Palatalization and Palatal Assimilation in Latvian
Olga Urek, CASTL, University of Tromsø
This paper provides representational and constraint-based analyses of palatalization
and palatal assimilation in Latvian. To date, Latvian has received very little attention in OT
literature, and to the best of my knowledge the formal analysis of these patterns has not been
attempted before. Data provided here constitutes strong evidence in support of a serialist OT
approach.
In Latvian, nouns of 2nd declension take a j-initial case suffix in singular genitive and
in all plural cases (1a). Stems ending with alveolar consonants show alternation in this
context: stem-final consonant gets palatalized and the glide of the suffix deletes (1b). If a
stem ends with an alveolar cluster, j-palatalization triggers palatal assimilation in the cluster
(1c).
1a.
/skap-ja/ à[ska.pja]
’closet, gen. sg’ cf. /skap-is/ à[ska.pis]
’closet, nom. sg.’
/buom-ja/à[buo.mja]
’pole, gen. sg’
cf. /buom-is/à[buo.mis]
’pole, nom. sg’
1b.
/la.s-ja/ à[la.ʃa]
’salmon, gen. sg’ cf /las-is/ à [la.sis]
’salmon, nom. sg’
/bri.d-ja/à[bri:.ʒa]
’moment, gen. sg’ cf /brid-is/à[bri:.dis]
’moment, nom. sg’
1c.
/pusl-ja/à [pu:.ʃʎa]
’bladder, gen. sg’ *[pu:sʎa] cf. [pu:.slis] ’bladder, nom. sg’
/kusn-ja/à[ku.ʃɲa]
’flux, gen. sg’
*[kusɲa] cf. [ku.snis] ’flux, nom. sg’
Crucially, palatal assimilation only applies if the trigger is located in the same syllable
as the target (2a). Syllabification in Latvian is governed by the sonority sequencing principle
(SSP, Clements 1990) and the maximal onset principle (MOP, Goldsmith 1990). However,
some onset clusters allowed by SSP are ruled out by language-specific phonotactics and
therefore should be broken up by a syllable boundary (2b).
2a.
/vals-ja/à [val.ʃa]
/gulsn-ja/à[gul.ʃɲa]
2b.
/su:tn-ja/à [sut.ɲa]
/bi:dn-ja/ à [bi:d.ɲa]
’waltz, gen. sg’
’tie, gen. sg’
’envoy, gen. sg’
’gauge, gen. sg’
*[vaʎ.ʃa]
*[guʎ.ʃɲa]
*[su:ʃ.ɲa]
*[bi:ʒ.ɲa]
cf
cf
[val.sis]
[gul.snis]
cf
cf
[su:t.nis]
[bi:d.nis]
’waltz,nom. sg’
’tie, nom. sg’
’envoy,nom. sg’
’gauge, nom. sg’
The problem arises for the forms like those in 2b, where the underlying alveolar
cluster is not a well-formed onset, while its palatalized counterpart is perfectly acceptable. In
rule-based phonology, such a pattern would be analyzed by having palatalization rules apply
to the output of syllabification rules.
Input
syllabification
palatalization
j-deletion
Output
/bri:d-ja/
bri:.dja
bri:.ʒja
bri:.ʒa
[bri:.ʒa]
/kusn-ja/
ku.snja
ku.ʃɲja
ku.ʃɲa
[ku.ʃɲa]
/sutn-ja/
sut.nja
sut.ɲja
sut.ɲa
[sut.ɲa]
Standard OT, however,
MAX( *Tj AGREE NoCoda
DepLink(
does not allow for
V(VV-place)
intermediate levels of
place)
place)SY
representation,
and
L
therefore the output
candidates that differ /sutn-ja/
*!
*
with
respect
to a. sut.nja
*!
*
palatalization
and b. sut.na
Lc.
sut.ɲa
*!
*
syllabification are all
*!
evaluated in parallel. d. su.tɲa
**
This leads to the Me. su.ʃɲa
situation where both /kusn-ja/
*!
*
*[su.ʃɲa] and [sut.ɲa] a. kus.nja
compete as an output b. kus.na
*!
*
for /sutn-ja/, and the c. kus.ɲa
*!
*
intended loser wins due d. ku.sɲa
*!
to MOP. It is illustrated àe. ku.ʃɲa
**
in the tableau, where
*Tj is a phonotactic constraint against coronal consonants immediately followed by a palatal
glide.
In this paper I propose the analysis of Latvian palatalization and palatal assimilation
based on Harmonic Serialism model (HS, McCarthy 2010). In HS, Gen constructs a set of
output candidates that are minimally different from the input. The winner selected by Eval is
then re-submitted to Gen. The loop continues until the output of Eval is identical with the
input. Following Clements (1991), I assume that palatal and palatalized consonants differ
from their plain counterparts in that the former have the feature [coronal] attached to the Vplace node. Therefore, j-platalization and palatal assimilation are treated as the leftward
spreading of V-place[coronal] feature within a specified domain (here: syllable onset).
On the first pass through Eval, only the candidates that minimally differ from the
input /sutn-ja/ are considered. Due to the high-ranking syllabification constraints, the output
of the first pass is the form |sut.nja|. This form is re-submitted to Gen, which creates a new
set of output candidates. Note that each output candidate can be different from the input
either in syllabification or featural composition, but not both (due to minimal divergence
requirement).
On the second pass through Eval, the form |sut.n[V-place-cor]a|, containing a floating
feature, wins due to MAX(V-place) constraint. On the third pass, the form |sut.ɲa|, where the
V-place[cor] attached to the alveolar consonant, is selected as optimal, and on the fourth pass
the evaluation converges with [sut.ɲa] as an output. Note that |su.ʃɲa| can not be a member of
the candidate set since it differs from the input both in terms of syllabification and featural
composition. Thus HS solves the problem outlined above by ensuring that forms *[su.ʃɲa]
and [sut.ɲa] never compete in the same candidate set.
Selected references:
McCarthy, John J. Submitted 2010. "An introduction to Harmonic Serialism". // Clements, G.
N. 1990. “The role of the sonority cycle in core syllabification.” In: John Kingston and Mary
Beckman (eds.), Papers in Laboratory Phonology I: Between the Grammar and the Physics
of Speech.// Clements, G. N. 1991. "Place of Articulation in Consonants and Vowels: a
Unified Theory," Working Papers of the Cornell Phonetics Laboratory, 5. // Goldsmith, John
A. 1990. Autosegmental and Metrical Phonology. Blackwell.