dissertation - chapter one-vowel-zero alternations_doc

27
1 Chapter One Vowel/Zero Alternations in Slavic 1.0 Introduction. One of the most interesting phenomena in the study of Slavic phonology is a series of vowel/zero alternations, in which a vowel may exist in some forms of a given word’s paradigm but not in the others. The alternations are most commonly found in nouns, short form adjectives, and a few past tense verbs. For the short form adjectives and verbs, the alternation is between the masculine, which contains the vowel, and the other forms, which do not. For the nouns, the alternation occurs when a vowel is present in the masculine nominative singular or the feminine or neuter genitive plural but is absent in the other forms of the paradigm. These alternations are commonly referred to as jers, in reference to a pair of high, lax vowels from which they are historically derived (see Section 1.3). Not all nouns, short form adjectives, or verbs exhibit any vowel/zero alternation. Examples of the alternations, or jers, in Russian are given in (1) below. 1 (1) a. otec masc. nom. sg. ‘father’ otca masc. gen. sg. b. lodka fem. nom. sg. ‘boat’ lodok fem. gen. pl. c. kreslo neut. nom. sg. ‘armchair’ kres’ol neut. gen. pl. 1 Examples are all Russian, unless otherwise stated.

Upload: jeanette-koenig

Post on 14-Aug-2015

34 views

Category:

Documents


1 download

TRANSCRIPT

1

Chapter One Vowel/Zero Alternations in Slavic

1.0 Introduction.

One of the most interesting phenomena in the study of Slavic phonology is a

series of vowel/zero alternations, in which a vowel may exist in some forms of a

given word’s paradigm but not in the others. The alternations are most commonly

found in nouns, short form adjectives, and a few past tense verbs. For the short form

adjectives and verbs, the alternation is between the masculine, which contains the

vowel, and the other forms, which do not. For the nouns, the alternation occurs when

a vowel is present in the masculine nominative singular or the feminine or neuter

genitive plural but is absent in the other forms of the paradigm. These alternations

are commonly referred to as jers, in reference to a pair of high, lax vowels from

which they are historically derived (see Section 1.3). Not all nouns, short form

adjectives, or verbs exhibit any vowel/zero alternation. Examples of the alternations,

or jers, in Russian are given in (1) below.1

(1) a. otec masc. nom. sg. ‘ father’ otca masc. gen. sg. b. lodka fem. nom. sg. ‘boat’ lodok fem. gen. pl. c. kreslo neut. nom. sg. ‘armchair’ kres’ol neut. gen. pl.

1 Examples are all Russian, unless otherwise stated.

2

d. bolen masc. short form adj. ‘sick’ bol’na fem. short form adj. e. žog impfv. past masc. sg. ‘burn’ žgla impfv. past fem. sg. The nature of the vowel in the alternation varies with the language: in

Russian, /e/ and /o/ regularly participate in vowel/zero alternations; in Polish and

Czech, the alternating vowel is /e/; in Slovak, /e/ and /o/ alternate; in Bulgarian, /e/

and / / alternate; and in Serbo-Croatian, the alternating vowel is /a/. This is shown in

(2) below.

(2) Reflexes of vowel/zero alternations in major modern Slavic languages

Russian [e]/[o] Polish [e] Czech [e] d’en’ dn’a den’2 dn’a den dne otec otca ojtec ojca3 otec otce ugol ugla w g’el w gla uhel uhla ogon’ ogn’a ogen’ ogn’a ohen’ ohne zamok zamka zamek zamka zamek zamku

Slovak [e]/[o] Serbo-Croatian [a] Bulgarian [e]/[ ] den dn’a no alt. den dni otec otca otac oca otec otci4 uhol uhla ugal5 ugla g l gli ohen’ ohna oganj ognja og n ogn’ove zamok zamku zamak zamka no alt. 2 Note on transcription: Polish orthography and pronunciation are complex and tend to obscure underlying phonemic relationships. Here, in order to normalize underlying segments to their correspondents in the other Slavic languages, I use the following system: /c/ = voiceless alveolar affricate; / / = voiceless postalveolar affricate (spelled “cz” in Polish); /š/ = voiceless postalveolar fricative (spelled “sz”); /ž/ = voiced postalveolar fricative (spelled “ ” ); /l’ / = soft lateral liquid; /l/ = labiovelar glide (pronounced [w], spelled “ ł” ); / / = voiced postalveolar fricative (pronounced [ž], spelled “ rz”); /t’ / = voiceless prepalatal affricate (spelled “ci” before vowels and “c” elsewhere); /d’ / = voiced prepalatal affricate (written “dzi” before vowels and “dz” elsewhere); / / = front nasal vowel; / / = back nasal vowel. 3 /tc/

� [c]

4 Archaic. 5 In Serbo-Croatian, /l/

� [o] at word-end.

3

In addition to the forms represented in (1), many vowel/zero alternations can

be found elsewhere, though their distribution is less systematic than that described

above, and they are not universally considered to be jers. For example, vowel/zero

alternations exist in some prepositions and verb prefixes, as in (3) and (4) below,

respectively. In (3), the noun also has a vowel/zero alternation; the same is true for

the verbal root in (4).

(3) v rot masc. nom. sg. ‘ into the mouth’ vo rtu masc. prep. sg. ‘ in the mouth’ (4) s- est pfv. infin. ‘count’ so- tu pfv. 1st sg. non-past s- ol pfv. past masc. sg. so- la pfv. past fem. sg. For a given preposition or verb prefix, the vowel or lack of it is often

consistent throughout a single word’s paradigm but variable if a different word or

root is used, shown below in (5) and (6). For prepositions, the vowel is more likely

to appear if the preceding consonant shares features with the initial consonant of the

following noun. For example, in (5-a) below, the /v/ has only the feature [+labial] in

common with the following /m/, and the vowel allomorph does not appear. In (5-b),

/v/ and /f/ are both labiodental fricatives, and the [vo] allomorph is used. A similar

situation occurs in (5-c & d).

(5) a. v Moskv’e ‘ in Moscow’ c. s utra do ve era ‘ from morning till night’

b. vo Florid’e ‘ in Florida’ d. so storony ‘on the part of’

4

Similarity of features plays little role in choice of verb prefix allomorph,

however, as shown in (6) below. Both the [-o-] and the [-∅-] allomorphs appear,

independent of whether the surrounding consonants are alike, and in (6-e), the

[vo-] allomorph appears even though the following element is an identical /o/.

(6) a. v-xodit’ ‘ to enter’ c. vo-dvor’at’ ‘ to settle’ b. v-vodit’ ‘ to lead into’ d. vo-vlekat’ ‘ to draw in’ e. vo-obražat’ ‘ to imagine’ In addition, sometimes a given stem will take the [-o-] allomorph (e.g., [so-],

not [s-]) of one prefix, but the [-∅-] allomorph (e.g., [iz-], not [izo-]) of a different

prefix, even though both variants exist elsewhere (e.g., [s-padat’ ] ‘ to fall down’ , but

[izo-šol] ‘he came from’). For example, with the root /-b#r-/ in (7) below (which

itself exhibits a vowel/zero alternation), the following variants are found:

(7) s-bor (⊗so-bor)6 ‘collection,’ so-bir-at’ ‘ to gather’ (impfv.) iz-birat’ (⊗izo-bir-at’ ) ‘ to select’ (impfv.) so-br-at’ ‘ to gather’ (pfv.) iz-brat’ (⊗izo-br-at’ ) ‘ to select’ (pfv.) Given that a form exhibiting a vowel/zero alternation usually yields the vowel if the

following syllable also has a vowel/zero alternation (see discussion below), it is

surprising to see the form izbrat’ in (7) above, rather than ⊗izobrat’ . Other roots,

however, are consistent in which prefix variant they take, regardless of whether they

themselves have a vowel/zero alternation, as in (8) below:

(8) so-jti ‘ to go down’ izo-jti ‘ to come from’ so-šol ‘he went down’ izo-šol ‘he came from’ so-šla ‘she went down’ izo-šla ‘she came from’

6 The symbol ⊗ is used here to mark incorrect or unattested forms, rather than the usual asterisk (* ), as the latter is used to denote unsyllabified consonants.

5

As shown in (7) above, vowel/zero alternations can also be seen in the roots

of some perfective/derived imperfective infinitive verb pairs. A three-way

alternation frequently exists between the perfective and imperfective infinitives and

the finite forms of the perfective, as in (9) below. This alternation involves the back

vowels /y/, /o/, and /∅/, or the corresponding front vowels /i/, /e/, and /∅/.

(9) a. na-zv-at’ ‘ to call’ (pfv.) b. za-stl-at’ ‘ to cover’ (pfv.) na-zov-u ‘ I will call’ (pfv.) za-st’el’ -u ‘ I will cover (pfv.) na-zyv-at’ ‘ to call’ (impfv.) za-st’ il-at’ ‘ to cover’ (impfv.) c. u-mer-et’ ‘ to die’ (pfv.) u-mr-u ‘ I will die’ (pfv.) u-m’ ir-at’ ‘ to die’ (impfv.) Some of the Slavic languages show other types of segment/zero alternations.

For example, in Polish, the choice between the comparative allomorphs /-š(i)/ and

/-ejš(i)/ usually depends on whether the stem ends in a syllabified consonant (as in

(10) below) or not (as in (11) below). (See 2.1 of Chapter Two and 4.0.2 of Chapter

Four for further discussion.) Here, the alternation is between /ej/ and ∅, rather than

between a single vowel and ∅ (examples from Szpyra (1992: 286)).

(10) tward-y ‘hard’ �

tward.7- ši ‘harder’ (Polish) (11) m dr-y ‘wise’

� m d. -ej- ši8 ‘wiser’

Slavic vowel/zero alternations are partially productive with foreign

borrowings, as in the Polish examples in (12) below, but they are not fully

productive, as in (13) (examples from Szpyra (1992: 279)).

7 Here, the /d/ is ultimately syllabified with the following syllable as part of a complex onset: [twar.dši] (Bethin (1992)). 8 The /r/ is palatalized by the front vowel in the comparative ending.

6

(12) rober nom. sg. masc. ‘ rubber’ (Polish) robr-a gen. sg. masc.

(13) rower nom. sg. masc. ‘bicycle’ rower-u gen. sg. masc.

The vowel/zero alternations illustrated in this section, summarized below in

(14-a-e), are caused by a variety of factors. Some appear to be connected with

syllabification (14-a & b), while others serve to break up consonant clusters sharing

one or more feature (14-c). Vowel/zero alternations in some verbal roots are

morphologically conditioned (14-d). The phonetic environments of the alternations

are identical: vowels precede the /b/ and follow the /r/, yet among the three forms a

different vowel appears between the consonants (or no vowel at all). What differs is

aspect and whether the verb is finite. Finally, vowel/zero alternations in prefixes

oftentimes appear arbitrary, as in (14-e), in which the phonetic environment is the

same (/v/ __ /v/), and both verbs are imperfective infinitives. The only thing that

differs is the root itself.

(14) a. lodk-a fem. nom. sg. ‘boat’ lodok fem. gen. pl. b. twar.d-ši ‘harder’ (Polish) m d. -ejši ‘wiser’ c. v Moskv’e ‘ in Moscow’ vo Florid’e ‘ in Florida’ d. so-br-at’ pfv. ‘ to gather’ so-ber-u pfv. ‘ I will gather’ so-bir-at’ impfv. ‘ to gather’ e. v-vodit’ ‘ to lead into’ vo-vlekat’ ‘ to draw in’

7

It is beyond the scope of this work to consider in detail every vowel/zero

alternation and the causes behind them. To be discussed here are primarily the

vowel/zero alternations most commonly derived from historical jers and triggered by

consistent phonological processes involving syllabification (to be discussed). As a

starting point, the alternation appears dependent on whether or not an overt vowel is

present in the following syllable. These include the alternations found in nouns,

short-form adjectives, and past tense verb forms illustrated in (1-a-e) (also (14-a)

above). Brief mention of the other types of alternations will be made when relevant

to the discussion.

1.1 Problems in representation.

Although these vowel/zero alternations have been widely studied, there is no

consensus on how they should be represented in the grammar and whether a unified

account can be presented for them all. About the only thing that can be agreed upon

is how they should not be represented. For example, it becomes clear upon

examination that the alternations in (1-a-e) cannot be represented as /∅/

underlyingly, with an epenthetic vowel appearing under certain conditions. Although

these vowel/zero alternations frequently break up consonant clusters, they cannot be

accounted for as instances of epenthesis. For one thing, the nature of the vowel is not

predictable in all Slavic languages (see Section 1.2).

For another thing, not all clusters are broken up by an “epenthetic” vowel, as

in the following well-known Russian examples in (15):

(15) a. lask-a/lasok-∅ fem. nom. sg./gen. pl. ‘weasel’ b. lask-a/lask-∅ fem. nom. sg./gen. pl. ‘caress’

8

If the vowel/zero alternation is represented as ∅ underlyingly, with

epenthesis under certain conditions, then the stems in (15-a) and (-b) above would be

the same: /lask-/. In (15-a), the cluster /sk/ is broken up by an /o/ in the genitive

plural form, which has a zero ending. The same cluster in (15-b) is not broken up,

even though the ending is the same. If the alternating vowel is indeed due to

epenthesis, we must wonder why the two seemingly identical stems /lask-/ differ in

its application. The only difference between (15-a) and (-b) is lexical: the form with

the alternating vowels refers to a mammal; the other refers to a gesture. This

indicates the stems in (15-a) and (-b) are not identical after all; the former must

contain something additional to allow the vowel to appear.

The unpredictable manner in which the vowel appears shows that the

vowel/zero alternations cannot be represented as /∅/ underlyingly with epenthesis;

however, examples (16) and (17) below show that the vowel cannot underlie the

alternation, with deletion occurring under certain conditions, either.

(16) a. oz’or-o/oz’or (not ⊗oz’ r-o) neut. nom. sg./gen. pl. ‘ lake’ b. r’obr-o/r’ob’or neut. nom. sg./gen. pl. ‘ rib’ (17) a. jezdok/jezdok-a (not ⊗jezdk-a) masc. nom. sg./gen. sg. ‘ rider’ b. pojezdk-a/pojezdok fem. nom. sg./gen. pl. ‘ trip’ Although the phonological environments in (16-a) and (-b) are alike

(CVRV),9 the vowel deletes in one but not the other. In (17-a) and (-b), the segments

surrounding the vowel are identical (/zd__k/), and again, deletion does not occur in

9 C = Consonant; V = Vowel; R = Liquid; S = Sonorant; N = Nasal consonant; G = Glide.

9

the former, even though it is clear from (17-b) that the cluster /-zdka/ is perfectly

acceptable. Thus, the alternation cannot be a case of deletion under certain

conditions, unless the vowel is specially marked in the lexicon as [+delete]. This

essentially requires that deletable /e/ and /o/ be phonemes separate from non-

deletable /e/ and /o/. However, this putative feature [+delete] is stipulative; it tells us

nothing illuminating about the vowel/zero alternations, and there is no evidence that

this feature exists in other grammars.

1.2 Predictability of jer reflex.

It appears that the distribution of Slavic vowel/zero alternations in the modern

languages, although increasing in productivity, is too idiosyncratic to be regarded as

a case of epenthesis or deletion. In addition, it is a matter of dispute whether or not

the nature of the vowel reflex (that is, its feature composition) is predictable.

Bulgarian and Slovak each have two vowel reflexes, whose distribution is lexically

determined. In Russian, both [e] and [o] alternate with zero, but because these two

phonemes occur in mutually exclusive environments ([e] occurs between soft

consonants; [o] occurs elsewhere), it can be argued that one underlying jer is

involved, and whether it becomes [e] or [o] is determined by default rules.

Polish is also disputable as one-jer or two-jer. The only vowel that alternates

is /e/, but because underlying front vowels cause palatalization of preceding

consonants, and the consonant preceding the jer is idiosyncratically palatalized, some

people have argued that Polish is a two-jer language (one front and one back) (e.g.,

Gussmann (1980); but see Bethin (1992) for a different view). Even in the other so-

10

called one-jer languages, a number of lexical exceptions exist. For example, Russian

has the alternations zajac/zajca (‘ rabbit’ ) , involving /a/, and odin/odna (‘one’ ),

involving /i/. Rubach (1993) shows that virtually every vowel in Slovak participates

in alternations with zero, though [e] is by far the most frequent. For example,

alternating with doska (‘board’ ) are the forms dosiek10∼dosa:k. The preposition

k ‘ to’ alternates with ku.

In addition, even when the phonetic nature of the vowel is predictible, one

should not be cavalier in positing “default” feature fill-in rules. Rules that

automatically fill in unspecified feature values (that is, everything that is not

predictable) would be expected to fill in the unmarked values. To suppose that

default rules would fill in the marked value of a feature is like expecting a ball to roll

uphill on its own. The feature combination [-high] [-low] is more marked than either

[+high] or [+low]. In three-vowel systems it is the mid-vowels that are excluded

from the inventory, not the high or low ones (see also Calabrese (1988)). It is then

somewhat surprising that the default features would result in /e/ as the alternating

vowel in some of the languages.

Because epenthesis and deletion are both unpredictable in application, as well

as the reflex of the vowel itself, there clearly must still be something present in the

underlying form of these alternations, some relic of the jers that vocalizes under

certain conditions and deletes under others. The nature of this phoneme (or

10 [ie] is a reflex of long /e:/ by a diphthongization process, according to Kenstowicz and Rubach (1987).

11

phonemes) is under dispute and will be examined here. To begin, a brief history of

the jers must be presented.

1.3 History of vowel/zero alternations.

The vowel/zero alternations in the modern Slavic languages are relics of a

pair of relatively short-lived high, probably lax, vowels called jers, which themselves

had descended from Early Proto-Slavic (EPSl) short /i/ and /u/. At some point in the

Late Common Slavic (LCS) period, phonemic length was exchanged for other

features, in the case of the jers, possibly [tense] (see discussion below in Section

1.3.2). The so-called “ fall of the jers” marked the end of the late LCS period, but it

did not happen at the same time over the entire Slavic speaking area. It began in the

southwest (i.e., Serbo-Croatian) around the 10th century and reached the northeast

(Russian) around the 12th century.11 The jers merged with other vowels or deleted

entirely, and, although there are exceptions in some dialect areas, the tendency was

for this to be determined by their position relative to each other within a word.

1.3.1 Havlík’s Law. Jers that merged with other vowels are referred to as “strong”

jers; the jers that deleted are called “weak.” The weakness or strength of a jer was

usually determined by its position relative to another jer. Scholars12 typically define

strong jers as those that are immediately followed by a syllable containing a weak jer.

Weak jers are thus found when they are (1) at word-end (because they are not

followed by any jer at all, strong or weak), (2) isolated, that is, in a syllable in which

11 Some accounts (e.g., Timberlake (1983) believe the fall of the jers began in the more central Slavic regions. 12 E.g., Shevelov (1965), Bethin (1988), Townsend (1996).

12

the syllables preceding and following contain full vowels (for the same reason as

(1)), and (3) odd-numbered jers in a series of successive syllables, counting

backwards from the right-most jer. This alternation of weak-strong jers is referred to

as Havlík’s Law.13 A jer at word-end or preceding a syllable with a full vowel is

weak, so a jer in the syllable preceding that would be strong. A jer in the syllable

before the strong one would then be weak. This means that generally there are never

consecutive syllables with both strong or both weak jers.

The actual vowel or vowels that the strong jers merged with became a

characteristic feature in the development of each Slavic language. In Russian, for

example, strong back jers became /o/ and strong front jers became /e/. (This /e/, in

turn, participated in a later change in which /e/ under certain conditions became /o/.)

Strong jers preceding a /j/, which are called tense jers, behaved slightly differently for

most Slavic languages. Tense back jers raised to /y/ (an unrounded high back

vowel), and tense front jers raised to /i/. Russian is an exception; tense jers usually

give the same reflexes as the other jers for this language. This is illustrated in (18)

below.

(18) LCS Rus14 Cz Pol SC Bg mUj- moj-u myj-i myj-em mij-em mij-a

13 Havlík’s Law should not be regarded as absolute—there are many exceptions to it at the time of the fall of the jers, and its application can be influenced by factors such as stress, syllabification, and morphological leveling. This will be examined in further detail in Chapters Four and Five. 14 Abbreviations used here are: Rus=Russian; Cz=Czech; Pol=Polish; SC=Serbo-Croatian Bg=Bulgarian.

13

The alternating pattern of weak/strong jers is the predecessor of the

vowel/zero alternations in the modern languages. If an inflectional ending contained

a jer, any jer in the stem (immediately preceding the ending jer) would be strong and

thus would merge with another vowel. (Henceforth, the term vocalize shall be used

to refer to a jer that merged with a vowel rather than delete.) If the inflectional

ending was a non-jer vowel, however, the preceding jer would be weak and would

delete. This is illustrated in (19) and (20) below. Front jers are transcribed here as

/I/; back jers are transcribed as /U/, in reference to their EPSl predecessors.

(19) (LCS) s U n- U �

(Mod Rus) son nom. sg. masc. ‘dream’ ∧ ∧ | | | weak strong (20) (LCS) s U n- a

� (Mod Rus) sna gen. sg. masc. ‘dream’

∧ | | weak The word-end jer in (19) was a nominative singular masculine ending, which

corresponded to other endings in other cases. Here, according to Havlík’s Law, this

jer, which is the right-most one, is weak and deletes. The jer to the immediate left is

therefore strong and vocalizes, resulting in Modern Russian [son]. In (20), the

genitive singular ending is not a jer; the root jer, which was strong in the nominative

singular, is now isolated and is thus in weak position. Hence, it deleted, resulting in

Modern Russian [sna].

14

The alternating pattern of weak/strong jers leads scholars to posit a bi-syllabic

requirement for jer vocalization; that is, for a jer to vocalize, there must be a second

jer in the immediately following syllable.

When successive jers occurred in more than two syllables, it was possible to

obtain multiple vowel/zero alternations, as in (21) and (22) below.

(21) LCS Old Rus

S m o l U n I s k- U �

Smolnesk masc. nom. sg. ^ ^ ^ (name of a Russian city) | | | weak | weak strong (22) S m o l U n I s k- a

� Smolenska masc. gen. sg.

^ ^ | | strong | weak Modern Russian no longer has these alternations because paradigmatic leveling

caused the Old Russian nominative form Smolnesk to reformulate the stems on the

basis of the oblique forms. This yielded Modern Russian Smolensk/Smolenska, with

no vowel/zero alternations. Multiple vowel/zero alternations are fairly rare in the

modern Slavic languages. In Modern Russian, they exist only in some prepositional

phrases and prefixed verbs, such as (3) and (4) above, respectively. Usually,

however, when a string contains multiple syllables of adjacent jers in derivational

morphology, many of the modern Slavic languages (e.g., Russian, Polish, Slovak)

vocalize all but the right-most jer (e.g., Polish /cukUr-Ik-U/ �

[cukerek];

15

/cukUr-Ik-a/ �

[cukerka], ‘candy’ masculine nominative/genitive singular).15

A notable exception to Havlik’s Law is where jers occur adjacent to liquids

(see Section 4.4 of Chapter Four). Here, we find great variation within and among

the Slavic dialects, although some general tendencies can be found. The South

Slavic regions (Serbo-Croatian, Bulgarian) eliminated the jer regardless of its

strength or position relative to the liquid (preceding or following), as in (23) below.

This resulted in a syllabic liquid, which since has, in many cases, decomposed into a

vowel or vowel/liquid combination. For instance, Bulgarian has inserted a mobile

vowel whose position is syllable dependent: [pr v.]/[p r.va] ‘ first’

(masculine/feminine singular short form).

(23) LCS �

Serbo-Croatian vIlkU vuk16 vIlka vuka krUvI k v17 krUvi k vi The Northeast Slavic regions replaced the jer with a non-alternating vowel,

also without regard to strength or position relative to the liquid, as in (24) below.

15 This statement has been simplified for ease of exposition and is based on the traditional approaches to jer vocalization, with which the reader is more likely to be familiar: a jer is said to vocalize if the following syllable contains a jer. This approach therefore assumes that the so-called zero-endings (masculine nominative singular and feminine/neuter genitive plural) are actually jers. However, Chapter Two will argue that jers vocalize when followed by unsyllabified consonants, and Chapter Five will argue that the zero-endings are indeed zero-endings, not jers. 16 A later change in SC: (syllabic) / / � u 17 Here, the symbols / / and / / are used for syllabic liquids.

16

(24) LCS � Russian vIlkU volk vIlka volka krUvI krov’ krUvi krovi Thus, in both of these extreme regions, Havlik’s Law does not apply for jers adjacent

to liquids. The middle regions, which include the Northwest Slavic areas (Polish,

Polabian) and Czech, show variable applicability depending on position relative to

the liquid. If it preceded the liquid (25-a), it was replaced by a non-alternating vowel

in the Northwest Slavic areas and a syllabic liquid (with no alternating vowel) in

Czech. If the jer followed the liquid, however, Havlik’s Law did apply (25-b) for

both Northwest Slavic and Czech.

(25) LCS � Polish Czech a. vIlkU vilk v k vIlka vilku v ka b. krUvI krev krev krUvi krvi krvi At this point, it is necessary to distinguish historical jers (henceforth termed

H-jers) from modern jers (henceforth termed M-jers), which underlie the vowel/zero

alternations under study here. Not all modern vowel/zero alternations come from H-

jers (e.g., LCS ogn’ I/ogn’a � Modern Russian ogon’ /ogn’a), and not all H-jers

yielded vowel/zero alternations (e.g., (23) and (24) above). That is, H-jers and

M-jers, while overlapping to a significant degree, are not completely congruent.

When the distinction between historical and modern jers is ambiguous or not

important, the term jer will be used by itself.

17

1.3.2 Marked feature combination. What was it about the H-jers that caused

them to stop being realized in surface form? As nearly a thousand years have passed

since the fall of the H-jers from Slavic vowel inventories, it obviously is impossible

to know their feature specifications with any certainty. However, some reasonable

conclusions can be drawn from the modern Slavic languages.

The H-jers, which existed for a relatively brief time in LCS, came from the

EPSl short [+high] vowels /i/ and /u/. At some point in the development of LCS, the

length feature lost its phonemic status and became redundant. Short /i/ and /u/ are

believed to have lowered (precisely how much lowering, and whether or not they

merged into one vowel, depends on the dialect); it may have been this difference in

height that became distinctive. However, Lightner (1972) suggests the new

distinctive feature was [-tense] (or [-ATR]), with loss of height being phonetically

redundant. This proposal allows us to avoid having to postulate four levels of height

in order to distinguish the LCS vowels. The [+high] vowels would have been the

only vowels for which the feature [tense] was contrastive, as illustrated below in

(26). It is not unheard of for a feature to operate distinctively over only part of the

vowel inventory; in LCS, the feature [labial] has contrastive function only among the

[+back] vowels, for instance; that is, /u/, which is [+back] [+high], contrasts with /y/

[+back] [+high] by the feature [round]. Although there is no phonemic distinction

among [-high] vowels for the feature [tense], the unmarked value is [-tense]. (In

Modern Russian, the vowel /e/ becomes [+tense] in certain phonetic environments,

but this occurs only in surface form.)

18

(26) LCS Vowels and Their Posited Feature Specifications18 high low back tense labial nasal i + - - + n/a (-) - y + - + + - - u + - + n/a (+) + - I + - - - n/a (-) - U + - + - - - e - - - n/a (-) n/a (-) - o - - + n/a (-) + - a - + + n/a19 - - 20 - -? - n/a - + - -? + n/a - + ? ? -? n/a ? - The Slavic vowel inventory underwent a collapse from eleven vowels at the

beginning of the LCS period to five or six vowels at the end. If a grammar is

decreasing the number of distinctive vowels in its inventory, it seems natural that the

most marked vowels would be the first to go. Vowels that are both [+high] and

[-tense] are relatively rare among languages and generally exist only in languages

with larger vowel inventories (see Calabrese (1988) for a hierarchy of marked feature

combinations). Thus, it is not surprising that the H-jers, with their unstable feature

combinations, were short lived.

18 Vowel inventory taken from Townsend (1996). 19 Lightner (1972) considers this vowel [+tense], assuming long � tense in all cases. Although the combination [+low] [+tense] is marked, this assumption does not crucially affect the analysis here. 20 The precise feature specifications for the nasal vowels is uncertain. Similarly, the features of / /, whose modern reflexes range from /i/ (Cz), /ie/ (SC), /e/ (Rus, Pol, Cz), /a/ (Bulg, Cz), cannot be established with any reliability.

19

1.4 M-jers in the modern grammar—absolute neutralization.

It is tempting to use the historical features [+high] [-tense] to explain the

vowel/zero alternations in the modern Slavic grammars. For example, one might

suggest (see, e.g., Lightner (1972); Gussmann (1980)) that a [+high] [-tense] vowel is

underlying, but some constraint prevents these features from appearing together in

surface form. This proposal entails a process of absolute neutralization, in which the

distinction between two underlying phonemes is neutralized in all contexts. An

abstract phoneme /z/ that never appears as such in surface form is posited to explain

an alternation between [x] and [y], where neither underlying /x/ or /y/ will adequately

account for the facts.

Under this proposal, vocalization would occur when one of the features

[+high] [-tense] is changed, either in response to an appropriate phonetic

environment or, possibly, to some morphophonemic rule. This would result in an

[α high] [α tense] configuration.21 Deletion would occur if none of the vocalization

rules could apply. See, for example, the Polish and Russian examples below in (27)

and (28), respectively. The conditions under which some M-jers vocalize while

others delete will be discussed in Chapter Two. As a starting point, it will be

assumed, in keeping with traditional analyses (e.g., Lightner (1972); Gussman

(1980); Bethin (1992); Rubach (1993)), that M-jers vocalize when immediately

followed by a syllable containing another M-jer. (Vocalization is not necessarily

21 Thanks to Michael Launer, personal communication, for pointing this out.

20

limited to this environment, however—it can possibly be triggered by morphological

rules as well.)

The [±high] [±tense] features of the alternating vowels in the Polish examples

in (27) below are in complementary distribution ([+high] [+tense] before /j/, [-high]

[-tense] elsewhere).22 It is thus perfectly natural to consider the same underlying

phoneme to be operating in both contexts. Lightner (1972) and Gussmann (1980)

propose the M-jer vocalizes if another M-jer is present in the following syllable.

Whether it vocalizes as [+high] [+tense] (e.g., [i]) or [-high] [-tense] (e.g., [e])

depends on whether it immediately precedes a /j/.

(27) 23 nom. sg. fem. (noun) nom. sg. masc. (adj.) nom. sg. masc. (adj.—s.f.) /policIj-a/ /policIj-In-y/ /vin’ -In-U/ Tense -- y -- -- -- Lower -- -- -- e -- Delete ∅ -- ∅ -- ∅ Surface: [policja] [policyjny] [vin’en] Lightner (1972) and Gussmann (1980) also assume a single [+high] [-tense]

M-jer underlies the three-way alternation seen in Russian perfective and imperfective

infinitives and non-past perfective verbs such as those in (28) below. Here,

vocalization, as well as whether it is Tensing or Lowering that applies, is conditioned

by morphological, rather than phonological, rules. This proposal allows the three-

way alternation to be collapsed into one underlying vowel. 24

22 An exception is in Russian, which has [-high] [-tense] nearly everywhere. 23 Data and derivations from Gussmann (1980: 55). 24 An alternative for morphologically conditioned vowel/zero alternations it to represent them as separate but related roots. Vowel alternations within a root, though not productive, are very common in Slavic for all parts of speech, again largely due

21

(28)25 Pfv. Inf. 1sg prfv. Der. Impfv. Pfv. Inf. 1sg prfv. Der. Impfv. /nazUvat’ / /nazUvu/ /nazUvat’ / /stIlat’ / /stIl’u/ /stIlat’ / Tense -- -- y -- -- i Lower -- o -- -- e -- Delete ∅ -- -- ∅ -- -- Surface: [nazvat’ ] [nazovu] [nazyvat’ ] [stlat’ ] [stel’u] [stilat’ ] A feature combination for the M-jers such as [+high] [-tense] is appealing.

Both features exist in other phonemes (just not with opposite +/- specifications), e.g.,

/o/ is [-high] [-tense], /i/ is [+high] [+tense],26 so the grammar does not have to be

weighed down with extra stipulative features to mark the deletable vowels.

Furthermore, the surface constraint against [+high] [-tense] together is plausible,

since it is a marked combination for such a small vowel inventory. It may well be

that these features do indeed underlie the vowel/zero alternations, at least for some of

the Slavic languages. (This may in part be dependent upon what alternations are

admitted into the inventory of M-jers.) However, absolute neutralization is

controversial; critics claim that it violates the “naturalness condition,” which states

that all underlying phonemes must be such that they exist in some surface form

without the intervention of phonological rules. In addition, a number of scholars

(e.g, Rubach (1986, 1993); Bethin (1992); Szpyra (1992)) have shown that this

feature combination alone is insufficient to account for vowel/zero alternations. This

will be discussed in Section 2.1 of Chapter Two.

to the relics of ablaut. E.g., Russian vzirat’ ‘ to look at’ ; zerkalo ‘mirror’ ; vzor (noun) ‘ look, gaze’ ; obozrenije ‘ review, survey.’ 25 Data from Lightner (1972: 32). Derivations from Lightner (1972: 34-35). 26 Note that [tense] is not necessarily contrastive elsewhere in the grammar. For /o/, [-tense] is a redundant feature associated with [-high].

22

1.5 Organization of thesis.

The purpose of this dissertation is to investigate under the framework of

Lexical Phonology the differences in jer vocalization patterns at the time of their fall

from LCS and in the modern languages. Focus is given primarily to the North and

Central Slavic regions (late LCS usually will be compared with Modern Russian,

although reference to some of the other modern languages will be made as well). It

is generally accepted that M-jers are still present in some form in the modern

languages, and recent work (e.g., Szpyra (1992); Cresti (1994)) ties vocalization to

the presence of unsyllabified consonants (but see Kenstowicz and Rubach (1987);

Rubach (1993); etc. for a different view).

Vocalization of H-jers, on the other hand, because of their alternating pattern,

is frequently assumed to be connected with a bisyllabic trochee (Bethin (1998)). In

such an analysis, the H-jers categorically dissociated from their morae. It is

suggested that an overarching prosodic reshaping of the Slavic languages, which

favored a bisyllabic trochee in terms of syllable prominence, also governed the

relinking of the H-jers to their morae, parsing from the right edge of the word to the

left. This would have occurred on a plane separate from the one on which syllable

prominence was determined. Non-H-jer vowels, which remained linked to their

morae, would reset the prosodic count on the relinking plane. Although syllable

repair may have been the triggering factor in Bethin’s analysis, it is not considered to

play an integral role in the relinking. Thus, the analysis requires one to consider

vocalization in the modern and historical grammars to be governed by separate

23

phenomena, that is, in the historical grammars, it is the trochaic foot that determines

vocalization, whereas in the modern grammars, it is unlicensed consonants that

determine it. The bisyllabic trochee also provides no account for the behavior of

H-jers adjacent to liquids.

I propose instead that vocalization is and was a part of syllabification, and

that the difference between the modern and historical patterns is due to the

component of the grammar where it occurs. I suggest, using an autosegmental model

that assumes feature matrices on one tier are linked to skeletal timing slots on a

separate tier, that at the time of the fall of the H-jers, they categorically dissociated

from their skeletal slots at the post-cyclic (i.e., word) level. This occurred as a result

of new constraints prohibiting the realization of their feature matrices, which were

highly marked, in surface form. The dissociation of the H-jers destroyed their

syllabic nuclei, which resulted in the adjacent consonants to their left becoming

unsyllabified. (Syllables in LCS had been predominately open, so the delinking of a

given H-jer did not result in an unsyllabified consonant to its right.27) If a string

contained multiple H-jers, this meant multiple unsyllabified consonants appeared

simultaneously. H-jer vocalization, which relinked the H-jers to their skeletal slots

and changed one or more of their melodic features (e.g., lowering), repaired the

problem. It occurred in conjunction with resyllabification, which in Slavic proceeds

27 Liquids, which frequently closed H-jer syllables, were an exception to this. Such an H-jer would have an unsyllabified obstruent to its left and an unsyllabified liquid to its right when delinked. See Chapter Four for more discussion. Also, an unsyllabified consonant would exist to the right of an H-jer, of course, if an H-jer was present in the immediately following syllable.

24

from right to left (See Chapter Four). The conditioning environment for vocalization

was an unsyllabified consonant to the right of the H-jer. An isolated H-jer did not

meet this condition, as the unsyllabified consonant would be to its left and would

resyllabify as the coda of the preceding syllable. As H-jers vocalized, they projected

nuclei with which to syllabify the adjacent stray consonants on either side, ending the

dominance of the open syllable. This bled the conditioning environment for

vocalization if an H-jer was present in the immediately preceding syllable, resulting

in the alternating pattern of weak and strong H-jers.

In most of the modern languages, I believe the M-jers are dissociated from

their skeletal slots, but this configuration has been lexicalized; that is, it has been

incorporated into the phonemic inventories of the Slavic languages. Thus,

vocalization occurs at the cyclic level, as new morphemes are added to the string and

are consequently syllabified. Many scholars assume that M-jers vocalize as long as

another one is present in the next syllable. This requires positing a M-jer in the

nominative masculine singular and feminine/neuter genitive plural endings, which

deletes at the end of the derivation. The motivation for this word-end M-jer is solely

to provide a consistent analysis for the vocalization of stem M-jers. Furthermore,

stress patterns in Russian cast serious doubts on their existence. This will be argued

in Chapter Five. In reality, I believe, it is simply that morphemes containing M-jers

bring unsyllabified consonants into the derivation, which, unlike in many other

languages, may not wait until the post-cyclic step to be prosodically licensed (see Ito

(1988, 1989) for a discussion of a different situation in Arabic). It is the need to

25

syllabify these consonants, cycle by cycle, that accounts for the difference in

vocalization patterns between late LCS and the modern languages.

The organization of this dissertation is as follows: In Chapter Two, Section

2.1 presents the evidence Rubach provides to argue that no linear representation of

M-jers is adequate, specifically, the Slovak Rhythmic Law and Polish comparative

allomorphy. Section 2.2 discusses the conditions for vocalization, which in recent

times has been tied to syllabification. Section 2.3 presents evidence that M-jers

prevent consonants on either side of them from being tautosyllabic. I show that M-

jer vocalization (under certain conditions) and deletion (elsewhere) must be separate

rules. Section 2.4 discusses the domains of application of vocalization and deletion

and presents Rubach’s evidence justifying why the M-jer-vocalization rule should be

considered cyclic. Section 2.5 gives a summary of what needs to be accounted for in

the distribution and representation of M-jers: they do not participate in syllabification

if they are not vocalized, they prevent the consonants on either side of them from

being tautosyllabic, and the vowel reflexes are not always predictable.

Chapter Three summarizes the main proposals for representing M-jers

autosegmentally. Section 3.0 gives a brief discussion of autosegmental theory.

Section 3.1 discusses the possible configurations and the advantages and

disadvantages of each. I conclude that M-jers are represented on both the skeletal

and melodic tiers, and I use data from Sloan’s (1991) dissertation to show that

material can exist on both tiers without being linked. I then demonstrate that the

vocalization of M-jers in Slavic can be characterized by a rule which links material

26

on the skeletal and melodic tiers when these tiers are followed by an unsyllabified

consonant.

Chapter Four discusses the nature of syllabification, a process central to my

analysis of jers. Section 4.0.1 argues that syllabification is cyclic, and Section 4.0.2

shows why it is necessary to assume that morpheme-final consonants are ignored by

the syllable parse until the next cycle, as well as the problems this assumption may

create.

Section 4.1 of Chapter Four is my extension of my analysis so far to the

historical grammars. Section 4.1.1 discusses delinking of H-jers at the surface

structure. Section 4.1.2 compares the differences in distribution patterns between

H-jers and M-jers, claiming these differences are due to where in the grammar

vocalization takes place. Section 4.1.3 shows that H-jer vocalization during the “ fall

of the jers” indicates syllabification is directional. Section 4.2 illustrates the

interaction of syllabification and YVoc over a string. Finally, in Section 4.3, I

discuss the problems of H-jers adjacent to liquids and argue that the varied

distribution of vocalization across the dialects (e.g., vocalization in all positions,

deletion in all positions yielding syllabic liquids, normal treatment of H-jers) is a

function of each dialect’s syllable type—how readily the dialect accepts syllabic

liquids and whether extrametricality of consonants is restricted.

Chapter Five discusses the implications of a M-jer’s structure on its behavior

in a string and in surface form. In Section 5.1, I suggest that as the H-jer’s structural

configuration at surface form is lexicalized and becomes phonemic, it should become

27

increasingly productive, incorporating other vowel/zero alternations not due

originally to H-jers. M-jers have also spread to other strings containing *C. As

M-jers become increasingly productive, the emergence of vowel/zero surface

alternations becomes more predictable. The more predictable the alternation, the

more likely it is to be reanalyzed as epenthesis or deletion in the grammar. The

spread of M-jers is illustrated with Russian “secondary jers” and the replacement of

the LCS imperative ending /-i/ with a M-jer. In Section 5.2, the effect of stress on H-

jer vocalization is considered. In Section 5.3, three cases of M-jer ambigity are

considered: verb imperative endings, noun case endings, and Russian secondary jers

in obstruent-sonorant stems.