ /ø y/ /ø e/ /y y/
dymen ‘wheel’ søgyt ‘willow’ øyle ‘thus’ ytyn ‘iron’
dygme ‘button’ gönyl ‘heart’ gönder ‘send’ yzyim ‘grape’
myspet ‘proven’ døvys ‘fight’ køpek ‘dog’ ‡ynky ‘because’

Analogous to the above, we can imagine a number of patterns with lexical associations. For exam-
ple, |I| can be missing in non-initial position or |U| can be invariable in non-initial position (as in
33), or the other way around (as in 34):

(33) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
U U U U
/i o/ /e u/ /e o/ /i u/
pilot ‘pilot’ mevzu ‘topic’ petrol ‘petrol’ billur ‘crystal’(rare)
cinko ‘zinc’ memur ‘official’ peron ‘platform’
sifon ‘toilet flush’ mebus ‘MP’ metot ‘method’
(34) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
U U U U
/u e/ /o i/ /o e/ /u i/
lutfen ‘please’ bobin ‘spool’ otel ‘hotel’ muzip ‘mischievous’
suret ‘manner’ polis ‘police’ rozet ‘collar pin’ kulis ‘stage wing’
kudret ‘power’ torik ‘blue fish’ model ‘model’ muhit ‘neighbour-hood’
15 H ARRY VAN DER H ULST

All these patterns are reported by CS as possible exceptions.

All other disharmonic patterns will involve either the presence of both harmonic elements on
one of the V-positions, or the presence of one harmonic element on a single V-position with the
other on both; twenty of such patterns are logically possible and of these only two cases occur that
we have not seen before:

(35) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
U > (*U) U > (U) U > (*U) U > (U)
*/y a/ */ø u/ */ø a/ */y u/
(36) ∀ -A A ∀ - A A- ∀ -∀
I I I I
U > (*U) U > (U) U > (*U) U > (U)
/u e/ */o y/ /o e/ */u y/
(see 21) (see 21)
(37) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
U U U U
*/y a/ */ö 1/ */ø a/ */y 1/
(38) ∀ - A A - ∀ A - A ∀ - ∀
I I I I
U U U U
*/1 ö/ */a y/ */a ø/ */1 y/
(39) ∀ - A A - ∀ A - A ∀ - ∀
I > (I) I > (I) I > (I) I > (I)
U U U U
?*/y e/ */ø i/ */ø e/ /y i/
— — — ymit ‘hope’
(see below) mymbit ‘fertile’
ymmi ‘illiterate’
(40) ∀ - A A - ∀ A - A ∀ - ∀
I > (I) I > (I) I > (I) I > (I)
U U U U
*/i ø/ ?*/e y/ */e ø/ /i y/
— — — tifys ‘typhus’
(see below) virys ‘virus’
bitym ‘bitumen’
It would seem desirable to find one generalization that rules out the marked vowels from dishar-
monic patterns (except the combination of /y/ and /i/). Somehow, we need to appeal to the fact that
the five vowels /i u e o a/ represents the most common 5-vowel system. The simplest solution would
VOWEL HARMONY IN T URKISH AND H UNGARIAN 16

be to show that /1 øy/ have something in common which is excluded to be ‘concentrated’in a single
position.
Goldsmith (1990, 304ff.) also argues for an analysis in which the fact that the vowels /i e a o u/
combine freely is not an arbitrary stipulation. He points out that this is a favored five-vowel system.
His analysis of the five vowels does not make use of the feature [±front] (or [±back]). Hence, stems
containing these vowels do not violate a rule governing palatal harmony, i.e. agreement for the
feature [front]. Whenever stems contain /y ø 1/, specification of this feature is required (i.e. [+front]
for /y ø/, and [-front] for /1/). Thus only ‘disharmonic’roots (in the traditional sense) containing any
of the three vowels are truly disharmonic in Goldsmith’s feature system. This is a clever solution,
cast in a binary-feature approach.
Given the representation of Turkish vowels in (20), the reasons for not allowing /1/ and /øy/
must be different. We could say that the vowel that has just |∀| functions as an ‘attractor’: if a
harmonic element is present, it cannot remain colorless. On the other hand for the vowels /ø y/
both harmonic elements are present and in this case, apparently, it is not possible to ‘restrain’ the
harmonic elements, i.e. vowels with two harmonic elements functions are ‘projectors’.
This suggests that combinations of /e/ and /y/ should also be acceptable, since here at least the |I|
harmonic element is not restrained to a single position. Indeed, Polgárdi (1998) observes that dishar-
monic patterns involving /e/ and /y/ are quite common (like /i/ and /y/) and states a generalization,
which I here formulate as follows:

(41) |I| and |U| can only be associated to a single position if the combination is harmonic for |I|

However, this seems to allow all combination in (39) and (40), so we would have to add: if both
positions share |∀|:

(42) |I| and |U| can only be associated to a single position if the combination is harmonic for |I|
and |∀|

But note that this then excludes combinations of /e/ and /y/ which Polgárdi (1989) claims are not so
bad. Aside from this problem, the absence of /1/ in disharmonic domains needs a separate statement:

(43) In the presence of I and/or U, bare ∀ is not permitted

However, there may be a way to achieve Goldsmith’s generalization if we represent the vowel /1/
differently, namely as a combination of |I| and |U| (which would be /W/ according to the table in 8):

(44)
/i/ /e/ /y/ /ø/ /1/ /a/ /u/ /o/
∀ ∀ ∀ ∀
A A A A
I I I I I U U
U U U

This would not extend to suffixes, however, where the representation with variable |I| and |U| is
required; see below.
17 H ARRY VAN DER H ULST

With the representation |∀ I U| for the vowel /1/, the basic statement concerning disharmonic
roots is that the combination |I U| cannot be restricted to a single position:

(45) IU combinations are not allowed if one or both are not harmonic, unless both I and ∀ are
harmonic

Note that this also excludes combinations of /e/ and /y/ as permissible under the unless-clause.
Note that this statement suggests an asymmetry among the two color elements since the har-
monic |IU| combination is not allowed if both |U| and |∀| are harmonic. We will later see whether
this can be seen as part of a larger generalization regarding an asymmetry between the elements |I|
and |U|.
We have now dealt with disharmonic roots internally. The next question is how roots of this sort
behave with respect to suffixes.

3.3 S UFFIXES
3.3.1 T HE REGULAR CASES
Most suffixes undergo regular harmony. High suffix vowels undergo both palatal and labial har-
mony. Consider the following set of representative examples:

(46) nom.sg poss. abl. nom.pl. poss./acc.pl.

‘room’ oda odas1 odadan odalar odalar1
‘end’ son sonu sondan sonlar sonlar1
‘pipe’ boru borusu borudan borular borular1
‘village’ køy køyu køydan køyler køylar1
‘worm’ kurt kurdu kurttan kurtlar kurtlar1
‘fox’ tilki tilkisi tilkiden tilkiler tilkileri
‘cow’ inek inei inekten inekler inekleri
‘river’ dere deresi dereden dereler dereleri
‘horse’ at at1 attan atlar atlar1
‘taste’ tat tad1 tattan tatlar tatlar1
‘girl’ k1z k1z1 k1zdan k1zlar k1zlar1
‘container’ kap kabi kaptan kaplar kaplar1
‘iron’ yty ytysy ytyden ytyler ytyleri

The fact that low suffix vowels do not undergo labial harmony follows from the constraint on non-
initial /o-ø/, which I already assumed for stem-initial harmony. The fact that after low vowels only
non-round vowels can appear, shows us that licensing in Turkish (and, I would submit: universally)
is local, i.e. involves vowel positions in adjacent syllables. Consider the underlying representation
of pullar1n ‘stamp nom.pl.’:

(47) U (U) (U)

p∀l - lAr - ∀n → [pul - lar - 1n]
VOWEL HARMONY IN T URKISH AND H UNGARIAN 18

The U harmonic element cannot license the variable element for the low vowel because of the
non-initial /o-ö/ prohibition (17c). As a result, the variable element in the second suffix cannot be
licensed either.

3.3.2 D ERIVED ENVIRONMENT EFFECTS ?

In other models that use unary primes and in which harmony is seen as ‘spreading’ a potential
problem arises when it comes to disharmonic roots. If a front vowel occurs in the stem-final syllable
it will spread to suffix vowel, whereas it cannot do that if it is associated to the stem- initial vowel:

(48) mezar-lar ‘auction’ and hotel-ler ‘hotels’

a. mA z A t ] l A r
I
b. hA t A l ] l Ar
I

To prevent a lexically associated |I| from spreading to the other vowel(s) of the stem, while allowing
it to go to the suffix vowel if it occurs in the stem-final syllable van der Hulst and Smith (1986)
and Polgárdi (1998) appeal to the notion derived environment. Kiparsky (1973) has shown that
phonological rules may be prevented from applying within morphemes if their effect is neutraliz-
ing. It is assumed that lexically associated elements are subject to this principle, whereas floating
elements are not because associating the latter does not eliminate a potential lexical contrast assum-
ing that a floating element is not distinct from a lexically associated element. Non- association is
a form of underspecification and underspecified representations cannot be taken to be distinct from
representations that differ only in adding association lines.
In the model developed here, no appeal to a derived environment condition is necessary because
the two cases in (48) would have vowels without the variable element. This being so, there is no
licensing ‘action’ in the first form:

(49) mezar-lar ‘auction’ and hotel-ler ‘hotels’

a. mA z A t ] l A r
I (I)
b. hA t A l ] l Ar
I >(I)

This leaves open whether derived environment condition may be motivated for other cases.

3.4 I RREGULAR SUFFIXES

CS discuss a number of exceptional suffixes11 :
11
Haiman (1972) also mentions the disharmonic diminutive –os. In various publications Kardestuncer argues that the
suffixes which fail to undergo harmony do not really qualify as suffixes. For example, in Kardestuncer (1983) the point
is made that /-iyor/ is not a suffix but a compound component, i.e. a stem. It also appears that the suffixes that are
exceptional to vowel harmony also have exceptional stress. However, this does not explain that suffixes can be partially
19 H ARRY VAN DER H ULST

(50) a. gel-iyor-um ‘I am coming’

koş-uyor-um ‘I am running’
gyl-yyor-um ‘I am laughing’
bak-1yor-um ‘I am looking’
b. yc ‘three’ yc-gen-ler ‘triangles’
alt1 ‘six’ alt1-gen-ler ‘hexagonals’
sekiz ‘eight’ seki-gen-ler ‘octagonals’
cok ‘many’ cok-gen-ler ‘polygonals’
c. arab-istan-1 ‘Arabia’
ermeni-stan-1 ‘Armenia’
mool-istan-1 ‘Mongolia’
tyrk-istan-1 ‘Turkestan’
d. gid-edur-sun ‘let him keep coming’
kos-edur-sun ‘let him keep running’
gyl-edur-sun ‘let him keep laughing’
bak-edur-sun ‘let him keep looking’
In two other suffixes, an invariant /a:/ occurs preceding an invariant front vowel:

(51) -a:ne : denominal, adjective-forming

-va:ri : denominal, adverb-forming

We can represent these suffixes as follows12 :

(52) /∀yAr / /gAn/ /∀stAn/ /Ad∀r/ /A:nA/ /vA:r∀/

(U)U I I (I) I I
(I) (U) (U) (U)U (U)(U) (U)(U)
In the third case, (-istan), a derived environment condition can not prevent rounding coming from
the stem. This is a problem for a spreading model, whereas in the present approach this is explained
by the absence of variable (U). There is, in addition, the suffix /-gil/, with a front non- alternating
vowel, which raises the same issue.

3.5 T HE DIRECTIONALITY OF LICENSING

Another problem for the derived environment approach is that the harmonic element of a dishar-
monic suffix cannot license a variable element in the stem. The fact that licensing cannot go from
suffix to root can be demonstrated with the following example:

disharmonic such as /-Iyor/ and /-Adur/.

12
Recall that all non-initial vowels have both color elements as variable unless they are disharmonic in which case they
lack it or have it invariably.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 20

(53) al t1 -gen -ler

∀
A A A
(I)/< I >(I)
(U) (U)
In this example, the variable (I) in the plural suffix is licensed. Crucially, the (I) on the final vowel
of the stem does not get licensed by the I-element of –gen. It could be said that what prevents
licensing from a suffix toward the final stem vowel is that the direction of licensing is rightward13 or
that licensing is ‘root-controlled’. This latter approach would stipulate that roots cannot harmonize
to affixes. Note, though, that ‘root’ here cannot mean ‘simplex root’, but rather ‘base’ since a vowel
in affix A cannot harmonize with the vowel of an affix B that is added to a base containing A:

(54) /i/ does not turn into /1/

[[[ root ] affix A ] affix B ]
X
-va:ri -. . . a. . .

Hence a better term would be ‘base-control’.

That licensing cannot go from suffix to root can be demonstrated with the following example:

(55) al t1 -gen -ler

∀
A A A
(I)/< I >(I)
(U) (U)
In this example the variable (I) in the plural suffix is licensed. Crucially, the (I) on the final vowel
of the stem does not get licensed by the I-element of –gen.
One reason for being suspicious of stipulating directionality is that in Turkish prefixes do not
integrate with stem like suffixes do; in that sense, prefixes behave like independent words, just like
members of compounds. I suspect that the deeper notion here is that vowel harmony in Turkish (and
many other cases) reflects base control which I here formulate as follows:

(56) Base control

Only base elements can be licensors

In so-called dominant-recessive harmony systems, base control does not apply. See van der Hulst
(ms. a) for further discussion of such systems. In such systems all morphemes contain the harmonic

13
Anderson (1980) argues that labial harmony in Turkish is directional, (i.e. left-to-right) given that the sequence /u a/ is
regular, whereas the sequence /a u/ is exceptional (even though such exceptions do exist). This only follows if labial
harmony is rightward directional. I would agree that this is the case for root-internal harmony and this is reflected in
the fact that I take the first syllable to be the privileged syllable.
21 H ARRY VAN DER H ULST

element as variable except for some (which can be roots or affixes) which thus cause all variable
elements to be licensed.

3.6 C ONSONANTS AND HARMONY

3.6.1 T HE PALATAL LATERAL
Roots ending in the palatal lateral take front suffixes irrespective of the preceding vowel:

(57) goL - y ‘goal ACC’ kol - u ‘arm ACC’

usuL - sLz ‘without a system jorul - du - k ‘we were tired’

However, if the palatal lateral occurs in the middle of a stem it does not enforce the next vowel to
be front (Kabak 2011):

(58) ysLub - u ‘style-ACC’

ma:Lum - dy ‘known PAST (ma:Lum-du is also possible)

We must also note that the palatal approximant /j/ does not cause a following suffix to be front. This
suggests that the |I| element of the palatal lateral has a different structural status than the |I| element
of /j/. I propose the following analysis. Let there be a redundancy rule which stipulates that a root
ending in a palatal lateral has a floating |I| element at the end of the root which can act as a licensor:

(59) ]I > (I)

U (U) (U)
u suL -sLz

This analysis explains why root internal laterals do not cause front vowels to their right, if we assume
that floating elements must occur at an edge.

3.6.2 F RONT SUFFIXES AFTER NON - PALATAL CONSONANTS

CS observe front suffixes after a number of stems ending in velars:

(60) ‘explosion’ infiLak infiLa:ki

‘alliance’ ittifak ittifa:ki
‘fasting’ imsak imsa:ki
‘real estate’ emLak emLaki

CS also note that some velars sometimes require suffix vowels to be back. They suggest that the
examples in (60) end in a velar that is ‘palatalized’. This palatalization is deleted when the velar
consonants are word-final. But Kabak observes that front suffixes may also show up after other
roots that do not end in palatal laterals or velars:

(61) saat-i ‘watch/clock’

harf-te ‘on the letter’
VOWEL HARMONY IN T URKISH AND H UNGARIAN 22

Also, Kabak (2011) suggests that the unexpected backness is specific to suffixes and thus not a
regular property of roots:

(62) tasdik - 1 ‘confirmation POSS 3sg’

but:
tasdik - ler ‘confirmation PLl’
tasdik - siz ‘confirmation DER’
It is not clear whether the unexpected behavior of suffixes after roots that do not end in a palatal
lateral is really due to a property of the root final consonant or an arbitrary property of the roots in-
volved. (In parallel to the analysis for the palatal lateral,) I suggest that ‘fronting roots’ are provided
with a floating |I| element14 :

(63) A A I > (I)

saat i (instead of saat1)
Clearly, the various cases of overapplication of palatal harmony require further research.

3.7 L ABIAL ATTRACTION

In the literature on vowel harmony, special status is sometimes assigned to the occurring pattern /a
CL u/, in which CL is a labial consonant. The unexpected rounding of the non-initial high vowel
is attributed to the preceding labial consonant. However, CS show that the pattern /a - u/ also
frequently occurs when the consonant is non-labial, while on the other hand the pattern /a CL 1/ can
also easily be found:

(64) a. marul ‘lettuce’

fatura ‘invoice’
yakut ‘emerald’
b. sab1r ‘patience’
kap1 ‘door’
kam1s ‘reed’
We must also note that the pattern is not productive in suffixation:

(65) kitap - 1n ‘of a/the book’ (not: kitap – un)

Kabak (2011) concludes that ’labial attraction’does not form part of the synchronic phonology of
Turkish.

14
The suggestion has been made in Kabak (2011) that the roots in question fail to trigger harmony and that frontness is
the default value. That suggestion is difficult to implement in the element approach since the prediction is that if roots
fail to induce harmony, we get the suffix vowel without the harmonic element, not with the harmonic element.
23 H ARRY VAN DER H ULST

3.8 C ONCLUSIONS
The case of Turkish is revealing with respect to various important issues. Firstly, we learn that low
vowels in non-initial syllables fail to receive labiality and act opaque. This suggests that vowels that
are incompatible with the harmonic element must act opaque. This fact is accounted for given our
use of unary elements and the claim that licensing is local.
Secondly, we learn that non-harmonic vowels in disharmonic roots cannot contain a combination
of the elements |I| and |U|, which motivates the representation |IU| for /1/ in stems.
Thirdly, I concluded that licensing is base-controlled which means that invariable elements of
disharmonic affixes cannot license into the stem that these affixes have been attached to.

4 H UNGARIAN

We now come to a second case which is also much discussed in the literature. In this chapter
I analyze palatal harmony in Hungarian, which, as will see, also has a limited amount of labial
harmony.

4.1 T HE BASIC FACTS

Hungarian has the following vowel system:

(66) [–back] [+back]

[–round] [+round] [–round] [+round]
[+high, –low] i(:) y(:) u(:)
[–high, –low] e: ø(:) o(:)
[–high, +low] E a: O
All high and mid round vowels occur short and long, but, as indicated, in case of non-high non-
round front vowels and in the case of low (back) vowels, there is a quality difference between the
long and the short vowel. In the former case, the difference is mainly one of height, whereas in the
latter case the difference involves lip rounding. Only the difference between /e:/ and Eappears to
have consequences for how the vowel harmony system works.
Let us first list the harmonic pairs (abstracting from length):

(67) Harmonic pairs Neutral

/y/ - /u/ /i/, /e:/, /E/
/ø/ - /o/
/a:/ - /e:/
/O/ - /E/
/i/ (long and short) is clearly neutral. Since /i/ is front and frontness (i.e. |I|) is the harmonic element,
we expect it to be transparent, and this will turn out to be the case, except in the case of a sequence
of /i/’s. The status of /e/ and /E/ is ambiguous, in particular that of /E/. First of all we note that in
suffixes /e:/ forms the harmonic counterpart of /a:/, and /E/ that of /O/:
VOWEL HARMONY IN T URKISH AND H UNGARIAN 24

(68) öröm - nek ‘joy’ (dat.) varos - nak ‘city’ (dat.)

öröm - ne:l ‘joy’ (ad.) varos - na:l ‘city’ (ad.)
These suffixes are specified as {|A(I)|}. Thus, these pairs ‘behave’ as the pair /a/ - /æ/ in Finnish.
In stems, however, /e:/ occurs both with front and back vowels, and, if preceded by back vowels, a
following suffix turns up as back. It is ‘transparent’ as shown in (70a):

(69) a. ta:nye:r - nak ‘plate’

ka:ve: - nak ‘coffee’
b. radi:r - nak ‘eraser’
taxi - nak ‘taxi’
The same behavior is displayed by /i/, as shown in (70b).
The vowel /E/ in a similar context triggers allows both suffix variants; this is called vacillation:

(70) A:gnes - nek/nak

szalamender - nek/nak

Ringen and Kontra (1989) and Kontra, Ringen and Stemberger (1991) have tested individual words
with either /e:/ or /E/ in the final syllable and there is a certain amount of unpredictability. For
example, in some cases (like okto:ber and a few others) it is claimed that only -nak can occur.
Before proceeding to a formal analysis, let us set up the vowel system in terms of elements.
There are two major challenges. The first involves the mid front unrounded vowels (long /e:/ and
short /E/) and here there are two issues.
Firstly, both vowels form the harmonic counterpart to the low vowel /a/, (phonetically [O]) or
/a:/. We have seen that /e:/ forms the harmonic counterpart of /a:/ and as such it will end up with a
representation lacking |∀|:

(71) ∀ ∀
A A A
I (I) > (I)
ö röm - ne:l ‘joy’ (ad.)
(72) ∀ ∀
A A A
I (I) > (I)
ö röm - nek ‘joy’ (dat.)
From this one might infer that the representation of long /e:/ and short /E/ does not contain the
element |∀|, but, both being mid vowels that seems wrong. Rather it would seem that /e:/ is high
mid, whereas /E/ is low mid:

(73) /e:/ /E/

∀ ∀
A A
(I) I
25 H ARRY VAN DER H ULST

We could remedy this as follows. We can adopt a rule that will insert the element |∀| for any segment
that has color. This would in fact be a constraint (see 85 below), so we must assume that the result
of harmony is ‘adjusted’ by inserting |I| (the constraint functions as what would be called a ‘linking
rule’ in SPE). I will assume that all structures must be in conformity with the constraint at the output
level of the lexical phonology:

(74) |I| → |∀|

We do not have to postulate headedness for the A+∀ combination. We could in fact assume that the
headedness is ‘spelled’ out at the phonemic level depending on the length of the vowel:

(75) If long: |∀| is the head, else |A| is the head

The second issue that we need to address has to do with how /e:/ and Ebehave within the stem.
The vowel Eis ambiguous. When occurring in stems with a preceding back vowel it can be either
predictable (acting transparently, as a non-licenser) or disharmonic (acting as a licenser):

(76) a. ∀
A A A
I >\ (I)
Ag nEs + nak
b. ∀
A A A
I > (I)
Ag nEs + nek

I propose to represent this vacillating behavior by postulating two different representations. When
Eacts ‘transparently’ this is the result of leaving out the predictable |I| element as in (78a). When this
vowel causes the suffix to be front, its |I| element is not predictable because we omit the |∀| element
(which makes this representation identical to the result of a fronted /a/ in suffixes):

(77) a. ∀
A A A
(I)
Ag nEs + nak
b. A A A
I > (I)
Ag nEs + nek

Both representations will be made complete at the phonemic level so that they conform to the con-
straints (see 85 below).
It is important to note that the /e:/ and Evariant of alternating vowels in suffixes are not transpar-
ent. Thus, if an alternating suffix follows a vacillating stem such as Agnes, the front variety of the
suffix must be followed by front varieties of suffixes (see 79a). But if a vacillating stem is followed
by a suffix with a non-alternating neutral vowel an alternating suffix following can still have two
VOWEL HARMONY IN T URKISH AND H UNGARIAN 26

varieties (see 79b):

(78) a. Agnes – je: - nek/*nak

Agnes – ja: - *nek/nak
‘that of Agnes – DAT’
b. Agnes - e: - nek/nak
‘his/her Agnes – DAT’

The -je:/-ja: suffix is alternating and both forms can be found for this stem, whereas the –e: suffix
is invariantly front. The point to note is that the choice between –je: and –ja: determines the
frontness of the vowel in the alternating dative suffix. But adding the non-alternating neutral vowel
/e:/ allows the vacillation to apply to the dative suffix. Thus –je behaves as harmonic, whereas /e:/
is ‘transparent’. In our account the two /e:/ have different representations:

(79) a. /–je:/ b. /e:/

∀
A A
I I

Recall that the representation in (80a) will be repaired by adding the element |∀|. Long /e:/ does
not display vacillating behavior. This vowel, like both long and short /i/ acts transparently when
preceded by a back vowel:

(80) a. Transparent /e:/

ta:n ye:r -nak ‘plate’
ka: ve: -nak ‘coffee’

∀
A A A
(I)

b. Transparent /i(:)/
ra di:r -nak ‘eraser’
tax i -nak ‘taxi’

∀
A A
(I)

The |I| element of /e:/ and /i(:)/ cannot act as a licensor because, being predictable, it is omitted. Not
only stems with Ein the final syllable may be vacillating. In addition, stems ending in more than
one neutral vowel (even if preceded by a back vowel) have a strong tendency to select a front suffix
vowel:
27 H ARRY VAN DER H ULST

(81) a na li: zis - nEk / -nak

as pi rin - nEk / -nak
∀ ∀
A A A
? > (I)
How can we account for the fact that a sequence of two /i/s somehow allows one or both of the /i/s
to act as a licensor. Now given that /i/ vowels are minimally specified, we might compare this sit-
uation to the prohibition on sequences of so-called empty-vowels, which, as argued in Government
Phonology, are ruled out by requiring that such vowels must be licensed by a non- empty vowel.
It is perhaps not a coincidence that a bare |∀| comes as close to being an empty vowel as possible.
Therefore15 it is reasonable to suggest that one of the bare vowels must be licensed by the other one
which requires the licensor to be fully specified:

(82) ... li: zis - nEk

∀ < ∀
A A A
I > (I)
Concluding, the vowel system at the phonemic level can be fully specified as follows.

(83) /i(:)/ /y(:)/ /u(:)/ /e(:)/ /ø(:)/ /o(:)/ /a(:)/

∀ ∀ ∀ ∀ ∀ ∀
A A A A
I I I I
U U U U
(84) Constraints
a. Aperture: ¬ Ø
∀→∀
b. Color: I → I
c. Cross-gestural:
(i) ¬ A → ∀
(ii) ∀ & ↔ (I v U)

For the vowel /i(:)/ and /e(:)/, the element |I| is predictable in accordance with (85cii): if a vowel
with the element |∀| does not have |U| is must have |I|.
I assume that the rounding of short /a/ is ‘phonetic’, as is the headedness of front mid unrounded
vowels which depends on length. This means that the special quality of short /a/ (as [O]) and short
‘/e/’ (as [E]) are both seen as allophonic and thus not specified at the phonemic level.

15
As Beata Moskal reminded me.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 28

4.2 A NTI - HARMONIC NEUTRAL VOWEL ROOTS : THE PROBLEM OF

UNDERAPPLICATION

Neutral vowel roots, i.e. having only neutral vowels, select front suffix vowels, with the exception
of a small group of 60 roots (almost all of which are monosyllabic), which take back suffixes. The
difference between both types of roots can be represented as follows16 :

(85) a. “front /i/”

∀
A
I > (I)
vi:z - nek

b. “back /i/”
∀
A
(I)
hi:d - nak
Regular neutral vowel roots can license (I) in suffixes because the first vowel of such roots is in a
licensed position, which means that its |I| element is licensed and thus present17 . The irregularity of
the anti-harmonic roots lies in the fact that the expected |I| element is not present even though this
predictable element occurs in a licensed position. This, I must conclude, implies that the presence of
a predictable element in a vowel where this element is positionally licensed can be a lexical choice.
It could be said that we have essentially ‘replicated’ the ‘abstract’ approach to antiharmonic
roots and to ‘transparent’ vowels. The question is whether allowing lexical representations to be
in disagreement with the phonemic constraints is ‘bad’. I suggest that it is harmless as long as the
lexical structures are substructures of the fully specified phonemic structures. Deviant structures are
allowed if we assume that lexical representations are not governed by ‘morpheme structure condi-
tions’. However, the resulting ‘freedom of the base’ is curtailed by requiring that the relationship
from lexical to phonemic level has to be ‘monotonic’ which means that elements can be added but
not deleted. That puts a limit on abstractness. For example, as suggested in van der Hulst (2011???),
it would prevent us from specifying low vowels as ‘ATR’ (i.e. with the element |∀|) if they do not
surface as ATR.
In conclusion, underapplication results from absence of the harmonic element. Recall that in
the analysis of Turkish we have seen that some roots with back vowels take front suffixes. Here
we were dealing with overapplication, which was represented in terms of the presence of a floating
element. In this connection we need to ask whether such floating elements need to be ‘deleted’ after
they have performed their licensing duty, or when such licensing is not required because there is
no suffix. Deletion, of course, has just been ruled out, so I will assume that these floating elements
16
There also antiharmonic roots with /e:/ which means that we also have to have two representations for this vowel. (There
is only one antiharmonic example with /E/; cf. Kontra 2011, footnote 13).
17
Recall that we already postulated the |I|-less varieties of /i/ and /e/ in roots where these vowels are preceded by back
vowels (see 81).
29 H ARRY VAN DER H ULST

remain in the representation, but fail to be phonetically interpreted because they are not linked to a
syllabic position.

4.3 D ISHARMONIC ROOTS

As in Turkish, we find disharmonic roots:

(86) soffö:r - nek

bü:ro - nak

As expected, it is the last vowel of the disharmonic roots which determines the quality of the suffix
vowel. There is no evidence that front rounded vowels ever act transparently, as in Finnish (see
Campbell 1980).

(87) sof fö:r - nek bü: ro - nak

∀ ∀ ∀ ∀
A A A A A
U U U U
I > (I) I (I)
The element |I| in the second example cannot license across /o/.

4.4 N ON - HARMONIC SUFFIXES

Here we may distinguish suffixes containing neutral vowels, and suffixes containing invariant non-
neutral vowels, or both18 :

(88) Suffixes
a. -i, -ig, -ni, -int, ke:nt, -e:rt, -ne:
b. -kor, -us, -u, -ko:, -a
c. -i:roz, -ista, -izmus, fika:l, -i:kus

We expect that category (89a) if followed by another suffix will act as transparent, whereas category
(89b) and (89c) ought to stop the harmonic process. Examples of transparency of a suffix belonging
to class (89a) are:

(89) poss plur

tükr e i hez ‘to his mirrors’
ablak a i haz ‘to the windows’
Disharmonic suffixes in groups (89b) will be represented as lacking the variable element. Finally,
the /i/s in groups (89c) will lack their predictable |I| element and have no variable element in the
vowel in their second syllable.

18
Note that there are no suffixes with short /E/.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 30

4.5 L ABIAL HARMONY

Hungarian has a limited amount of apparent rounding harmony affecting suffixes which have short
Eas on of their alternants. There are three types of suffixes with mid short vowels19 :

(90) a. E - a (i.e. nek/nak)

b. E - y - o (i.e. hez/ höz/hoz)
c. E - a - ø - o (i.e.ek/ak/ök/ok)

I have already adopted the representation {|A|} for two-way alternators (like –nak/-nek). Let us ask
what the representation of a three-way alternator might be. The distribution of the three alternants
is as follows:

(91) 3-way alternators

a. E after harmonic i, e
b. ø after y, ø
c. o after u, o, a and anti-harmonic i, e

The first case, (92a), refers to the occurrence of the suffix vowels after active neutral vowel roots
(such as vi:z in 86a). Thus, this concerns /i/’s and /e/’s that cause fronting. In the case of three- way
alternators, we cannot see the round variants as the result of spreading of |U| because we find the /o/
variant after stem-final /a/. I propose that three-way alternators are specified as {|A,∀, (I), (U)|} so
that we can derive the variants as follows:

(92) E after harmonic i,e licensing of (I): {|A, ∀, I, (U)|}

ø after ü, ö licensing of (I) and (U): {|A, ∀, I, U|}
o after u, o licensing of (U): {|A, ∀, (I), U|}
o after a and anti-harmonic i, e U by default: {|A, ∀, (I), U|}
Note that the default rule in the suffix system supplies |U|, while the default rule for stems supplies
|I|. This may be a stem – suffix difference or be due to the fact that |U| is the default in the presence
of the element |A|20 .
Let us now turn to the 4-way alternators. The basic issue here is that a certain class of stems,
called ‘lowering stems’ in Vago (ms), treat ternary suffixes as binary suffixes. (These stems also
have other phonological properties that distinguish them from regular stems.) Lowering stems are
a closed class of stems that cause the low vowels /a/ or E(the choice being dependent on whether or
not the stem has a spreading I element) to appear in the initial vowels of 4-way alternators such as:

19
Note that suffixes that alternate just between /o/ and /ø/ do not exist (cf. Polgárdi and Rebrus 1998).
20
Polgárdi and Rebrus (1998) also suggest that the element |U| is a property of the suffix to begin with, but as invariant,
which entails an analysis in which |U| must be deleted in case the suffix shows up with the vowel E(i.e. after a and
anti-harmonic i,e).
31 H ARRY VAN DER H ULST

(93) Plural -Vk

Accusative -Vt
1st person poss. -Vm
A derivational suffix -Vs

All of these start with a ‘linking vowel’ that appears when the stem ends in a consonant. If the
linking vowel appears after a regular (non-lowering stem), it alternates three-ways just like suffixes
that have three alternants only.
Kontra (1991) suggests that lowering stems impose the element |A| onto the linking vowel.
Following the spirit of this idea and implementing it within the current model, we could say that
the linking vowel is ‘empty’and that lowering stems come with the floating combination {|A (I)|}.
We then need a rule which spells out the empty vowel as the combination {|A, ∀, (I), (U)|} after
non-lowering stems.

4.6 C ONCLUSION
In this section, I have provided an analysis of Hungarian vowel harmony patterns. Firstly, alternat-
ing suffixes with /a ∼ e/ point to the need to have a rule adding the element |∀| to the fronted variant.
Secondly, it has been concluded that anti-harmonic roots can most naturally be represented as lack-
ing the element |I| (which mimics the old abstract analysis). Thirdly, a representational proposal for
ternary suffixes points to |U| as the default element for suffix vowels (or vowels specified with the
element |A|).

5 G ENERAL CONCLUSIONS

The analyses proposed here are still ‘sketchy’, but I believe that the model proposed in section 2
offers a very strict guideline for how to analyze both the general and the specific aspects of vowel
harmony systems. It has been assumed that lexical representations can be minimally specified and
that omitted predictable elements will be added at the end of the lexical derivation where the rep-
resentation must be fully specified phonemically. This allows a degree of abstractness given that
predictable elements can, but need not occur in specified in privileged positions so that both har-
monic and anti-harmonic neutral vowel roots can be represented in Hungarian. Many details need
to be worked out and no doubt problems will emerge. However, it has been shown that there is
no need for binary features such as [±back] and [±round], or, more specifically, that [+back] and
[–round] are not needed in the analysis of palatal and labial systems. The model also makes the cor-
rect prediction regarding the behavior of neutral vowels as either opaque or transparent, supporting
the claims in van der Hulst and Smith (1986).

R EFERENCES

Anderson, J. and Ewen, C. (1987). Principles of dependency phonology. Cambridge University

Press.
VOWEL HARMONY IN T URKISH AND H UNGARIAN 32

Anderson, S. R. (1980). Problems and perspectives in the description of vowel harmony. In Vago,
R. M., editor, Issues in vowel harmony, pages 1–48. John Benjamins.

Campbell, L. (1980). The psychological and sociological reality of finnish vowel harmony. In Vago,
R. M., editor, Issues in vowel harmony, pages 245–270. John Benjamins.

Clements, G. N. and Sezer, E. (1982). Vowel and consonant disharmony in turkish. In van der Hulst,
H. and Smith, N., editors, The structure of phonological representations, part II, pages 213–255.
Foris Publications.

Goldsmith, J. (1993). Harmonic phonology. In Goldsmith, J., editor, The last phonological rule,
pages 21–60. The University of Chicago Press.

Haiman, J. (1972). Phonological targets and unmarked structures. Language, 48:265–377.

Harrison, D. K. and Kaun, A. (2001). Patterns, pervasive patterns and feature specification. In Hall,
T. A., editor, Distinctive deature theory, pages 211–236. Mouton de Gruyter.

Kabak, B. (2011). Turkish vowel harmony. In van Oostendorp, M., Ewen, C. J., Hume, E., and
Rice, K., editors, The Blackwell companion to phonology, volume 5, pages 2831–2854. Wiley-
Blackwell.

Kardestuncer, A. B. (1982). Theoretical implications of Turkish vowel harmony. PhD thesis, Uni-
versity of Connecticut.

Kardestuncer, A. B. (1983). Vowel harmony and gerundive compounds in Turkish. Acta Linguistica
Hafniensia, 18(1):55–64.

Kaun, A. (2004). The phonetic foundations of the rounding harmony typology. In Hayes, B.,
Kirchner, R., and Steriade, D., editors, Phonetically based phonology, pages 87–116. Cambridge
University Press.

Kaye, J., Lowenstamm, J., and Vergnaud, J.-R. (1985). The internal structure of phonological
elements: A theory of charm and government. Phonological Yearbook, 2:305–328.

Kaye, J., Lowenstamm, J., and Vergnaud, J.-R. (1990). Constituent structure and government on
phonology. Phonology, 72:193–232.

Kiparsky, P. (1968/1973b). How abstract is phonology? In Fujimura, O., editor, Three dimensions
of linguistic theory, pages 5–56. Taikusha.

Kiparsky, P. (1973a). Abstractness, opacity, and global rules. In Fujimura, O., editor, Three dimen-
sions of linguistic theory, pages 57–86. Taikusha.

Kontra, M., Ringen, C., and Stemberger, J. P. (1991). The effect of context on suffix vowel choice
in hungarian vowel harmony. International Congress of Linguists, 12.
33 H ARRY VAN DER H ULST

Kornai, A. (1990). Hungarian Vowel Harmony. In Kenesei, I., editor, Approaches to Hungarian 3:
Structures and arguments, pages 183–240. JATE.

Nevins, A. (2010). Locality in vowel harmony. MIT Press.

Polgárdi, K. (1999). Vowel harmony and disharmony in Turkish. The Linguistic Review, 16(2):187–
204.

Polgárdi, K. and Rebrus, P. (1998). There is no labial harmony in Hungarian: A Government

Phonology analysis. In de Groot, C. and Kenesei, I., editors, Approaches to Hungarian, vol. 6,
Papers from the Amsterdam Conference, pages 3–20. JATE.

Ringen, C. and Kontra, M. (1989). Hungarian neutral vowels. Lingua, 78:181–191.

Vago, R. M. (Ms.). On the analysis of lowering in the inflectional system of hungarian.

van der Hulst, H. (1988). The dual interpretation of |i|, |u| and |a|.

van der Hulst, H. (2011a). Dependency-based phonologies. In Goldsmith, J., Yu, A., and Riggle, J.,
editors, The handbook of phonological theory, pages 533–570. Blackwell.

van der Hulst, H. (2011b). A framework for vowel harmony. In Botman, B. and Noske, R., editors,
Festschrift for Norval Smith. Max Newmeyer.

van der Hulst, H. (Ms.a). Asymmetries in vowel harmony. University of Connecticut.

van der Hulst, H. (Ms.b). An outline of Radical cv Phonology. University of Connecticut.

van der Hulst, H. (Ms.c). Phonology: from scratch to theory. University of Connecticut.

van der Hulst, H. (to appear). Yoruba vowel harmony.

van der Hulst, H. and Smith, N. (1986). On neutral vowels. In Bogers, K., van der Hulst, H., and
Mous, M., editors, The phonological representation of suprasegmentals, pages 233–279. Foris
Publications.

van der Hulst, H. and van de Weijer, J. (1991). Topics in turkish phonology. In Boeschoten, H. E.
and Verhoeven, L. T., editors, Turkish linguistics today, pages 11–59. Brill.