Is Turkish a tongue-twister? Nicolas Royer-Artuso
[email protected] The existence of large numbers of exceptions to proposed rules is often a sign that a system has not been well understood Clements and Sezer (1982: 213)
Abstract. In this paper, I claim that Turkish Vowel Harmony (TVH) is no longer a phonological process of Turkish. TVH and, more importantly, the exceptions to it, have been a subject of intense discussions. Different models have been proposed but all of them are faced with disharmonic phenomena that are hard to handle coherently. This observation brings our attention to the sort of implications that should be drawn from the inexpediency between a more or less well understood concept and a mass of facts that are reluctant to get domesticated. In order to arrive at the conclusion that TVH is not or no longer a phonological process of Turkish, I first show some logical implications of the previous analyses of TVH. I start with the traditional account of TVH in order to present the facts that need to be handled by a model of TVH. Some alternative models of TVH are then presented. I show some of the weaknesses of these proposals. The logical consequences of these proposals then bring me to present my theoretically agnostic conclusions, something that should pave the way to a more realistic analysis, compatible with different theoretic models of language description. Keywords: Phonology; Turkish phonology; vowel harmony; Turkish Vowel Harmony.
1. Introduction In this paper,1 I show some logical implications of the previous analyses of Turkish Vowel Harmony (TVH). TVH and more importantly, the ex1
I want to thank Luc Baronian and Jaïmé Dubé for discussions on some of the ideas that are developped here. I also want to thank all my students at the Marmara University in Istanbul when I was still teaching there for all what they taught me on L2 phonology,
Nicolas Royer-Artuso
42
ceptions to it, have been a subject of intense discussions. Different models have been proposed but all of them are faced with disharmonic phenomena that are hard to handle coherently. Even if not stated directly, footnotes always remind us of some irregular cases that do not fit the models. This observation brings our attention to the sort of implications that should be drawn from the inexpediency between a more- or less(?)2well-understood concept and a mass of facts that are reluctant to get domesticated. After repeated attempts to find a solution to TVH, it might be time to start considering the option that TVH is not (anymore?) a (regular?) phonological process in Turkish. This is the suggestion that I will make in what follows. But before getting there, I have to show in what sense the proposals given until now have failed to give us a coherent picture of TVH. Allah’tan ‘thanks to God’, Kabak and Vogel (2011) did a lot of the job for me, although their conclusions differ from mine. I recommend this paper to everybody wishing to dwell into the problem of exceptionality and its analysis in modern phonology, especially the problems encountered with Lexical Phonology and Optimality Theory when used in the description of TVH or other phonological phenomena. The discussion runs as follows: I start with the traditional account of TVH in order to present the facts that need to be handled by a model of TVH. Some alternative models of TVH will then be presented, models that tried to cope with TVH’s exceptions by proposing alternative ways of dealing with it. I show some of the weakness of these proposals. I then present some logical consequences of these proposals that, if I am right, are really problematic on a (meta) theoretical level. I finally present my theoretically agnostic conclusions, something that could pave the way to a more realistic analysis, compatible with different theoretic models of language description.
even when not aware of teaching it to me. And of course, I want to thank Rajendra Singh for all that he taught me and continues to teach me throught the amazing work he did. I miss him a lot. 2
See Anderson (1980) more than 30 years ago.
Is Turkish a tongue-twister?
43
2. Traditional analyses of TVH3 The Turkish vowel system is the following: Ex. (1)
High Rounded Unrounded ü i u i
Front Back
Non-high Rounded Unrounded ö e o a
Turkish Vowel Harmony (TVH) is understood as a phonological process whereby a vowel harmonizes with a preceding vowel. This process explains why all vowels in a phonological word share some properties (roundness, frontness, etc.). In TVH, this claim is valid for vowels inside a root and for morphologically complex words, the results of affixation (Turkish being, apart from some borrowed prefixes, largely an agglutinative language that uses suffixation). Table 1 shows the distribution of vowels in native Turkish words. Table 1. Distribution of vowels in native Turkish words. (Exceptions not mentioned.)
Syllable 1
Syllable 2 i
a
u
o
i
e
ü
ö
i
h
h
*
*
*
*
*
*
a
h
h
*
*
*
*
*
*
u
*
h
h
*
*
*
*
*
o
*
h
h
*
*
*
*
*
i
*
*
*
*
h
h
*
*
e
*
*
*
*
h
h
*
*
ü
*
*
*
*
*
h
h
*
ö
*
*
*
*
*
h
h
*
Note : h = harmonic; * = disharmonic.
3
Göksel and Kerslake (2005), Kornfilt (1997) and Lewis (2000) serve as references for the traditional model of TVH. I follow Göksel and Kerslake for the presentation.
Nicolas Royer-Artuso
44
The distribution is explained this way: TVH consists of: (1) (2)
Fronting harmony (assimilation in frontness) Rounding harmony (assimilation in roundness)
1 and 2, with the addition of the following restriction (3) are sufficient to take care of the facts. (3)
O and ö only occur in the initial syllable of a word (to explain why o and ö do not respect the pattern “followed by X and itself”).
Fronting harmony operates in all but a few native roots. Rounding harmony also, with some exceptions sometimes due to an intervening labial consonant’s influence: savun ‘to defend oneself’, tavuk ‘chicken’, etc. There are 2 kinds of suffixes (but see exceptions presented below): the I-type suffix and the A-type. (1)
The I-type suffix’s vowel is a high one. Fronting and rounding harmony determine the form that it will have relatively to the last vowel of the word it is suffixed to.
Ex. (2) kiz ‘girl’ kas ‘muscle’ diz ‘knee’ el ‘hand’ mum ‘candle’ kol ‘arm’ yüz ‘face’ göl ‘lake’ (2)
-i ‘ACC.’ kizi kasi dizi eli mumu kolu yüzü gölü
-di ‘PAST’ kizdi kasti dizdi eldi mumdu koldu yüzdü göldü
-miʃ ‘PAST’ kizmiʃ kasmiʃ dizmiʃ elmiʃ mummuʃ kolmuʃ yüzmüʃ gölmüʃ
The A-type suffix’s vowel is unrounded and non-high. The frontness property of the last vowel of the word it is suffixed to determines the form it will have.
Is Turkish a tongue-twister?
45
Ex. (3) kiz kas diz el mum kol yüz göl
-a ‘DAT.’ kiza kasa dize ele muma kola yüze göle
-lar ‘PL.’ kizlar kaslar dizler eller mumlar kollar yüzler göller
-dan ‘ABL.’ kizdan kastan dizden elden mumdan koldan yüzden gölden
There are a lot of exceptions to TVH because of (1) the large number of loans that entered the language; (2) compounds and their lexicalization; (3) some exceptional suffixes in loan words; (4) some foreign suffixes and prefixes that were borrowed; (5) some native suffixes that do not vary (do not participate in vowel harmony); and (6) acronym formation (this picture might not be complete). These exceptions are one of the problems facing anybody trying to propose a coherent account of TVH. These facts are generally taken out of the analysis by putting some sort of diacritic on them, generally of the [+ foreign] or [−native] type.4 Table 2 presents the distribution of vowels found in Turkish when (1–6) are taken into account. 3. Some other accounts of TVH Given the facts presented above and the traditional account of TVH, I will now present other types of models that have been proposed to deal with these exceptions coherently without the recourse to the diacritic [+foreign]. Note that the 2 following models are using Table 3 for their descriptions. 4
The recourse to the native/non‐native criterion is a problematic one for Turkish (see Kabak and Vogel 2011): some non‐native words are harmonic (lise ‘secondary school’, haram ‘forbidden’ etc.), while some native words are not (anne ‘mother’, elma ‘apple’, etc.). There is no way we can separate the lexicon using this criterion. I will come back later to this problem and related ones.
Nicolas Royer-Artuso
46
Table 2. Distribution of vowels in Turkish including “exceptions”.
Nucleus 1
Nucleus 2 i
a
u
o
i
e
ü
ö
i
h
h
✓
✓
✓
✓
✓
✓
a
h
h
✓
✓
✓
✓
✓
✓
u
✓
h
h
✓
✓
✓
✓
✓
o
✓
h
h
✓
✓
✓
✓
✓
i
✓
✓
✓
✓
h
h
✓
✓
e
✓
✓
✓
✓
h
h
✓
✓
ü
✓
✓
✓
✓
✓
h
h
✓
ö
✓
✓
✓
✓
✓
h
h
✓
Note : h = harmonic; ✓ = attested but disharmonic.
Table 3. Distribution of vowels in Turkish in Clements and Sezer’s (1982) analysis.
Nucleus 1
Nucleus 2 i
a
u
o
i
e
ü
ö
i
h
h
*
*
*
*
*
*
a
h
h
✓
✓
✓
✓
*
*
u
✓
h
h
*
✓
✓
*
*
o
✓
h
h
*
✓
✓
*
*
i
*
✓
✓
✓
h
h
✓
*
e
*
✓
✓
✓
h
h
*
*
ü
*
*
*
*
✓
h
h
*
ö
*
*
*
*
*
h
h
*
Note : h = harmonic; ✓ = attested but disharmonic; * = not attested.
Is Turkish a tongue-twister?
47
3.1. Van der Hulst and van de Weijes’s proposal Van der Hulst and van de Weijes (1991) propose that TVH follows the traditional analysis in term of frontness and roundness. Any vowel may appear in the first syllable of a word. o and ö do not appear in non-initial syllables with some exceptions explained below. Their analysis utilizes elements: Lowness is an intrinsic property of vowels. Frontness and Roundness are not lodged in individual vowels, but rather seem to be word properties. Frontness and Roundness are considered word properties and described as prosodies. Their model, it is said, can predict the distribution of vowels in Turkish, stating the ones that can appear in disharmonic roots and the ones that cannot. Taking the distributions given in Table 3 from Clements and Sezer (1982), their aim is to show that we can explain the fact that some vowels combine freely in stems, while others comprising ü, ö or i cannot except for stems combining i and u. All vowel primitives are regarded as privative in Trubetzkoy’s sense. For Turkish vowel inventory we have the following: Ex. (4) i Low Front Round
F
e L F
ü F R
ö L F R
i
a L
u
o L
R
R
Vowel position (V) is taken as a primitive; i is represented as a bare Vposition. F, L, R and V are therefore the primitives needed for the theory. Regular stems get a “free ride” on the harmony rules which means that for them, prosodic properties are not associated to V-positions. The property Low is not predictable: we associate it to specific positions. Ex. (5) V L i a hitta ‘province’
L V a i alti ‘six’
L L a a kara ‘black’
V V i i kisim ‘part’
Nicolas Royer-Artuso
48
F
F
F
F
:
.
:
.
:
.
:
.
V
L
L
V
L
L
V
V
i
e
e
i
e
e
i
i
ince ‘thin’
yedi ‘seven’
kere ‘time’
gibi ‘like’
Attested disharmony in frontness is the result of pre-association to the first or second V-position (lexically associated is another way of seeing the phenomenon). The prosody therefore cannot associate to any other vowels. For example: Ex. (6) F
F
F
F
x
x
x
x
V
L
L
L
L
L
L
V
i
a
e
a
a
e
a
i
siyah ‘black’
elma ‘apple’
haber ‘news’
tatil ‘vacation’
Combinations /e i/, /i, i/, /i, e/ and/i, i/ are not attested. They could be possible exceptions, but would produce empty V-slot on the surface, which is considered ill-formed. Front and Round when associated lexically still spread to suffixes but Low does not. Roundness has another property: it does not associate with non-initial L.
Is Turkish a tongue-twister?
Ex. (7) R : V L
R : L
u a tuhaf ‘strange’
o u yorgun ‘tired’
. V
R : L
L
o a oda ‘room’
49
R : V
. V
u u kuru ‘dry’
Exceptional patterns arise for example when R associates with either of the two V-positions: Ex. (8) L
R x V
a u marul ‘lettuce’
L
R x L
a o karakol ‘police office’
R x V
L
u a tuhaf ‘strange’
R x L
L
o a oda ‘room’
/i, o/, /i, u/, /o, i/ and /u, i/ would again result in an empty V-position, and therefore be ill-formed. Ü and ö have 2 prosodies. The regular pattern is the following: Ex. (9) F : . V L : R
F : L :. R
ü e dümen ‘wheel’
ö ü gönül ‘heart’
. V
F : L : R
. L
ö e gönder ‘send’
F : V :. R
. V
ü ü ütü ‘iron’
Nicolas Royer-Artuso
50
R does not associate to L. If it was, we would find /ü o/ and /ö o/ patterns but we do not. With lexical association we can predict the following disharmonic words: Ex. (10) F x V L x R i o pilot ‘pilot’
V x R
F x L
u e kudret ‘power’
F x L
V x R
F x L
L x R
e u e o memur ‘official’ petrol ‘petrol’
L x R
F x V
o i bobin ‘spool’
L x R
F x L
o e otel ‘hotel’
F x V
V x R
i u billur ‘crystal’
V x R
F x V
u i kulis ‘stage wing’
With this model, 20 other kinds of lexically associated patterns can be predicted, but only 2 of them actually occur. The only attested cases are /ü i/ and /i ü/. The conclusion of all these results is that (1) if there is one prosody present, no vowel receives less than one property and (2) if there are two prosodies present, no vowel receives more than one prosody (except if one vowel is /i/). But the model still has problems explaining some non-variable suffixes like -a:ne ‘transforms an adjective into a noun’ and -va:ri ‘like’:
Is Turkish a tongue-twister?
51
they should be associated in frontness with the preceding stems but are not. The explanation provided is that they behave in a compound-like manner. The problem with this explanation is that Turkish compounds do not always behave this way: stress is assigned to each word forming the compound until lexicalization. On the contrary, morphologically complex words (less non-lexicalized compounds) follow the normal stress pattern, which is on the last syllable of the (resulting) word. Even if we accept this explanation for this troublesome class of facts, it does not help the overall argumentation in the sense that we still have to explain why other suffixes do not react in a compound-like manner. If van der Hulst and van de Weijes are right in saying that Clements and Sezer went too far while saying that TVH does not hold in stems anymore and that “synchronic harmony on suffix vowel is independently needed (van der Hulst and van de Weijes 1991: 42)”, blocking the lexicon with association lines will not do the job if we still need at the end of the analysis to propose that some non-variable suffixes are compound-like in character. Generally, a class of objects tends to behave in the same manner. Why not propose that all suffixes behave in a compound-like manner? Synchronic harmony on suffix would then not be independently needed. We will get back to this later, but I think the message is clear. 3.2. Clements and Sezer’s proposal Clements and Sezer (1982) is one of the most radical approaches to TVH. They take the problem of exception very seriously and propose that the only possible conclusion we can arrive at is to say that TVH does not operate inside roots anymore (due to the impact of massive borrowing of words not following TVH), but still does operate in suffixation. Exceptions to TVH are specified in the lexicon with the recourse to opaque vowels that block the effects of TVH. Opaque segments “can be characterized as […] vowels that are underlyingly associated with autosegmentally-represented features” (Clements and Sezer 1982: 214). Opaque vowels are segments (1) not undergoing, (2) blocking and (3) spreading the effects of TVH. Root vowels are opaque but a constraint still exists: “disharmonic distributions involving /i, e, a, o, u/ are well-
52
Nicolas Royer-Artuso
formed, but those involving /ü, ö, i/ are not” (Clements and Sezer 1982: 227) less /i, ü/ that may occur in either order. They conclude that all morphemes are constrained the same way. TVH is thus harmony in backness only. From there we only need to postulate a distinction between opaquevowel suffixes and non-opaque-vowel suffixes to accommodate the facts. The A-type suffix given above (Example 3), for example, has an opaque vowel underlyingly. Some other suffixes must also be specified the same way. The immense role given to opaque vowels is evident from the above discussion. It is said that a theory that provides natural means for expressing such notions as ‘opaque segments’ embodies the hypothesis that such segments may exist, and creates a strong presumption that they should be instantiated in particular linguistic systems. The empirical justification for the notion of opaque segments derives from the fact that the recognition of this segment type allows for simple, straightforward descriptions of commonly recurring phenomena that require complicated or ad hoc accounts under alternative theories. (Clements and Sezer 1982: 214)
I am not sure if this argumentation is valid, at least for the case we are dealing with. It looks a lot more like the type of reasoning against which they seem to be arguing. To put matters differently: when in a description we are faced with facts that do not fit our model, instead of hurting the core of the latter, it is better to find some devices that take care of the exceptions. In this context it means a device that would block the regularity of some processes that we have described before. Then we can call this device natural on a formal basis and propose that this device could be used later to explain exceptions to a similar process in another language. More importantly, there are some facts not taken into account in Clements and Sezer’s proposal (this applies to van der Hulst and van de Weijes’s proposal’s as well, as they took Clements and Sezer’s distribution as their source of analysis). When they propose that TVH is harmony in backness only, they discard distribution of vowels that are “very rare” (Clements and Sezer 1982: 223), “extremely infrequent” (Clements and Sezer 1982: 224), etc. to assess this rule. Even if this is not
Is Turkish a tongue-twister?
53
completely satisfying on a methodological level, some phonological and word-formation processes in Turkish are not taken into account in the discussion: epenthesis in loan words, acronym formation and compounding, processes that create disharmonic sequences (these processes will be discussed here later). The proposed restrictions on ü, ö and i do not stand if we take those processes into account (see Table 2 where I added distributions that are actually attested even if “very rare” or “extremely infrequent”). What those two strategies also have in common is a proposed blocking device rendering some facts immune to TVH. Kabak and Vogel also conclude that there is no other way around it: if we do not prespecify disharmonic words (DWs) in some way, there is no way of saving TVH models from the counterexamples facing them. 3.3. Other problematic devices Kabak and Vogel also mention the problems facing Lexical Phonology, Optimality Theory and models using co-phonologies when it comes to exceptions. I will not present a detailed account of these models because only the consequences of their architecture are important for the discussion. (1) Lexical Phonology (LP) was a reaction to SPE-types of models and the way they treated exceptions (diacritics, boundary symbols, etc.). LP is an attempt to “enrich the structure of the phonology of the language to recognize and formalize the role of exceptions in the make-up of the language as a whole” (Kabak and Vogel 2011: 61– 62). Different phonological rules are associated with different levels of representation but we still have to stipulate which rule works at which level and still end up with proposed levels that correspond to the distinction native/non-native, in our case, a level for the harmonic words and an other one for the disharmonic one. Two related problems exist: (1) how many levels should we propose and on what ground do we establish these levels and (2) ordering paradoxes: how do we order all these devices to make the model fit the facts?5 5
Although Kiparsky’s most recent incarnation of the theory provides answers to these questions, the general dislike for a model with levels persists in the field of phonology.
54
Nicolas Royer-Artuso
(2) Co-phonologies are devices introduced to save the lexicon from being divided into different levels. Different phonologies are in charge of taking care of the behavior of different classes of patterns. But associating items with a specific phonology creates problems in the sense that some words might react exceptionally to a rule but still react normally to others, in this way being subject to two different co-phonologies. The main problem is again one of constraining the number of co-phonologies that are needed and explaining with what criteria these different co-phonologies are proposed. (3) Optimality Theory treats exceptions (1) in terms of the nature of the representation or (2) in terms of the ordering of constraints. In the former case we protect some properties with a Faithfulness Constraint. In the latter case, we specify specific words for a particular ordering of constraints. Specification is therefore done on an item basis without concerns for classes of facts that behave similarly. We are back to diacritical marking. Even if very different from each other, these three devices share the same consequences: faced with exceptions to their proposed model, they need to prespecify in some way the items that will not get harmonized. These devices are chosen according to the model’s architecture, but the reasons of the incompatibility between the model and the facts are exactly the same. I will come back to this below. 3.4. Government Phonology’s account of TVH Since the importance Government Phonology (GP) has taken (especially in Turkey) in recent discussions of Vowel Harmony and, in the present case, of Turkish Vowel Harmony (TVH), it seems important to establish the power this model has in explaining the facts. Critical discussions of the above models have been done before, but a discussion of this type is not found for GP. For this reason, I will present a more complete picture of this model than I did for the others. Marküs Pöchtrager’s (2010) “Does Turkish diss harmony?” (DTDH) is the focus of this section as it present itself as the final word on TVH framed in GP, an even, as the final word on TVH, dot.
Is Turkish a tongue-twister?
55
It is suggested in what follows that it actually does not qualify in its present form as a final word on TVH since (1) a lot of exceptions can be found to what is proposed and – more importantly – (2) some of its logical consequences are hard to maintain in the face of what is said by the author. Even if Pöchtrager presents probably one of the most powerful and coherent models dealing with Turkish vowel harmony (TVH) and its exceptions, some problems still surface that are compromising for the entire framework in which it is couched. Pöchtrager proposes that traditional analyses of TVH went the wrong way and for this reason are faced with exceptions. A correct analysis of TVH (his, for instance) would, it is said, automatically get rid of these exceptions. My discussion runs as follows: after a description of the proposal, I discuss some exceptions to his claims. Some alternative explanations will be given for his analysis of certain general aspects relative to TVH, explanations that touch on some fundamental aspects of the analysis. Some consequences of its approach will finally be shortly discussed, some of them being hard to reconcile with the proposal. The goal of this section is (1) to show that the proposal doesn’t offer what Pöchtrager pretends it offers; (2) to show that some of the facts discussed may be handle more realistically; and (3) to show that working with this kind of model even if it provides interesting conclusions and predicts facts that were not taken into account in the paper cannot in its present version count as the final word on TVH. As we have seen, there are a lot of exceptions to vowel harmony, because of: (1) the large number of loans that entered the language; (2) some exceptional suffixes in loan words; (3) some foreign suffixes and prefixes that were borrowed; (4) some native suffixes that do not vary (do not participate in vowel harmony); and (5) compounds. These exceptions are one of the problems facing anybody trying to propose an account of TVH. Generally, analyses simply take these words out by putting some sort of diacritic on them. Pöchtrager’s analysis does
Nicolas Royer-Artuso
56
not. Instead, it is a proposal that faces these exceptions directly by offering a model where these exceptions are explained: they are actually not exceptions. They are totally conform to TVH if we understand what TVH really is about. 3.4.1. The proposal Pöchtrager’s analysis is couched in the framework of Government Phonology (GP) “which takes a very clear stand on exceptions – it is built on the assumption that there aren’t any, that phonology is exceptionless” (Pöchtrager 2010: 458). This is also called the Minimality Hypothesis (MH), which states that “processes apply whenever their conditions are met” (Pöchtrager 2010: 458). This criterion has a lot of consequences for the type of analysis that we do, and regarding the analysis of Turkish phonology, can also serve to discard the majority of accounts that has been given to TVH until now. Pöchtrager’s proposal is a development of earlier analyses of TVH that have been done in the GP framework. The details of some of the critiques he addresses to his colleagues won’t be developed here. Only his final proposal will be taken into account as it is that final version which is important for the present discussion. The phonological expressions for Turkish vowels are said to be the following (we will slightly modify that statement below): Ex. (11) i ({ } __ ) a ({}A ) u ({ } U ) o ({A} U ) i ({ } I ) e ( {A} I ) ü ( {I} U ) ö ( {I, A } U ) Note : the underlined element is the head.
Is Turkish a tongue-twister?
57
TVH is usually understood as the spreading of I and/or U to the right, which gives the following alternations in suffixation if we posit that Turkish has two types of suffixes: (1) one empty nucleus suffix type ( { } __ ) (-i, -di, -ci, etc.), corresponding to the I-type suffix of traditional analysis; and (2) one ( { } A ) nucleus suffix type (-a, -da, -ca, ect.), corresponding to the A-type suffix of traditional analysis. The description of TVH looks as follow (for the examples presented above) : Ex. (12) kiz -i kizi ( { } __ ) ( { } __ )
kiz -a kiza ( { } __ ) ( { } A )
kas -i kasi ( { } A ) ( { } __ )
kas -a kasa ({}A) ({}A)
diz -i dizi ( { } I ) ( { } __ ) --- I --->
diz -a dize ({}I) ({}A) ----- I --->
el -i eli ( { A } I ) ( { } __ ) -------- I --->
el -a ele ( { A } I ) ( { } A) ---- I --->
mum -i mumu ( { } U ) ( { } __ ) ----- U ---->
mum -a muma ({}U) ({}A)
kol -i kolu ({ A } U ) ( { } __ ) ---- U --->
kol -a ({A}U )
yüz -i yüzü ( { I } U ) ( { } __ ) ---- U --->
yüz -a yüze ( { I } U )( { } A ) ------- I ----->
göl -i gölü ({ I, A } U ) ( { } __ ) ---- U --->
göl -a göle ( { I, A } U ) ( { } A ) ---- I ---->
kola ({}A)
Nicolas Royer-Artuso
58
One remark is added: “U fails to spread […] when the target position is already headed” (Pöchtrager 2010: 462). It explains muma, kola, yüze, and göle. The problem is, as in every model of TVH, that a lot of exceptions exist to it, exceptions which are referred generally as “disharmonic words” (DWs). The list is long but let’s give Pöchtrager’s examples: bira ‘beer’, elma ‘apple’, dünya ‘world’, mühim ‘important’, kilo ‘kilo’. If we take bira, the first syllable’s element I should, according to the model, spread to the right and give bire: the a ( { } A ) of bira should become e ( { I } A ) but it does not. Ex. (13) bi ra ({}I)({}A) Thus an exception and, according to MH, a necessity to revise the model6. After a discussion concerning some of the problems found in other proposals, the basic observation that leads to his own proposal is presented: if we look at the attested vowel distribution in Turkish (Table 4), the most striking fact is the restrictions that we have on the vowels that precede i. The only vowels allowed to do so are i and a: “Since the vowel i, the realization of an empty nucleus, does not occur after a vowel containing I or U, we have to assume that i is always harmonized. If there is an I or U preceding it, this I/U will always spread” (Pöchtrager 2010: 465). Some exceptions do exist: new loans from English such as hemstir ‘hamster’, blendir ‘blender’, etc. but “given their questionable status i.e whether they are treated as real Turkish words”, it is better not treating them as “real counterexamples […] about the distribution of i” (Pöchtrager 2010: 465).
6
Words like hafif, arzu, etc. present a different kind of problem, but not for Pöchtrager’s model. Because of the revisions he made to older models, we won’t enter in the details discussed. His critiques of the models involving “dummy” morphology will also not be discussed for the same reason.
Is Turkish a tongue-twister?
59
Starting from that observation, Pöchtrager adds three claims which, if taken seriously, would according to him, eradicate all the problems associated with DWs: Claim 1 (C1): Non-initial positions can contain any PE. This claim concerns representations: according to the model, words do not need to be completely specified for the second vowel when the results are predictable given the proposed spreading conditions. C1 is needed for words like hafif ‘light’, arzu ‘desire’, etc. A-i and a-u are disharmonic sequences. If, by C1, any element in a non-initial nucleus is given permission to appear, then those words become perfectly well-formed. The word hafif, for example, is well-formed because its i is there from the beginning and A doesn’t spread. Ex. (14) ha fif ({}A) ({}I) Claim 2 (C2): I only spreads into empty-headed positions. This explains why words like kilo are well-formed. The non-initial vowel o ( {A} U ) is therefore headed and the I of the initial vowel cannot spread for that reason. Ex. (15) ki lo ( { } I ) ( {A } U ) “C1 and C2 taken together explain most of the details of TVH” (Pöchtrager 2010: 467). C2 is said also to make “important predictions about a” (Pöchtrager 2010: 467). It helps us predict that there is in fact not one a but two: one that is headed and therefore blocks the spreading of I (bira, elma, etc.), and one that is non-headed and therefore do not block the
Nicolas Royer-Artuso
60
spreading (böl ‘divide’ + suffix -ma ‘negative imperative 2nd pers. Sg.’ ( {A} __ ) = bölme ‘do not divide!’). A lot of DWs also have a long vowel as their second syllable. Long vowels are headed which explains why words like dünya, ceza ‘punishment’, ebru ‘marbling’, etc. are well-formed: their second syllable is long, thus headed and “immune to TVH” (Pöchtrager 2010: 468). Ex. (16) bi ra ({}I)({}A) ce za ({A}I)({}A )
böl ma → bölme ( { I, A } U ) ( { A } __ ) ----- I ---> eb ru ({A}I)({}U )
Claim 3 (C3): The element U spreads into empty expressions only. This explains why kol-a doesn’t surfaces as kol-o, and at the same time explains why words like jüri ‘jury’ are well-formed. Ex. (17) jü ri ({I}U) ({}I) All this, taken together, explains the attested distribution of Turkish vowels shown in Table 4. 3.4.2. Counterexamples Unfortunately, it is quite easy to find counterexamples to Pöchtrager’s proposed distribution. (1) Let’s take the “non-attested” distribution o-ü. A very “popular” word is popüler ‘popular’; bromür ‘bromide’ is another, etc. According to
Is Turkish a tongue-twister?
61
Table 4. Distribution of Turkish vowels in Pöchtrager’s analysis.
Nucleus 1
Nucleus 2 i
a
u
o
i
e
ü
ö
i a u
h h *
h h h
* ✓ h
✓ ✓ ✓
✓ ✓ ✓
✓ ✓ ✓
* ✓ ✓
* ✓ ✓
o
*
h
h
✓
✓
✓
*
✓
i e
* *
✓ ✓
✓ ✓
✓ ✓
h h
h h
✓ ✓
✓ ✓
ü ö
* *
✓ ✓
✓ ✓
✓ ✓
✓ ✓
h h
h h
✓ ✓
h = harmonic; DWs: ✓ = does exist; * = does not exist.
C1, ü is fine where it is. According to C3, U spreads only to an empty position. So we can simply rectify the table and say that o-ü is a possible sequence in Turkish. This is not hurting the model. (2) Let’s now take the “non-attested” i-{ö, ü, u} sequences. Here too we find exceptions: kilören ‘family name’, kilükal ‘gossip’, hirgür ‘fight’, Sirru’l Esrar ‘The secret of secret’, etc. But again, the model predicts the possibility of these words: i, because of its empty nucleus, cannot spread. C1 might also be invoqued (?). So again, we’ll have to modify the table and say that those words show that the distribution i-{ö, ü, u} is an attested one. (3) Finally, even if finding words that have a sequence of the “nonattested” {o, ü, u, i, e, ö}-i type is very difficult, we still can find some: Belkis ‘first name’, for example. 7 Some authors would say (Yavaş 1978, for example) that the i is triggered not by the preceding vowel, but by the k, like in the following suffixed words: faik-i ‘hight’, şevk-i ‘desire’, fevk-i ‘highness’, etc., which are also excep7
I guess that given MH, this exception would be enought to discard the model, even more if we remember that it is this kind of distribution that forms the core of the argumentation.
Nicolas Royer-Artuso
62
tions to Pölchtrager’s model8. But it is interesting to see the words that Pöchtrager himself gives as counterexamples to his own theory: blendir, hemstir, etc. Here, no k is present to produce a consonant assimilation.9 The explanation might not lie in the consonant. The strategy used by the author to evacuate those problematic words is to say that they sound non‐native to the native speakers. The third kind of exception will be the basis of what follows since the two other kinds do not violate the proposal. It is also, as we saw before, the main observation on which Pöchtrager elaborates his model. But, before, a word on the native/non-native criterion. We already saw that the criterion native/non-native does not stand for Turkish (see note 3 and below). When we say, following MH, that a process occurs whenever its conditions are met, the origin of words should not count. It is in fact exactly, even if not presented this way, the same problem facing Pöchtager in his speculations (his word) at the end of the paper. Why, he asks, “are there hardly any native DWs” (Pöchtrager 2010: 470)? It is basically the same problem facing all theories that have been presented so far. Pöchtrager’s model is one more model dealing with the complexity brought into the Turkish phonological system by the large amount of loans from languages that do not share its phonological restrictions. His answer runs as follows: we can probably argue that the vast majority of non-monosyllabic Turkish words were originally suffixed monosyllabic words that became lexified. Because, as he suggests, suffixes contain almost only a nuclear expression of the ( { } _ ) and ( { A } _ ) types, “the scarcity of native disharmonic words follows [...] [D]isharmony never had much a chance to occur” (Pöchtrager 2010: 470). Disharmony thus comes from outside the system. But because 8 9
This would also mean that consonant assimilation wins over vowel harmony.
We also have to say that the suffixed examples given above also have variants with i, so it is probably false to say that it is k that triggers i. There are also, of course, other words that we could give as counterexamples: fosforişi ‘phosphorescent’, menkibe ‘some kind of story’, zikiymet ‘precious’, nütrişin ‘nutrition’, opereyşin ‘operation’, speküleyşin ‘speculation’ (Lewis 1999: 135 for these English loans), etc., and the search was far from being exhaustive. Those words are words and not compounds, according to the accent pattern they have, as Pöchtrager would also admit.
Is Turkish a tongue-twister?
63
“[t]here is no way that certain words are marked as exceptions to spreading, as this would be a violation of the MH” (Pöchtrager 2010: 459), Pöchtrager is forced to develop a model that is complex enough to handle those non-native words: the blendir type DWs are thus a real problem for the model.10 But let’s say that they aren’t. Let’s now look at some facts which are a lot more disturbing for Pöchtrager’s proposal. It is well known that L1 phonology influences the way words of L2 will be accepted in L1’s lexicon (see Calabrese and Wetzels 2009; Major 2001; van Coetsem 1989 for examples of transfer processes in the SLA and Loan Phonology literature). It is quite difficult with this model to explain why komünist ‘communist’ (taken from French as komünist) tends to be changed to kominist (this is in fact true in all models as Turkish is supposed to have only left-to-right spreading). Within this framework, o is made of ( { A } U ). There is no i that could spread to the second syllable. Where is this i coming from? The ü is totally fine according to the model, in that it shares the U with o and that U doesn’t spread to non-empty position. The model is in fact not able to account for changes that occurred to the form of a lot of loan words. Why were the following words transformed? Ex. (18) minibüs (from French) ‘minibus’ → minübüs mumkin (from Arabic) ‘maybe’ → mümkün imperator (from Serbo-Croat) ‘emperor’ → imparator Traditional analyses do have something to say (it fits or it is closer in some way to what should be expected Turkish harmonic vowel sequences in their account). Pöchtrager cannot account for it (those sequences are perfectly acceptable according to his analysis). Let’s look at another problematic case: in its accusative form, sanat ‘art’ used to be pronounced sanati, which is fine by Pöchtrager’s TVH 10
Children pronounce blendir without asking themselves where that word is coming from: it is simply a form found in their linguistic environment.
64
Nicolas Royer-Artuso
because A does not spread. But we now hear sanati. What happened? Why is the perfectly correct sequence a-i not here anymore? Traditional analyses have no problem explaining this case. The sequence a-i is not a well-formed one according to TVH so harmonization in backness occurs. The problem of free variation can also be approached in a similar manner. Pöchtrager discusses the relativiser -ki suffix. This suffix sometimes varies (dün-kü/dün-ki both meaning ‘yesterday’s’). The only way to explain this (it is said) is to suppose a second -ki (optionality is not “expressible in GP” (Pöchtrager 2010: 470)) which has the same function, meaning and distribution but a different underlying i (one -ki has an empty nucleus, one has a nucleus that contains I). We find the same kind of variation with the future suffix -acak: Ex. (19) ol ‘be’ olacak/olucak yap ‘make, do’ yapacak/yapicak konuş ‘talk’ konuşucak/konuşacak Traditional analyses do have a problem explaining this type of variation but still have something to offer: this variation still conforms to their account of TVH. In Pöchtrager’s model, there is no way to explain what is happening. Finally, look at the following case (Lewis 1999: 97): hazir ‘ready’ isek ‘if we are’ varies with the contraction hazirsak ‘if we are ready’. Should we propose one isek word, and one -sak suffix? If not, what triggers the change from e to a, if we posit that i, as an empty nucleus vowel, cannot spread? These kinds of questions are what we presume the theory should also account for. 3.4.3. Distribution: Another explanation Pöchtrager might be right in saying that explaining the distribution of i is the crux of the problem and probably also the key to the solution. But maybe not for the reasons he proposed. I’ll give a tentative solution to it
Is Turkish a tongue-twister?
65
in (1) and (2) below. Paragraph (1) also presents facts that are falsely represented and others not taken into account in the proposal. (1) The explanation for its distribution may be searched in Pöchtrager’s discussion of the native/non-native words’ responsibility in the appearance of DWs. The reason why native polysyllabic words are (almost) never disharmonic is to be found in the fact that Turkish suffixes generally didn’t give the opportunity to develop disharmonic sequences (see the discussion above). First, the fact that some native and non-native suffixes have the power to create disharmonic sequences is an important fact. Secondly, phonological and morphological processes other than suffixation have also this power. The following processes will now be discussed:11 (a) suffixation; (b) acronym formation; (c) epenthesis; and (d) compounding, a process that is discussed for other reasons earlier in Pöchtrager’s discussion (this might not be a complete picture of the relevant facts: there might be other processes). (a) Some native suffixes actually create disharmonic sequences: -(i)yor ‘present 3rd pers. Sg.’, -gil ‘and his/her/their family’, -(i)mtrak ‘resembling X’, -ken ‘while X’, etc. Ex. (20) ütü ‘to iron’ ütüyor isin ‘to warm up’ isiniyor beklet ‘to keep waiting’ bekletiyor çekinir ‘to drag’ çekinirken bulunur ‘to find’ bulunurken savaşir ‘to fight’ savaşirken
Anil ‘person’s name’ Anilgil Ülfet ‘person’s name’ Ülfetgil Uygur ‘person’s name’ Uygurgil ekşi ‘sour’ ekşimtrak beyaz ‘white’ beyazimtrak aci ‘hot, spicy’ acimtrak
The vowels of these suffixes are immune to TVH. Pöchtrager’s analysis of -ken is that it has an e in its nucleus thus blocking spreading. Logically, the same explanation works for the other 11
I benefited from examples and analysis from Baturay (2012) and Kabak (2011).
Nicolas Royer-Artuso
66
native suffixes presented and for non-native invariable suffixes as well: -istan ‘country of X’, -izm ‘theory, ideology’, etc. Again, it is possible to give a representation that would block spreading according to the model. Ex. (21) -ken ( {A } I )
-imtrak ( { } __ ) ( { } A ) -istan ({}I)({}A)
-iyor ( { } __ ) ( { A } U )
-gil ({}I)
izm ({}I)
These suffixes do not produce some of the unattested sequences in Table 1 but can produce some of the attested disharmonic ones. (b) Acronym making is a productive process in Turkish which sometimes results in attested and non-attested DWs according to Table 1: Ex. (22) TMMOB (Türk Mühendis ve Mimar Odalari Birliği) timop/tümop/timop ITÜ (Istanbul Teknik Üniversitesi) itü AKP (Adalet ve Kalkinma Partisi) akepe TÜBITAK (Türkiye Bilimsel ve Teknolojik tübitak Araştirma Kurumu) IKSV (Istanbul Kültür Sanat Vakfi) ikaseve (c) Turkish has strong restrictions on the possibility of consonant clusters. One of them is that it is not possible to have one at the beginning of a word. When words that do have such clusters are borrowed, Turkish vowel epenthesis breaks the cluster by separating these consonants. A lot of the time, with a vowel that conform to traditional TVH accounts. But a lot of the time with a i that does not respect those traditional accounts.
Is Turkish a tongue-twister?
67
Example (23) Epenthesis respecting TVH grup ‘group’ → gurup blok ‘block’ → bulok kral ‘king’ → kiral
Epenthesis disrespecting TVH spor ‘sport’ → sipor glikoz ‘glucose’ → gilikoz grip ‘flu’ → girip
Epenthesis thus has the power to create attested and non-attested DWs. (d) Compounds show a specific stress pattern: the first word’s final syllable takes the stress, contrary to the normal stress pattern in Turkish where stress appears on the last syllable of the word. But, as time goes by, certain compounds receive their stress on their last syllable, this fact showing that they’ve been lexicalized. Through this process, some DWs with exceptional sequences containing an i are this way given permision to slowly enter the lexicon. For example, from two native words: bakir ‘copper’ + köy ‘village’ → Bakirköy ‘name of a place’. Note that in the first step, bakir has the stress on its second syllable, but that in the second step, stress is assigned to the last syllable, köy. Ex. (24) ata ‘father’ + türk ‘Turk’ → Atatürk yan ‘side’ + kesici ‘cutter’ → Yankesici ‘pickpocket’ fosfor ‘phosphorus’ + işi ‘light’ → fosforişi ‘phosphorescent’ Paragraphs (b), (c) and (d) explain some of the exceptions to the “non-attested” disharmonic sequences of Table 1 above, including the rare disharmonic sequences including i. It explains also the apparition of a lot of other “attested” disharmonic sequences. Examples in (a) do not explain the distribution because the results are not lexicalized. But (a–d) do not explain everything.
Nicolas Royer-Artuso
68
(2) The fundamental cause of the scarcity of sequences containing i and some vowels other than a and itself is probably the following: the languages that were borrowed from (mostly Arabic, Persian and French) do not possess a vowel that could be transposed as an i. This is why we do not find DWs containing an i: this vowel, having not been borrowed with loan words “diss-respecting”12 TVH, was not given the chance to move from it’s “well behaved” 13 role in the Turkish phonological system. Now that English has become the international language, and that borrowing has started to come from that direction, there is a need for using this vowel to express the sound er that we find in blender, hamster, printer ‘printer’, and the sound en that we find in speculation, operation etc. We can probably predict that, as more English words will enter the Turkish lexicon, sequences containing i with any other vowel will become a lot more frequent.14 This finally brings me to the next discussion, something that I find a lot more important than the above exercise in “searching the dictionary for minimal syllable pairs” trying to find exceptions to an author’s proposal. 3.4.4. Some theoretical consequences of DTDH Pöchtrager’s proposal, even if faced with a lot of problems, has its strengths: it predicts distributions that are not attested in Table 1 (o-ü and i-{ö, ü, u} ). It also predicts that the distribution {o, ü, u, i, e, ö}-i will not appear, and this is almost true. More importantly, it is able to deal with the majority of the DWs in a coherent way without recourse to the kind of diacritics found in the majority of works dealing with TVH. On a theoretical level, the most striking consequence I see with Pöchtrager’s type of proposal is that what we have at the end of the analysis, is not at all what the concept of VH traditionally referred to. 12
See Pöchtrager’s note 1 on the role of hip hop music in “creating the slang clipping to dis(s) from to disrespect”. 13 14
See Pöchtrager’s part 1.2 where he talks about the “well-behaved part of TVH”.
Very recently, when confronted with the “surface form” sedil, I didn’t understand it till the context of yoga unfolded the meaning of this new Turkish word: saddle is the name of a yoga position that I did not know at that moment.
Is Turkish a tongue-twister?
69
Nothing similar to a feature or property is shared anymore by the vowels inside a word. This, however, was precisely the idea behind the concept of VH. If we keep that in mind, a theory that gives us a way to predict how vowels will behave, but where the vowels inside a word share nothing anymore at the end of the analysis (except the fact of being vowels) is not something we would like to call a theory of VH, and a language that would obey these kinds of rules would probably not be called a harmonic language. But DTDH’s consequences might well have an impact on how we look at TVH in the future. On one side, TVH might not be a regular harmonic process but something else (a very restricted type of assimilation process?) that looks quite similar in its consequences. On the other side, it is possible that our understanding of VH is wrong and that Pöchtrager’s type of proposal will help us understand more deeply what kind of processes are at work in TVH and more generally in VH. But I still see some flaws in the analysis. For example, I am pretty sure that aspiratör is on the same level of “foreignness” as blendir, even if the claim “A does not spread” in conjunction with C2 makes the word aspiratör a well‐formed word within Pöchtrager’s analysis. Even more: according to the model, ikolelüpona-da-ki-ler if there was a company name having IKOLLÜPONA for its acronym (-dakiler ‘the people of’ Ikolelüpona) would be considered harmonic and so represented as more “native” than blendir. I am sure that this is not the case: the blendir class of words cannot thus be put aside in the discussion, especially when we see how important borrowing is to explain the attested and not-attested distributions that are given in Table 1 and the role that the sequence containing i has in Pöchtrager’s model. As I have said, this sequence will probably become more frequent with the arrival of English as the new donor language. Free variation is also explained with the addition of new vowels, which may be a positive point for the model (demonstrating that the set of vowels in Turkish is bigger than traditionally postulated), but a negative one as well, in that it can be seen as a way of getting rid of compromising counterexamples. I am also not sure if it is a richness of the model to prove that in Turkish we have two differently represented suffixes which have the same meaning, function and distribution and the same surface form. The proposal’s major consequence is that the distribution in Table 4 becomes explainable. But, in Turkish, if we take Pöchtrager's proposed
70
Nicolas Royer-Artuso
TVH, in conjunction with the distribution in Table 4, MH and the exceptions that I gave, every word composed with sequences of vowels taken from the set {Turkish vowels} is a possible word in Turkish. (1) This consequence is paradoxical in a discussion of VH and (2) the only facts that the model has problems explaining are adjustments that are found, for example in loan words (we have seen that traditional models are better at explaining them). If, as it is clear from SLA or Loan Phonology studies, L1 influences L2 in the form that words will take when they are borrowed, so that those processes are in fact a good place to search for some relevant aspects of a language’s phonology, then not being able to explain why some loans that conform to a model’s phonology are transformed in the process of borrowing is a big handicap for the model. The model has a problem explaining it for the simple reason that there is no reason, if we look at the model and its logical consequences, to fix something if everything is already OK. This problem might in fact just have to do with positing C1: from the moment when, in a harmonic language, any vowel is allowed to enter the non-initial nucleus, we might start to have doubt about the power of the proposed model. For those reasons I am inclined to say that more traditional analyses, even if they face tremendous problems with the exceptions they encounter, cannot be replaced by a theory of this type. As Kabak and Vogel (2011) and Kabak (2011) propose, there might be no other way than prespecify words (DWs) that do not conform to TVH (but see below). As we saw, even in the GP framework they are the ones who create problem for the analysis. 4. A note on loan adaptation It is well known that L1 phonology influences the way words from L2 will be accepted in L1’s lexicon (Calabrese and Wetzels 2009; Major 2001; van Coetsem 1989 for examples of transfer processes in the SLA and loan phonology literature). Loan words are often a good place to look in order to understand L1’s phonology. The following loans are problematic for all models dealing with TVH:
Is Turkish a tongue-twister?
71
(1) Minibüs ‘minibus’ → minübüs and imperator ‘emperor’ → imparator presuppose some form of right-to-left spreading. Even this doesn’t explain much in that the resulting words are still disharmonic. An explanation based on perceptual factors might be better to handle this type of cases. (2) Mumkin ‘maybe’ → mümkün and Omar ‘first name’ → Ömer are some of the few adapted loans where the result fits TVH’s description. But the processes involved in the adaptation are mysterious, since none of the resulting vowels are there at the beginning of the process. (3) Halal ‘permissible’ → helal and harakat ‘movement’ → hareket are interesting in that totally harmonic words become disharmonic through borrowing. A lot of loans are in the same category, and they are very difficult to explain. These kinds of facts are actually what we presume a theory should account for. Should we propose that TVH is a process that also works from right to left? And what about the third case, when harmonic words according to TVH become disharmonic? Given what has been said of the role of L1 in borrowing, we are left with a lot of unanswered questions about our L1 (Turkish) phonology. 5. Why are we there? The general conclusion is the following one: no model is saved from exceptions. Exceptions are accepted in the whole picture of TVH but handled by prespecification or some other diacritical devices, generally in the lexicon, this way becoming immune to TVH’s consequences; this, according to the proposed model and the constructs that are available for doing so. But the fundamental problem with those accounts has to do with the consequences of this type of analysis: trying to save the notion of TVH by any means leaves us with a coherent (?) formal description of the facts but with, on an empirical level, a rather large set of facts left out of it. Kabak and Vogel, for example, do a good job of presenting the prob-
72
Nicolas Royer-Artuso
lems facing different models but do not reach the only conclusion that is, for me, logically the one we should reach. This is I guess the one that is more or less assumed by Clements and Sezer: no matter how we can save TVH from its exceptions, the encountered forms in modern Turkish are so far from what we would like to call possible forms in a harmonic language that it becomes difficult to see in what way the proposed models bring something to the discussion. And this, not only for roots but, contra Clement and Sezer, also for suffixes, which is the phenomenon that probably still triggers analyses referring to harmonic processes in Turkish. Aspiratördekiler ‘those inside the vacuum cleaner’ (aspiratör ‘vacuum cleaner’ -deki ‘that is inside’ -ler ‘pl.’) is not a form we would like to describe as harmonic (vowels sharing some sort of property in a specific domain other than the fact of just being vowels). The real question is the following, given what has been presented so far: why are those models still in need of the concept of TVH? It seems that it is for the well known presupposition that a morpheme should be represented as invariant and should get his surface varied forms from processes handled by the phonological component. Keeping in mind this theoretical constraint, it would be difficult without postulating TVH to account for the fact that the suffix -di ‘past’ alternates in four different ways depending on the vowel that precedes it. We also do not want to say that suffix-related harmonic processes take not only phonological information but also morphological and/or lexical information into account. When Clements and Sezer propose that TVH applies only in the context of suffixation they are saying exactly this: inside roots, postulating the phonological rule in (25). Ex. (25) i → i, i, u, ü / (a, i) (e, i) (o, u) (ö, ü) respectively This would bring very damaging consequences to the model because of the large number of exceptions that exist to it. But for suffixation, this rule works quite well (if we still prespecify some suffixes). From that moment, each morpheme can have its own deep representation and this way, we can leave to phonology the role that was hers until now. This constraint has probably to do with a methodological presupposition related to economy of descriptive devices. Filling the lexicon with
Is Turkish a tongue-twister?
73
different variants of the same morpheme is not economical if we can propose some phonological processes that can deal with the encountered variation. But marking the lexicon for each exception is probably no more economical except if we use the diacritic [−Vowel Harmony], the most economical of all, but also the one with the most circular effects at the conceptual level (see Kabak 2011: 2837) for the presentation of this interesting device). 6. Explaining disharmony differently When authors try to maintain that TVH is operating, they think about results of the following type (I will repeat some of the examples discussed above for practical reasons): Ex. (26) bronşit ‘bronchitis’ (from French) → buronşit train ‘train’ (from French) → tiren where the epenthetic vowel respects TVH. But this kind of analysis does not stand when we see initial consonant clusters broken in the following way: Ex. (27) sport ‘sport’ (from French) sipor klüb ‘club’ (from French) kulüp where the epenthetic vowels create sequences of vowels that do not conform to TVH. This is therefore not the kind of place where to find concluding proofs of the existence of TVH. Some native processes other then the ones having to do with borrowing do also create disharmonic words: the explanation therefore has not only to do with external factors but also internal ones. (1) Acronym formation is a productive process in Turkish which sometimes results in attested and non-attested DWs:
74
Nicolas Royer-Artuso
Ex. (28) TMMOB (Türk Mühendis ve Mimar Odalari Birliği) timop/tümop/timop ITÜ (Istanbul Teknik Üniversitesi) itü AKP (Adalet ve Kalkinma Partisi) akepe TÜBITAK (Türkiye Bilimsel ve Teknolojik tübitak Araştirma Kurumu) IKSV (Istanbul Kültür Sanat Vakfi) ikaseve Note that the first one conforms to final consonantal devoicing, another productive phonological process in Turkish (look below for the discussion on nativeness). (2) Compounds show a specific stress pattern: each word of the compound is stressed on its last syllable, contrary to the normal stress pattern in Turkish where stress appears on the last syllable of the (morphologically complex) word. But as time goes by, certain compounds receive their stress on their last syllable only, showing that they’ve been lexicalized: their disharmonic character cannot be shown as being related to their particular status in the system. Through this process, some DWs with exceptional sequences are this way given permision to slowly enter the lexicon. For example, from two native words: bakir ‘copper’ + köy ‘village’, we form Bakirköy ‘name of a place’. Note that in the first step, bakir has the stress on its second syllable, but that in the second step, stress is assign to the last syllable, köy. Ex. (29) ata ‘father’ + türk ‘Turk’ → Atatürk yan ‘side’ + kesici ‘cutter’ → Yankesici ‘pickpocket’ fosfor ‘phosphorus’ + işi ‘light’ → fosforişi ‘phosphorescent’ Perhaps Turkish once had the harmonic character some still suppose it has. But borrowing so many foreign words (in conjunction with some
Is Turkish a tongue-twister?
75
native word-formation processes) may have transformed its phonology progressively (this is a good subject for a future work), but anyway completely and (maybe? probably?) irreversibly. Looking at the different distribution tables used for the different analyses can be revealing in the present context: the sequence Table 1 – Table 3 – Table 4 – Table 2 gives us an insight on the history of the language.15 Bilinguals (multilinguals) were probably in the avant-garde of the process. Because they were mastering the languages that were borrowed from, they had enough L2 phonology to be able to cope with foreign words without needing to modify them. Code-switching16 without nativization of the loan words was an actual possibility for them. For the nonbilinguals, it was not. Borrowed words that were disharmonic according to TVH probably went through a phase of nativization: transforming the L2 consonants and vowels into L1 phonological inventory, devoicing voiced consonants at the end of the word, harmonizing, etc. But slowly, with the help of habituation, schooling, etc., new generations adopted those words as they were produced. Constraints relative to harmony disappeared and we ended up with a language where every vowel sequence is permitted. This consequence can be seen in the adoption of new words in Turkish: none of them passes through the filter of the supposed TVH (anymore). The examples above show that certain transformations are made sometimes, but (1) the results are not always more harmonic than before and (2) sometimes they surface in an even worse shape. I suspect that a lot of these transformations are explainable in terms of perception (an allophone of L2 being closer to the allophone of an other phoneme in Turkish as in halal → helal) and therefore would not be considered results of production and constraints relative to it (look at this one: Salam maleykum! → Selam maleyküm!).
15
This has implications for the traditional view of TVH: one of the core elements of the analysis is that o and ö only occur in the initial syllable of a word (to explain why o and ö do not respect the pattern “followed by X and itself”); borrowed words that do not follow this constraint in a sense destroyed the constraint. Another explanation is thus needed for the asymmetry. 16
I take code-switching as a general concept which also includes borrowing from an L2 or L3 in the course of a conversation: to be able to do that means that in some ways we have mastered two codes.
Nicolas Royer-Artuso
76
7. Final discussion I claim that Clement and Sezer did not go far enough. They should have completely given up the concept of TVH, not only inside roots. Turkish suffixal morphology is also full of exceptions to TVH: the list is long and it seems that it is getting longer. What will we do with the following invariable suffixes: (1) the diminutives -o of İso ‘İsmail’, İbo ‘İbrahim’, and -oş of Fatoş ‘Fatma’, Memoş ‘Mehmet’; (2) -ko ‘?’ of şişko and the -man of şişman ‘big’, kocaman ‘immense’; (3) -matik of bankamatik, paramatik ‘cash machines’; (4) -istan ‘the country of the X’ of Hindistan ‘India’, Magyaristan ‘Hungary’; (5) -i ‘player of X’ of udi ‘lute player’, tamburi ‘tambur player’; (6) -kolik ‘addict’ of işkolik ‘workaholik’, sekskolik ‘sex addict’, etc.; (7) -im mi ‘should I X’ of yapim mi ‘should I do ?’, gülim mi ‘should I laugh?’; (8) -men/man of uzman ‘specialist’ and öğretmen ‘teacher’ but the disharmonic sportmen ‘sportsman’, barmen ‘barman’, etc. Even if some of these suffixes are not totally productive, they are sometimes used in creative ways when Turkish speakers play with their language. Prespecifying these suffixes is already a way of consenting that the analysis does not stand. The general proposal is the following: modern Turkish is not (anymore?) a harmonic language. We cannot describe in any way this language as a language where phonology constrains the distribution of certain types of vowel sequences. No vowel distribution is forbidden (anymore?) inside the domain of a word (see Table 2). Even if “the constraints on vowel co-occurrences in nearly all roots of native origin are almost the same as those in suffixes in Turkish” (Kabak 2011: 2839), leaving the problematic “nearly” and “almost” of this statement, some native word-formation processes do create problems for the analyses. And considering what has been said of L1 phonology in regard to borrowing, the fact that “any combination of vowel seems to be legitimate as long as the donor language permits it” (Kabak 2011: 2844) is not militating in favour of a harmonic phonotactics in Turkish. What must be considered is not the origin of words (etymology is not relevant in a synchronic description) but the fact that native speakers are actually able to pronounce those sequences:17 I think this is the best way to describe what phonotactical constraints should account for. For 17
I guess that we should add: without suffering, etc.
Is Turkish a tongue-twister?
77
example, when we talk about Turkish phonology, origin of words doesn’t count when we analyze consonant behaviour: certain types of consonant assimilation take place when a morphological operation triggers it, or when a foreign word enters the lexicon, etc. Turkish has, for example, a phonotactic constraint that forbids a voiced consonant to occur after a voiceless one. An assimilation processes unvoices the first voiced consonant of a suffix when it is suffixed to a root finishing with a voiceless consonant. The suffix -ci ‘somebody dealing with X’ will serve as an example: Ex. (30) Native words yalan ‘lie’ yalanci çanta ‘bag’ çantaci yol ‘road’ yolcu
Non-native words akordeon ‘accordeon’ akordeoncu gazete ‘journal’ gazeteci futbol ‘football’ futbolcu
But Ex. (31) ip ‘rope’ ipçi at ‘horse’ atçi etek ‘skirt’ etekçi kiliç ‘sword’ kiliççi
kasap ‘type of meat’ kasapçi sikayet ‘complain’ sikayetçi kadilak ‘Cadillac’ kadilakçi maç ‘match’ maççi
Note : c = dj; ç = tʃ The distinction native/non-native is not relevant here: there is no alternative for the speakers, it is totally automatic. All the models presented are trying to deal with the complexity brought into the Turkish phonological system by the large amount of loans from languages that do not share its phonological restrictions. Cophonology is the only device that is conceptually close enough to this constatation: Turkish is now a mixed language and we have to account for this fact in our descriptions. But this type of device is not very different in spirit from the [+ foreign] diacritic. The use of this kind of devices gives an intriguing picture
Nicolas Royer-Artuso
78
of Turkish speakers: native speakers of Turkish are code-switching with languages they do not know (anymore). Or, even more: they are speaking their own language with a foreign accent. Turkish being a mixed language, we have to treat it this way in our description: the phonological module does not work the same way the original one was working. The problems facing descriptions of mixed languages always remind me of the following paradox: if we have a boat and every now and then we change one of its pieces, do we have the same boat at the end of the day? It seems that all the proposed models of TVH are an affirmative answer to this question. It probably also has to do with concerns related to typology and to the history of the language we are dealing with. Some implications coming from studies on perception can be used in the present discussion to broaden the debate on the harmonic character of Turkish. It is well known that the phonological system of a language influences or even biases the way its speakers perceive the linguistic signal (see, for example, Flege 1981; Strange 1995; and Escudero 2007). Speakers of a harmonic language should therefore tend to perceive words that are not harmonic in a harmonic way, unconsciously transforming the signal making it fit the constraints of the phonological component. We have much proof that this actually happens: in Turkish (as in a lot of other languages with phonotactical constraints on consonant clusters, for example Korean, Japanese, etc.; see e.g. Dupoux et al. 1999; Kabak and Idsardi 2007), the epenthesis that was described above is actually not “added” at the level of production, but before, at the level of perception. The speakers are therefore not producing forms with epenthesis but just merely producing the form they are hearing. If this is a correct description of what is actually happening, we can suppose that a harmonic phonology would tend to produce the same effects: perceiving only harmonic forms and therefore, producing what is perceived.18 And if the most radical version of these kinds of theory, the Motor Theory of Speech Perception (Liberman and Mattingly 1985), is right in saying (very roughly) that we cannot perceive what we cannot produce, an implication drawn from the fact that it postulates a link between the 18
We also saw that epenthesis do not always create harmonic surface forms. In the present context, that would mean that their phonology do not only make the Turkish speakers perceptually fill the clusters with epenthetic vowels, but also make them fill the clusters with epenthetic vowels that do no necessarily fit “TVH”.
Is Turkish a tongue-twister?
79
speech signal and the organs that are involved in producing it at the level of perception, in the present case, there would be no chance whatsoever to find disharmonic forms in these types of languages. If this is a correct view on phonology, this discussion could provide another proof against the harmonic character of Turkish: either we say that it was never harmonic, or we say that in some way, its (actual) mixed character has completely transformed the phonological component. We can also look at acquisition for a different perspective on the subject. The debate on the perceptual accuracy of children is full of insights for the present discussion (see Smolensky 1996 for a good introduction). Generally, it is claimed that accurate perception predates accurate production. Children with Turkish as L1 do not have problems with disharmonic forms. Following Smolensky’s model, we could therefore propose that “TVH”, if it exists, disapears quite early19 if, as he says, the perception of adults’ produced forms is accurate for children (adults’ production being full of disharmonic words). Even more: I have neither heard nor observed children harmonizing disharmonic words, but have observed a lot of other non-standard results of child phonology (consonant deletion, consonant switching, etc.). A dichotomy does exist between the production/perception processes. But to my knowledge, it never involves harmonic processes. Another fact militates against a harmonic analysis: phonological processes sometimes go beyond the domain where they usually apply. I am thinking of processes like affrication, consonant assimilation, etc., that sometimes exceed the domain of the word and apply to the next word if its first phoneme allows it. I suppose that TVH viewed as a spreading process should often apply to the first vowel of the next word in fast tempo speech, slips of the tongue and other contexts of the same type. But I have never observed something resembling that and never heard about a mention of it20. 19
If we consider that TVH is a natural process (in the sense Natural Phonology uses the term, e.g. Stampe 1979, 1987; Major 1987) or some sort of repair strategy, as in Singh (1990). In Optimality Theory this role is given to Markedness Constraints which have, in a way, the same theoretical role. The problem is that it is not always (actually generally not) clear if TVH or even VH is some sort of initial process or if it is some sort of rule-system that needs to be acquired. 20
I have actually heard about “speech errors” that are the result of VH but I was far from being convinced. The research I have seen actually gives extra proof of what I am
Nicolas Royer-Artuso
80
There is a final problem: what will we do with suffixation? This is the place where the remains of TVH are still observable. Certain invariable suffixes do not create problems if we get rid of TVH. But for variable ones, we will have to suppose as many deep forms as is needed for the description: certain suffixes will need four variants, others two.21 Characterizing this variation as morphological, morphophonological or the result of analogical processes is not the subject of this paper. It depends on the model with which these results will be processed. The conclusion is that what was phonologically (phonetically?) motivated is not so anymore. On a meta-theoretical level, the conclusion of this discussion can more or less be illustrated in the following way. Grammars are generally organized in a way that reflects more or less the steps in the acquisition of language. Starting with phonology, we get to morphology (the chapter needed in between, lexicon, is handled in dictionaries) and then syntax. When we open a Turkish grammar, one of the first things we learn is that words need to respect TVH. Then comes a question: for the child saying here. Look at the following cases presented in Sofu (2001): çok yoğlu (= çok yağlı) ‘very greasy’; fuur-una (= fuar-ina) ‘its fair’ from French foire; banim başimın belasi (= benim) ‘my head’s curse’; gölmekten ölür (= gülmekten ölür) ‘laughing to death’; herkas hata (= herkes hata yapabilir) ‘everybody can make mistakes’. The first two present something like TVH (a process that goes from left to right). The problem is that the sequences of vowels is already correct /o a/, /u a/ (in the fuuruna/fuarina case, the resulting “harmony” occurs after fuar becomes fuur and from there suffixation takes its input). The last three follow a sort of anticipatory assimilation, something very common in every language, even those that do not share the “harmonic character” of Turkish. The results of Altan’s experiment (Altan 2011) are even more welcome. Some Turkish native speakers were tested in harmonic/disharmonic/mixed conditions in an experiment where they had to learn an invented “language” (which was closer to word lists than an actual language). She found that the context (harmonic/disharmonic/mixed) had an influence on the type of mistakes and on the production efficiency of these learners. Those in the harmonic condition tended to harmonize, those in the disharmonic condition didn’t, those in the mixed condition harmonized and disharmonized in their “speech errors” randomly. Having already commented so much on the mixed character of the Turkish lexicon, it is easy to speculate that its speakers will react in the same way the subjects in the mixed condition reacted to the word lists. And if we remember the discussion about loans and remember that the subjects are Turkish native speakers, it is very problematic that they were actually not influenced by their supposed TVH phonotactics in the context of this experiment. 21
This is more or less the solution that Skousen (1972) proposed for Finnish for completely different reasons than the ones I am dealing with here.
Is Turkish a tongue-twister?
81
acquiring this language, will mastering this type of phonology not become a severe handicap for the following task of learning this language’s lexicon, given what has been said so far (look at Table 2)? The only conclusion that I can think of is the following one: when a speaker develops the capacity to produce sequences he could not produce before, we can only propose that his phonological module has changed in some way (Natural Phonology has probably very interesting things to say on this matter22). Often, bi- or multilinguals will be at the source of this phenomenon, and the fact they can insert “foreign” L2 words (not so foreign for them) into their L1 without transforming them makes those words, in their surface form (L2 form), good candidates for adoption by the speech community. Take the case of Anglophones’ rendez-vous or Montrealers’ parking (pronounced with an “English r” but the rest of it following Quebec French phonology). For this last case, there is no other solution than to propose that a second r phoneme has entered the Montreal inventory through words containing it. It is not code-switching anymore: these words enter the Montreal French lexicon and are treated the same way “native” words are treated in morphological operations – inflexion, syntactic behaviour, etc. The linguistic setting of the city (strong bilingualism, the necessity of learning English to get a job, strong presence of Anglophones, etc.) explains the predisposition of the language to adopt many features of the donor language. The case of the English speaker’s rendez-vous is different from the Montrealer’s parking. It does not have an influence on English phonology, probably because this kind of codeswitching represents a rather rare phenomenon. For bilinguals, it is phonetically close to a French rendition, save the r, which – if produced with a French one – would sound very awkward for many. For monolinguals, there is no other choice then to produce it with the English phonology that is available. Turkish, if we take the proposed models seriously, is seen as a language where at least a third of its lexicon23 (after the language reform 22
See Lovins (1974) and more importantly Major (1987): they do not talk directly of this phenomenon but their discussions touch on matters very closely related and their implications are important for the present discussion. 23
The actual number of “foreign words” in Turkish is the subject of a big debate. It is tainted with so much emotion and ideology that it is hard to get a neutral perspective on the subject. The given numbers (and a lot of related things; maybe even “TVH” itself)
Nicolas Royer-Artuso
82
that got rid of a lot of “foreign” words) behaves the way rendez-vous works for Anglophones pronouncing it à la française… For some periods in the history of a language, there is no way to predict how this language will evolve, especially when intense language contact and multilinguism are implied. Some words will be borrowed according to L1 phonology. Some words are borrowed in their original form.24 Some sociological explanations do exist, but I will not touch on this subject. For new generations, these borrowed words do not create any problem: they are part of the linguistic environment and taken as “normal” words. The consequence for Turkish is that there is in this situation no possible way of developing a harmonic phonology. Results of earlier phases of the language still survive and this is why harmonic patterns can still be observed. But we should always be aware of not confusing processes with results.
are often a good indication of the political orientation the people giving them (see Lewis 1999 for good examples). The ways the statistics are produced are never clear. If one counts dictionary entries, etymologically “foreign words” are already computed this way (origin being part of the entry) for the formally (ideologically?) accepted ones (I developed some perspectives on this subject in Royer-Artuso (2012a); Royer-Artuso (2012b), and for the phonological aspect of some (created) socio-political dilemmas see RoyerArtuso (2013)). On the other hand, some dictionaries are devoted entirely to “non-native Turkish words”, this way permitting the reification of this division, its status of presupposition becoming a societal fact this way. If one counts tokens, I am pretty sure we get a totally different picture (a lot more than a third). When we listen to Turkish, even when we do not understand it, we slowly get accustomed to the şey ‘thing’ and yani ‘it means/I mean’ that punctuate every sentence (two “Arabic” words; note that the second one is “disharmonic”). A lot of the everyday life objects (clothes, kitchen instruments, accessories, etc.) are referred to with “foreign words”. The vocabulary of religion, of science, of art, etc. is almost totally foreign. 24
As always, Singh (1985 : 269) had some interesting things to say on the matter: “‘Borrowed words’ should be reserved for words that are not so domesticated. ‘Adapted words’ are borrowed words not for speakers but for historical linguists. Undomesticated, unadapted words raise questions regarding bilingualism and its transmission as well as regarding the possibility of phonotactic change as a result of language-contact and bilingualism”.
Is Turkish a tongue-twister?
83
References Altan, A. 2011. “The influence of vowel harmony on Turkish speakers learning an artificial language.” Hacetepe Universitesi Türkiyat arastirmalari dergisi 14. 27–44. Anderson, S.R. 1980. “Problems and perspectives in the description of vowel harmony”. In: Vago, R.M. (ed.), Issues in vowel harmony. Amsterdam: John Benjamins. 1–48. Baturay, S. 2012. “Loan word adaptation and vowel harmony in Turkish: A Government Phonology account”. Proceedings of ConSOLE XX. 1–22. Calabrese, A. and W.L. Wetzels (eds.). 2009. Loan phonology. Amsterdam: John Benjamins. Clements, G.N. and E. Sezer. 1982. “Vowel and consonant disharmony in Turkish”. In: van der Hulst, H. and N. Smith (eds.), The structure of phonological representations (Part 2). Foris: Dordrecht. 213–255. Dupoux, E., K. Kakehi, Y. Hirose, C. Pallier and J. Mehler. 1999. “Epenthetic vowels in Japanese: A perceptual illusion?” Journal of Experimental Psychology – Human Perception and Performance 25(6). 1568–1578. Escudero, P. 2006. “Second language phonology: The role of perception”. In: Pennington, M.C. (ed.), Phonology in context. New York: Palgrave Macmillan. 109–134. Flege, J.E. 1981. “The phonological basis of foreign accent: A hypothesis”. TESOL Quarterly 15. 443–455. Göksel, A. and C. Kerslake. 2005. Turkish. A comprehensive grammar. London: Routledge. Lovins, J. 1974. “Why loan phonology is Natural Phonology”. Papers from the Parasession on Natural Phonology. Chicago: CLS. 240–250. Kabak, B. 2011. “Turkish vowel harmony”. In: van Oostendorp, M., C. Ewen, B. Hume and K. Rice (eds.), The Blackwell companion to phonology. Blackwell Publishing. 2831–2854. Kabak, B. and W.J. Idsardi. 2007. “Perceptual distortions in the adaptation of English consonant clusters: Syllable structure or consonantal contact constraints?” Language and Speech 50. 23–52. Kabak, B. and I. Vogel. 2011. “Exceptions to vowel harmony and stress in Turkish: co-phonologies or prespecification?” In: Simon, H. and H. Wiese (eds.), Expecting the unexpected: Exceptions in grammar. Berlin: Mouton de Gruyter. 59–94. Kornfilt, J. 1997. Turkish. London: Routledge. Hulst, H. van der and J. van de Weijer. 1991. “Topics in Turkish Phonology”. In: Boeschoten, H.E. and L.T. Verhoeven (eds.), Turkish linguistics today. Leiden: Brill. 11–59. Lewis, G.L. 1999. The Turkish language reform. Oxford: Oxford University Press. Lewis, G.L. 2000. Turkish grammar. Oxford: Oxford University Press.
84
Nicolas Royer-Artuso
Liberman, A.M. and I.G. Mattingly. 1985. “The motor theory of speech perception revised”. Cognition 21(1). 1–36. Major, R. 1987. “The natural phonology of second language acquisition”. In: James, A. and J. Leather (eds.), Sound patterns in second language acquisition. Foris: Dordrecht. 207–224. Major, R. 2001. Foreign accent. Mahwah, NJ: Lawrence Erlbaum. Pöchtrager, M.A. 2010. “Does Turkish diss harmony?” Acta linguistica Hungarica 57(4). 458–473. Royer-Artuso, N. 2012. “Une approche réaliste du phénomène hétérophonique”. RTMMAM 6. 119–128. Royer-Artuso, N. 2012. “Rousseau, la République et l’esthétique musicale turque”. In: Ertem, C. (ed.), Numéro special Jean-Jacques Rousseau (numéro consacré au colloque international Rousseau et la Turquie). Istanbul: Littera Edebiyat yazilari. 113–123. Royer-Artuso, N. 2013. “Pour une phonologie comparative des phénomènes de contact des musiques de l’aire du maqām”. RTMMAM 7. Singh, R. 1985. “Prosodic adaptation in interphonology”. Lingua 67. 269–282. Singh, R. 1990. “Vers une theorie phonotactique generative”. Revue Québecoise de Linguistique 19(1). 131–163. Sofu, H. 2001. “Dil sürçmeleri”. In: Demircan, Ö. and A. Erözden (eds.), XV. Dilbilim Kurultayi bildirileri. Istanbul : Yildiz Üniversite basim yayin merkezi. Stampe, D. 1979. A dissertation on natural phonology. New York: Garland Stampe, D. 1987. “On phonological representations”. In: Dressler, W.U. et al. (eds.), Phonologica 1984. Cambridge: Cambridge University Press. 287– 300. Strange, W. 1995. “Cross-language study of speech perception: A historical review”. In: Strange, W. (ed.), Speech perception and linguistic experience: Issues in cross-language research. Baltimore: York Press. 3–45. Skousen, R. 1972. Substantive evidences in phonology. The Hague: Mouton. Smolensky, P. 1996. “On the comprehension/production dilemma in child language”. Linguistic Inquiry 27(4). 720–731. Van Coetsem, F. 1989. Loan phonology and the two transfer types in language contact. Dordrecht: Foris. Yavaş, M. 1978. “Borrowing and its implications for Turkish phonology”. Kansas Working Papers in Linguistics 3. 34–44.
