. 3
( 17)


recommended pronunciation
context example (Avanesov 1972) Usage (Krysin 1974)

ZD˛ cl†kfnm ˜do™ [z˛ d˛ ] 38%
[z˛ b˛ ] ∼ [zb˛ ]
ZB˛ bp,«nm ˜beat™ 32%
[v˛ b˛ ] ∼ [vb˛ ]
VB˛ d,t;’nm ˜run in™ 16%
VD˛ dl†kfnm ˜set™ [vd˛ ] ---

than before front. Because back vowels have a lower F2 , their F2 is affected more
by palatalized consonants than is the F2 of front vowels, whose high F2 has less
room to change in the vicinity of palatalized consonants.

2.3.4 Palatalization assimilation
In sequences of two consonants in which the second is palatalized, the ¬rst may
or may not be palatalized by assimilation. This is just a question of the timing of
the articulatory gesture of palatalization. If the raising of the blade of the tongue
occurs anticipatorily as the ¬rst consonant is formed, assimilation has taken
place; if raising occurs within the sequence of consonants, then assimilation has
not occurred. Whether palatalization extends over both consonants or begins in
the middle of the cluster depends on the extent to which the two consonants
are articulatorily linked in other respects. The more linked the two consonants,
the more likely it is that palatalization will extend throughout the cluster. There
is variation, and the trend is very much towards losing assimilation.49
One way to approach the variation is to examine the recommendations of
Avanesov (1972) for one morphological context in which most combinations
occur, speci¬cally the context of pre¬x and following root. To see the effect of
place of articulation, we may examine combinations of fricative plus stop in
Avanesov™s recommendations and compare them with Krysin™s (1974) survey of
usage, in which younger speakers (the last two decades, born between 1930--39
and 1940--49) represent half of the speakers interviewed.
Avanesov does not explicitly mention the combination of labial followed by
dental, nor does Krysin (1974) consider it, an indication that assimilation is out
of the question in this context. From Table 2.6 we derive a hierarchy of likelihood
of assimilation: TT ≥ TP ≥ PP PT.50 Comparing the ¬rst two terms to the last
¸ ¸ ¸≥ ¸

49 See Drage 1967[a], 1967[b], 1968, on factors. Contemporary speakers have rather less -- if any --
assimilation than was reported by Drage and Krysin (in the mid-1960s).
Krysin (1974:82) states the hierarchy as TT ≥ PP ≥ TP (and then presumably ≥ PT), based on the over-
50 ¸ ¸ ¸ ¸
all incidence of palatalization in all types of morphological contexts. The hierarchy artefactually
re¬‚ects the kinds of examples tested. Many of the examples of dental plus labial involve pre¬xes
62 A Reference Grammar of Russian

Table 2.7 Palatalization assimilation and manner of articulation

recommended pronunciation
context example (Avanesov 1972) Usage (Krysin 1974)

[zv˛ ] ∼ ± [zv˛ ]
ZV˛ bpdby«nt ˜excuse!™ 35%
[z˛ b˛ ] ∼ ±[zb˛ ]
ZB˛ bp,«nm ˜beat™ 32%
[db˛ ] ∼ ?[d˛ b˛ ]
DB˛ jn,«nm ˜repel™ ---
[dv˛ ] ∼ ?[d˛ v˛ ]
DV˛ gjld=k ˜subsumed™ 04%

two, we note that dentals, as targets, undergo assimilation better than labials.
Comparing the ¬rst two terms (TT ≥ TP leads to the result that the same place
¸ ¸)
of articulation in the source and target consonants favors assimilation, because
there is no shift in the place of articulation internal to the cluster.
Before velars assimilation is restricted. Labials no longer assimilate; thus in
k’grb ˜paws™, the pronunciation [pk˛ ] that occurred at the end of the nineteenth
century gave way long ago to [pk˛ ]. Assimilation of dentals to velars is out of
the question: nf,k†nrb ˜tablets™ [tk˛ ], uk’lrbvb ˜smooth™ [tk˛ ].51 Velars before velars
once assimilated (vz ´urbq [x˛ k˛] ˜soft™), but the tendency is again towards hardness
([kk˛ ]).
Table 2.7 shows the effect of manner of articulation.
Avanesov™s discussion of these combinations of dentals and labials implies
a two-way grouping of ZV ≈ ZB ≥ DB ≈ DV .52 His discussion of combinations
¸ ¸ ¸ ¸
involving labials implies VV ≥ VB ≥ BB ≈ BV , and his discussion of combinations
¸ ¸ ¸ ¸
´ryenm ˜dry up™ [s˛ s˛]) ≥ ST
involving only dentals implies a hierarchy of SS (bccz ¸ ¸
(hfpl†k ˜division™ [z˛ d˛] ∼ [zd˛ ]) ≥ TT (gjllth;’nm ˜support™ [d˛ d˛] ∼ ±[dd˛ ]) ≥ TS (jnc†xm
¸ ¸
˜hack off™ [ts˛ ]). Combining the various kinds of information leads to the hierarchy
(using the symbols for dentals as general symbols): SS ≥ ST ≥ TT ≥ TS. That
¸ ¸ ¸ ¸
hierarchy encodes two principles: fricatives are more likely to assimilate than
stops (the ¬rst two terms of the hierarchy as opposed to the last two), and
consonants that have the same manner of articulation assimilate better than
those that have heterogeneous manner (the ¬rst and third terms as opposed to
the second and fourth). Thus identity of manner, when there is a single elongated
articulation without an internal change in manner, favors assimilation.

or even prepositions (,tp d«krb ˜without a fork™, c g«djv ˜with beer™), in which no more than
10 percent of speakers use palatalized dentals. These examples depress the extent of palataliza-
tion with dental targets. Among morphologically comparable examples, the 16 percent of d,t;’nm
(the only example of labial fricative before labial at a pre¬x boundary) compares unfavorably with
bpdby«nt (35%), bp,«nm (32%), or even djpd=k (22%).
51 Matusevich 1976:203.
52 Trubetzkoy (1975:184) noted in 1930 that there was no palatalization across pre¬x boundaries in
jnd=hnsdfnm, though there would be assimilation internally in ,hbndtyysq.
Sounds 63

In combinations of dentals, dental stops do not assimilate to a following lateral
[l˛] (assimilation to [r˛] is out of the question), because there is a shift to a different
mode of articulation (lateral) within the cluster. Dental obstruents assimilate
better to dental nasals [n˛ ], presumably because the oral component of a dental
nasal is effectively just [d˛ ].
Additional factors have emerged in other investigations. Clusters in which voic-
ing is maintained throughout seem to assimilate better (pd†hm ˜beast™ 30%, ld†hm
˜door™ 30% in Krysin™s survey) than clusters in which voicing switches and intro-
duces an internal articulatory boundary (nd†hm 17%) or than in voiceless clusters
(cg«yrf ˜back™ 15%). Intervocalic position favors assimilation over absolute initial
position (ktcy«r ˜forester™ 49%, dj cy† ˜in sleep™ 54%, but cy†u ˜snow™ 28%).
The position before [j] is a special case. Dentals within words assimilate well to
[j]. Assimilation to [j] of a dental in a pre¬x is possible but not obligatory (c(†k ˜ate
up™ [s˛ j], bp(znm ˜extract™ [z˛ j] ∼ [zj], gjl(=v ˜ascent™ [dj]) and infrequent in a prepo-
sition (bp z ´vs ˜from the pit™: [¬zjamï], outmoded [¬z˛ jamï], only jn =krb ˜from
5„ 5„
the ¬r tree™ [ tjolk˛ ¬]). With labials before [j] within words, assimilation still pre-
dominates (over 50% of speakers with gj,m=v ˜we™ll beat™ and djhj,mz ˜sparrow™),´
but assimilation is unlikely in pre¬xes (j,(†[fnm ˜drive around™ [ bj†x´t˛]). ö

2.3.5 The glide [j]
The glide [j] has realizations ranging from strong to weak to weakest.54 It is pro-
nounced as a relatively strong, more consonantal [j] before a stressed vowel: z ´vf
˜pit™ [jam´], z
5„ ´rjhm ˜anchor™ [jak´r˛]. In other positions it is a weaker, less conso-
nantal [i]: zpßr ˜language™ [i¬z˝!k] (initially before unstressed vowel), l†kf/n ˜they
8 8
do™ [d˛ el´i√t] (medially before unstressed vowel), [jpz
5„ 8 ´qrf ˜mistress of the house™
[xøz˛ a„ik´] (after vowel before consonant), cn’hjq ˜old™ [st’r´i] (after a vowel, not
55 8 8
before a consonant).
There is a third, even weaker, pronunciation, and that is nothing. The glide
[j] ∼ [i] is, after all, just an extended [i]-like transition to or away from a vowel.
It remains a segment only if it is distinct for a signi¬cant interval of time. The
glide [j] merges into the adjacent vowel. It is normally lost in verbs of the e-
Conjugation: py’tim ˜you know™ [zna¬s], l†kftim ˜you do™.55 It is often inaudible
5„ ‹
in declensional endings: c edf;†ybtv ˜with respect™ [¬i´] ∼ [¬´]; cn’hjt ˜old™ [´i´] ∼
8 8
´pyjt ˜threatening™ [´i´] ∼ [´´]; jh©;bt ˜weapon™ [ïi´] ∼ [ï´].
[´´]; uhj 8 8
The glide is also absorbed after a vowel before a following stressed [í].56 Forms
like vjz ˜my™ [møja]„ , cnj· [støju]„ ˜I stand™ imply stems {moj-}, {stoj-} including
5 5

53 In reference to hard [vo ]: “the pronunciation [. . .] [dj™uf] cannot be considered correct” (p. 127), a
statement which applies to a third of the population, including those with higher education.
54 55 Avanesov (1971:367) restores the [i] only in careful speech.
Isaˇenko 1947:145--48, 1959.
56 SRIa 1.109.
64 A Reference Grammar of Russian

[j], but that [j] is not pronounced before [í]: vj« [m í], cnj«im [st ís]. However,

˛ ‹jí], djhj,m«
[j] is maintained after a consonant before stressed [í]: xm« ˜whose™ [c
˜sparrows™ [b˛ jí].
In words that begin with {i}, there is no [j] left at all. As a result, when initial
{i} is put after a pre¬x or independent word ending in a consonant, the vowel
that is pronounced is [í] (unstressed [ï]): d b[ lj
- ´vt ˜in their house™ [vï dom¬], 5„ ˛
l’k bv ˜he gave to them™ [d’lïm], d Bylb/ ˜to India™ [v˝!nd˛ ¬i ]. Interestingly, [j] is
maintained before [¬] that derives from a non-high vowel -- Zhjck’dkm [¬¬r sl’vl˛],8
— [vïvj 57
to= ˜still™ [¬¬s˛ ‹ o]„ , d tuj [vi¬vj not
85 ´ 8 ´], ´].

2.3.6 Affricates
The affricates [c] and [c˛ ‹] begin, like stops, with a sudden initial closure, which is
followed by a static interval of closure, but the closure is released more gradually
than with an ordinary stop, in a fashion similar to the release of a fricative.
To indicate their mixed character as part stop, part fricative, it is sometimes
convenient to write the affricates as combinations of two symbols: [c] as [ts],
[c˛ ‹] as [t˛s]‹˛ .58 Affricates are not, however, simply clusters. They are not appreciably
longer than fricatives [s s]. The affricate [c] does not palatalize before {e} (d rjyw†)

as might be expected if it were composed of [t] plus [s], inasmuch as [s] does
(j k†ct). The affricate [c˛‹] does not condition a vowel in unstressed imperatives
like true clusters: gk’xm ˜cry!™, yt v©xm ˜don™t torment!™.
While affricates in Russian are units, clusters of consonants result in phonetic
sequences like affricates.59 Word-internally, a dental stop [t] that is followed by
[c] or [s] ([s˛ ]) will become a single consonantal complex consisting of a stop
onset, a long static interval of closure (written here as “tt”), and a fricative-like
release: gen sg ,h’nwf ˜chap™ [bratts´], cnhtv«nmcz ˜strive™ [tts´], identically 3sg prs
° °
cnhtv«ncz [tts´]. Similarly, a dental stop [t] plus [c˛ ‹] becomes an affricate with an
5 tts˛ ‹
elongated closure: dj ´nxbyf ˜patrimony™ [voˇ¬n´]. If such a combination is placed
before an obstruent, the long closure will be shortened, becoming equivalent to
the affricate [ts] = [c]: Gtnhjpfdj ´tsk] = [vj
´lcr [vj ´ck].
When combinations of stops and fricatives arise at pre¬xes, they maintain
the duration of the fricative of the following root while the preceding hard stop
develops the release of an affricate: jncbl†nm ˜sit out™ [cs] = [tss˛], yflpbh’ntkm
ˇ ˇ
˜overseer™ [Zz˛ ] = [dzz˛ ], jni«nm ˜rebuff™ [cs]‹ = [tss]‹ , jn;«nm ˜become obsolete™ [Zz] =
ˇ ˇ ˇ ˇ ˇ‹
This [Zz] or [dzz] -- a dental stop onset of normal duration followed by the release
ˇ‹ ˇ‹
of an affricate to a full hard alveo-palatal fricative -- is the recommended pro-
nunciation for orthographic ¤l;≥ in borrowings: l;fp [dzzas], l;tv [dzzem].60
ˇ‹ „ ˇ‹ „
57 Trubetzkoy (1975:237).
In other systems of notation, one could write [c] = [ts ], [c˛ ‹] = [t˛s˛ ‹ ] or [ˇ˛ ].
58 59
t ˛S SRIa 1.106--7.
ˇ ˇ
60 Avanesov 1972:166, Jones and Ward 1969:102.
Sounds 65

There is, then, a range of complex articulations of stop and fricative, which
can be ranked in order of increasing duration: true affricates [c c˛‹] = [ts t˛s],˛‹
which are usually lexical (also derived from clusters of [t] and [s] before con-
sonants: ,h’ncrbq ˜brotherly™); affricates with long closures derived from stops
ˇ ˇ˛‹
followed by fricatives or affricates, [tts tt˛s]; and complexes with full fricative
releases, [tss˛ tss˛‹ dzz].61
ˇ ˇ ˇ‹ ‹

2.3.7 Soft palatal fricatives
The sound represented by the letter ¤o≥ derives etymologically from Common
Slavic palatalizations (— sk before front vowel and — stj); it is also the Russian in-
terpretation of the Church Slavonic re¬‚ex of — tj. Earlier it was pronounced with
an internal closure: [sc˛‹] or, equivalently, [sts]‹˛ . Throughout most of the Russian
˛‹ ˛‹ ˛
dialect area, this older pronunciation has lost out to a Muscovite pronunciation
in which the internal stop closure has weakened, resulting in a more or less
homogeneous long, soft alveo-palatal fricative [s˛‹ ]: ,j ´ho ˜borsch™ [bj ˛‹ ]. In the
sociolinguistic survey of the 1960s, [s˛ ‹ ] was used by close to 80 percent of speak-
´ho and o«, the most favorable lexical items.62 And
ers born in 1940--49 in ,j
although [sc˛‹] is often said to be a Petersburg variant, another survey from the
same period had ninety percent of (then) young Leningrad natives born after
the war using the national variant [s˛ ‹ ].63
In addition to lexical instances of [s˛‹ ] (,j
´ho, etc.), this sound also arises produc-
tively in combinations of dental fricatives [s z] with [c˛‹].64 Dental fricatives [s z]
often assimilate in place of articulation to palatals across pre¬x and preposition
boundaries: ci«nm ˜suture™ [ss], c ;tyj ˜with the wife™ [zz], ,tp ;«hf ˜without
‹‹ ‹‹
fat™ [zz]. These fricatives also assimilate to a following [c˛ ‹] in place of articula-
tion and, since [c˛‹] is palatalized, for that feature as well: bcx«ckbnm ˜calculate™,
c x†cnm/ ˜with honor™ [sc˛‹]. As a further stage, the stop closure in the middle
of the complex can be lost: [sc˛‹] = [st˛s]‹˛ > [s¦]. Which variant occurs, whether [sc˛‹]
˛‹ ˛‹ ˛‹ ˛‹
or [s˛ ‹ ], depends on how cohesive the two units are: the weaker the morpholog-
ical boundary, and the more lexicalized the combination, the more likely the
further stage of [s˛ ‹ ] is. By now [s˛ ‹ ] is usual in suf¬xal derivatives (hfccr’pxbr
˜raconteur™) and in idiomatized pre¬x--root combinations (cx’cnmt ˜happiness™,
bcx†pyenm ˜disappear™); it is possible with free pre¬x-root combinations of the
type bcx«ckbnm, hfcxbo’nm ˜clean™, bcxthn«nm ˜sketch out™, ,tcx†cnysq ˜dishonor-
able™. In the 1960s, on the order of 10 to 20 percent of all speakers surveyed used
[s˛‹ ],65 and it is not uncommon now for speakers under forty. Loss of closure is
rare with preposition and noun, though it occurs in idiomatic combinations:

61 Trubetzkoy (1975:182), however, allows that these distinctions are blurred in allegro style.
62 Krysin 1974:100.
63 Ivanovna-Lukianova 1971. Similar observations in Baranova 1971, Drage 1968:377--79.
64 65 Krysin 1974:102--3.
And in principle palatal fricatives [s ‹ z], as in gtht,†;xbr.

66 A Reference Grammar of Russian

c xtkjd†rjv ˜with a person™ [sc˛‹], c x†cnm/ ˜with honor™ [sc˛‹] ∼ ±[s¦], and idiomatic
˛‹ ˛‹ ˛‹
c xtuj ˜why, from what™ [sc˛‹] ∼ [s¦]. ˛‹ ˛‹
It is dif¬cult to assign an unambiguous phonemic analysis to [s˛‹ ] if one expects
to de¬ne a set of features that distinguish it invariantly from all other sounds.66
What necessary property would distinguish [s˛‹ ] from [s]? If [s˛ ‹ ] were viewed as

the soft counterpart of [s], one might expect [s]‹ to become [s˛‹ ] before the {-e} of

the (dative-)locative. It does not: j rfhfylfi† [s†], not — [s˛ ‹ e]„ or — [se]. Further, [s˛ ‹ ]
˛5‹ „

is often phonetically long, and it conditions a vowel in the imperative (hßcrfnm
˜roam™, imv hßob; vj ´hobnm ˜pucker™, imv vj ´hob), as is characteristic of clusters.
De¬ning [s˛ ‹ ] as the soft counterpart of [s]‹ would not motivate its characteristic
length. But length cannot be its necessary property, because the length some-
times disappears. As a third possibility, it might be tempting to think that [s˛ ‹ ]
in general derives from a cluster -- from [sc˛‹] or [sc˛‹] or, with an abstract fricative,

from [Sc˛ ‹], inasmuch as [s˛‹ ] arises productively from clusters of dental or palatal
fricative and [c˛ ‹] (hfccr’pxbr). This analysis violates invariance in another way.
It is usually assumed that [c˛‹] differs from [s]‹ by not being continuous. If all [s˛‹ ]
derive from [c˛ ‹], then [c˛ ‹] has an allophone [s]‹˛ which is continuous, in violation of
this invariant property. There seems to be no analysis which would not violate
one or another axiom of structuralist phonemics and, accordingly, no option
other than simply restating the facts: [s˛‹ ] is a soft alveo-palatal fricative; it is
historically a long consonant, though it sometimes shortens; it does not form a
canonical pair with [s]; and it can arise from combinations of fricatives with [c˛ ‹].

Super¬cially parallel to [s˛‹ ], there is also a voiced [z˛‹ ], which, however, differs in
certain respects.67 With [s˛‹ ], softness is maintained in all contexts, regardless of
whether length is maintained. In contrast, the soft pronunciation of [z˛ ‹ ] is yield-
ing to a hard pronunciation [z ‹ ], on a lexeme-by-lexeme basis. In the 1960s, [z˛ ‹ ]
was used by over half of the speakers of all ages in lhj ´;;b ˜yeast™, the word with
the greatest incidence of [z˛ ‹ ], after which came ,hßp;tn ˜gush™, dbp;’nm ˜squeal™
(a third), then †p;e ˜I drive™, gj ´p;t ˜later™ (a quarter), and ¬nally vj;;td†kmybr
˜juniper™ (15%).68 Nowadays [z˛ ‹ ] is quite limited among speakers under forty. In-
cluded in the set of relevant words should also be lj ´;lm ˜rain™, gen sg lj;lz ´
(likewise, dj ´;lm ˜leader™, dj;lz which allows either this pronunciation (that is,
[z˛ ‹ ] or, with devoicing, [s˛‹ ]) or one with a palatal fricative and dental stop (that
is, [z˛ ‹d˛] or devoiced [st]). The pronunciation with a stop has become usual; only a
˛‹ ˛
¬fth of speakers surveyed still used [z˛ ‹ ] in the 1960s.69
In the most explicit register, [s˛‹ ] is generally pronounced with length, but it
is often shortened to [s]. Table 2.8 lists most environments.
The table suggests the following observations. Intervocalic position (<a>) pre-
serves length. (A sonorant intervening between a vowel and post-vocalic [s˛ ‹ ] does
66 67 Zinder 1989. 68 Krysin 1974:85.
See Avanesov 1948, Panov 1967, Flier 1980.
69 ˛¸
Avanesov transcribes it with a hard fricative. Jones and Ward (1969:142) imply [zd ].

Sounds 67

Table 2.8 Degemination of [s˛‹ ]

context pronunciation (possible) syllable structure

<a> [ss]˛ ‹
˛‹ ˛‹ ˛‹
/V V Vs.sV
<b> [ss]˛ ‹
˛‹ ˛‹ ˛‹
/# V .ssV
<c1> [ss]˛ ‹ ∼ ± [s]˛ ‹
˛‹ Vss.˛ ‹
/V #
<c2> ±[ss]˛ ‹ ∼ [s]˛ ‹
* ˛‹ Vss.˛ ‹
/V #
<d1> ±[ss]˛ ‹ ∼ [s]˛ ‹
˛‹ ˛‹ ˛‹
/V CV Vs.sCV
<d2> * [s]˛ ‹ ˛‹ ˛‹
/V CV Vs.sCV
<e> / [s]˛ ‹ ˛‹ ˛‹
/VC V VC.ssV

± = less acceptable variant

not shorten it: yjc«kmobr ˜porter™ [l˛s˛‹ ], j,v’yobr ˜deceiver™ [n˛ s˛‹ ].70 ) Length is
also preserved in absolute initial, prevocalic position (<b>). Before a following
consonant ([n] or [n˛ ]), length is often lost: <d2>, ,tcgj ˛‹
´vjoysq ˜helpless™ [sn˛ ],
´dboysq ˜monstrous™ [sn˛], though its length may be preserved after a stressed
vowel: <d1> bpz ´oysq ˜elegant™ [s˛ ‹ n] ∼ ±[sn˛ ]. A preceding obstruent shortens [s˛ ‹ ]:
<e> ufhlthj ´,obr ˜cloakroom attendant™ [ps]‹˛ , ,hfrj ´dobr ˜sorter™ [fs]‹˛ . In absolute
¬nal, post-vocalic position, length is also vulnerable: <c1> njd’hbo ˜comrade™
[s˛ ‹ ] ∼ ±[s]‹˛ , gj
´vjom ˜help™ [s˛‹ ] ∼ ±[s]‹˛ , though less so after a stressed vowel: <c2>
´om ˜might™ [s˛ ‹ ].71
These regularities might at ¬rst glance seem to follow from syllable structure.
Context <a> suggests length is preserved when [s˛ ‹ ] is ambisyllabic between vow-
els, and context <b> suggests length is preserved in syllable-onset position,
while <c1, c2> suggest that the coda position is less than ideal. Up to this
point the distribution is consistent with principles of syllable structure. But if
<d1, d2> is ambisyllabic [Vs.‹˛ snV], [snV] should be an acceptable onset, yet length
˛‹ ˛‹
tends to be lost. In <e>, length is lost despite the fact that the cluster could
be in syllable-onset position, as [VC.ssV]. In fact, in <d> and <e> it does not
˛‹ ˛‹
matter which syllable [s˛ ‹ ] belongs to. The constraint is whether [s˛‹ ] is adjacent
to another consonant. The adjacency of another consonant -- regardless of its
syllable allegiance -- is suf¬cient to shorten [s˛‹ ]. The regularities of Table 2.8 do
not follow from syllable structure.

2.3.8 Geminate consonants
Clusters of identical consonants, as they are written in the orthography, are
likely to be simpli¬ed in speech, depending on various factors: tempo, register,
familiarity of the word, the ability of the geminate cluster to occur in native
Russian words. When geminates arise at the boundary between pre¬x and root,
70 According to Avanesov 1972. Panov 1967 transcribes such words with [s˛]‹ .
71 Avanesov 1972 has length here, Jones and Ward (1969:139) shortness.
68 A Reference Grammar of Russian

they are normally maintained. At internal boundaries, only [nn] and [ss] occur.
Geminate spellings involving most consonants occur in foreign borrowings, and
then the gemination may or may not be maintained in pronunciation.72
Certain segments tolerate gemination more readily than others. Geminate [rr]
is quite unlikely. Geminate stops occur occasionally (uh©ggf ˜group™ 55%, [jrr†q
˜hockey™ 34%). Geminate [ss], [nn], [mm], and [ll] are the most likely. With respect
to position in the word, geminates are best maintained between vowels, and they
are maintained better following a stressed vowel than following an unstressed
vowel. Compare: v’ccf ˜mass™ (88%), r’ccf ˜cashier™ (85%), rjv«ccbz ˜commission™
(48%) as opposed to rkfcc«xtcrbq ˜classic™ (44%), ,fcc†qy ˜pool™ (44%). In fact,
no less of a native informant (and phonologist) than Trubetzkoy wrote in a
letter to Jakobson that geminate [n ] was possible only after stressed vowels; the
sequence would be shortened after unstressed vowels.73 In word-¬nal position
after a vowel, geminates are often simpli¬ed but can be maintained (nom sg
r’ccf ˜cashier™ 85% [s ] vs. gen pl r’cc 52% [s ]). Before a following consonant (in
the context VCi Ci Cj V), geminates are lost, as is sometimes re¬‚ected in spelling:
jgth†nnf ˜operetta™ but jgth†nrf, h©ccrbq [sk], dim uh©ggrf [pk], dim ghjuh’vvrf
[mk].74 Just as with [s˛ ‹ ], if a potential geminate is adjacent to another consonant,
it loses length.
In general, maintaining gemination (including [s˛ ‹ ]) requires an environment
which grants a large measure of duration to the interval of obstruents. Intervo-
calic position, when the geminate is supported on both sides by vocalic intervals,
is more favorable to maintaining gemination than absolute initial (prevocalic)
position. Both are more favorable to geminates than absolute ¬nal (post-vocalic)
position. Position adjacent to an obstruent is inferior. Position after a stressed
vowel favors maintaining the geminate.

2.3.9 Voicing of consonants
In Russian some obstruents are voiced, some voiceless. Voiced obstruents such
as [d], [g], [z˛ ] are produced with the vocal cords taut and therefore vibrating
through much of the duration of the obstruent -- in Russian, voiced stops are
voiced through more of their duration than voiced stops in English.75 Voiceless
consonants such as [t], [k], [s˛ ] are produced with spread vocal cords that do
not vibrate. Almost all obstruents come in pairs that differ only by voicing.
Both kinds of obstruents can occur in the context before vowels and distinguish
words: ,«nm [b˛ ] ˜be™ vs. g«nm [p] ˜drink™, l†kj [d˛ ] ˜matter™ vs. n†kj [t˛] ˜body™.

72 Avanesov 1972:128--38, statistics from Glovinskaia 1976. Kuz mina 1976 treats the related problem
of simpli¬cation in consonant clusters.
73 74 SRIa 1.107--8.
Trubetzkoy 1975:237.
75 Initial [d] is voiced in English only 60 percent of its duration, Russian [d] 90 percent (Heffner
Sounds 69

The obstruents [c c ‹ x] are normally voiceless, though they do become voiced
preceding a voiced obstruent within a minimal domain: jn†w ,ßk ˜father was™
> [Zb], nr’x ,ßk ˜the weaver was™ > [Z˛ ‹ b], ktx,’ ˜healing™ [Z˛‹ b], lde[ly†dysq
˜two-day™ [ d˛ n˛]. The voiced phones [Z Z˛‹ ] do not normally occur before vow-
els by themselves.76 (Historically, [ ], a Slavonic pronunciation, could occur au-
tonomously, in certain lexical items such as ,kfuj- ˜well-™, <j ˜God™, uj ´u ´cgjlm
Sonorants are intrinsically voiced. They do not come in pairs that differ by
voicing, although sonorants become voiceless in speci¬c contexts. Vowels some-
times become voiceless when they are unstressed between voiceless consonants.
In addition to the paradigmatic constraints on voicing, there are syntagmatic
constraints, or “rules,” governing how voicing is distributed in connected speech.
Because voicing involves vibration of the vocal cords, absence of voicing is gen-
erally taken to be articulatorily less complex and less marked than voicing. But
speech might be viewed as basically voiced, and absence of voicing -- voiceless-
ness -- as an interruption of the ¬‚ow of intrinsically vocalic, and voiced, sound.77
An interlude of voicelessness must be initiated by an active gesture of opening
the glottis (“ ”) and terminated by a gesture of closing the glottis (“ ”). These
gestures coincide more or less with the oral gestures that de¬ne the boundaries
of segments; for example, in making [p] in jgz the glottis spreads (and voicing
stops) as the lips close and the glottis is closed (and voicing resumes) as the lips
Vowels tolerate a boundary of voicelessness on either side. In n=nz ˜aunt™,
the stressed [o„] tolerates the onset of voicelessness on its right margin and the
end of voicelessness on its left: [ t˛ o „ t˛ ´]. Sonorants (= “R”), intrinsically voiced,
extend the vocalic domain, in that they allow a boundary of voicelessness on
either side. For example, [r] allows the onset of voicelessness on its right margin
in v’hrf ˜stamp™ [már k ´]; and before a vowel, [v˛ ] and [l] tolerate the end of
voicelessness immediately before them in cd=rkf ˜beet™ [ s v˛ o „ k l´]. Sonorants
followed by vowels allow preceding obstruents to distinguish voicing: lkz ˜for,
on behalf of™ and nkz ˜beetle™, which differ by initial [d] and [t], or nk†nm ˜rot™ > [tl˛]
vs. lk«yysq ˜long™ [dl˛], gm· ˜I drink™ > [pj] vs. ,m· ˜I beat™ [b˛ j]. Before a vowel, the
two members of the small class of W, voiced labio-dental approximates [v v˛ ], also
allow both voiced and voiceless obstruents to precede (ldj«[<gen> ˜a pair™ [dv]
vs. ndj«[<genpl> ˜your™ [tv]), just as sonorants do. The distribution is recursive, so
that a series of these sounds (sonorants R or W) before a vowel permit obstruents
of either type: jn vyj ´ujuj ˜from much™ [tmn] vs. gjl vyj ´ubv ˜under much™ [dmn],
76 As emerges below, the relationship of /f f˛/ to /v v˛ / is not the same as that between /b˛ / and /p/, and
it might be justi¬ed to include /f f˛/ in the list of consonants that are unpaired for voicing.
77 Browman and Goldstein (1986[a], [b]) argue that voiceless consonants in English and French are
marked by an active “glottal closing-and-opening gesture,” gestures which voiced stops lack.
70 A Reference Grammar of Russian

hfpdhfn«nm ˜dissipate™ [zvr] vs. jndhfn«nm ˜repel™ [tvr]. In this respect the class W
behaves like sonorants.
Russian has two syntagmatic rules of voicing that apply obligatorily and al-
most exclusively to obstruents: voicing assimilation and word-¬nal devoicing.
Within a sequence of obstruents, all obstruents must have the same voicing as
the last segment of the sequence; for example, the sound corresponding to ¤;≥
is voiceless [s]‹ before voiceless [k] in yj ´;rf ˜knife [dim]™ (it surfaces as [z]‹ only in
the gen pl yj ´;tr), while the sounds corresponding to ¤c≥ and ¤nm≥ are voiced
in gfcnm,’ ˜pasturage™ [z˛ (d˛ )b]. Viewed in terms of gestures, voicing assimilation
is the constraint that no boundary of voicelessness can fall between obstruents;
´;rf — [z ‹ k], nor can
voicelessness cannot begin between the two obstruents of yj
the end of voicelessness fall in the middle of the obstruent interval of gfcnm,’
— [s˛ (t˛) b].
Voicing assimilation is thought to apply without exception within words and
at the boundaries of pre¬xes or prepositions and words: jn,«k ˜repelled™ [db˛ ], c
,†htuf ˜from the shore™ [zb˛ ], gjlgbc’nm ˜sign™ [tp], l†drf ˜wench™ [fk]. Assimilation
is usual before enclitics beginning with voiced obstruents (lj ,s ˜daughter ´xm
might™ [Z˛ ‹ b], jn†w ;t ˜father indeed™ [Zz], dj ,s ˜well now™ [db]), possible but not
‹ ´n
obligatory in compounds (gfhn,bk†n ˜party card™ [tb˛ ] ∼ [db˛ ]), and occasional be-
tween independent words, at least in close syntagms (and with connotations of
colloquial register): y’i pyfvty«nsq (r«tdcrbq nj ´hn) ˜our famous (Kiev tort)™ [zz], ‹
rjy†w uj ˜end of the year™ [cg] ∼ [Zg], v’nm c ltnmv« ˜mother with children™ [t˛z˛d˛]
∼ [d˛ z˛d˛], lj dljdß ˜daughter of a widow™ [c˛ ‹vd] ∼ [Z˛‹ vd], nh©lyjcnm pfrk/x’kfcm
˜dif¬culty consisted of™ [s˛ (t˛)z] ∼ [z˛ (d˛ )z], p©, ,jk«n ˜tooth hurts™ [pb] ∼ [bb], lj ´xm
,sk’ ˜daughter was™ [c˛ ‹b] ∼ [Z˛ ‹ b], jn†w ,ßk ˜father was™ [cb] ∼ [Zb].78 When assimi-
lation occurs, it appears that there is neutralization. That is, a lexical [s], when
voiced, is identical to a lexical [z], and conversely, a lexical [d˛ ], when devoiced,
is identical to [t˛].79
Final devoicing pushes the beginning of voicelessness as far back into the
word from the end of the word as possible. Thus the sound corresponding to
¤lm≥ in ntnh’lm is voiceless [t˛¬ t ra „ t˛ ], and, by voicing assimilation, the onset of
voicelessness is pushed back to include both obstruents in udj ´plm ˜nail™: [gvo „ s˛ t˛ ].
Devoicing occurs without exception in phrase-¬nal position, normally in the ¬rst
word of a phrase consisting of two independent words (l†l ei=k ˜grampa left™
[t], nh©l k/l†q ˜labor of people™ [tl˛]), usually but not always at the end of a word
78 Some information is given in Paufoshima and Agaronov 1971.
79 There are still questions to be investigated in the phonetics of consonants participating in voic-
ing rules. Drage 1968 noted some occasional exceptions to the rules. Barry (1988) considered the
possibility that ¬nal devoiced obstruents maintain some properties characteristic of voiced conso-
nants, but does not ¬nd consistent evidence of a phonetic difference. Burton and Robblee (1997),
examining assimilation, found that consonants neutralize.
Sounds 71

before a clitic beginning with a sonorant or a vowel (l†l e; ˜grampa already™
[t], p©, kb ˜the tooth?™ [pl˛]). Primary prepositions maintain voicing before sono-
rants and vowels (gjl k’vgjq ˜under the lamp™ [dl], ,tp vyj ´ub[ ˜without many™
[zmn]). However, root and pre¬xal prepositions have a stressed vowel, marking
them as autonomous words (,k«p ˜near™, crdj ˜through™, ghj ´nbd ˜against™, dck†l
˜following™), and their ¬nal consonants devoice: crdj k†c ˜through the forest™
[s˛ l].
As noted, sonorants (R) and labio-dental approximates (W) are normally voiced.
However, they are not completely inert with respect to voicing rules. W is less
inert than sonorants. The relevant contexts are these.
<a> V #: In ¬nal open position after a vowel, the two members of W de-
voice and become identical to the voiceless obstruents [f f˛]: rhj ˜roof™ [f], rhj
´d ´dm
˜blood™ [f˛]. Sonorants, in contrast to W, are expected to remain voiced. Yet devoic-
ing, partial or complete, occurs, [r˛] being the most susceptible: rj ˜measles™ >
[kor˛] . . . [koˇ˛] . . . [kor]. (It is convenient to write the ligature sign to indicate a
5 ü˛
5„ 5 r˛r
„ü „
consonant of normal duration over which some feature such as voicing changes
its value.) Sonorants may devoice partially after a vowel before a ¬nal voiceless
obstruent, as in c†hg ˜sickle™ and c†h, ˜Serb™ [rrp]. ü
<b> VC #: After a preceding voiceless obstruent at the end of a word, ¬nal
sonorants are usually devoiced: nt’nh ˜theater™ [a„ tr ].80 After a previous voiced
obstruent, the sonorant may acquire an anaptyctic vowel (h©,km ˜ruble™ [b™l˛]).81
Or, in less than standard speech, it may devoice, partially ([bll˛ü ˛ or completely
ü˛ ü˛
([bl]), and then pass on voicelessness to the preceding obstruent (([pl]),), and (in
dialects) even be identi¬ed as an obstruent: [zÈs˛ t] for ;«pym ˜life™. In this context,
the constraint of ¬nal devoicing attempts to move the onset of voicelessness back
towards the margin of the previous vocalic domain, in the process potentially
affecting a sonorant.
When W follows an obstruent at word end, both the W and the obstruent
apparently devoice: [jh©udm ˜banner™ [kf˛], nh†pd ˜sober™, h†pd ˜frisky™, vyj z ´uj ´pd
˜many sores™ [sf].82
<c> C CV: Internally between obstruents, W behaves as an obstruent. It par-
ticipates in voicing assimilation: when W precedes a voiced obstruent, a previous
obstruent remains voiced (gjl dljdj ˜under the widow™ [dvd]) or becomes voiced
(r dljd† ˜to the widow™ [gvd]). Before a voiceless obstruent, W devoices and passes

80 Jones and Ward 1969:189, Matusevich 1976:188, 198, SRIa 1.105--6.
81 See Reformatskii 1971, Liubimova 1975, Barry 1989, Flier 1990, 1993, with references.
82 But according to Reformatskii (1975), devoicing is not complete: though the W of vyjuj z is ´ ´pd
devoiced, the preceding /z/ can remain partially voiced ([zsf]) or fully voiced ([zf]). At the same
time, a /z/ before an /f/ is said to devoice, in an abbreviation concocted by Reformatskii: ZPA. If
so -- if [zsf], [zf], or [zv] is pronounced in z instead of [sf] -- it would show simply that the behavior
ˇ ´pd
of W in this position is not completely that of an obstruent.
72 A Reference Grammar of Russian

on voicelessness to a preceding obstruent, which remains voiceless (jn dnjhj ´uj
˜from the second™ [tft]) or becomes voiceless (gjl dnjhßv ˜under the second™ [tft]).
In this context, W forms part of an extended obstruental interval that does not
permit changes in voicing within the interval.
With sonorants between obstruents, it is possible to insert an anaptyctic vowel
and make the sonorant syllabic, in which case the sonorant can accept bound-
aries of voicelessness. The interesting question -- a question on which there is
some disagreement -- is what happens if the sonorant does not become syllabic.
Four sub-contexts can be distinguished.
<c1> D D: Between voiced obstruents, sonorants remain voiced, and may
become syllabic: gjl km;«dsv ˜under false™ [dl˛(™)z].

<c2> T D: After a voiceless and before a voiced obstruent, sonorants most
probably leave the preceding voiceless obstruent untouched, whether or not
they acquire an anaptyctic vowel: jn kml«ys ˜from the ice-¬‚oe™ [ø t (™)l˛d˛¸nï], „
Gh;td’kmcrbq [ p (™)rz]. There is, however, some uncertainty on this point, dis-

cussed below.
<c3> D T: After a voiced obstruent before a voiceless one, a range of variants
is possible. In, for example, v©lhcndjdfnm: ˜act wise™, both obstruent and sonorant
can maintain voice ([dr st v]), or the sonorant can devoice partially ([dr rst v] or
completely ([d rst v]), or both can devoice, as is not uncommon in jrn„¦,hmcrbq
[ pr˛sk˛ ]. The sonorant may be lost. If the sonorant acquires an anaptyctic vowel,
as is possible at preposition boundaries, the obstruent is unaffected: bp K[’cs
<c4> T T: Between voiceless obstruents, the sonorant is hemmed in by
voicelessness on both sides without the aid of a supporting vowel. The sonorant
can insert an anaptyctic vowel, as it does usually at a pre¬x boundary: jn MXATf
˜from MKhAT™ [ø t ™m x a„ t ´]. Or it can devoice, creating a single extended in-
terval of voicelessness without internal shifts in voicing (Cegh’ckmcrbq [ slsk˛ ]).
Or it may be lost altogether.
The behavior of W and sonorants, especially in these environments, has gen-
erated something of a controversy about the nature of the voicing rules. It is
Jakobson who is credited with ¬rst observing the unusual behavior of W in
particular.83 In his original article in 1956, Jakobson characterized the voicing
rule so that the ¬nal obstruent in a cluster was held wholly responsible. As-
similation occurs between two obstruents, “regardless of whether one follows
directly after the other or v comes between them.” On this view, W is a perme-
able membrane that transmits voicing from a following obstruent to a preceding
83 Jakobson 1956/1971[a]. On Jakobson™s treatment of W in this position, see Shapiro 1966. Shapiro
1993 provides an overview of the problem of voicing.
Sounds 73

The subsequent tradition saw in Jakobson™s observation the possibility that
W is to be classi¬ed as a sonorant. One hypothesis was that W is intrinsically
a sonorant that becomes an obstruent in weak environments, when it merges
with [f f˛].84 This solution maintains the assumption that active participation in
voicing rules is limited exactly to the class of obstruents; true sonorants would
have to be excluded on this account.
Another tack was to ask whether true sonorants behave the same as W in the
vicinity of obstruents.85 Sheveroshkin, citing Gh;tdfkmcrbq, states that “voicing
of p- <...> does not occur <...> It can hardly be suggested that the sonorant [r] is
syllabic.” He notes that, in bp K[fcs, devoicing is possible, provided the [l] itself
devoices: [sl x]. Zalizniak (1975) claims that sonorants generally do not transmit
voicing, even if they remain non-syllabic. He states categorically that a voiceless
obstruent does not become voiced across an intervening sonorant in assimila-
tion to a voiced obstruent. Thus Tsvetaeva™s line pfcnhf[jdfyyjcnm эnb[ k,jd
˜insurability of these foreheads™ could only be pronounced without assimilation
as [ x lb], never with assimilation as — [ lb]. If so, sonorants differ from W or
obstruents, which would affect a preceding consonant in this combination: ¦nb[
dlj ˜of these widows™ [ vd], ¦nb[ ly†q ˜of these days™ [ d˛ n˛ ]. Zalizniak mentions
that if the following obstruent is voiceless, voicing assimilation -- devoicing --
could occur.
After the appearance of these studies, Jakobson responded by pushing the par-
allelism between W and sonorants. In his last summary discussion (1978/1985),
he insisted that voicing assimilation is passed through sonorants, both when the
obstruent after the sonorant is voiceless (bp Vw†ycrf ˜from Mtsensk™ [smc]) and,
remarkably, when the obstruent after the sonorant is voiced ([dl˛d˛] in jn kml«ys).
(Jakobson does not explicitly say whether the sonorant itself would have to be
devoiced before a voiceless obstruent in order to communicate voicelessness;
his transcriptions do not indicate that the intervening sonorant is devoiced.) To
judge by his examples, sonorants behave like W: they are also permeable mem-
branes that transmit voicing. In extending this property of permeability from
W to all sonorants, Jakobson makes a substantive claim that differs from those
made in other sources: his r k;«dsv ckjd’v ˜to false words™ [glz]‹ and jn kml«ys
[dl˛d˛] seem incompatible with Sheveroshkin™s Gh;tdfkmcrbq [prz]‹ and Zalizniak™s
´nb[ k,j [xlb]. While Jakobson claims that there is complete parallelism in the
¦ ´d
context CRC and CWC regardless of whether the second obstruent is voiced or
voiceless, other investigators point to the likelihood that there is an asymmetry
in contexts: voicelessness may be passed on, but only if it is imposed on the

84 The solution proposed by Andersen (1969) and subsequently reinvented in other places.
85 Es kova 1971:245, Sheveroshkin 1971 (especially 282).
74 A Reference Grammar of Russian

sonorant; voicing will not be passed through a sonorant to a preceding voice-
less obstruent.86 Thus Jakobson™s factual observations differ from those of other
To understand voicing, it is useful to think of voicelessness as a feature with
a temporal life. Voicing and voicelessness are not entirely symmetrical. Vocalic
domains (vowels themselves, and vowels extended by sonorants) are intrinsically
voiced.88 Boundaries of voicelessness are tolerated on the margins of vocalic
domains and only there; consonantal intervals are either voiced or voiceless
throughout with no internal change -- no internal beginning of voicelessness
(— [VD N V]) and no internal end of voicelessness (— [V N DV]). Whether an inter-
val is voiced or voiceless is determined by the last obstruent. Sonorants adjacent
to vowels extend the vocalic domain and therefore tolerate adjacent voiceless-
ness. But sonorants are not completely inert in voicing rules. Complications arise
when a sonorant is next to a word boundary or is sandwiched between obstru-
ents. In such contexts, the class of W behaves like an obstruent. Unambiguous
sonorants -- liquids and nasals -- can do the same if the following environment
imposes voicelessness, but they can also act as an autonomous domain and
tolerate the cessation of voicelessness immediately preceding them, in the con-
¬guration N RD (Gh;tdfkmcrbq, jn kmlbys).

2.4 Phonological variation

2.4.1 General
Most -- perhaps all -- phonetic processes in Russian exhibit variation in their
application. While each process deserves its own description, the processes are
governed by analogous factors. The factors can be grouped into three classes:
systemic factors (those motivated by intrinsic properties of the sounds involved);

86 The assumption that both voicelessness and voicing are passed through sonorants is critical for
the theoretical studies of Halle and Vergnaud (1981) and Hayes (1984). Robblee and Burton 1997,
examining the duration and amplitude of consonants in clusters in which a sonorant is between
two obstruents, could ¬nd no instrumental con¬rmation that voicing is transmitted through
sonorants. Kavitskaya (1999), a phonologist from Moscow, states that in her speech there is no
assimilation in either context, bp Mw†ycrf ˜from Mtsensk™ or jn ku©ymb ˜from a liar™.
87 Jakobson claims (1968/1971[a]) that [f f˛] do not assimilate in word-¬nal position before an enclitic
or word beginning with a voiced obstruent: thus uh’a ;t ˜the graf, though™ remains [fz], not ‹
[vz]. A half century earlier, before any controversy about the status of W had arisen, Cherrnyshev

(1908:37) transcribed the phrase rfk«a ,s gj,t;l=y ˜the caliph was defeated™ as rfk«d ,ßk,
indicating voicing assimilation.
88 Though voiceless vowels do occur, the optimal environment being unstressed between voiceless
obstruents at word end, e.g., i=gjn [ˇop´t] (Panov 1967:131). Jones and Ward (1969:191--92) say that

a voiceless vowel occurs “not infrequently,” in contexts adjacent to voiceless obstruents, such as
dßcnfdrf ˜exhibition™, x†htg ˜skull™, l†deitr ˜girls™ [gen pl].
Sounds 75

factors of idiomaticity (those having to do with the morphological and lexical
constraints on processes); and sociostylistic factors.

2.4.2 Phonological variation: idiomaticity
Variation depends in large measure on the extent to which the given combina-
tion of sounds is conventionally pronounced together. The more the two sounds
that participate in the change are associated and linked in usage, the more
likely they are to show the effects of phonetic interaction. This factor might be
termed the criterion of i d i o m a t i c i t y .
The most familiar aspect of idiomaticity is the hierarchy of morphological
“boundaries.” By measuring the degree of cohesion vs. autonomy of constituent
units, boundaries in effect measure syntagmatic idiomaticity. As is familiar, the
weaker the boundary, the more likely it is that segments on either side of the
boundary will interact. As a consequence, phonological processes apply most
readily within morphemes, a little less regularly across boundaries of deriva-
tional suf¬xes, less across in¬‚ectional boundaries, and with decreasing willing-
ness across pre¬x, preposition, and word boundaries.89 Most processes in Russian
are sensitive to boundaries, though the cut-off points are different for different
processes. Processes are summarized in Table 2.9, in which boundaries are listed
from weak to strong along the horizontal axis, and processes are listed along the
vertical axis from restricted to general. There is no reason to think that bound-
aries are becoming more prominent or less prominent over time; there is no sin-
gle direction of development. Any rule can be sensitive to boundaries, whether it
is expanding or receding. The generalization is that, as a rule changes -- whether
it expands or recedes -- it will expand or be maintained better when weak
rather than strong boundaries intervene between the sounds that interact.
The change whereby [c˛‹] loses closure in clusters of [s] plus [c˛ ‹], resulting in
[s˛ ], is regular at suf¬x boundaries, but less regular at pre¬xes and uncom-
mon with prepositions. Palatalization assimilation has been disappearing. Its
retrenchment has been following the hierarchy of boundaries. Assimilation,
even for Avanesov, was unlikely between preposition and head word; it was
somewhat more likely at pre¬x boundary, and regular only within morphemes.
Palatalization of velars before {i e} applies within words and across in¬‚ectional
boundaries (between stems and in¬‚ectional endings), her’ ˜hand™ [k’], gen sg
her« [k˛´], but normally does not cross preposition or word boundaries: r «vtyb
˜to the name™ [k˝!]. The alternation of [i] and [˝!] is less restricted; it crosses pre-
¬xes (csuh’nm ˜play™ [sÈ]) and prepositions (jn «vtyb ˜from the name™ [t˝!]) and, not

89 Shapiro 1967.
76 A Reference Grammar of Russian

Table 2.9 Boundaries and variation

internal derivation in¬‚ection pre¬x preposition enclitic word
√ — — — —
CC > CC (current) ±
¸ ¸¸ ?
√ √ — —
CC > CC (older) ± ±
¸ ¸¸ ?
√ √ √ — —
{Ki} > [Ki] ˛ n.a. n.a.
√ √ —
{sc˛} > [s ˛ ‹ ] ±
˛‹ ‹ n.a. ? n.a.
√ √ √ √ —
{sc˛} > [sc˛] ˛‹ ‹ n.a. n.a.
‹ √ √ √ √ √ √
{s z}{ˇ ˇ} > [ˇ ˇ ]
sz sz ?
√ √ √ √ √ √
{C o i} > [Co È] ±
√ √ √ √ √ √
{C C } > [CC ] ±
// //
$ˇ ˇˇ

= process applies regularly, without restriction
± = process applies less frequently, with some restrictions
? = process applies exceptionally, with signi¬cant restrictions
— = process does not apply
n.a. = not applicable

infrequently, occurs between closely linked words within a phrase (i=k buh’nm
˜went to play™ [lÈ]). Assimilation of [s z] to [s˛ before [c˛‹] or to [s ‹ z]‹ before root
[s ‹ z]‹ is regular across pre¬xes and prepositions and can occur between words in
a phrase (vjhj ;=cnrbq ˜a harsh frost™ [z ‹ ]). Voicing assimilation easily crosses
the boundary between words within phrases.
Variation is affected by other considerations that are, however, dif¬cult to en-
code as boundaries. Processes apply to words to the extent that the conditions
for a process are present in all forms of the paradigm of the word, in¬‚ectional
and sometimes derivational. Palatalization assimilation is less likely if the trig-
ger consonant is palatalized only in some forms. Palatalization is less likely in
d ,«ndt ˜in battle™ (only [tv˛ ]), because [v˛ ] is palatalized only in certain cases, than
in uniformly palatalized d†ndb ˜branches™ (possible [t˛v˛]).90
How regularly a process applies may depend on the relationship between a
particular form and the rest of the morphological paradigm to which it belongs.
The zero ending -- a fecund environment for changes, since consonants are not
supported by a following vowel -- exhibits different effects depending on which
“zero” it is. For example, geminates are often maintained in the genitive plural
(gen pl v’cc [m’s ] ˜of the masses™), because the genitive plural is under paradig-
matic pressure from other weighty members of the paradigm in which a vowel
follows (nom sg v’ccf [m’s ™]). Geminates are often lost in the nominative sin-
gular, a more autonomous form which is less subject to pressure from forms
with vowels (nom sg rjyuh†cc [s] ˜congress™, not — [s ]).
90 Krysin 1974:61 cites 13 percent for d†ndb but an even paltrier 5 percent for d ,«ndt.
Sounds 77

The paradigm of verbal forms has less cohesion among its forms than the de-
clension of nouns. Palatalized labials are well maintained before the zero ending
of the nominative singular of nouns, because in the rest of the paradigm the
labial is before a vowel and is palatalized: palatalized [p] in uj ´ke,m<nom sg> ˜dove™
is supported by [b˛ ™] in uj
´ke,z<gen sg> . Palatalized labials are beginning to be lost
in the imperative in the substandard register: ghbujnj ´dmnt ˜prepare!™ standard [f˛],
substandard [f], even despite the fact that there are other forms with palatalized
labials in prevocalic position (inf ghbujnj ´dbnm, 2sg ghbujnj ´dbim). In the mas-
culine singular of the past tense of verbs, [l] was lost after fricatives: — nesl(>
y=c, — vezl(> d=p. Analogous phonological combinations have been maintained
in nouns (sometimes by insertion of a vowel): cvßck ˜sense™, ©ptk ˜knot™ because
related forms have following vowels (gen sg cvßckf, gen sg ©pkf). Again, the
zero form of nouns underwent less extreme change than the zero form in verbs
because this context in nouns is better integrated in a paradigm of forms.
In many of these processes the target (the segment that is potentially affected)
is situated before a boundary in either case; the context is syntagmatically the
same. What is different is the paradigmatic context: the allegiance of the partic-
ular word form to other word forms. A process is retarded when a word form
with the proper phonological context is related to other word forms lacking the
phonological context for the process.
The principle at work here is the paradigmatic analog of the syntagmatic
constraint of boundaries. A word or morpheme will try to remain uniform and
not change its shape, even down to the level of the allophonic shape of the
segments of which it is composed. To the extent that two otherwise independent
units are conventionally pronounced together, their autonomy is overridden. On
both the paradigmatic and syntagmatic axes, rules apply to the extent that the
proper phonological context is regular, conventional, idiomatic.

2.4.3 Phonological variation: systemic factors
Speech results from various articulatory gestures, scripted in time. Though ges-
tures often line up to de¬ne segments (for example, [m] involves almost simul-
taneous labial closure, raising of the tongue, and opening of the velum), each
articulatory gesture has its own pro¬le in time. Assimilation is just the spread
of a gesture across segments.
Speech is evidently composed of alternations of vocalic domains and conso-
nantal interludes. Consonantal interludes require sanctioning from the vocalic
domain. As a consonant cluster grows in complexity, the presence of additional
consonants reduces the duration available for other adjacent consonants in the
same interlude. As noted in the discussion of vowel duration, vowels are shorter
before multiple consonants than before single consonants, and shorter before
78 A Reference Grammar of Russian

single obstruents than before no obstruent. These facts suggest that consonants
have a negative valence -- they remove duration from the vocalic interval. Simi-
larly, voicelessness and stop articulations (both involving energetic and precipi-
tous gestures) remove duration.91
Vowels vary in duration as a function of stress.92 Stress also affects nearby
consonants. The general principle is that stress sanctions extra length in the
vowels themselves and also extra duration in adjacent consonants, both before
and after the vowel.
The variation of [i ∼ È] sometimes crosses between words, when one word ends
in a hard consonant and the next begins with {i}. Matusevich says explicitly that
the pronunciation of [È], which involves assimilation of the vowel to the hard
consonant of the preceding word, is more common when the vowel is unstressed
(bl=n B ´h«yf ˜there goes Irina™ [tï]) than when the vowel is stressed (bl=n B ´hf
˜there goes Ira™ [tí] ∼ [t˝!]).
Normally in Russian, consonants are palatalized before {e}, though borrow-
ings allow non-palatalized consonants. The consonant is able to avoid palatal-
ization before {e} in borrowings more readily when the vowel is stressed than
when it is unstressed.93 The extra duration of stressed vowels allows more dis-
tinctions in the transitions between consonants and vowels, and in particular
allows either a palatalizing transition (with raised F2 ) or a non-palatalizing tran-
sition (with low F2 ); unstressed vowels, with greater cohesion between the vowel
and consonant, as a rule allow only the palatalizing transition.
The glide [j] is said to have two allophones, a more consonantal [j] as opposed
to a less consonantal [i]. The more consonantal [j] occurs before stressed vowels,
because the glide has more time for elaboration before stressed vowels than
before unstressed. Also, [j] tends to be absorbed before the homorganic vowel [i].
It is less likely to be absorbed before a stressed [í], because a stressed vowel allows
more time for an elaborate transition: stressed cdbym« gen sg ˜pig™ > [n˛ jí] ∼ ±[n˛ 8í]
but unstressed ku©ymb gen sg ˜liar™ > [n˛ ¬] ∼ ±[n˛ 8¬]. Vowels that are stressed permit
more elaborated transitions between consonant and vowel.
Stress also affects consonants in post-vocalic position, especially when they
are not also prevocalic. Gemination -- maintaining a single articulatory con¬g-
uration over an extended time -- is permissible to the extent that extra dura-
tion is sanctioned by adjacent vowels. Stressed vowels sanction more duration

Browman and Goldstein (1986[a]) examined duration in C— VC complexes (C— = a singleton, dou-

bleton, or three-consonant cluster) in English and documented that there is a constant duration
measured from the temporal center of the C— through the vowel to the onset of the post-vocalic
singleton consonant. If the interval from the temporal center of the cluster to the end of the
vowel yields a stable value, then as the consonant cluster increases in complexity and duration,
it must do so at the expense of the vowel.
92 93 Glovinskaia 1976.
Bondarko, Verbitskaia, and Zinder 1960.
Sounds 79

(rightward) in the following consonantal interlude than do unstressed vowels,
and hence ghjbpytc=yysq ˜pronounced™ is more likely to have a geminate [nn]
than jnj ´hdfyysq ˜torn off™. Palatalized labials, endangered when no vowel fol-
lows, are better maintained after stressed vowels: c†vm ˜seven™ [s˛ö† ˛ ] but dj
öm ´ctvm
˜eight™ [vos˛¬m]. In palatalization assimilation, palatalization is better preserved
after a stressed vowel than otherwise, hence better preserved in gj ´lkt ˜alongside
of™, in Avanesov™s norm [pod˛l˛¬] ∼ [pj ¬], than in ,jlk«dsq ˜prone to butting™,
5„ ´dl˛
only [b dl˛ívÈi].
Thus, stressed vowels sanction greater duration both in prevocalic consonants
and in post-vocalic consonants. Position preceding a vowel gets more duration
than position after.
It might appear that these effects could be described by appealing to syllables.
There is more than one algorithm for determining syllable structure. The ma-
jor point of difference concerns what to do with multiple consonants between
vowels, which may be assigned all to the following vowel or split between the
preceding and the following syllable according to some principle. Avanesov 1956,
for example, allows closed syllables only when a sonorant precedes an obstru-
ent: compare closed r’hnf [r.t] ˜map™, ljycrj [n.sk] ˜of the Don™, but open kj
´q ´,pbr
˜fret-saw™ [.bz˛ ], rjul’ [.gd] ˜when™, jnl’nm [.dd] ˜gave away™, cjyk«dsq [.nl˛] ˜drowsy™,
lkbyy’ [.nn] ˜long™ [fem sg]. This approach has the result of minimizing closed
syllables. Other approaches, not speci¬c to Russian, might be more tolerant of
internal closed syllables of the type VCi .Cj V.
Whatever algorithm is invoked, syllable structure does not account for the
variation described above. Palatalized labials in the imperative are more likely
to be lost in the plural, when a consonant follows (эrjyj ´vmnt), than in the
singular, when no consonant follows (эrjyj ´vm). The palatalized labial (here [m]) ˛
would be in syllable-coda position in both instances according to any algorithm
of syllable structure. If the behavior of sounds were based strictly on the position
in syllable structure, [m] should behave the same in both forms; the presence of
another obstruent after the syllable-¬nal [m] should be irrelevant.
Palatalized consonants cause a preceding vowel to become more front; they
do so whether they belong to the following syllable (gen sg uj ˜grief™ [go.„ r˛™]) or 5
form the coda of the syllable (e.g., uj ´hmrj ˜bitterly™ [gor˛.k™]). Similarly, consonants
are labialized in the vicinity of labialized vowels, and this process does not
respect syllable boundaries.94 Nor does voicing assimilation. These processes,
then, pay no attention to syllable boundaries.
Finally, we might consider the algorithm for syllable structure of L. V.
Shcherba. According to Shcherba, a syllable coda is possible only after a stressed

94 Bondarko 1977:130--37.
80 A Reference Grammar of Russian

vowel, as in cd’lm,f [svad˛.b™] ˜wedding™ but not ujym,’ [g .n˛ b’] ˜pursuit™.
Shcherba™s algorithm, because it refers to stress, might seem relevant to the pro-
cesses discussed above. Shcherba™s algorithm leads to an odd result with respect
to assimilations that affect post-vocalic consonants. In palatalization assimila-
tion, for example, if syllable structure were assigned according to Shcherba™s
principle, one would expect assimilation to be less regular after stressed vow-
els, because the post-vocalic consonant would be assigned to the same syllable
as the vowel, and not to the following syllable that contains the consonant
that is the source of palatalization. Thus one would expect less palatalization
in gj´lkt [pod.l˛¬] ˜alongside™ than in ,jlk«dsq [b dl˛ívïi] ˜prone to butting™. In
5„ 8
fact, Avanesov observed the opposite. Also, Shcherba™s algorithm has nothing
to say about consonants in the position before vowels, since preceding conso-
nants would be treated as syllable onsets regardless of whether that vowel
is stressed. As noted, stress allows more elaboration in consonants preceding
Thus models that rely on syllable structure do not describe the variation that
relates to the stress of vowels. We might attempt to describe these facts directly
in a temporal model of phonetic interaction. Speech is an alternation of vocalic
and consonantal domains. Consonants can be understood as a kind of negative
space between the positive articulatory intervals of vowels. Vowels have positive
valence proportional to their own duration (at least insofar as duration is a
function of stress). The longer the vowel, the more duration is granted to the
adjacent consonantal interludes. Consonantal domains are not self-suf¬cient;
they require the support of vocalic domains; they consume duration supplied
by vowels. Asymmetrically, consonants get more support from following than
from preceding vowels.
Consonants have negative valence: they limit the duration available in the
context (in adjacent, especially preceding, vowels and in adjacent consonants
in either direction). The longer the consonantal interlude, the less duration is
available for neighboring vowels. The systemic (phonological) factors that govern
variable processes can be formulated in terms of durational valence.
Sonorants seem neutral or, possibly, variable. In prevocalic position, sonorants
behave as an extension of the vocalic domain: in Ci RV(Cj V) contexts, the initial
stressed vowel is nearly as long as the corresponding ¬rst vowel in Ci V(Cj V) con-
texts. Further, sonorants in the position before vowels have the same behavior
as vowels with respect to voicing -- they tolerate the cessation of voicelessness
on their margins. In post-vocalic position, sonorants do not shorten a preceding
vowel and they permit a following [s˛ ] more elaboration (j,v’yobr [n˛ˇ˛ ]) than an
ˇ s
obstruent in the same position would (ufhlthj ´,obr [ps˛ Sonorants after vowels,
then, extend the vocalic domain.
Sounds 81

The overall view is that vowels have duration, especially in proportion to stress,
and offer duration to surrounding consonants. Consonants consume derivation.
Sonorants are more or less neutral; they extend the domain of vowels.

2.4.4 Phonological variation: phonostylistics and Old Muscovite pronunciation
For most processes in which there is variation, variants are correlated with
different stylistic values: characteristically one variant will be evaluated as con-
servative and explicit, the other as more casual; variation may be correlated
with tempo as well. Moreover, as the sociolinguistic investigations of the 1960s
documented repeatedly, the conservative variant is the variant preferred (in sta-
tistical terms) by the higher social classes, while the innovative, casual variant
is that used (by a statistical margin) by workers.
There is a collection of unconnected phonological traits that have been iden-
ti¬ed in Russian phonological literature as “Old Muscovite” features, features
dating back to the residual population of Moscow before the October Revolution
of 1917. For the most part, they have been overridden by the national norms of
twentieth-century Russian.95
Some Old Muscovite features are the following. In vocalism, a more open
vowel, conventionally transcribed [ ], is used for non-high vowels after hard con-
sonants in ¬rst pretonic position rather than [ ]: cnjkß [st l˝!]. Velars in the nom-
inative singular masculine of adjectives remained hard, and after them atonic [™]
is used (nz´;rbq ˜dif¬cult™ [k™i]); in this instance the Old Muscovite pronunciation
is more original; the national norm of [k˛ ] in such adjectives is a spelling pronun-
ciation. The imperfectivizing suf¬x begins with [™] (again, without palatalizing
a velar) rather than [ï]. Pervasive use of [ï] for orthographic ¤f≥ after ¤i≥ and
¤;≥ is Old Muscovite.
In consonantism, maintenance of [z˛ ‹ ] in lhj ´;;b ˜yeast™ and the like is Old
Muscovite. In Old Muscovite pronunciation, a palatalized [r˛] used to occur in
the position after {e} before a following consonant. This ancient pronunciation
(it derives from a progressive palatalization of the r in — Ci mrCj sequences when
Cj was not a hard dental) occurred in words such as d†h[ [v˛ö† x] ˜top™, g†hdsq
[p† vïi] ˜¬rst™. Old Muscovite had prevalent spirantization of stops in clusters: in
öör˛ 8
lexical items yj ´unb ˜nails™ OM [xt˛], rnj ˜who™ OM [xt], rjul’ ˜when™ OM [ d], and
even in combinations of pre¬xes and lexemes, r lj ´ve OM [ d] ˜to the building™.
Long ago, [c˛ ‹] lost closure before [n] and was reinterpreted as hard [s]. This [sn]
‹ ‹
is still maintained in certain high-frequency lexical items such as rjy†xyj ˜of
course™ and cr©xyj ˜boring™, but in general this pronunciation is receding in
favor of the new national norm, [c˛ ‹ n]. Thus, older speakers have [sn] in ,©kjxyfz

95 See Shapiro 1968, Matusevich 1976, Panov 1990.
82 A Reference Grammar of Russian

˜bakery™, gh’xtxyfz ˜laundry™, while younger speakers use [c˛‹ n]. (Some words have
consistent [c˛ ‹ n]: rbhg«xysq ˜brick™.) Maintaining [sn] is a feature of Old Muscovite

pronunciation, as speakers are aware. Lidiia Chukovskaia, referring to the word
gjlcd†xybrb ˜candlesticks™, exclaimed that Anna Akhmatova “so magni¬cently
pronounces ¤gjlcdtiybrb≥ [= [sn˛ ], AT]. I adore that venerable Russian i, which

has not been replaced on her lips by modern x.”96

2.5 Morpholexical alternations

2.5.1 Preliminaries
Sounds occur only in combination with other sounds in conventionalized, lexical
combinations. Related forms of a word or related words share most of the same
sounds, but not all. It is then possible to example the correspondences of sounds
from one word or word form to another, which can be termed m o r p h o l e x i c a l
(or “morphophonemic”) alternations.

2.5.2 Consonant grades
Aside from automatic alternations involving voicing, consonant alternations go
back to the palatalizations of Slavic and East Slavic. C o n s o n a n t g r a d e s --
the different forms consonants can take -- are summarized in Table 2.10. We can
de¬ne the form of the consonans not affected by any alternation (speci¬cally,
by palatalization) as the basic grade, or C0 .
There are two major patterns of alternation. Firstly, an alternation of C0 with
Cj , re¬‚ecting the historical ¬rst palatalization of velars before — j and iotation of
dentals and labials, occurs in verbs of the type {CVC-a : CVCj -|e|}, among which C0
occurs in the past-in¬nitive stem, Cj in the present stem: gbc’nm ˜write™, 1sg gbi©,
2sg g«itim and gk’rfnm ˜cry™, 1sg gk’xe, 2sg gk’xtim. Secondly, C0 alternates
with Ci , which re¬‚ects ¬rst palatalization of velars and “bare” palatalization of
other consonants before front vowels, within the “middle” forms of the present
tense of obstruent stems: compare [s] in 1sg ytc© ˜I carry™ vs. [s˛ ] in 2sg ytc=im or
[k] in 1sg gtr© ˜I bake™ vs. [c˛‹] in 2sg gtx=im. In i-Conjugation verbs, Cj in the ¬rst
singular and past passive participle alternates with Ci elsewhere: 1sg vjkjx© ˜I
thresh™, passive participle -vjkj ´xty versus vjkjn«nm, 2sg vjkj ´nbim. It could be
noted that C and C have the same values for velars. Clusters of fricative and
j i

stop (= ST) have developed the unique sounds [s˛‹ z˛‹ ] (or [ˇ ]: §2.3.7): uhecn«nm ˜be
sad™, 1sg uheo©, gjkjcr’nm ˜rinse™, 1sg gjkjo©, †plbnm ˜ride™, 1sg †p;e.
Ci has been subject to changes that have led to the development of secondary
patterns. Many derivational suf¬xes that now begin with consonants originally

96 Lidiia Chukovskaia, Zapiski ob Anne Akhmatovoi, vol. II (Moscow, 1997), 437.
Sounds 83

Table 2.10 Consonant grades

Co Cj Ci Ci Ci C

Po ={p b f v m} Po Po
Pl˛ P˛ P˛
To ={t d s z} To
{c‹ z‹ s ‹ z} T˛ T˛ T˛

Ko ={k g x} ˇ ={c‹ s ‹ z} C={c‹ s ‹ z} C={c‹ s ‹ z} C={c‹ s ‹ z}
ˇ ˇ ˇ
C K˛
‹ ‹ ‹ ‹
Ro = {n r l} R˛ ={n˛ r˛ l˛} R˛ ={n˛ r˛ l˛} {n r l˛} R˛ ={n˛ r˛ l˛} R˛ ={n˛ r˛ l˛}
ST = (st sk zd} SC = {s¦ s¦ z¦}
ˇˇ ˛‹ ˛‹ ˛‹

began with — m. As a front vowel, — m conditioned palatalization (hence Ci ) in the
consonant preceding the suf¬x before it disappeared, leaving behind a conso-
nant cluster. Palatalization has been restricted in consonants in the position
before other consonants, a development which has reduced the scope of Ci and
led to an alternative pattern, labeled the Ci grade in Table 2.10. Thus before
adjectival {-n-} the consonants are not palatalized in hß,ysq ˜¬sh™, v†cnysq
˜local™, e,ßnjxysq ˜unpro¬table™, ldthyj ˜pertaining to a door™. (The exception
is [l˛], which is preserved: jnl†kmysq ˜separate™.) As a variation on Ci , labeled
Ci in Table 2.10, palatalization can be maintained in dentals before labials and
velars while being lost in dentals before dentals and lost in labials: rjymr’ (nom
sg rjy=r ˜hobbyhorse™), dim z ´,kjymrf ˜apple-tree™, [jlm,’ ˜walking™, htpm,’ ˜carv-
ing™, ujym,’ ˜pursuit™. In another minor variation on Ci (very restricted, and so
not recorded in Table 2.10), the suf¬x {-sk-}, which conditioned Ci in the conso-
nant preceding the suf¬x, allows the dental sonorants over and above [l˛] to be
palatalized: b·ymcrbq ˜of June™, jrnz´,hmcrbq ˜of October™, as well as gjhneu’kmcrbq
˜Portuguese™. This minor pattern would be: {Po , To , {c ‹ s ‹ z}, R
‹ ¸}.
These variations on Ci involve tinkering with how well palatalization is pre-
served before suf¬xes beginning with consonants. The original Ci grade has also
developed in another direction. The Ci grade of velars, historically palatal conso-
nants, has been yielding to palatalized velars by analogy to palatalized dentals
and labials in certain contexts: note substandard dee ,thtuz ˜protecting™, stan-
dard 2sg nr=im ˜you weave™, substandard ;u=im ˜you burn™, gtr=im ˜you bake™,
analogous to ytcz ytc=im, uht,z uht,=im. This variation on Ci , in which {K} re-
´, ´,
places {C} while Ci is maintained for dentals and labials, is the pattern used in
adjectives formed productively with the suf¬xes {-icesk-} and {-icn-}. These suf-
‹ ‹
¬xes evoke palatalized velars rather than palatals: gcb[«xtcrbq ˜psychological™,
pjjkju«xtcrbq ˜zoological™, gtlfuju«xysq ˜pedagogical™, fyfh[«xysq ˜anarchic™.
In addition to Cj as stated, there is a special variant with the Church
Slavonic re¬‚exes of — tj and — dj, a pattern that is not recorded in Table 2.10.
The third row would then be {s˛ zd} (or {[s˛ zd
ˇ‹ ˇ ‹ ¸]}): djpdhfn«nm/djpdhfo’nm
˜return™, gj,tl«nm/gj,t;l’nm ˜vanquish™. The statement of alternations in
84 A Reference Grammar of Russian

Table 2.10 makes no provision for the re¬‚exes of the Slavic second and
third palatalization of velars: if the alternation of consonants that derives
from these changes had been maintained, there would be a synchronic series
{c z (z˛ ) s (s˛ )} corresponding to {k x}, but the possible alternations resulting
from this palatalization have been eliminated.

2.5.3 Types of softness
Much in Russian phonology depends on whether consonants are palatalized or
not. A notion of softness (to use the informal term) is relevant at different
levels. It may be useful to review the role of softness on different levels.
The concept in the ¬rst place is phonetic. Phonetically soft consonants are
those with a certain articulatory con¬guration. That con¬guration in¬‚uences
adjacent vowels by co-articulation in the same way in different consonants, and
in this effect all phonetically soft consonants are equal. By this criterion [T], [P ], ¸¸
and [R are phonetically soft, and so are [c˛ ‹], [s:], [z˛‹ ], and [j]. In contrast, ordinary
o o o
[T ], [P ], [K ], and [c s ‹ z]‹ are hard.
The pairs [To ] vs. [T] and [Po ] vs. [P ] are capable of occurring in the same syn-
¸ ¸
tagmatic environment and are capable of distinguishing words. In this respect,
[T] and [P ] are phonemically soft. The pair [K] vs. [Ko ] is moving in this direction.
¸ ¸ ¸
Moreover, exactly these sets of consonants are paired in the sense that the hard
sounds are replaced by the soft ones before the {-e} of the locative singular of
Declension<Ia> and Declension<II> (and the dative singular of Declension<II> ):
;ty† ˜to the wife™ [n˛ †], j xtkjd†rt [k˛ ¬].
The details of stressed vocalism depend on the phonetic softness of consonants.
Unstressed vocalism is also sensitive to phonetic softness, but the picture is more
complex. In pretonic position in roots, [c] behaves as an ordinary hard consonant
with respect to {a} (gen sg wfhz [c ]). After the unpaired hard consonants [s ‹ z], ‹
there is variation between the inherited [ï] (;fk†nm [zï ] ˜pity™) and the innovative

[ ] (;’h ˜heat™, ;fh’ [z ‹ ]). The innovative [ ] is what one expects after a hard con-
sonant. The high quality of [ï] rather than the lower [ ] is merely a back version
of [¬]. By virtue of eliciting this high variant [ï], [s ‹ z]‹ are morphophonemically
The different criteria for de¬ning softness are summarized in Table 2.11.
Softness is additionally relevant in two other places. First, most nouns of
Declension<Ia> use one of two overt endings in the genitive plural. Stems ending
in paired hard consonants, including velars, take {-ov}, while stems ending in
paired soft consonants take {-ej}. On this basis, one might de¬ne any stem-¬nal
consonant that takes {-ov} as morphologically hard, and any that takes {-ej}
as morphologically soft. The phonemically unpaired consonants [c˛ ‹], [s¦], [s ‹ z]‹ ˛‹
Sounds 85

Table 2.11 Types of “softness”

phonetic independent morphophonemic morphophonemic morphophonemic
softness {a} > [¬]
softness (phonemic) softness gen pl softness
{-ej} {… ∼ Ci {† ∼ ¬}}
¸ softness ([ï])
√ √ √ √ √
[T˛ ]
√ √ √ √ √
¸ √ √
¸] n.a. n.a.
√ √ √
— —

— — — —
√ √ √

¸ n.a.
√ √ √

[s ˛ ‹ z ‹ (z˛ ‹ )] n.a.

— — ±
[s ‹ z]‹ n.a.

= counts as soft
— = fails to count as soft
n.a. = not applicable, indeterminate
± = partial, to some extent soft
take {-ej}, and in this respect are morphologically soft. Curiously, [j] and [c] take
{-ov}, making them by this criterion morphologically hard.97
In certain word forms, vowels alternate with the absence of a vowel. The overt
vowel can take different forms under stress. One variant (labeled {º ∼ Ci {† ∼ ¬}}:
§2.5.5) occurs consistently before following paired palatalized consonants:
ptv†km ˜land [gen pl]™, htv†ym ˜belt™. It also occurs before [j] (cdby†q ˜pigs [gen
pl]™, hex†q ˜brook™ and [c] (jn†w ˜father™, rjy†w ˜end™), which according to this test
would be evaluated as morphophonemically soft. (The alternation of vowels does
not occur before [c˛ ‹ s˛ ‹ s ‹ z].)98

Thus, each process de¬nes a set of consonants as soft, but the de¬nitions
are not entirely consistent for different processes. A consonant may behave as
morphophonemically soft, even if it is not phonetically or phonemically soft in
synchronic terms.

2.5.4 Vowel grades
Alternations in vowels date from various time periods. The oldest derive from
Slavic re¬‚exes of Indo-European ablaut relations. Only residual traces remain
of the q ua l i t a t i v e a b l au t of — e and — o (dtpn« ˜lead<dt> ™, djp«nm ˜lead<id> ™,

The motivation is presumably historical: {-ej} derives from the original i-stem declension (from
— -mjm). Some words in the i-stem declension ended in [c˛ ], [s˛ ], [s z], and so {-ej} was understood to
‹ ‹
occur after these consonants. As the masculines that belonged to the i-stem declension moved
over to Declension<Ia> , they brought the ending {-ej} with them.
98 Hard labials have both: cel†, (newer c©lt,) but yf=v.
86 A Reference Grammar of Russian

djp ˜wagon™). There was another type of ablaut, between full grade and reduced
grade, and traces of this r e d u c t i v e a b l au t (in Slavic, — e ∼ —* > — e ∼ — m;
* ±
— a ∼ — u > — o ∼ — () remain in the allomorphy of verbs such as 1sg ,th© (full
* *
grade) ∼ inf ,h’nm ˜take™ (reduced grade). Somewhat more productive is the
re¬‚ex of q ua n t i t a t i v e a b l au t in the formation of secondary imperfectives.
The original alternation of — a ∼ — a > — o ∼ — a is still visible in, for example,
* ¯
jnnjkry©nm/jnn’krbdfnm ˜push away™ (in which it must be an extension, since the
o is not original). The alternations in the stems of gjlj,h’nm/gjl,bh’nm ˜pick up™,
dßpdfnm/dspsd’nm ˜call out™ also go back to quantitative Ablaut of the reduced
grade (—* ∼ —¯ > — m ∼ i > — º ∼ [Ci]; — u ∼ — u > — ( ∼ — y > º ∼ [Co È]). This alternation
* ¯
± ± ¸
can be described synchronically as an alternation of no vowel with {i}, either
with palatalization in the preceding consonant (gjl,bh’nm<if> ˜pick up™ {-b˛ ir-aj-})
or without (dspsd’nm<if> ˜call forth™ {-ziv-a(j)}). Such residues of earlier ablaut
can be treated as part of the lexical allomorphy of verbs.

2.5.5 Morphophonemic {o}
Hundreds of years ago, — e changed to o under certain conditions -- under stress
before hard consonants (nom pl — s†la > c=kf ˜villages™, nom pl —ˇ†ny > ;=ys
˜wives™) and in ¬nal open position (— pitm† > gbnm= ˜drinking™), while — e remained
unchanged under stress before a palatalized consonant (— selmsk(jm > c†kmcrbq
˜village™s™). Unstressed, this — e did not change to — o, and the re¬‚ex of unstressed
— e is realized as [¬] after soft consonants (— seló > ctkj [s˛ ¬lj and as [ï] after
´ ´])
—ˇen’ > ;ty’ [zïn’]).
hardened palatals ( z ‹
If the original pattern had been preserved without change, it would have
resulted in a synchronic pattern of {CóCo ∼ C†C ∼ C¬} (or {So óCo ∼ So †C ∼ So ï}
ˇ ˇ¸ ˇ
¸ ¸¸ ¸
after hardened palatals ˇo ); in simpli¬ed terms, the pattern would be {óCo ∼ †C
S ¸
∼ ¬}. The original distribution has been eroded in various ways. Original — ˇ dide
— e, leaving many tokens of
not change to o and then subsequently merged with
[†] before hard consonants that derive from — ˇ: n†kj ˜body™, v†cnj ˜place™, w†ksq
— e was preserved as [†] before formerly soft consonants that
˜whole™. In addition,
have hardened before other consonants: ;†ycrbq ˜female™, ex†,ysq ˜teaching™. As
a result, we now ¬nd [†] as well as [ó] before hard consonants.
The earlier pattern {óCo ∼ †C ∼ ¬} is still preserved in some derivational
nests, but there has been a tendency, gradual and long-term, to generalize [ó] at
the expense of [†]. Thus hti†nxfnsq ˜latticed™ can now be hti=nxfnsq ˜latticed™,
in deference to nom pl hti=nf ˜grates, lattice™. Pd=plxfnsq ˜starry™ has already
adopted [o] from pl=pls, itself derived by analogy to the alternation of {óCo ∼
†C ∼ ¬} in ;ty’ ˜woman™, nom pl ;=ys.
Within nominal paradigms, alternation has been eliminated (except for nom
sg x=hn, nom pl x†hnb ˜devil™). In particular, the [ó] does not revert to [†] before
Sounds 87

Table 2.12 Reflexes of the {óCo ∼ †C ∼ ¬} pattern in verbs
{CVC -’- : CVCj -} [¬] / inf, 1sg prs [†] / prs [ó] / if, psv

˜hew™ -ntc’nm -n†itim -n=csdfnm, -n=cfy
˜tousle™ -nhtg’nm -nh†gktim -nh=gsdfnm, -nh=gfy
˜scratch™ -xtc’nm -x†itim -x=csdfnm, -x=cfy
˜lash™ -[ktcn’nm -[k†otim -[k=cnsdfnm, -[k=cnfy

the palatalized consonants of the (dative-)locative singular (j v=lt ˜about honey™,
j rk=yt ˜about the maple™) or before palatalized velars (o=rb ˜cheeks™).
Matters are complex in the root vocalism of verbs; it depends on the class of
verb. There are two classes of verbs in which the alternation {óCo ∼ †C ∼ ¬} is ¸
still visible.
One group is obstruent-stem verbs. Stressed [ó] is found in the masculine past
(l-participle), e.g., — pekl(> g=r ˜he baked™, — nesl(> y=c ˜he carried™. This is one of
the few forms of such verbs in which the root vowel is actually stressed. The
past active participle at one time had [†], but now has [ó], e.g., ghby†cibq >
ghby=cibq ˜having brought™. In velar-stem verbs, the in¬nitive is also stressed
and the vocalism is [†] (e.g., g†xm), while [ó] appears in the masculine singu-
lar (g=r), implying {óCo <msc sg pst> ∼ †C<inf> ∼ ¬<elsewhere> } for velar-stem verbs.
This pattern has been imposed on verbs with etymological — ˇ, which other-
wise should have become [†]: edk=r ˜he carried away™, ghtyt,h=u ˜he neglected™.
Recently c=r has become possible as the masculine past of c†xm ˜hack™, in a
root with etymological — ˇ. To judge by warnings in normative manuals, a pro-
nunciation with [o] has long been an alternate pronunciation for pfghz ˜he
„ ´u
The one other class of verbs in which the alternation {óCo ∼ †C ∼ ¬} occurs
is the type {CVC -’- : CVC -}. In these verbs, [¬] occurs in the in¬nitive, past, and
0 j

¬rst-singular present; [†] occurs in the other forms of the present; and [o] is found

in the derived imperfective and passive participle, as illustrated in Table 2.12.
Other verb conjugations do not have a three-way alternation in vocalism. Verbs
with ¬xed stress on the stem have [†] or [o] but no alternation: -l=hyenm ˜jerk™,

-l=hytim, -l=hubdfnm, -l=hyen; -l†kfnm ˜do™, -l†kftim, -l†ksdfnm, -l†kfy. Verbs with
retracted stress in the imperfective and passive participle have an alternation of
unstressed [¬] with either [o] (-vtny©nm ˜cast™, -vtny=im, -v=nsdfnm, -v=nyen) or [†]

(-cktl«nm ˜follow™, -cktl«im, -ck†;bdfnm, -ck†;ty), but again there is no three-way
alternation. The upshot is that the original pattern {óCo ∼ †C ∼ ¬} that arose
out of the change of — e > o / Co is virtually moribund; it has remained only in
quite speci¬c lexical groups.
88 A Reference Grammar of Russian

2.5.6 Null- and full-grade vocalism
In certain words there is an alternation at the end of the stem between the
presence of a vowel and the absence of a vowel: nom sg vj gen sg v[f ˜moss™;
´;rf ˜foot™; nom sg ,j,=h, gen sg ,j,h’ ˜beaver™.
gen pl yj ´;tr, nom sg dim yj
The absence of vowel, or null grade, is found when a vowel follows, as hap-
pens in most in¬‚ectional forms. The overt vowel, or full grade, occurs when no
in¬‚ectional vowel follows, when the ending is “zero.”
These “¬‚eeting” vowels come from two historical sources. Some stems ended
in jer vowels, whether in the root (— m(x() or a derivational suf¬x (the adjectival
suf¬x — -mn-, the diminutive suf¬x — -mk-/— -(k-). Whenever a vowel other than a jer
followed the stem, as happened in most in¬‚ectional endings, the stem jer was
“weak” and was lost. It was “strong,” and kept, only if the following in¬‚ectional
ending contained a weak jer. Such weak jers were the source of zero endings in
declension that now elicit the full grade in the root. This happens in the declen-
sion of nouns in the nominative singular of Declension<Ia> and Declension<IIIa>
(there also in the instrumental) and the genitive plural of Declension<Ib> and
Declension<II> .
This source of vowel alternations was supplemented by a tendency to break up
clusters of obstruent and sonorant at the end of words by inserting an anaptyctic
vowel. Such clusters arose in the same morphological environments as those in
which jers were vocalized. For this reason there are two historical sources of
alternation between what might be termed null grade (no vowel) and full
grade in the same morphological contexts. The synchronic re¬‚exes of these two
sources are similar and can be discussed together with a little caution.99


. 3
( 17)