Uniwersytet Warszawski Wydział Neofilologii Marek Krzemiński Nr albumu 282 742 Dental Obstruent Palatalization in Polish from the Perspective of Modern American Phonological Theories Praca magisterska na kierunku filologia w zakresie filologia angielska Praca wykonana pod kierunkiem prof. zw. dr. hab. Jerzego Rubacha Instytut Anglistyki Wydział Neofilologii UW Warszawa, czerwiec 2015
61
Embed
Dental Obstruent Palatalization in Polish from the Perspective of Modern American Phonological Theories
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Uniwersytet Warszawski
Wydział Neofilologii
Marek Krzemiński
Nr albumu 282 742
Dental Obstruent Palatalization in Polish from the Perspective of Modern American Phonological Theories
Praca magisterska na kierunku filologia
w zakresie filologia angielska
Praca wykonana pod kierunkiem prof. zw. dr. hab. Jerzego Rubacha
Instytut Anglistyki Wydział Neofilologii UW
Warszawa, czerwiec 2015
2
Oświadczenie kierującego pracą
Oświadczam, że niniejsza praca została przygotowana pod moim kierunkiem i stwierdzam, że
spełnia ona warunki do przedstawienia jej w postępowaniu o nadanie tytułu zawodowego.
Data Podpis kierującego pracą
Oświadczenie autora (autorów) pracy
Świadom odpowiedzialności prawnej oświadczam, że niniejsza praca dyplomowa została
napisana przez mnie samodzielnie i nie zawiera treści uzyskanych w sposób niezgodny z
obowiązującymi przepisami.
Oświadczam również, że przedstawiona praca nie była wcześniej przedmiotem procedur
związanych z uzyskaniem tytułu zawodowego w wyższej uczelni.
Oświadczam ponadto, że niniejsza wersja pracy jest identyczna z załączoną wersją
elektroniczną.
Data Podpis autora (autorów) pracy
3
Streszczenie
„Palatalizacja zębowych obstruentów w języku polskim
w ujęciu współczesnych amerykańskich teorii fonologicznych”
Niniejsza praca magisterska ma na celu przedstawienie palatalizacji obstruentów zębowych w
języku polskim z perspektywy wybranych teorii fonologicznych. Rozdział pierwszy przybliża
podstawowe założenia teoretyczne teorii SPE (Chomsky i Halle 1968), teorii Geometrii Cech
In (1), each column represents a feature complex and specifies the articulatory properties of
every phone of the word tak ‘yes’. It is important to note that the columns of features have no
internal structure. There is no hierarchy of features and features do not have a set order. Also,
every segment is composed of all the existing features. Different phones are the results of
setting the value on the features to either positive or negative.
12
In SPE, for example, rules of grammar are designed to affect features rather than
separate units of speech sounds. Polish has a rule of Final Devoicing, which affects all voiced
obstruents by changing them into voiceless ones in a word-final position. Consider a
schematic formulation of Final Devoicing in (2).
(2) Final Devoicing
b v d z dz dʒ ʒ dʑ ʑ g → p f t s ts tʃ ʃ tɕ ɕ k / __ #
The rewrite rule in (2) makes reference to particular speech sounds of Polish by listing the
segments in the input, the output and the context of the rule. Although rule (2) is accurate
from a descriptive point of view, it can be replaced by a much simpler rule by referring to
distinctive features rather than phonemes. This is illustrated in (3).
(3) Final Devoicing
[+obstr] → [–voice] / __ #
The formulation of Final Devoicing in (3) is not only more economical than the rule in (2),
but also has a greater predictive power than its schematic counterpart. Namely, the rule in (2)
is limited to the Polish set of voiced obstruents and fails to predict devoicing of non-native
segments. Polish learners of English devoice voiced dental fricatives //ð// in words such as
bathe [beɪð], even though the segment does not belong to their native repertoire of sounds.
The formulation of Final Devoicing in (3), however, rightly predicts devoicing of any voiced
obstruent by a native speaker of Polish.
13
1.1.2. Feature Geometry
Feature Geometry (Clements 1985, McCarthy 1988, and others) is a theory of representations
where distinctive features are hierarchically organised. The hierarchy is represented in the
form of a tree diagram. The tree is composed of constituents of three kinds: nodes, which are
monovalent and distinctive features, which are binary or privative. The relationship between
constituents is represented by lines linking the elements. Consider the Halle-Sagey model of
Feature Geometry (Sagey 1986, Halle 1992, 1995) presented in (4).
(4)
ROOT
LAR [±cons][±sonor][±nas][±later] SL
[±voice] [s.g] [c.g.] PL [±cont]
[±strid]
LAB COR DOR
[±rd] [±ant] [±distr] [±high] [±back] [±low]
In FG, phonological processes are limited to delinking and/or spreading operations. Thus, the
rule of Voice Assimilation involves an operation of delinking of the [+voice] feature and
subsequent spreading of the feature from another, neighbouring segment. This is presented
schematically in (5).
14
(5) Voice Assimilation:1
ROOT ROOT ROOT ROOT
[–son] [–son] [–son] [–son]
LAR LAR LAR LAR
[–v] [–v] [+v]
1.1.3. Lexical Phonology
Lexical Phonology is a theory of rule organisation. It differentiates between three classes of
rules: cyclic, postcyclic and postlexical. Cyclic and postcyclic rules operate inside the lexical
component. The scope of application of lexical rules is limited to the domain of words.
Postlexical rules apply after the rules of syntax have formed the sentence. If follows from the
model in (6) that rules that apply to words take precedence over the rules that apply across
word boundaries. Inside the Lexical Component, cyclic rules apply in tandem with word
formation rules and postcyclic rules apply to morphologically complete words.
(6) Lexical Phonology mechanism
1 Alternatively, Voice Assimilation could be represented as a process of laryngeal node delinking and laryngeal node spreading. Both operations have the same effect for Polish.
WFRs cyclic rules
Lexical roots
postcyclic rules
syntax
postlexical rules
[s] [s] [z] [d]
a) b) c)
15
1.1.4. Optimality Theory
The backbone of Optimality Theory is the assumption that all languages share the same
underlying principles. These universal principles are believed to determine possible structures
and patterns in any language. In Optimality Theory, phonological processes in a language
result from the interaction between sets of constantly conflicting markedness and faithfulness
constraints. Markedness constraints mandate to impose changes on the output forms while
faithfulness constraints strive to preserve the correspondence between the input and the output
forms. It is assumed that all languages share the same set of constraints. Cross-linguistic
diversity stems from language-specific constraint ranking. The candidates which incur
violations of the high-ranked constraints are penalized by exclusion from the evaluation.
Because it is impossible for a candidate to satisfy all constraints, it is the candidate that
violates the least number of constraints, or the candidate with lower-ranked violations, that
emerges as the optimal output.
The process of selecting the optimal output begins in the LEXICON from where the
underlying forms of words are taken and submitted as the inputs to evaluation. Next,
GENERATOR generates a list of possible outputs called Candidates, which are subsequently
submitted for evaluation by EVALUATOR, which finally appoints the optimal output form.
This is illustrated by the diagram adapted from Kager (1999) in (7).
(7) OT mechanism
LEXICON Input GEN
CandidateA
CandidateB
Candidate…
EVAL Optimal
Output
16
The evaluation of candidates is presented in the form of a tableau shown in (8) below.
(8) Schematic evaluation
//input// C1 C2 C3 C4
a. ☞ candidate a
candidate a
* * b. candidate b * *! * c. candidate c **! d. candidate d *!
In tableau (8), Candidate (8d) violates the highest-ranked C1 and is excluded from the further
evaluation. The ranking is marked by a solid line separating C1 from other constraints,
violation is marked by ‘*’ and exclusion by ‘!’. Candidates (8c), (8b) and (8a) all violate C2.
Only Candidate (8c) is excluded because it violates C2 twice. The remaining Candidates, (8b)
and (8a), are passed on for further evaluation because each has an equal number of violations.
The broken line between C3 and C4 denotes the lack of internal ranking between the two
constraints. This means that the ranking could be either C3 >> C4 or C4 >> C3 as both rankings
reject Candidate (8b). Candidate (8a) violates C4, but the violation is of minimal importance
hence (8a) emerges as the optimal output. The winning candidate is marked by ‘☞’.
It is important to note that evaluation is strictly parallel in OT and does not have
intermediate stages. This assumption of standard OT (strict parallelism) is rejected by
Derivational Optimality Theory, which admits separate phonological levels. The levels in
DOT resemble levels proposed in Lexical Phonology. Namely, DOT distinguishes between
the stem level, the word level and the sentence level, which correspond to the cyclic, the
postcyclic and the postlexical components in Lexical Phonology. Each level may introduce
changes to the ranking of constraints, but the changes must be minimal. It must be noted,
however, that the evaluation within each of the levels remains strictly parallel.
17
1.2. Descriptive background
This section offers a descriptive background for the following chapters. It discusses Polish
soft and hard dental obstruents, prepalatal obstruents and the surface inventory of Polish
vowels.
1.2.1. Polish dental and prepalatal obstruents
Polish consonants can be either soft or hard. Namely, a consonant can be articulated with an
additional gesture of the body of the tongue. If the tongue moves towards the hard palate, we
obtain softening, that is, a palatalizing effect. The hardening, or velarization, is the movement
of the back of the tongue towards the soft palate (Wierzchowska 1963, 1971). In feature
terms, palatalization is expressed by the feature [–back] and velarization by [+back]. Table (9)
presents hard and soft obstruents relevant for the subsequent analyses.2
(9) Polish dental obstruents and prepalatals.
s t s’ t’ ɕ tɕ
contin + – + + + –
anter + + + + – –
strid + – + – + +
back + + – – – –
2 For reasons of space, I show voiceless obstruents only.
18
In (9), it is important to note that the soft segments [s’ t’] differ from [ɕ tɕ] not only in the
manner of articulation (this is expressed by the feature [+strid]), but also in the place of
articulation. This is indicated by the feature [–anter] for the prepalatals and [+anter] for the
dentals.
The Polish inventory of vowels and their featural representation is provided in the form of a
feature matrix below.3
(10) Polish vowels
Vowels: a ɛ ɔ u i ɨ
high – – – + + +
back + – + + – +
low + – – – – –
round – – + + – –
The table in (10) shows that Polish has two front vowels: [ɛ] and [i]. The frontness of the
vowels is encoded by the feature [–back]. All remaining vowels are phonologically [+back].
3 The inventory of Polish vowels can be further extended by two yers. For more information, see Rubach (1984).
19
Chapter 2
Rule-based phonology
This chapter considers palatalization effects on dental obstruents in Polish from the
perspective of rule-based phonological frameworks. The analysis and the reasoning are based
on the observations made by Rubach (1984) and Gussmann (1978). Section 2.1 begins with a
schematic and articulatory description of Coronal Palatalization. Next, the section discusses
the SPE account of Coronal Palatalization presented in Gussmann (1978). Section 2.2.
investigates low phonetic effects of Surface Palatalization in Polish. A phonetic description of
this process is followed by an analysis in the framework of SPE phonology. Section 2.3.
analyses the same facts from the point of view of Lexical Phonology. In addition, the section
considers description of the relevant processes in the Feature Geometry Theory. Section 2.4.
offers partial conclusions.
2.1. Coronal Palatalization
This section provides a schematic description of Coronal Palatalization from the articulatory
point of view. Selected data are used to illustrate the process. Section 2.1.1. discusses
formalisation of Coronal Palatalization in the SPE framework. It examines selected cases
from Gussmann (1978) and Rubach (1984).
Polish has a productive palatalization rule which transforms dental obstruents
//s z t d// into prepalatals [ɕ ʑ tɕ dʑ] in the context of a front vowel.4 The alternations
generated by the process are demonstrated in (1).
4 The rule of Coronal Palatalization affects all coronal segments of Polish //s z t d n r ɫ//, but this dissertation deals only with obstruents. For a more competent discussion of the topic, see Rubach (1984).
The changes illustrated in (1) involve three kinds of operations. First, the affected segments
exhibit the change of the place of articulation. Specifically, the sounds in the input are
retracted to a prepalatal position. Second, all sounds receive secondary place of articulation
expressed by the movement of the blade of the tongue towards the hard palate. Third, dental
stops exhibit a change in the manner of articulation and become prepalatal affricates. The
alternations are accounted for by a rule of Coronal Palatalization, which is stated
schematically in (2).
(2) Coronal Palatalization: s z t d →ɕ ʑ tɕ dʑ / __ i, ɛ
The question is how rule (2) should be stated in the SPE framework. This question is
addressed in the next section.
2.1.1. The SPE framework
Gussmann’s (1978:86) generalization, quoted in (3), predicts palatalization of all [+anterior]
segments in the context of a front vowel. In keeping with the assumptions of the model
5 The letter <i> serves here as a way of denoting palatalization of the consonant. It does not represent a high front vowel. Cf. Rubach (1984). 6 In Polish voiced obstruents are devoiced in a word-final position by a rule of Final Devoicing. Thus, the final //z// of mróz ‘frost’ becomes [s] and the //d// of trud ‘hardship’ changes into [t].
21
developed by Chomsky and Halle (1968), the feature [+anterior] encompasses both dentals
and labials.7
(3) Anterior Palatalization: + anterior → + high− back
/ __ −conson− back
The above formulation of the rule predicts a structure-adding rather than structure-changing
process that transforms dentals //s z t d// into their palatalized counterparts [s’ z’ t’ d’] by
adding the features [+high] and [–back]. Under Gussmann’s analysis, the outputs of the rule
in (2) are achieved by later rules that add [+strid], [+distrib] and [–anterior] to the structure of
the affected segments.
Because the rule in (3) applies in an unrestricted manner, with no regard to
morphological information, it is beset by a number of problems. These difficult cases are
recognised and analysed by Rubach (1984) and Gussmann (1978). One of such problematic
instances is presented in (4), where non-palatals appear phonetically in the immediate context
of front vowels, a combination, which, in theory, should not be attested due to the rule in (3)
and later spell-outs.
(4) też ‘also’ [tɛʃ] (adv.)8
dech ‘breath’ [dɛx] (masc. nom.sg.)
sen ‘dream’ [sɛn] (masc. nom.sg.)
zez ‘squint’ [zɛs] (masc .nom.sg.)
7 In the Halle-Sagey model of Feature Geometry (Halle 1992, Sagey 1986), the feature [anterior] depends from the coronal node, thus it can only apply to coronals and not labials. In the following discussion only dental obstruents will be examined. 8 Due to Final Devoicing, //ʒ// is devoiced to //ʃ//. Simirarly, //z// is devoiced to //s// in the examples below.
22
To account for the conundrum, Gussmann (1978) proffers the following solution. In every
instance of [ɛ] preceded by a non-palatal consonant, the mid-front vowel must be derived
from a back yer, which, being a back vowel, lacks palatalizing properties. It is crucial that [ɛ]
is not derived before Anterior Palatalization applies. The change of [!̌] to [ɛ] is accomplished
by the rule in (5) taken from Rubach (1984).
(5) Lower: +syll+high−tense
→ [–high] / __ C! +syll+high−tense
To present the application of Lower, consider the derivation of zez ‘squint’ [zɛs]
(masc. nom.sg.) given in (6).
(6) UR //z!̌z// N/A Anterior Palatalization N/A Lower /z!̌s/ Final Devoicing SF *[z!̌s]
The above derivation failed to yield the correct surface form because the context for Lower
was not met. In order for Lower to apply, the underlying representation of zez ‘squint’ //z!̌z//
(nom.sg.) must be modified by postulating a word-final yer. Thus, the UR of zez ‘squint’ has
to be //z!̌z+!̌//. The final year is assumed to be the nom.sg. suffix. Now the final yer provides
the context for Lower and successfully derives [ɛ] from [!̌]. The word-final yer, which
remains unaffected by the rule in (5), is finally deleted by the rule in (7).
(7) Yer Deletion: +syll+high−tense
→ ∅
23
The derivation in (8) shows that the yer-blocking strategy requires the existence of two rules,
which are crucially ordered after Anterior Palatalization.
(8) UR //z!̌z+!̌// N/A Anterior Palatalization /zɛz+!̌/ Lower /zɛz/ Yer Deletion /zɛs/ Final Devoicing SF [zɛs]
To sum up, under Gussmann’s analysis words such as zez ‘squint’ [zɛs]
(masc .nom. sg.) would have to be derived from the underlying form //z!̌z+!̌// by Anterior
Palatalization applying vacuously due to the lack of palatalizing context, Lower lowering the
first yer to [ɛ] and finally Yer Deletion deleting the final yer of the nom.sg suffix.
This interpretation becomes problematic when we consider other cases, e.g. the
gen.sg. of zez ‘squint’, zeza [zɛz+a]. Because the UR of zeza //z!̌z+a// does not include a final
yer, the context for Lower is not met and consequently, zeza //z!̌z+a// surfaces as *[z!̌z+a].
What is more, Gussmann’s analysis fails to account for disyllabic and polysyllabic words that
do not have a penultimate yer in their underlying representation, as exemplified in (9)
b. głos Edwarda [gwɔs ɛdvarda] ‘Edward’s voice’, uraz Edwarda [uras ɛdvarda]
‘Edward’s injury’, trud Edwarda [trut ɛdvarda] ‘Edward’s hardship’, but
Edwarda [but ɛdvarda] ‘Edward’s shoe’
c. partia [part’ja] ‘political party’, diwa [d’iva] ‘Diva’, singiel [s’iŋg’ɛl] ’single’,
zjazd [z’jast] ‘rally’
The rule is stated schematically in (15).
(14) Surface Palatalization:11 C→C’ / __ i,j
2.2.1. The SPE framework
Gussmann in his research on consonantal changes disregarded “low phonetic
regularities.” (1978: 27). The alternations shown in (13a) and (13c) are not accounted for in
his analysis. Gussmann’s original Anterior Palatalization, which yields soft dental obstruents,
cannot be employed to explain the changes in (13). First, the rule of Anterior Palatalization is
excluded from applying across word boundaries and, therefore, cannot account for the
changes in (13a). Second, its application to the words in (13c) would result in ill-formed 11 Surface Palatalization affects all consonants, not just those presented in (13a) and (13c). See, for example, Rubach (1984).
28
surface forms. Anterior Palatalization and the later spell-out rules would incorrectly predict
*[ɕ] instead of [s’] in words such as singiel ‘single’ (masc.nom.sg.). Third, the rule operates
in the context of i and ɛ, and the data in (13) show that the rule must be confined to i and j, as
there is no palatalization of the words in (13b). Another generalization is warranted.
One possible solution to this problem is to postulate a separate palatalization rule. The
scope of the application of the rule must be restricted to stem-internal segments and to
neighbouring segments across word boundaries. In SPE terms, this is achieved by referring to
a word boundary, ‘#’, which stands for a feature bundle [–seg, –FB, +WB]. The rule
Palatalization (12) VA ------ ------ /-s’# idɨ/ /d’iva/ Surface
Palatalization (15) SF [-ɕitɕ] [tɛz+a] [tɛʑ+ɛ] [-s’# idɨ] [d’iva]
12 Based on Rubach (1984: 246).
29
Because the application of Anterior Palatalization is restricted to segments not spanning word
boundaries, it affects only głosić ‘to preach’ [gwɔɕitɕ] (verb) and tezie ‘thesis’ [tɛʑ+ɛ]
(fem. loc.sg.) in derivation (17). The rule yields [ɕ] and [ʑ] from //s+i// and //z+ɛ//
respectively. The //t// in teza ‘thesis’ [tɛz+a] (fem. nom.sg.) and the //d// in diwa ‘diva’ [diva]
(fem. nom.sg.) remain unffected by Anterior Palatalization, because both segments in //tɛ//
and //di// are tautomorphemic, thus Anterior Palatalization cannot apply. In addition, the
application of Anterior Palatalization is stymied by the word boundary which separates the
//s// and the //i// in głos Idy ‘Ida’s voice’ [gwɔs#idɨ]. Next, Surface Palatalization applies to
głosić ‘to preach’ /gwɔɕitɕ/ (verb) by adding the features [–back,+high] to the structure of [ɕ].
Because the prepalatal fricative already is [–back,+high], the application of Surface
Palatalization is vacuous. Surface Palatalization has no influence on the words teza ‘thesis’
[tɛz+a] (fem. loc.sg.) and tezie ‘thesis’ [tɛʑ+ɛ] (fem. loc.sg.) because the rule has only i in its
context. Finally, the //s// in głos Idy ‘Ida’s voice’ [gwɔs#idɨ] and the //d// in diwa ‘diva’ [diva]
(fem. nom.sg.) change to [–back,+high] segments. Although the derivation in (16) places
Anterior Palatalization before Surface Palatalization, the ordering of the rules for this analysis
is of no relevance, as both arrangements yield the correct surface forms.13
To conclude, the analysis of Polish palatalization must make reference to
morphological structure of words. In SPE, this is accomplished by referring to word and
morpheme boundaries in the formulation of the rules. Such approach, albeit successful in this
investigation, fails to account for more complicated areas of Polish phonology.14
13 The ordering of Surface Palatalization before Coronal Palatalization would result in gło//s+i//ć going through an intermediate stage of [s’+i] before arriving at [ɕ]. 14 The discussion of this issue falls outside of the scope of this dissertation. For a detailed discussion, see Rubach (1984).
30
2.3. Lexical Phonology
This section investigates palatalization effects of dental obstruents within the framework of
Lexical Phonology (Kiparsky 1982, Booij and Rubach 1987). The main focus of the section is
the interaction of morphology and phonology in the derivation of palatalization rules. I
initially assume the SPE theory of representation and investigate the relevant cases based on
Rubach (1984). Section 2.3.1 considers the effects of Coronal Palatalization and Surface
Palatalization in the Halle-Sagey Feature Geometry model of representation (Sagey 1986;
Halle 1992) and summarises Rubach’s (2006) analysis, which shows the need for an
intermediate stage in the derivation of prepalatals.
To account for the two kinds of palatalization processes described in the previous
section, Rubach (1984) postulates two rules, stated schematically in (17) and (18).
(17) Coronal Palatalization: s z t d → ɕ ʑ tɕ dʑ / __ i, ɛ
The rule in (18), in a similar vein to the one proposed in (12), changes //s z t d// directly to
[ɕ ʑ tɕ dʑ] without an intermediate stage of /s’ z’ t’ d’/.
(18) Surface Palatalization: C → C’ / __ i,j
The formulation of Surface Palatalization in (18) is almost identical to the one given in (15).
The rules in (17) and (18) differ from those in (12) and (15) in that the latter pair makes no
reference to morphological boundaries (a morpheme boundary and a word boundary,
respectively). The problem of rule convergence presented in the Section 2.1. is resolved here
by assigning each rule to either the lexical or the postlexical component. Coronal
31
Palatalization must be restricted to the lexical component for the following reasons.15 Frist, it
is excluded from applying across word boundaries as in głos Edwarda ‘Edward’s voice’
[gwɔs#ɛdvarda]. Its application must take place before the rules of syntax have formed the
phrase, thus the rule cannot be postlexical. What is more, Coronal Palatalization applies only
to strings spanning morpheme boundaries, as in tezie ‘thesis’ [tɛʑ+ɛ] (masc. loc.sg.), where
the stem-initial [tɛ] is tautomorphemic and the [t] remains unpalatalized. This shows that
Coronal Palatalization must be governed by the Derived Environment Constraint in its
morphological sense. The stem //tɛz// is taken from the lexicon hence it is not derived, with
the consequences being that the dental stop remains intact. In contrast, the word tezie ‘thesis’
[tɛʑ+ɛ] (masc. loc.sg.) is derived by a WFR adding the loc.sg. morpheme //ɛ// to the stem,
which allows for the application of Coronal Palatalization. A parallel derivation of głosić,
teza, tezie, głos Idy and diwa in the Lexical framework is given in (19).16
(19)
15 The rule is actually cyclic and, as shown later in the discussion, it must be restricted to derived environments. See, Rubach (1984). 16 For reasons of space, only the relevant parts of the words are included in the derivation.
‘głosić’ ‘teza’ ‘tezie’ ‘głos Idy’ ‘diwa’
UR //-s+i-// //tɛz+a// //tɛz+ɛ// //-s// //i-// //div+a//
The weakness of this analysis lies in the fact that it requires the admission of an
abstract transitional phase for which there is no supportive phonological evidence.
34
Namely, dental obstruents need to go through an intermediate stage of /s[–bk] z[–bk] t[–bk] d[–bk]/
before they surface as [ɕ ʑ tɕ dʑ].
Surface Palatalization is a much simpler process to formalize from the point of view
of Feature Geometry. This change is accomplished by a spreading-cum-delinking procedure
where the DORS node, along with its dependants [–back] and [+high], spreads onto the
consonant. The process is fully assimilatory and does not involve abstract spell-out rules as in
the case of Coronal Palatalization. A Feature Geometry account of Surface Palatalization is
given in (24).
(24) Surface Palatalization: C → C’ / __ i
ROOT [+cons] ROOT [–cons] PLACE PLACE DORS DORS [+back] [–back] [+high]
A point that begs discussion is the partial convergence of Surface Palatalization and
General Palatalization. The similarity between the two rules is striking. It might therefore be
tempting to minimize the number of rules by incorporating one into the other, and thus
simplifying the grammar. Rubach (2006) notices that this is not possible as the context and
the output of the two rules differ. General Palatalization operates in the context of i and ɛ,
whereas the application of Surface Palatalization is limited to the context of i. General
Palatalization spreads the feature [–back] because the feature [+high] is not included in the
make-up of ɛ. Surface Palatalization spreads the whole DORS node along with its dependants:
35
[+high] and [–back]. The conclusion is that General Palatalization and Surface Palatalization
must remain separate rules, despite their similarity (Rubach 2006).
As for the distribution of the rules presented above, General Palatalization and the
later spell-out rules tandem must be confined to the lexical component. The reasoning is fully
parallel to the one provided in Section 2.3. Namely, (20) and (22) have to be lexical to protect
phrases such as but Idy ‘Ida’s shoe’ [but’ # idɨ] from surfacing as *[butɕ#idɨ]. Surface
Palatalization must remain postlexical.
2.4. Partial conclusions
The aim of this chapter was to review the palatalization process of dental obstruents from the
point of view of selected rule-based frameworks. SPE phonology fails to capture the
assimilatory nature of palatalization. As discussed in Rubach (1984), Coronal Palatalization is
a rewrite rule which changes //s z t d// directly to [ɕ ʑ tɕ dʑ]. This process involves an
arbitrary addition of the features [+strid], [–anter] and [+distr]. What is more, an SPE strategy
of including morphological and lexical information unnecessarily complicates the rules. This
problem is readily resolved in the analysis carried out in Lexical Phonology, where
generalizations are allocated to different phonological components and no reference to word
or morphological boundaries is necessary in the structure of the rules.
In Feature Geometry, the processes of Coronal Palatalization and Surface
Palatalization are fully assimilatory. Both processes involve the spreading of the feature
[–back] onto the affected consonant. The surface forms of Coronal Palatalization are derived
with the aid of spell-out rules. This requires the admission of an intermediate stage of
/s[–bk] z[–bk] t[–bk] d[–bk]/ before they surface as [ɕ ʑ tɕ dʑ]. Moreover, General Palatalization and
Surface Palatalization cannot be amalgamated into one rule despite their conspicuous
alikeness. The rule-based models fail to capture the functional unity of the processes, as they
36
have no formal way of stating the ‘output goal’, which is the agreement in backness between
hard dental obstruents and front vowels. The issue of functional unity is resolved in the next
chapter, where I recast the abovementioned analyses in Optimality Theory.
37
Chapter 3
Constraint-based approach
In this chapter, I investigate how Optimality Theory (Prince & Smolensky 1993, McCarthy &
Prince 1995) (henceforth, OT) accounts for the facts of Polish palatalization of dental
obstruents on the basis of selected examples. The analysis is based on observations made by
Rubach (2003, 2006). Section 3.1. looks at how the driving force behind palatalization can be
expressed in terms of OT constraints drawing on typological evidence from Slavic languages.
The section also shows that an OT analysis of Polish palatalization demands a broader view
of the matter. Namely, palatalization is only one of the possible strategies satisfying the
demands dictated by PAL constraints. Alongside palatalizations, fronting and retraction are
also discussed. Subsequently, it is shown that OT fails to provide an adequate analysis of the
relevant examples due to ranking paradoxes arising from its strict parallelism. In section 3.2.
the parallel one-level approach is contrasted with Derivational Optimality Theory (Kiparsky
1997, 2000, Rubach 1997, 2000a, b, 2005, 2007), which allows for level distinction much in
the same way as Lexical Phonology does. Section 3.3. provides a summary of Chapter 3 and
presents partial conclusions.
3.1. Palatalization in Optimality Theory
This section discusses Palatalization of dental obstruents in the framework of OT. It opens by
establishing the main propellers of the changes related to palatalization processes in Polish.
Further, the section investigates possible strategies of satisfying PAL constraints, and
38
develops constraint rankings for both types of palatalization, Coronal Palatalization and
Surface Palatalization, as well as retraction and fronting.
The problems arising from the limitations of rule-based models can be resolved by
adapting the framework of Optimality Theory. Let us consider once again the rules of Coronal
and Surface Palatalization. For the reader’s convenience, the schematic formulations of the
rules are repeated below in (1) and (2).
(1) Coronal Palatalization: s z t d → ɕ ʑ tɕ dʑ / __ i, ɛ
(2) Surface Palatalization: s z t d → s’ z’ t’ d’ / __ i,j
In the past research, as shown in the previous chapter, rules (1) and (2) were treated as
separate, seemingly unrelated processes despite their noticeable alikeness. The rules converge
in two places. First, the environment in which they apply is partially shared by both rules, as
they both apply before //i//. Second, both processes yield [–back,+high] segments. In the
framework of OT these facts should be expressed by a single underlying driving force.
3.1.1. Driving force behind palatalization
Rubach (2003) proposes two markedness constraints as the main propellers of the alternations
stated in (3) and (4).
(3) PAL-i: 18 A consonant and a following high vowel must agree in backness.
(4) PAL-ɛ: A consonant and a following mid vowel must agree in backness.
18 I assume after Rubach (2006) that PAL-i covers both [i] and [j] in its jurisdiction.
39
The rationale behind postulating two separate constraints triggering palatalization is of
typological nature.19 One of the basic tenets of OT is that constraints must be universal for all
languages and that languages may differ only in the way the constraints are ranked. Rubach
(2003) shows that different languages allow different triggers of palatalization.20 For example,
Ukrainian palatalization is limited to the context of i but excluded from applying before ɛ,
whereas Russian and Polish palatalize before both i and ɛ. One general constraint, PAL,
mandating agreement in backness of a consonant neighbouring on any front vowel would fail
to account for the facts of Ukrainian. It must therefore be assumed that palatalization in Polish
is driven by two separate markedness constraints. It is also worth noting that, given the
universality of constraints in OT, reduction of the number of generalizations, constraints in
OT, in a bid to simplify the grammar does not hold water. All constraints are shared by every
language of the world, therefore the question of their number in phonological analysis is
irrelevant.
3.1.2. Strategies to satisfy the PAL constraint
The dictates of PAL-i and PAL-ɛ in the matter in hand may be satisfied by a number of
operations. The vowel may assimilate to the hard consonant by retraction, //Ci// → [Cɨ], or
the converse, the consonant may undergo palatalization, //Ci// → [C’i]. The question of which
of the strategies is preferred in a language is determined by faithfulness constraints, militating
against the changes in the output. Rubach (2003) proffers the following set of IDENT-IO
constraints:
19 See Chen (1973).
20 Rubach’s (2003) analysis is restricted to Slavic languages exclusively.
40
(5) a. IDENT-V([-back]): Input [–back] on a vowel must be preserved as output
[–back] on that vowel.
b. IDENT-C([+back]): Input [+back] on a consonant must be preserved output as
[+back] on that consonant.
The role of IDENT-V([–back]) is to preserve the frontness of the vowel. IDENT-C([+back])
fights to preserve the backness of the consonant. The ranking IDENT-V([–back]) >>
IDENT-C([+back]) predicts palatalization of a consonant when followed by a front vowel. In
such scenarios, the underlying cluster of //Ci// is rendered [C’i], as it is more important to
preserve the [–back] feature of the vowel than to protect the [+back] feature of the consonant.
The reverse ranking, IDENT-C([+back]) >> IDENT-V([–back]), brings about the change in the
backness of the vowel. Thus, //Ci// converts into [Cɨ], as the retention of the backness of the
consonant takes precedence over the retention of the [–back] feature on the vowel. The
schematic evaluation of palatalization and retraction is given in (6) and (7), respectively.
(6) Palatalization: //Ci// → [C’i]
PAL IDENT-V ([–back])
IDENT-C ([+back])
a. Ci *! ☞ b. C’i * c. Cɨ *!
In the above evaluation, Candidate (6a) is ruled out by the high-ranked PAL as it contains a
string of vowel and a consonant which do not agree in backness. Candidate (6c) incurs a fatal
violation of IDENT-V([–back]), which bans retraction. This leaves Candidate (6b), which
violates the lowest IDENT-C([+back]), but complies with the demands of PAL and IDENT-
V([–back]), as the optimal output.
41
(7) Retraction: //Ci// → [Cɨ]
PAL IDENT-C ([+back])
IDENT-V ([–back])
a. Ci *! b. C’i *! ☞ c. Cɨ *
In Tableau (7), similarly to the evaluation in (6), Candidate (7a) is excluded due to the
violation of the high-ranked PAL. Candidate (7b) is rendered inadmissible as a consequence
of the violation of IDENT-C([+back]). Finally, although Candidate (7c) violates the least
important constraint IDENT-V([+back]), it conforms with the requirements of the higher
ranked constraints and, thus, emerges as the winning candidate.
Both strategies of satisfying PAL-i, retraction and palatalization, are valid solutions in
Polish phonology. Rubach (1984) shows that borrowings containing strings of //s z t d//
followed by i may either undergo the allophonic palatalization presented in (2), or may be
subject to retraction, where the [–back] value of the vowel is changed to [+back].21 Rubach
(2003) demonstrates that in low class Polish retraction is the default strategy of choice. In
high class Polish, on the other hand, retraction is optional and Surface Palatalization usually
takes precedence. The data in (8) are representative examples.
21 Rubach (2003) points out that retraction is triggered by hard coronal segments and does not apply to labials or velars. Thus, French pilote → pilot [p’ilɔt] ‘pilot’ and English hippy → hipis [x’ip’is] ‘hippy’ show no retraction of the vowel.
42
(8) Low Class Polish High class Polish
English striptease [strɨptɨs] [str’ipt’is] ∼ [strɨptɨs]
Latin maximum [maksɨmum] [maks’imum] ∼ [maksɨmum]
Latin optimum [ɔptɨmum] [ɔpt’imum] ∼ [ɔptɨmum]
BUT: only [ɨ] in maksymal+n+y [maksɨmal+n+ɨ] ‘maximal’ and optymal+n+y
[ɔptɨmal+n+ɨ] ‘optimal’
Standard OT predicts that a language may employ either retraction or palatalization, but not
both, to satisfy PAL-i. The fact that Polish has both palatalization and retraction as possible
solutions to the demands of PAL-i is problematic from the point of view of a parallel analysis.
The problem of contradictory rankings in standard OT, which allows only a parallel
evaluation, is resolved by the admission of derivational levels. Each of the rankings in (6) and
(7) is assigned to one of the levels.22
Another set of constraints is required to account for the phonemic changes in (1). The
change of the place of articulation and affrication in the case of palatalized dental stops is
driven by Posterior and Stridency constraints proposed by Rubach (2006). Soft coronals are
retracted to a prepalatal position by POST and soft dental stops are turned into affricates by
STRID. Rubach (2006) remarks that STRID and POST are the OT equivalents of the spell-out
rules presented in (22) in Section 2.3.1. of Chapter 2.
(9) a. POST: Palatalized coronals must be [–anter].
b. STRID: Palatalized coronals must be [+strid].
22 The discussion of the organisation of the rankings is postponed until Section 3.2.
43
Typological evidence form Slavic languages shows that (9a) and (9b) cannot be unified under
one simple constraint. Rubach (2006) evokes data from Slovak and Vilnius Polish, a dialect
of Polish spoken in Lithuania, to back this claim. The palatalization strategy in Slovak
involves retraction of the main place of articulation to a prepalatal position. A string of
underlying //ti// is simultaneously palatalized and retracted to [t̠’] without affrication. In
Vilnius Polish the reverse is true: palatalization goes hand in hand with affrication but not
retraction, thus producing a soft coronal fricative [ts’] from the underlying string of //ti//. Had
there been one constraint mandating retraction of the place of articulation and affrication, the
alternations in Slovak and Vilnius Polish would have been impossible to derive. Given that,
POST and STRID must remain separate constraints.
In order to derive the correct surface forms of Slavic languages, POST and STRID
must interact with faithfulness constraints militating against the abovementioned changes.
This is resolved by the following IDENT-IO constraints (McCarthy and Prince 1995).
(10) a. IDENT([+anter]): [+anter] on the input consonant must be preserved in the
output correspondent of that consonant.
b. IDENT([-strid]): [–strid] on the input consonant must be preserved in the
output correspondent of that consonant.
IDENT([+anter]) blocks retraction of //t// to a prepalatal position mandated by POST, and
IDENT([–strid]) militates against the affrication which results from the dictates of STRID.
Now, the question of which form surfaces as the optimal output is a matter of constraint
ranking. To portray the interplay of the constraints in (9) and (10), consider Tableau (11),
where I present the effects of Coronal Palatalization.
44
(11) Polish: //t+i// → [tɕi]
PAL-i POST IDENT ([+anter]) STRID IDENT
([–strid]) a. ti *!
b. t’i *! * c. ts’i *! * d. t̠’i * *!
☞ e. tɕi * *
In Tableau (11), Candidate (11a) is eliminated as it incurs a fatal violation of the
highest-ranked PAL-i. Candidates (11b) and (11c) are discarded from further evaluation by
POST. Because all the remaining Candidates, (11d) and (11e), are in violation of
IDENT([+anter]), that is, there is a tie between them, they are moved on to the next constraint
in the hierarchy for further evaluation. STRID penalizes Candidate (11d) leaving Candidate
(11e), which violates only the lowest-ranked IDENT([strid]), as the optimal output.
To accommodate the effect of Surface Palatalization, the ranking must be reversed.
Namely, IDENT([+anter]) and IDENT([+strid]) must be undominated to block the change in
place of articulation and affrication, respectively. This is shown in (12).
(12) Polish: //ti// → [t’ i]
PAL-i IDENT ([+anter]) POST IDENT
([–strid]) STRID
a. ti *! ☞ b. t’i * * c. ts’i * *! d. t̠’i *! * e. tɕi *! *
Candidate (12a) incurs a fatal violation of the high-ranked PAL-i and is removed from further
evaluation. Next, Candidates (12d) and (12e) are killed due to the violation of
IDENT([+anter]). Candidates (12b) and (12c) tie on POST and the burden of evaluation is
45
passed on to IDENT([–strid]), which excludes Candidate (12c). Candidate (12b) commits a
violation of lower-ranked STRID and emerges as the winning candidate.
In Polish //ti// may palatalize lexically to [tɕi] and postlexically to [t’i]. Alternations as in
knot ~ knocić ‘botch’, [ˈknɔt] → [ˈknɔtɕitɕ], are obtained by ranking POST above
IDENT([+anter]) and STRID above IDENT ([–strid]). The allophonic type of palatalization, as
in but Idy ‘Ida’s shoe’ [but’ # idɨ], is achieved by the reverse ranking, where IDENT ([+anter])
is ranked higher than POST, and IDENT ([strid]) must dominate STRID. Note that the
evaluation of //s// → [s’] and //s// → [ɕ] is fully parallel to that of //t// → [t’] and //t// → [tɕ].
STRID and IDENT([–strid]) play no part in the evaluation of //si//, as the underlying segment
is already [+strid]. The above analysis poses a rather nettlesome question. How the two
rankings in (11) and (12) can coexist within one phonological system since they are mutually
exclusive? Both rankings are necessary to account for Polish palatalization, yet they cannot
coexist in standard OT. Again, a strictly parallel evaluation is incapable of producing optimal
outputs of Polish palatalization.
Another possible scenario where PAL plays the lead part involves strings of
underlying soft segments followed by a back vowel. Under the influence of PAL, an
underlying sequence of //C’ɨ// might emerge as [C’i], thus satisfying PAL by fronting the
vowel. An alternative strategy is to derive [Cɨ] from //C’ɨ// by the hardening of the soft
consonant. Consider the data in (13) below. The underlying suffix //ɨ// of the fem. gen.sg.
retains its backness when it is preceded by a hard consonant (13a). The same suffix emerges
as [i] when preceded by a soft segment, as in (13b). It must be noted that the soft segments in
(13b) are underlying as they appear in contexts that do not trigger palatalization (Rubach
analogy to Lexical Phonology, differentiates between phonological levels. The levels are of
three kinds: the stem level (root plus suffixes), the word level (stem plus prefixes) and the
postlexical level (sentences) (Kiparsky 2000). The following subsections investigate the
organisation of the rankings offered in the previous sections within the framework of DOT.
3.2.1. Evaluation at Level 1
Level 1 in DOT corresponds to the lexical level in Lexical Phonology in that its application is
limited to the domain of words. This means that phrases such as but Idy ‘Ida’s shoe’ [but#idɨ]
are not evaluated at this level. What is more, the changes introduced at Level 1 must be
restricted to derived environments. It is at this level that words such as trudz+ie [trudʑɛ]
‘hardship’ (loc.sg.) should be evaluated as they are derived in the morphological sense: a
PAL-i IDENT-C ([–back])
IDENT-V ([+back])
☞a.C’i * b.Cɨ *! c.C’ɨ *!
48
suffix is added, which creates a new derived structure. Words containing tautomorphemic
clusters of //tj//, such as in partia [part’ja] ‘political party’, remain unaffected due to the non-
derived environment blocking.
The driving force behind palatalization at Level 1 must be the tandem of PAL-i and
PAL-ɛ, which collectively have j, i and ɛ as their environment. The words in (1), Chapter 2,
show that palatalization across morpheme boundaries is triggered by both i and ɛ. If only
PAL-i were responsible for palatalization at Level 1, the correct form of trudz+ie [trudʑɛ]
‘hardship’ (loc.sg.) would be impossible to achieve. Similarly, if PAL-ɛ were the sole driver
of palatalization at this level, words such as knoc+i+ć [knɔtɕitɕ] would fail to exhibit
palatalization. For the sake of compactness, in the remaining part of Section 3.2.1., PAL-ɛ and
PAL-i will be referred to collectively as PAL-i,ɛ.24 As for the strategy satisfying PAL-i,ɛ
palatalization, (6) has priority over retraction, (7). PAL-i,ɛ and IDENT-V([–back]), must be
ranked above IDENT-C([+back]) ensuring that the vowel remains front and the consonant is
palatalized, as in (6).
Rubach (2003) shows that the full effects of Coronal Palatalization (1), that is the
changes in the place and manner of articulation, must be postponed until Level 2. The
rationale behind this position is cursorily discussed below.25
Coronal Palatalization is shown to be inextricably intertwined with Velar
Palatalization. The latter rule changes velar obstruents //k g x// to [tʃ dʒ ʃ]26, respectively,
when followed by a front vowel in words such as in krocz+y+ć ‘to step’ //krɔk+i+tɕ// →
[krɔtʃɨtɕ]. The context for the rule is fully parallel to the one required by Coronal
Palatalization, where front vowels are the triggers. Rubach (2003) demonstrates that the
24 PAL-i and PAL-ɛ must remain separate constraints. The following sections provide more evidence to support this claim. 25 For a more detailed account see (Rubach 2003). 26 Phonetically these segments are retroflex.
49
derivation of kroczyć ‘to step’ in the framework of DOT requires an intermediate stage.
Namely, //k g x// are palatalised to / tʃ’ dʒ’ ʃ’ / at Level 1 and subsequently hardened to [tʃ dʒ
ʃ] at Level 2. This must be the case since clusters such as [tʃɨ] which originate from //k+i// are
impossible to obtain at Level 1 due to the high-ranked IDENT-V([–back]). The ranking of the
constraints responsible for obtaining /tʃ’ dʒ’ ʃ’/ at Level 1 rules out /ɕ ʑ tɕ dʑ/ as a possible
output by ranking *[–distr] above *[+distr]. If both //t+i// and //k+i// were to palatalize to [tʃ’i]
at Level 1, there would be no way of distinguishing between //tʃ’i// that derives from a
coronal segment and that which comes from a dorsal segment at Level 2. Consequently, all
//tʃ’i// would eventually surface as [tʃɨ] at Level 2, resulting in unattested outputs such as knot
//knɔt// → knoc+ie *[knɔtʃɛ], instead of [knɔtɕɛ]. For this reason, the transformation of //s z t
d// → [ɕ ʑ tɕ dʑ] cannot be carried out completely at Level 1. It is concluded that Coronal
Palatalization must yield /s’ z’ t’ d’/ at Level 1 to avoid collision with Velar Palatalization.
The correct evaluations of //t+i// and //t+ɛ// at Level 1 are given in (16) and (17).27
(16) Level 1 //t+i// → [t’i]
IDENT-V ([–back]) PAL-i,ɛ IDENT
([+anter]) POST IDENT ([–strid]) STRID
a. ti *! ☞b. t’i * * c. ts’i * *! d. t̠’i *! * e. tɕi *! *!
f. tɨ *!
27 The evaluation of //s z d// is fully analogous. As pointed out earlier in the discussion, STRID is mute on //s z// as the segments are already [+strid].
50
(17) Level 1 //t+ɛ// → [t’ɛ]
IDENT-V ([–back]) PAL-i,ɛ IDENT
([+anter]) POST IDENT ([–strid]) STRID
a.tɛ *! ☞b. t’ɛ * * c. ts’ɛ * *! d. t̠’ɛ *! * e. tɕɛ *! *! f.təә *!
The high-ranked PAL-i,ɛ enforces agreement in backness between the vowel and the
consonant. In both evaluations this demand is flouted by Candidates (16a) and (17a), which
are immediately ruled out from the evaluation. All the remaining candidates satisfy PAL-i,ɛ,
but violate other high ranked constraints in a bid to do so. Candidates (16f) and (17f) conform
to PAL-i,ɛ by retracting the vowel, a strategy penalized by the high-ranked IDENT-V
([–back]). Candidates (d) and (e) in (16) and (17) are excluded from the evaluation because
they shift from anterior to posterior position. This manoeuvre is blocked by IDENT([+anter]).
Finally, Candidates (16c) and (17c), which, additionally to palatalization, show affrication,
lose by violating IDENT([–strid]). Candidates (16b) and (17b) emerge as the optimal outputs
at Level 1, and are submitted as the inputs for evaluation at Level 2.
As presented in the previous section, Polish has a rule of fronting which transforms
underlying //t’+ɨ// into [t’+i], as shown by the data in (13b). The strategy of satisfying PAL-i,ɛ
at Level 1 is to keep the [–back] feature of the underlying soft consonant rather than [+back]
feature on the vowel. Thus, IDENT-C([–back]) must outrank IDENT-V([+back]) to allow for
this change to happen. The underlying form of postac+i //pɔstat’+ɨ// (gen.sg.) surfaces
eventually as [pɔstatɕ+i]28 and not *[pɔstatɨ]. IDENT-C([–back]) and IDENT-V([+back]),
responsible for fronting, do not interact with IDENT-V([–back]) and IDENT-C([+back]),
28 I assume after Rubach (2006) that morpheme internal soft segments [ɕ ʑ tɕ dʑ] are derived from underlying //s’ z’ t’ d’//, respectively. All underlying //t’//, are spelt out to [tɕ] at Level 2.
51
which are in charge of palatalization. The former pair of the constraints is mute on underlying
strings of //Ci// and the latter is mute on //C’ɨ//. To sum up, fronting and palatalization can
coexist at the same level provided that IDENT-C([–back]) and IDENT-V([–back]) outrank
IDENT-V([+back]) and IDENT-C([+back]). The tableau showing the evaluation of //pɔstat’+ɨ//
→ [pɔstatɕ+i], where for the reason of space only the relevant part of the word is examined, is
given in (18) below.
(18) Tableau for postac+i //t’+ɨ// → [t’i]
PAL-i,ɛ IDENT-C ([–back])
IDENT-V ([–back])
IDENT-C ([+back])
IDENT-V ([+back])
☞a. t’i * b. tɨ *! c. t’ɨ *!
Candidate (18c) is rejected because it does not conform to the demands of the high-ranked
PAL-i,ɛ. Candidates (18a) and (18b) show agreement with respect to the feature [±back], thus
complying with PAL-i,ɛ. Candidate (18b) is rejected due to the violation of
IDENT-C([–back]). Candidate (18a) emerges as the optimal candidate, while violating
IDENT-V([+back]), which is the lowest-ranked constraint.
3.2.2. Evaluation at Level 2
Evaluation at Level 2, similarly to Level 1, remains limited to the domain of words. The
differences between Level 1 and Level 2 are three. First, the main difference consists in that
the derived environment condition is inactive at Level 2. This means that the effects of PAL
constraints are active morpheme internally. Second, PAL-ɛ is inactive at Level 2 and PAL-i
takes over the role of the driver for palatalization and related processes. This is formally
52
expressed by the following ranking: IDENT-C([+back]), IDENT-V([–back]) >> PAL-ɛ. The
above assumption is borne out by the fact that words such as teza [tɛza] ‘thesis’ (nom.sg)
contain a hard segment followed by a mid front vowel morpheme-internally and show no sign
of palatalization. Finally, the preferred strategy of satisfying PAL-i at Level 2 is vowel
retraction rather than consonant palatalization. The effects of palatalization at Level 2 are
limited to spell-out operations of Level 1 soft outputs: //s’ z’ t’ d’// to [ɕ ʑ tɕ dʑ].
The process of retraction was illustrated by the examples of borrowings in (8). Rubach
(2003) shows that words such as optymalny ‘optimal’ [ɔptɨmal+n+ɨ] and maksymalny
‘maximal’ [maksɨmal+n+ɨ] always surface with a mid back vowel [ɨ]. Recall from Section
3.1.2. that vowel retraction is limited to coronal segments. Labials and velars display
palatalization.
(19) IDENT-Coron([+back]): Input [+back] on a coronal must be preserved as
output [+back] on that coronal.
The relevant constraints governing vowel retraction must therefore be IDENT-V([–back]) and
IDENT-Coron([+back]) (Rubach 2003). It is important that IDENT-Coron([+back]) be ranked
above IDENT-V([–back]) as the observed change is in the quality of the vowel and not of the
consonant. The evaluation of optymalny is shown in Tableau (20), where only the relevant
fragment of the word is considered.
53
(20) Tableau for optymal+n+y: //ti// → [tɨ]
PAL-i IDENT-Coron ([+back])
IDENT-V ([–back])
a. t’i *! ☞b. tɨ * c. t’ɨ *!
Level 2 completes the effects of Coronal Palatalization from Level 1 by adding
affrication and shifting the place of articulation of the soft dentals. The ranking of the
constraints at Level 2 gives precedence to POST over IDENT([+anter]), ensuring that soft
/s’ z’ t’ d’/ receive a prepalatal place of articulation. The change in the manner of articulation
of /t’ d’/ is brought about by the reranking of STRID above IDENT([–strid]). Notice that
IDENT-Coron([+back]) from the previous section has no bearing on the evaluation of //t’//
since the segment is [–back] and not [+back]. Thus, the Level 1 output of knoc+i+ć
//knɔt+i+t’//, a verb from knot [ˈknɔt] ‘botch’, is /knɔt’+i+t’/. This form is admitted as an
input to Level 2 where it is further spelt out to [knɔtɕitɕ]. The evaluation of knoc+i+ć is
tabulated in (21).
(21) Level 2 evaluation of knoc+i+ć /knɔt’+i+t’/ → [knɔtɕitɕ].
PAL-i IDENT-V ([–back])
IDENT-C ([–back]) POST IDENT
([+anter]) STRID IDENT ([–strid])
a. ti *! * b. t’i *! *
c. ts’i *! * d. t̠’i * *! ☞ e. tɕi * *
f. tɨ *! *
54
Candidates (21a) and (21f) are rejected due to the violation of the high-ranked PAL-i and
IDENT-V([–back]), respectively. Their incursions of IDENT-C([–back]) are of no relevance as
the candidates are outside the evaluation due to the earlier incursions. Candidates (21b) and
(21c) lose by violating POST. Candidate (21d) is excluded by STRID as it fails to comply with
the demands of the constraint. The winning Candidate, (21e), violates the low-ranked
IDENT([–strid]) due to the change in the value of the feature [strident].
At Level 2, morpheme-internal clusters, such as //sj zj tj dj//, are no longer impervious
to the demands of PAL-i because operations at Level 2 are not governed by the derived
environment condition. It has been shown that words such as partia ‘party’ [part’ja] and
similar forms in (13c) eventually surface with a soft segment. However, this sound change
cannot be introduced at Level 2 due to the high-ranked IDENT-Coron([+back]), which blocks
palatalization of hard coronal segments. As presented in (20), the preferred strategy at Level 2
is vowel retraction rather than consonant palatalization. Given that, [tj] clusters should surface
with a retracted version of [j]. Because the representation of a palatal glide [j] on the melodic
tier is equal to that of a nucleusless front vowel [i], the logical candidate at Level 2 should
contain a nucleusless [ɨ], which eventually surfaces as a central unrounded back glide [ɨ̯].
Although a candidate such as [tɨ̯] obeys PAL-i, it is ultimately doomed, as the central
unrounded glide does not belong to the repertoire of Polish sounds. It can be therefore easily
eliminated by postulating an undominated segment inventory constraint, *ɨ̯, banning it from
surfacing. Another, more probable outcome of the constraint interaction at Level 2 is
retraction of [i̯] to [u̯], which eventually surfaces as [w]. The current ranking of the constraints
predicts partia //partja// to surface as *[partwa] instead of [part’ja]. To prevent the change of
//tj// to [tw] at Level 2, Rubach (2000) puts forward the following constraint:
(22) *ONSET ([u]): The melodic segment [u] cannot be in the onset.
55
Consider Tableau (23) showing the evaluation of ‘partia’.
(23) Level 2 evaluation of /partja/ → /part’ja/
IDENT-Coron ([+back]) *ONSET[u] PAL-i
☞ a. tj
* b. t’j *!
c. tw *!
It is crucial that *ONSET ([u]) outranks PAL-i. Since in the evaluation above, the violation
incurred by Candidate (23a) must be regarded as less serious than that committed by
Candidate (23c) and (23b). The optimal output /tj/, with the violation of the least important
PAL-i, is admitted as the optimal input to Level 3.
3.2.3. Evaluation at Level 3
Finally, Level 3 assumes a strategy similar to that of Level 1 in that it favours palatalization
over retraction. Again, it is more important to preserve the frontness of the vowel than to
preserve the backness of the consonant. Level 3 must, therefore, rank IDENT-V([–back])
above IDENT-C([+back]) and IDENT-Coron([+back]) to allow palatalization of the
consonant. This means that clusters such as //Ci// surface as [C’i], rather than [Cɨ]. The
difference between Level 1 and Level 3 consists in the fact that the latter employs PAL-i
instead of PAL-ɛ as the main propeller of palatalization. The latter constraint is inactive and
must be therefore ranked below faithfulness constraints. Thus, teraz ‘now’ and but Edwarda
‘Edward’s shoe’ are free from palatalizing at Level 3 since the environment of PAL-i is not
met. What is more, the scope of the application of Level 3, which is a postlexical level, is
56
unrestricted. That is, its jurisdiction is stretched from words to word phrases. It is at this level
that but Idy ‘Ida’s shoe’ //but#idɨ// and partja ‘party’ //partja// are finally palatalized.
The palatalization strategy at Level 3 differs immensely from the one employed at
Level 2 in one more aspect. Namely, the change of the manner and place of articulation
alongside palatalization is dispreferred at this level. That is, //s z t d// palatalize to [s’ z’ t’ d’]
rather than to [ɕ ʑ tɕ dʑ]. This is formally expressed by the reranking of IDENT([+anter]) and
IDENT([–strid]) above POST and STRID. The correct evaluations of but idy ‘Ida’s shoe’ and
partia ‘party’ are presented in tableaux (24) and (25), respectively.
(24) Level 3 evaluation of but idy ‘Ida’s shoe’. /ti/ → [t’i]
IDENT-V ([–back]) PAL-i IDENT
([+anter]) POST IDENT ([–strid]) STRID
a. ti *! ☞b. t’i * * c. ts’i * *! d. t̠’i *! * e. tɕi *! *
f. tɨ *!
(25) Level 3 evaluation of partia ‘party’. /tj/ → [t’j]
IDENT-V ([–back]) PAL-i IDENT
([+anter]) POST IDENT ([–strid]) STRID
a. tj *! ☞b. t’j * * c. ts’j * *! d. t̠’j *! * e. tɕj *! *
f. tw *!
In the above tableaux, Candidates (24f) and (25f), which retract the originally front vowel or
glide, fatally violate IDENT-V([–back]). Candidates (24a) and (25a) are penalized by PAL-i
57
for the lack of agreement in backness. Candidates (24d), (24e), (25d) and (25e), which show
retraction of the original consonant, are ruled out by IDENT([+anter]) and are excluded from
further evaluation. Candidates (24b), (24c), (25b) and (25c) tie on violating POST and are
moved for further evaluation to the lower-ranked constraint. IDENT([–strid]) penalizes
Candidates (24c) and (25c) for not preserving the original feature [–strid] on the consonant.
As a result, Candidates (24b) and (25b), with the minimal violation of STRID, emerge as the
optimal outputs.
3.3. Partial conclusions
This chapter has discussed the formalisation of the effects of Polish palatalization of dental
obstruents within two major constraint-based phonological theories: Optimality Theory and
Derivational Optimality Theory. Section 3.1. showed OT’s superiority over the rule-based
phonological theory which consists in uniting seemingly unrelated processes of palatalization,
retraction and fronting in a bid to comply with the general requirement of the PAL constraints.
Although OT makes immense advances in the way it formalises palatalization, the theory fails
in the way it organises its generalizations. Section 3.1.2. proved that a strictly parallel analysis
of Polish palatalization is impossible due to contradictory rankings.
Section 3.2. turned to Derivational Optimality Theory, which admits phonological
levels, thus allowing for a neat organisation of the rankings developed in Section 3.1. It has
been shown that DOT’s distinction between lexical and postlexical derivational levels is
effective, and correctly predicts the attested surface forms in a transparent manner.
58
Chapter 4
Conclusions
The aim of this thesis was to review the palatalization of dental obstruents from the point of
view of selected rule-based as well as constraint-based frameworks. In Chapter 2, it was
shown that each of the theories, SPE, Lexical Phonology and Feature Geometry, is able to
capture the facts of Polish palatalization correctly. However, these theories show different
degree of aptness in the ways they formalize the processes.
One of the crucial conclusions of this thesis is the fact that dental obstruents exhibit
two different kinds of palatalization. Allophonic Surface Palatalization, which changes
//s z t d// to [s’ z’ t’ d’] and phonemic Coronal Palatalization, which changes //s z t d// to
[ɕ ʑ tɕ dʑ]. Both rules change hard consonants into their soft counterparts. Although the two
rules, to a degree, converge in their outputs and in the contexts in which they apply, they must
remain separate and cannot be amalgamated into one general palatalization rule. Surface
Palatalization is triggered by i and j, whereas Coronal Palatalization applies before ɛ and i.
What is more, Surface Palatalization applies across the board in contrast to Coronal
Palatalization, which is limited to applying to segments spanning morpheme boundaries.
Chapter 2 has also shown that the palatalization rules need to make reference to
morphological and lexical information in order to correctly predict the facts of Polish
palatalization. Thus, the rule of Coronal Palatalization needs to be excluded from applying to
the [t] in but Edwarda [but ɛdvarda] ‘Edward’s shoe’ as it is separated from the ɛ by a word
boundary. Similarly, the [t] in teraz [tɛraz] ‘now’ is free from palatalizing into [tɕ] because
the dental and the vowel do not span a morpheme boundary. Surface Palatalization must
apply across word boundaries as in głos Idy [gwɔs’ idɨ] ‘Ida’s voice’ and to tautomorphemic
59
segments, as in partia [part’ja] ‘political party’. SPE captures this relation by referring to
word boundaries, ‘#’, and morpheme boundaries, ‘+’, which are expressed by abstract
distinctive features, in the composition of the rules. Lexical Phonology provides a neater
solution to the same problem by allocating each rule to different phonological components.
The consequence is that reference to word or morphological boundaries is no longer
necessary.
Chapter 3 recasts the analyses in Optimality Theory, which proves most successful in
capturing the formal unity of palatalization in Polish. The demands of PAL, which enforce the
agreement in backness between hard dental obstruents and front vowels, unite the processes
of palatalization, retraction and fronting. However, OT fails in the way it organises its
generalisations. Namely, the ranking of the constraints producing Surface Palatalization and
the ranking producing Coronal Palatalization are contradictory. The principle of strict
parallelism of OT makes it impossible for the two processes to coexist in one language.
Level distinctions in DOT resolve the problem of contradictory rankings by assigning
each ranking to a different phonological level. It is concluded that OT must admit level
distinctions in order to be able to account for the processes of Coronal Palatalization and
Surface Palatalization.
60
References
Booij, Geert E., and Jerzy Rubach. 1987. Postcyclic versus postlexical rules in Lexical
Phonology. Linguistic Inquiry 18:1–44.
Chomsky, Noam. and Halle, Morris. 1968. The Sound Pattern of English. New York: Harper
and Row.
Clements, George N. 1985. The Geometry of phonological features. Phonology Yearbook 2.
225–252.
Chen, Matthew.1973. Predictive power in phonological description. Lingua 32.173–191.