WAFL 2003 Andrew Nevins and Bert Vaux

Consonant harmony in Karaim

 Background on the Karaites
• “The Karaim settlements strewn along the length of the eastern border of the one-
time Polish kingdom must be considered the farthest northwestern outpost of
Turki[c]. Only five of these have survived to the present day…in Poniewież (Karaim
Pon’ev’ež), Troki, Wilno, Łuck, and Halicz. The colony of Poniewież is now in
Lithuania; the remaining four in Poland. The largest and strongest is the colony of
Troki (actually Nowe Troki, Karaim Troχ), a small city…” (Kowalski 1929:vi)
• “For more than six hundred years, Karaim has been spoken as a community
language in the territory of today's Lithuania and the Ukraine. Due to political
measures taken by the post-war Soviet regime, the communities are now dispersed
and the maintenance of their language has become endangered. The number of
Karaims in Lithuania is about two hundred, but only a fourth of them, mostly
members of the eldest generation, still have a communicative competence in the
language. In the Ukraine, the number of Karaim speakers is not more than six.
Karaim belongs to the Kipchak branch of the Turkic language family, more closely
to the sub-branch also including Karachay, Balkar, Crimean Tatar and Kumyk.
Three dialects of Karaim are distinguished. The eastern dialect, spoken in the
Crimea till the beginning of this century, is now extinct. There are two western
dialects. The north-western, Lithuanian, dialect, is often called the Troki (Polish) or
Trakai (Lithuanian) dialect. The other one, the south-western dialect of Galicia in
today's Ukraine, is now only spoken in the town of Halich.” (Csato 1997)
• Our data comes from Kowalski 1929 and the recordings of Csato on the
Spoken Karaim CD project (to appear)
 Karaim displays nonlocal harmony
• NW Karaim has palatal consonant harmony that spreads
[–back] without necessarily affecting the intervening vowels
• Locality: two elements A and B are local if there is no element
C of type T that intervenes between A and B
• Two families of phonological representations
– Strict Locality: Type T is the phoneme (or root node)
– Relativized Locality: Dependent Instantiation of Type T; could be
vowel, consonant, sonorant, A position, A’ position, …
– Strict Locality: No relation (spreading, feature movement) can be
established across a phoneme P without P participating in that
relation
– Consonant Harmony constitutes a direct empirical challenge to Strict
Locality
 Why is everything violable except
locality?
• Counterexamples argued to be something else that doesn’t have to obey locality:
– Consonant spreading in Semitic, which is irreducibly non-local, claimed to be
something else, i.e. reduplication by Gafos 1998, even though there is no
morpheme of which the putative reduplication is the exponent.
– Bantu nasal harmony (Ao 1991, Odden 1994), claimed to be a Correspondence
relationship by Walker 1999
• Strict Locality: anything that looks nonlocal is not at the relevant stage of the
derivation – Ní Chiosáin and Padgett (2001:50, 54) propose that “transparent
vowels actually undergo harmony at an abstract level of representation; however, the
offending feature cannot appear at the surface…another mapping to a final output is
required”.
• Would anyone propose this in syntax: a wh- phrase moves through every XP and then
its traces are deleted from all non A’ positions afterwards?
• “If there is no upper limit to the length and depth of structural representations, a
fundamental core of syntactic processes are bound to apply in local domains…[one
way to investigate this] is to assume that the process cannot apply across an
intervening element of the designated kind, which could in principle be involved in the
process” (Rizzi 1990)
Claims about Consonant Harmony
• Flemming 1996: “the only consonantal gestures which can be extended
through a vowel without significantly affecting its quality are the
differences in tongue-tip posture that can distinguish coronals”
• “the spreading of single C-Place features (major articulations) to
nonadjacent consonants appears to be restricted to [coronal], and in all
known cases of [coronal] spreading, the target must also be [coronal].”
(Clements and Hume 1995:305)
• “consonantal place features do not…spread long-distance” (Ní
Chiosáin and Padgett 2001:2)
• XXX FIND SOMETHING GAFOS SAID XXX
 Kowalski p.30
• “Departing from the general rule just stated, loanwords often maintain
intact their original phonic form. As a result of the above vowel shift, the
original vowel harmony is destroyed; and in its place there appears a
kind of consonant harmony in that inside a sound group (a word) only
hard (nonpalatalized) or only soft (palatalized) consonants are
admitted. We thus have, e.g., k’un’l’ardan’ ‘from the days next’ to
kunlardan ‘from the servants’. The vowels in the two words are identical;
the consonants, however, are palatal in the former case, and not
palatal, in the latter case. In place of two contrasting vowel series we
have in Northwest Karaim two contrasting consonant series. The
appearance of harmony has thus been transferred from the domain of
the vocalism into that of the consonantism.” (emphasis added,
translation by M. Halle).
• Kowalski, Tadeusz. 1929. Karaimische texte im dialekt von Troki. Prace
Komisji Orjentalistycznej Polskiej Akademji Umiejetnosci nr. 11. Cracow.
 Kowalski quote 2
• This phenomenon is unique in the Turkic languages, to the best of my
knowledge…in the Turkic languages, we find in Gagaus (cf. V. MoSkov,
Proben der Volksliteratur der türkischen Stämme, X, p. xxvii) a clear
palatalization of all consonants under the influence of front vowels,
without change in the nature of these vowels. One can therefore say
that in Gagausian both vowel harmony and consonant harmony exist
simultaneously next to one another. Northwest Karaim takes one further
step in that it gives up vowel harmony in favor of consonant harmony.
 Comrie 1981:63
• “Although the distinction palatalised versus velarised consonants is in
general less salient phonetically than that between front and back
vowels, and can therefore be considered purely allophonic, there are
some Turkic languages where the distinction has shifted from the vowels
to the consonants, so that one has a tendency towards consonant
harmony rather than vowel harmony. In both Gagauz and Karaim, front
vowels are backed after a palatalised consonant, so that in Karaim we
have k’oz’um’d’a for expected közümde ‘in my eye’ and öz’an’ for
expected özen `stream’. It will be seen from these examples that the
front-back distinction is not carried completely by consonants: words
with an initial vowel still retain initial front, including front, rounded
vowels, since the backing takes place as a historical process only after
consonants.”
 Hamp 1973
• “To Baskakov’s remarks on Trakai Karaite I would offer the following
analysis and obervations which also bear on more general themes. It
seems clear that the mechanism of vowel harmony in these languages
has been the factor which has facilitated or at least permitted at small
cost the structural shift of vowel quality into consonantism that we
observe. That is to say, for an earlier vowel harmony we find consonant
harmony, but with the form of the phonological rule changed in little
detail…Now, I would formulate what has happened in Karaite as being
a change in the composition of the harmony rule whereby [α front] has
been transferred from the [-cons] to the [+cons] segments of the
word. In its mechanism the rule change seems to be as simple as that;
and that simplicity is what permitted it perhaps to be so pervasive. After
that of course, the vowels simply neutralize.”
 Kowalski’s transcription scheme
[+back] [-back]
ł l
r r’
d d’
t t’
n n’
z z’
š š’
č č’
m m’
g g’
k k’
u ü
o ö
a e
y (we use Æ, ï ) i
 Fundamentals of Karaim phonology
• Illustrative pair:
– [back]: bar-a-mÆn (I go, K45) vs.
– [front] b’er’-a-m’in’ (I give, K45)
• K’org’uz’d’uy (du hast gezeigt, K70)
• Backness harmony affects all consonants. Non-initial [a o u] are
transparent even though they have front counterparts
• Kowalski:
– Front rounded vowels only in absolute initial position: ölg’an’ (dead,K44)
– No ï in absolute initial position, despite harmony: imïrtxa (egg,K81)
– [e] found only in initial syllable (see b’er’-a-m’in’ )
– The only vowels that alternate in non-initial syllables are i/ï (see b’er’-a-m’in’)
– [j] doesn’t trigger harmony: yoł (road, K108)
 Further cases of harmony
• YOUR HANDOUT WILL INCLUDE A WORD LIST:::
• K33 deverbal /-V/ Æ [v, v’, uv, uv’]
– k’el-uv’ ‘arrival’
– ač-uv ‘anger’
– č’org’a-v’ ‘belt, sleeve’
– K108 t’uz’u-v’-č’u-d’an’ ‘author-abl.’ vs. boł-uš-uv-ču ‘helper’
• t’uz’u ‘build, form, create’ + -V- + agent –CI- + abl. -DaN
• /-GaN/ ‘subject participle’
– K44 öl-g’an’ ‘dead’, k’om’ul-g’an’ ‘buried’
• K61 öl-um’-un’-d’a ‘in his death’ [from /öl-IM-IN-Da/]
• K69 tor’a-s’iz’-lig’-im-dan’ ‘from my injustice’
– The m and d are presumably palatalized
 Was Kowalski’s hearing faulty?
• Troki system is departure from the backness harmony
Turkologists would expect (cf. mishearing of Khalkha by
Turkologists, discovered by Rialland & Djamouri, Svantesson)
• Kowalski noted VH and the problematic vowels (e, ö, ü) for
other dialects, in the same book
• Kowalski notes these vowels (and their contrasts with a o u) in
initial syllables
– Cf. 42 üč’un’ ‘wegen’
• There may well be coarticulatory effects from neighboring
consonants, as in Russian, Abkhaz and Irish, but this is a
separate low-level phenomenon occurring in phonetic
implementation...
 Csato transcribes differently..
• Csato has three informants:
– one (Firkovich) matches (sometimes, roughly) with her
transcriptions, complete with central vowels
– One (the head of the community, Firkovicius) has something
resembling the Kowalski system
– Firkovich also has a variant of the Kowalski system
It is worth remembering that language preservation issues require
standardization of orthography despite interspeaker variation

Csato Kowalski-an gloss sound IPA

t’ụz’ụm’ạ t’uz’u-m’a restore-INF tHJuzJumJa
s’ụv’ạr’lạr s’uv’-ar’-lar’ lovers sJuvJaRJIlJaRJI
č’ọp’r’ạn’i č’op’r’a-n’i yeast-ACC tSJopJRJanJi
Speaker 1: Mykolas Firkovicius
• Led community for 50 years
• Died recently in his 80s
• 6 songs recorded by Csato
• From the Kïbïn Song:
b’iz’-g’ạ üv’r’ạt’-t’i-l’ạr ‘they taught us’ [bJizJigJa yvJrJatJ˘ilJAr]
we-DAT teach-PAST-PL

iš’-l’ạ-d’-l’ạr ana-lar ‘mothers made [them]’ [iSJlJAdJIlJAr AnA…Ar]

work-denom-PAST-PL mother-PL

• From another song:

k’ọt’-ụr’-m’ạ ‘to raise’ [kJotJ¨RJ¨mJA]
rise-CAUS-INF
Mykolas Firkovicius, part 2
• From a Psalm: super-short
k’ọt’ụr’-ụl’-g’ụn’ ‘lift yourself up’ [kHJçtHJ¨RJ¨lJ¨gJunJ]
lift-MIDDLE-2sg.IMV N.B. palatalized
despite preceding
back vowel

kH J ç tHJ ¨ RJ ¨ lJ ¨ gJ u nJ

high F2
low F2
 Jozef Firkovich
• From a narrative about the Karaim street in Trakai:
– Anda ed’i yụv’ụ alarnïn, anda ed’i baxčasï d´,
b’ič’ạn’l’ig’i d’´.
– ‘They had a house, a garden, and also a pasture’
• yụv’-ụ ‘house-ACC’ (cf. Turkish ev) [juvJu]
 Jozef Firkovich
• More: Notice F2 dips during both vowels
• g’ọl’-ụ ‘lake-ACC’ (cf. Turkish göl) [gJPlJu]
 Jozef Firkovich
• More: Notice F1/F2 separation visible on ü, not on ụ
• üs’n’ụ ‘on’ [YsJnJU]
 Jozef Firkovich
• More:
• uruv-u-nun ‘family-3SG-GEN’ [uRuvunun]
 Interim Summary
• Kowalski’s careful description, auditory confirmation of
samples, and visual inspection of F2 reveal that:
– Fronting of consonants is categorical
– Fronting of [a, u, o] is not categorical, but may happen in the
phonetics as a coarticulatory effect
– Degree of coarticulation depends on two factors:
• If flanked by palatal consonants on both sides
• Duration of vowel
– One possibility is that i/Æ do not alternate at all in the phonology, but
are simply more susceptible to coarticulation due to their status as
the shortest duration vowels (well-noted for high, front vowels; cf.
Japanese for well-known case; Shademan 2003 on
duration/assimilation susceptibility)
 Analytical Desiderata
• [j] (obviously –back) doesn’t initiate harmony:
– Harmony cannot be caused by specification of initial segment, but
rather must be due a specification of the root morpheme as [+back]
or not (Clements & Sezer 1982)
• Distribution of vowels (no initial y, no non-initial E, Ü Ö) must
be explained (via positional faithfulness/markedness or due to
phonologization)
• Fact that i/y are only vowels that alternate must be explained
 From Vowel Harmony to Consonant
Harmony: Phonologization
• Ohala 1981: Listener as a Source of Sound Change
• Csato 1987: Contact phenomena with Slavic languages induced interpretation of
Karaim has having palatalized consonants
• Proto-Turkic 9-vowel system interpreted as secondary articulations on consonants
• Colarusso (1992:28): “The two-vowel system of Kabardian has arisen by a
historical process in which the normal vowel colorings of the syllable peak have
been reinterpreted as belonging to the consonants and glides of the syllable
margins. Only +low cannot be reinterpreted, and so the process stopped at two
vowels”
• Kochetov (2002) agent-based parent/child simulation with random noise for
production and perception goes from C’üC’ to C’uC’:
– “a grammar that allows multiple contrasts in backness in both vowels and secondary
articulations is highly unstable because it cannot be well replicated by the learner…there are
certain default states at which the grammar naturally arrives. The first one allows multiple
secondary articulation contrasts at the expense of vowel distinctions”.
 How did Karaim arise?
Phonologization of Positional Effects
• Barnes (2003 et preq): Initial segment is significantly
lengthened, leading to greater inventory of contrasts (ü ö)
• Barnes: Proto-Turkic (Poppe 1960) had initial syllable stress,
leading to greater inventory of contrasts (e)
• Relationship between duration and coarticulation: [i] is
shortest vowel, coarticulatory velarization to the point of
categorical [y] in all positions except absolute initial (due to
increased length, hence less susceptibility)
 Synchronic Analysis
• Didactic point that must be made: the issue of non-local representations
is totally separate from that of derivational versus parallel computation!
• Whether a syntactician believes in actual covert movement or Agree, she
will not dispute the existence of (non)-intervention effects
• Because our ultimate goal here is to argue against the fallacy of strict
locality, we will present both a traditional spreading analysis and a
penalty-minimization analyses...
 Spreading Analysis
• There are four underlying vowels in Karaim: /o u a y/ (two round, two
low, no back contrast). Consonants besides j underlyingly non-palatal
• The root bears a marking indicating it is back or not
– If +back:
• All consonants retain underlying specification
• Vowels retain underlying specification
– If –back, the following featural changes are enacted:
• All consonants receive palatalized specification
• Absolute initial o,u become fronted
• Initial syllable a becomes fronted
• y becomes fronted
• All other vowels retain underlying specification
 A Static Agree Analysis in OT
Three positional markedness constraints:
• *[-bk, +rd] in non-initial syllables
• *e in non-initial syllables (except in loans)
• *Æ / # _

All of these outrank AGREE[bk]: Assess a violation for every

segment in the word-domain that does not bear an identical
[bk] specification to that of the first consonant

AGREE[bk] >> IDENT-IO[bk]

 Contributions
• NW Karaim provides a case of consonant harmony for the feature [bk],
disproving the typological claims that such phenomena do not exist
• If it was all just one ungapped gesture, there would be no way to explain
the difference in categorical assimilation in the case of y/i but only
gradient centralization in [a o u])
• The existence of non-local identity in feature specifications shows that
harmony cannot be reduced to pure articulatory optimization, but rather
must involve operations on features at a more abstract level of
representation
• The Karaim pattern represents an interesting diachronic case of
listener-reinterpretation of harmony, and of phonologization of phonetic
prominence into purely idiosyncratic distributional statements in the
synchronic grammar
 References
Comrie, Bernard. 1981. The languages of the Soviet Union.
CUP.
Csató, Eva. 1997. Turkic and Slavic language contacts: the
case of Karaim in Lithuania and the Ukraine.
HAMP REF
Kowalski, Tadeusz. 1929. Karaimische texte im dialekt von
Troki. Prace Komisji Orjentalistycznej Polskiej Akademji
Umiejętności nr. 11. Cracow: Nakładem Polskiej Akademji
Umiejętności.