Turkish Stress A Review
Turkish Stress A Review
Turkish Stress A Review
This work evaluates an argument recently made in these pages by Kabak & Vogel
(2001) to the effect that the analysis of Turkish which they develop is superior on
theoretical grounds to that of past accounts. Kabak & Vogel explicitly contrast
their account to that offered in two recent, comprehensive discussions of Turkish
stress by Inkelas & Orgun (1998) and Inkelas (1999). Careful consideration of the
data discussed by Kabak & Vogel and by Inkelas & Orgun, as well as some ad-
ditional data introduced in this paper, shows that the original Inkelas & Orgun
analysis achieves greater empirical coverage while using less theoretical machinery.
1 The facts
The essential facts of stress in Turkish are these.1 Every word has exactly
one main stress. The default stress position for words is final, e.g. arabá
‘ car’, araba-lár ‘car-PL ’, araba-lar-dán ‘car-PL-ABL ’; bırák ‘leave !’, bırak-
acák ‘leave-FUT ’, bırak-acak-lár ‘ leave-FUT-3PL ’.2 Final stress is over-
ridden in words containing lexically stressed roots, lexically stressed or
pre-stressing suffixes, certain types of compounds or the type of subcon-
stituent referred to by I&O as a ‘ Sezer’ stem because of its distinctive
stress pattern, first described by Sezer (1981). Most compound types have
main stress on the first member (baş+bak-anpbáşbakan ‘head look
1 This summary draws from Inkelas (1999) and Inkelas & Orgun (1998). Earlier
treatments of subparts of the Turkish stress system include Lees (1961), Swift
(1963), Lewis (1967), Foster (1970), Zimmer (1970), Underhill (1976), Lightner
(1978), Sezer (1981), Kardestuncer (1982), Poser (1984), Hameed (1985), Kaisse
(1985, 1986), Dobrovolsky (1986), Hammond (1986), Sebüktekin (1986), Halle &
Vergnaud (1987), Barker (1989), van der Hulst & van de Weijer (1991), Idsardi
(1992), Halle & Idsardi (1995), Hayes (1995) and Kornfilt (1997).
2 Examples in this work are given in Turkish orthography, with the addition of a
length mark to indicate long vowels, a palatality mark for phonemically palatal
velars (ff), an acute accent to mark the stressed syllable, and hyphens to mark mor-
pheme boundaries. ffi is the orthographic representation of the so-called soft g (his-
torically a velar fricative), which is realised either as zero or as lengthening of the
preceding vowel. Following standard practice, vowels which are subject to vowel
harmony and consonants which are subject to voicing assimilation are represented
with uppercase letters in underlying representation. We indicate pre-stressing suf-
fixes by a circumflex accent preceding their underlying form (e.g. /- ^mE/). The
data represent the speech of the second author, and have been checked with a
66-year-old native speaker of the standard Istanbul dialect of Turkish, who has
lived in Turkey all his life (GO, to whom we are infinitely grateful). In cases of
discrepancy, we have used GO’s pronunciations, which are consistent with other
descriptions of Istanbul Turkish as well as the data in K&V, whose alphabetically
first author is also a speaker of standard Istanbul Turkish. Dialect variation in word
stress exists but does not factor into the present discussion.
Abbreviations of suffix names are as follows : 1=1st person, 2=2nd person,
3=3rd person, ABL=ablative, ACC=accusative, ADV=adverbial, AOR=aorist
(=imperfective), ASSOC=associated with, CAUS=causative, COND=conditional,
COM=instrumental/comitative, COP=copula, ECOP=epistemic copula, FUT=future,
IMP=imperative, INF=infinitive, INT=interrogative, MIT=mitigative, NEG=nega-
tive, PL=plural, POSS=possessive, PROG=progressive, PRT=particle, REL=relati-
viser, SG=singular, VBL=verbaliser.
3 K&V are rather inaccurate in stating (p. 339) that phrasal stress falls on the first
word in the phrase in Turkish. In fact the general pattern is for stress to be either
phrase-final or preverbal (see e.g. Kornfilt 1997). Phrase-final stress is the norm
when all of the verb’s complements are definite and none is in narrow focus. Thus,
to use an example which is similar to K&V’s (39a), the sentence Orhan açı-yı ölç-
tü´ ‘Orhan angle-ACC measure-PAST=Orhan measured the angle’ has phrase-final
stress. Preverbal stress occurs when a definite argument is in narrow focus ; pre-
verbal position is the spot for focus in Turkish and attracts stress. Indefinite argu-
ments are also attracted to preverbal position, and stressed, even under broad focus.
K&V consider only verb-final sentences consisting of two words, thereby conflating
initial and preverbal position. Their four examples all involve preverbal indefinite
arguments under broad focus, e.g. Sü´t beyaz-dır ‘ milk white-ECOP=milk is white ’
(their (38b)) or Açı́ ölç-er-mi ? ‘ angle measure-AOR-INT=does (it) measure an
angle ?’ (38a). Simply adding an overt subject to an example like Açı́ ölç-er-mi?
makes it clear that the correct analysis is preverbal, rather than initial, stress: Orhan
açı́ ölç-tü ‘ Orhan angle measure-PAST=Orhan measured angles ’.
4 Inkelas (1999: 158) identifies three suffixes as stress-bearing : progressive -Íyor,
adverbial -ÉrEk ‘ by ’ and adverbial -ÍncE ‘ when ’. K&V inaccurately characterise
this set as ‘ several aspect and modality markers ’ : they also imply that Inkelas treats
the aorist suffix in bul-úr-um ‘ find-AOR-1SG-I find ’ as stressed (p. 323) ; it is not. Not
only does Inkelas (1999) correctly classify the aorist suffix as unexceptional in its
stress behaviour (p. 140), she also highlights the generalisation that no monosyllabic
suffix is ever lexically stressed (pp. 255, 274–276 ; see also § § 1.1, 4.1 of this paper).
The reason that the aorist suffix appears stressed in K&V’s bul-úr-um example is, as
K&V themselves later point out, that the agreement suffix is pre-stressing.
5 K&V classify some of the pre-stressing endings as clitics, rather than suffixes ; as in
I&O, we avoid this terminological distinction, which correlates with no phonological
criteria, nor with any morphological criteria relevant to the discussion at hand.
Table I
Combinations of stressed and/or pre-stressing morphemes.
1.4 Compounds
Compounds subdivide into two types phonologically : stressed and un-
stressed. This phonological distinction correlates with a morphological
distinction.
Unstressed compounding, which has no effect on stress, is limited to
compounds of a phrase-like structure in which the second member is a
predicate of which the first member is an argument. Following Swift
(1963), Inkelas & Orgun (1998 : 376) cite the example hünffar ‘sultan ’
+beğen-di ‘like-PAST ’phünffar-beğend ‘pot roast of lamb with eggplant
puree ’ (lit. ‘ the sultan liked it ’), whose behaviour under further suffix-
ation (e.g. hünffar-beğendi-niz-dén ‘pot roast of lamb with eggplant puree-
2PL.POSS-ABL ’) shows that the compound lacks inherent stress of its own.
The better-known type of compound from a stress perspective, and the
only type discussed by K&V, is the stressed compound. In stressed com-
pounds, the second member is stressless (according to I&O; K&V say that
its stress is demoted, which may mean the same thing). The first member
of the compound is stressed as it would be if it were an independent word,
conforming to the patterns described above for word stress. Inkelas &
Orgun cite examples like ye-mek ‘eat-INF ’+oda ‘ room’pyemék oda-sı
‘ dining room ’ (1998 : 375), in which yemek surfaces with the same final
stress it has when it stands alone as a word, and Kandlli+cadde
‘ street ’pKandlli cadde-si ‘ Kandilli street ’ (1998 : 384), in which
Kandlli, a Sezer place name formed from kandil-li ‘oil lamp-ASSOC ’,
surfaces with the Sezer stress pattern. (The possessive suffix -(s)I in both
The I&O analysis accounts in a theoretically economical way for all the
facts discussed by K&V, as well as for the stress of some morphological
constructions that K&V do not discuss (e.g. unstressed compound for-
mation and morphologically derived place names ; on the latter, see es-
pecially § 3.2).
K&V criticise the account of Turkish stress in I&O on grounds of excess
machinery, proposing an alternative analysis that they argue is simpler.
The components of their analysis of word stress are as follows. They
3 Comparison
K&V argue that their analysis is simpler than that of I&O in requiring
fewer lexical distinctions and fewer rules and in making no reference to
multiple levels or co-phonologies (see e.g. p. 353). The point-by-point
comparison, provided below, shows that K&V’s reasoning is faulty. The
only respect in which the K&V analysis could be said to use less theoretical
apparatus is that it accounts for less data. K&V fail to consider a class of
data (Sezer stems) that I&O argue must be accounted for on indisputable
grounds of morphological productivity; in addition, K&V’s analysis (as
stated) makes incorrect predictions for two other sets of data, one of which
I&O analyse and the other of which is not discussed by I&O but is cor-
rectly predicted by the I&O analysis.
7 K&V’s treatment of pre-stressing suffixes resembles that of Poser (1984), who treats
pre-stressing suffixes as extrametrical and assumes that, in Turkish, extra-
metricality extends to any suffixes to the right (p. 90). Barker (1989) adopts this
analysis but, because his analysis is cyclic, does not require Poser’s novel extension
of extrametricality ; see Barker (1989 : 29–32). Barker’s analysis, adopted by van der
Hulst & van de Weijer (1991) and discussed by K&V, is elegant in many respects
but makes incorrect predictions for disyllabic suffixes, sequences of monosyllabic
pre-stressing suffixes and stressed stems which combine with pre-stressing suffixes.
Turkey. For internal consistency, we have used GO’s stress patterns for these place
names, rather than those of the speaker whose transcriptions are currently available
from the TELL website. Development of the TELL database and website was
supported by NSF awards #SBR-9514355 and #SBR-9911003.
10 Some internally compounded place names which are used frequently can be trun-
cated to their first member ; in this case the first member exhibits the same stress
it has in the compound (e.g. Sana;y Sitesi ‘ industrial site ’ p Sana;y, Cumhu;riyét-
köy ‘ republic village ’ p Cumhu;riyét). Such forms exemplify compound stress,
rather than counterexemplifying Sezer stress.
11 The examples cited in this table are foreign place names; there happen to be no
place names formed by -istan in the TELL database.
12 There are no place names zero-derived from stems ending in -mEz in the TELL
database.
13 Orgun (1996) and Orgun & Inkelas (2002) use the term ‘ co-phonological allomor-
phy ’ for construction sets of this kind in which the morphological function of the
constructions is the same, but the associated co-phonologies differ.
Table II
Stress patterns found in morphologically complex place names.
14 As suggested earlier, it could be that appositives pattern with stems ending in -lEr
or -mEz in being zero-converted to place names via the non-stressing place-name
forming construction.
15 Though not in the TELL database, there is another place name that has the same
stress neutral suffix and unexpected antepenultimate stress: Kı́nalı (< /kına-lı/
‘ henna-ASSOC ’ ; cf. kıná, kına-lı́). We thank Jaklin Kornfilt for bringing this example
to our attention.
17 To preserve the unstressed lexical status of the place name Anadolu, Inkelas &
Orgun (1998) proposed that Anadolu undergo the non-Sezer stressing place-name
forming construction. This is also morphologically redundant.
18 On the association of different co-phonologies (constraint rankings) with subsets of
roots in a language, see e.g. Itô & Mester (1995, 1999). Inkelas et al. (1997) argue
against the use of positing co-phonologies just to capture phonological differences
among bare roots, but the co-phonology in question here – namely the Sezer co-
phonology – exists independently in the grammar, and would not have to be created
just to capture root stress patterns.
Table III
Comparison between the K&V and I&O analyses.
Table IV
K&V’s analysis of the interactions among stressed roots, stressed suxes
and pre-stressing suxes.
5 Conclusion
Once K&V’s analysis is augmented, as required for descriptive adequacy,
with Sezer-stem capability and a principle of Innermost Wins, it becomes
nearly indistinguishable from the I&O analysis which K&V claim to be
supplanting. Inkelas (1999) uses stress prespecification to account for all
types of exceptional stress. K&V also use stress specification (or, rather, its
indirect equivalent) for all but one type of exceptional stress. Inkelas &
Orgun (1998) use Leftmost Wins to account for stressed compounds ;
K&V do the same. The only bona fide difference in theoretical apparatus
between the analyses is that K&V propose for pre-stressing suffixes the
distinct theoretical mechanism of PW adjunction. And this difference does
not favour K&V. PW adjunction is ad hoc, lacking independent motiva-
tion in Turkish ; as K&V point out, neither word-level syllabification nor
vowel harmony, for example, is bounded by the particular PWs needed to
generate pre-stressing suffix effects. Moreover, the stipulation that all
suffixes to the right of a PW-adjoiner are excluded from PW status is a
theoretical innovation needed only for these data and is at odds, as K&V
acknowledge in their note 12, with cross-linguistically common patterns
in which recursive PW formation would seem to be the norm, especially
when multiple PW adjoiners are concerned. Since lexical stress specifi-
cation is needed anyway for stressed roots and stressed suffixes, using it
for pre-stressing suffixes as well (as in Inkelas 1999) is more theoretically
parsimonious than introducing PW structure in an essentially diacritic
19 K&V misgloss the adjective güzel as a noun, ‘ beauty’ ; the corresponding abstract
noun is actually güzel-lik. They also misgloss verbalising -leş as an adjectivaliser,
though they translate its product, güzel-leş, correctly as ‘ beautify ’.
REFERENCES