<<

Phonotactic constraints on the distribution of palatalized consonants* Alexei Kochetov University of Toronto

This paper investigates the neutralization of plain-palatalized contrasts in Russian. It argues that the attested distribution patterns and phonotactic restrictions on clusters can be adequately explained only if we consider acoustic and articulatory properties of palatalized consonants, and particularly, the cues available in release, burst, and approach. It shows that the analysis based on Russian extends straightforwardly to the patterns attested in other . The paper supports the view that phonetic detail plays an important role in , and particularly, in preserving and neutralizing phonological contrasts.

What are the crucial factors that condition neutralization of phonological contrasts? The view that is common in current phonological theory holds that the distribution of features is controlled by their syllabic position, whether it is an onset or coda (Licensing by Syllable Position). The phonological constraint CONTRASTIVECODA, or Coda Condition (Prince&Smolensky 1993, Pulleyblank 1997, etc.) requires codas to be minimally specified. The motivation for this universal constraint is based on an observation that "codas make poor hosts for a number of features: voicing, place distinctions, articulations involving more than one articulator and so on" (Pulleyblank 1997:81). An alternative approach derives neutralization from phonetic perceptual and articulatory factors (Steriade 1997, Flemming 1995, Hamilton 1996, Silverman 1997, etc.). In this view phonological contrasts are neutralized in environments poor in terms of phonetic cues and are licensed in positions that are high on a scale of perceptibility (Licensing by Cue). The latter is determined by the relative number of cues, their relative duration and perceptual salience. The acoustic factors interact with articulatory constraints specifying gestural magnitude and overlap timing. What follows from this view is that phonetic detail, and particularly, knowledge of phonetic implementation, plays an important role in distribution of phonological contrasts. This approach has been successfully tested on the distribution of laryngeal and retroflex contrasts (Steriade 1997) and phonotactic patterns in Australian aboriginal languages (Hamilton 1997). This paper investigates the distribution and neutralization of palatalized stops in Standard Russian. It argues that the attested distribution patterns and phonotactic restrictions on clusters can be adequately explained only if we consider phonetic

* am grateful to Keren Rice, Elan Dresher, and Naomi Cull for their multiple comments and suggestions which have hopefully led to some improvement of the paper. I also benefited from the discussion of the paper at the University of Toronto Phonology Group. Special thanks to Nila Friedberg for being an informant in an acoustic experiment.. All errors are my own. This research was funded by SSHRC research grant number 410-96-0842. 172 TORONTO WORKING PAPERS IN LINGUISTICS acoustic and articulatory properties of palatalized consonants, and particularly, the cues available in release, burst, and approach. I propose a harmonic contextual markedness hierarchy based on the availability of perceptual cues to palatalized segments. If the contrast is maintained in a less favourable environment, it will be also preserved in all more informative contexts of the hierarchy. I further demonstrate that the analysis based on Russian extends straightforwardly to the patterns attested in other languages. I illustrate this by the types of palatality neutralization attested in some Slavic, Uralic, and Celtic languages. The typology follows directly from the proposed harmonic constraint hierarchy. The paper is organized as follows. Section 1 outlines the theoretical assumptions of Licensing by Cue, providing some phonetic background, as well as articulatory and perceptual description of palatalized consonants (based on Russian). In this section I discuss specific predictions about more and less likely environments for plain-palatalized contrasts. In section 2 I test the predictions analyzing the distribution patterns of Russian palatalized consonants. Section 3 extends the conclusions to several other languages determining the distribution of the contrast in a typological aspect.

1. Phonetic cues to palatalized consonants

1.1 Auditory representations: Background I will begin with a brief introduction to phonetic cues and their properties. Here I follow Flemming 1995. The main point is that sounds have auditory properties that are perceptually significant and thus make a particular sound distinctive to the human ear. Among these properties are frequencies of of the sound spectre, noise frequency and noise intensity. These properties consist of a number of acoustic dimensions. The first two formants of the spectre are the primary dimensions of the quality of a vowel. The first , F1, corresponds to vowel height, the second formant, F2, correlates with the front-back dimension (roundness lowers F2). Front unrounded vowels have the highest F2, back rounded vowels have the lowest value. F3 is high for high, front unrounded vowels, and low for front rounded vowels back and low vowels. The formant frequency adjacent to a stop is characteristic of that stop and provides important cues to its place of articulation. Dentals and alveolars have high F2. In general labials have lower F2 and F3 than the following vowel. Velars tend to have an F2 close to that of the following vowel. Relative values of F2 for some vowels and consonants are presented in Table 1 (based on Flemming 1995). ALEXEI KOCHETOV 173

Table 1. Formant frequency (F2): vowels and consonant release vowels release of consonants i ´ aouptkêƒsçx highest F2 + ------+ - - + - high F2 + + - - - - + - + - + + - low F2 - - - + + + - - - + - - - lowest F2 - - - - + ------

Fricatives differ in the frequencies where energy is concentrated in the spectrum, or noise frequency. It is high for strident , especially for [s]. It is zero for stops (during closure) (Table 2).

Table 2. Noise frequency stops fricatives ptkêƒ sçx high - - - - - + - - low - - - - + - + +

Noise intensity distinguishes sibilants from other fricatives. Sibilants, e.g. /s/, have high intensity, while /ç/, /x/, /ƒ/ are characterized by lower intensity (Table 3).

Table 3. Noise intensity stops fricatives ptkêƒ sçx - - - + + + + + strident - - - + - + - -

Since the focus of this paper is on plain-palatalized contrasts in stops, it is relevant to discuss the acoustic and perceptual features that characterize these segments. If we consider an intervocalic consonant, acoustically it consists of four parts: approach, closure, burst, release. APPROACH is the final interval of the preceding vowel that contains V-C transitions. RELEASE is the beginning of the following vowel that has C-V transitions. The CLOSURE is the main part of the consonant, which is silent for stops. It is followed by a shorter period of BURST, a short interval of frication noise, similar to a fricative of the same place of articulation. Thus, an intervocalic [t] can be presented as in (1) (after Flemming 1995). Short intervals of approach and the release of the alveolar/dental [t] are characterized by [+high F2] and [-low F3] (see Table 1). The closure is [silent], and a short period of burst after it has the feature of [+high] Noise Frequency, which is similar to a sibilant [s] (see Table 2). This sequence of acoustic events is perceived by a listener as an intervocalic stop [t] (1b). 174 TORONTO WORKING PAPERS IN LINGUISTICS

1. Sequence [ata] a. Acoustics [a] transition silence noise transition [a] |------|--|------|---|---|------| ↑↑ ↑↑ approach closure burst release [+high F2] [+silent] [+high NF] [+high F2] [-low F3] [-low F3] b. Percept at a |------|------|------|

Note that closure, while being silent, cannot perceptually differentiate stops. The cues to their place of articulation are in the other three components: approach, burst, and release. Thus having a [+low F2] in the approach and release will signal a labial place of articulation. It follows that with the removal of any of the three significant cues to a stop (no following vowel, no burst, or no preceding vowel) the perception of a stop will deteriorate.

1.2 Articulatory implementation and acoustic cues

1.2.1 Inventory and palatalized contrasts: Russian In this section I will show what phonetic cues are employed in the differentiation of palatalized consonants from the plain ones and what the best environments for their realization are. Standard Russian can be considered as fully representative of the typology of palatalization: it makes extensive use of the plain-palatalized phonemic distinction involving all places of articulation: labials, coronals and velars (Halle 1959, Jones&Ward 1969, etc.).1 Postalveolar fricatives and dental /ts/ are not paired and always pattern as plain consonants. The postalveolars /ê/ and /ß:/ always behave as palatalized segments. Among velars only /kj/ is considered to be phonemically independent. The sounds [gj], [xj] are traditionally treated as phonetically conditioned, and, thus, will not be considered here. The inventory is given in Table 4.2

1The term ‘plain’ is used here for convenience. It has been noted that non-palatalized consonants are velarized to some extent (Skalozub 1963, etc.). 2Cf. Halle (1959), Jones & Ward (1969), Timberlake (1993), etc. ALEXEI KOCHETOV 175

Table 4. Consonant inventory of Standard Russian Labial Coronal Velar plain pal. plain pal. plain plain pal. pal. pal. plain

p pj t tj ts ê kj k b bj d dj [gj] g f fj s sj s≥ ß: [xj] x v vj z zj z≥ m mj n nj r rj l lj j Note: pal. = palatalized.

For convenience I will consider the contrast in stops disregarding the laryngeal distinction between them. Now the question is what is different between plain and palatalized stops both in terms of their articulation and acoustics?

1.2.2 Palatalized consonants: Articulation A palatalized consonant is characterized articulatorily as having a primary gesture (Lips, Tongue Tip, or Tongue Body) with a superimposed onto it. The secondary gesture (Tongue Body) is described as a raising of the front of the tongue toward a position similar to that for [i] (Ladefoged&Maddieson 1996). Compare drawings made from X-ray tracings of plain and palatalized labials in Figure 1.

Figure 1. [p] [pj]

Articulation of plain and palatalized labial stops (from Bolla 1981)

The secondary articulation, as a rule, is not absolutely simultaneous with the primary one: it is more apparent at the release rather than at the formation of the primary constriction. Labials differ from other places of articulation in the fact that 176 TORONTO WORKING PAPERS IN LINGUISTICS two gestures (Lips and Tongue Body) are independent. For coronals, Tongue Tip and Tongue Body are dependent parts of the tongue and the conflict between them may lead to the change of the primary place of articulation (Ladefoged&Maddieson 1996). The same may happen with velars, since both their primary and secondary constrictions are implemented by the same gesture, Tongue Body. The gestural representations contrasting plain and palatalized labials, based on gestural representations adopted in Articulatory Phonology (Browman&Goldstein 1989, Zsiga 1997, 1998; cf. Kochetov 1998), are given in (2). It is apparent from these representations that a palatalized consonant, that involves two gestures, is articulatorily more complex, or marked.

2. a. [p] Lips labial

Time ⎯⎯⎯⎯⎯⎯→ b. [pj] Lips labial

palatal TB Time ⎯⎯⎯⎯⎯⎯→ Note: TB = Tongue Body

1.2.3 Palatalized consonants: Acoustics

1.2.3.1 Release and approach As mentioned in the introduction, C-V and V-C formant transitions contain important information for identification of the place of articulation of a consonant. This is also true for the palatalized contrast. A comparison of the F2 values at the release of Russian plain and palatalized consonants (Table 5) demonstrates that this is in fact an important transitional cue in the identification of a consonant (based on Bolla 1981:116-122). The F2 values of palatalized segments are almost double those of the plain ones. Notice also the difference between plain and palatalized segments in values of F1, which is much less significant. Given this we can consider the feature [+highest] F2 as characteristic of all palatalized consonants. ALEXEI KOCHETOV 177

Table 5. F1 and F2 values of the release of plain and palatalized stops F1 (Hz) F2 (Hz) p 400 1000 pj 300 2000 t 250 900 tj 200 1900 k 250 1500 kj 200 2000

How do these values signal the presence of a palatalized consonant? A spectrogram of a Russian word pjatj ‘five’ in Figure 2 shows that immediately at the release of [p] the value of F2, similar to that of [i], begins to fall to the level of the vowel [a]. In terms of our features it is a change from [+highest] to [+low] (see Table 1). This rapid change is perceived as an [i]-type off-glide. The presence of a following palatalized consonant, the coronal stop in our case, is indicated by a much shorter on-glide at the approach, raising F2 from [+low] to [+highest] (cf. Ladefoged&Maddieson 1996:364).3 It follows from these facts that an intervocalic palatalized consonant will be signaled by both the approach on a preceding vowel and the release on a following vowel. It should be kept in mind, however, that the V-C transition at the approach of a palatalized consonant is much less acoustically prominent than the C-V transition at the release. Consider the vowel [a] in tjatja 'daddy' that has both the release of the preceding and the approach of the following palatalized segments (Figure 3). The release that is phonetically composed of [i] changing to [´] constitutes more than a third part of the overall vowel duration (35%). At the same time the approach is a short period of 6 ms (6%) (Kuznetsova 1977:73).4 This acoustic fact falls from the articulatory representation in (2b): the gesture of TB of a palatalized consonant is delayed, and thus overlaps with the gesture of the following vowel to a greater extent.

3The word is pronounced by Nila Friedberg. 4 The release and approach duration is considered to be the period with F2 typical for [i] and [´]. 178 TORONTO WORKING PAPERS IN LINGUISTICS

Figure 2.

[pj atj ] A spectrogram of the vowel [a] in pjatj ‘five’

Figure 3.

release approach

0 50 100

[i] [´] [a] [i]

Duration of release and approach: vowel [a] in tjatja 'daddy'

Tests of relevance of F2 value in perception of palatalized consonants (the fricative [sj]) are described in Derkach 1975. In one of the tests the [i]-like beginning of the second vowel was removed and the remaining parts of the word were pasted ALEXEI KOCHETOV 179 together. 80 percent of palatalized sequences [asja] with removed beginning of the second vowel were perceived as plain consonants. Then, the second vowel in the sequences with plain consonant was substituted by one after a palatalized consonant. As a result, all consonants were perceived as palatalized. This reiterates the fact of the crucial importance of the F2 transition at the release of the palatalized consonant and its sensitivity to environments, which is much more significant than for plain consonants. In sum, reference to values of F2 of transitions indicates that the presence of vowels is crucial for the identification of a palatalized consonant. The following vowel (or vowel at the right edge) provides a more important cue for this identification than the preceding vowel (or the vowel at the left edge) (3).

3. Release » Approach

1.2.3.2 Burst F2 is not the only parameter that distinguishes palatalized consonants from the plain ones. A description of acoustic parameters of Russian sounds is given in Bolla (1981), Halle (1959), and Skalozub (1963). One of them is the burst of fricative noise that follows the closure of a stop. Compare the sequence of acoustic events in a plain and palatalized labials (Figure 4). In addition to the overall duration of the segment (the palatalized one is longer), the two segments differ in the duration of the burst: 5 ms for [p] and 20 ms for [pj] (based on Bolla 1981).

Figure 4.

Coronal Burst Velar Closure Labial

0 100 200

Closure and burst durations: plain and palatalized labials

The duration of burst enhances not only the contrasts between plain and palatalized stops, but also serves to distinguish palatalized stops of different places of articulation. Consider Figure 5. While the bursts of palatalized labial and velar stops are 20 ms and 55 ms respectively, the burst of the coronal stop is 98 ms (Bolla 1981). This is close to the acoustic structure of an . Thus, the palatalized coronal stop emerges as most salient. 180 TORONTO WORKING PAPERS IN LINGUISTICS

Figure 5.

Palataliz ed labial Burst

Plain Closure labial

0 100 200

Closure and burst durations: palatalized coronal, velar, and labial

Thus, burst duration enhances not only the contrasts between plain and palatalized consonants, but also those between palatalized consonants of different places of articulation. The coronals emerge here as most salient. The quality of the burst is another, no less important factor. As it was mentioned before, the burst for stops is similar to fricative noise of the same place of articulation. Note that it differentiates plain and palatalized consonants of the same place of articulation (Table 6). In addition to the burst characteristic for a particular place of articulation, all voiceless palatalized consonants have the palatal [ç] component which represents the assimilated part of the [i]-glide. In Standard Russian the palatalized [tj] is also accompanied by slight affrication, i.e. by the palatalized element [sj] (Avanesov 1972, Kuznetsova 1969).5 From the previous discussion we know that fricatives differ in Noise Frequency and Noise Intensity. The values of these features for bursts.

Table 6. stop burst quality noise frequency noise intensity [p] [ƒ] low fricative [pj][ƒjç] low fricative [t] [s] high fricative-strident [tj][sjç] high-low fricative-strident [k] [x] low fricative [kj][xjç] low fricative

Again, coronals seem to be the most salient among palatalized consonants (and plain ones too), having the [high] value for Noise Frequency and [strident] for Noise Intensity. Note also that a coronal affricate would be an ideally salient stop, having the longest strident burst. In sum, both the quality of the burst and its duration are important cues to the palatalized contrast in Russian. The prominence of burst as a cue allows one to

5Presence or absence of this component and its length varies across Russian . In some of them the palatalized dental is enhanced to a true affricate (Kuznetsova 1969). ALEXEI KOCHETOV 181 consider it as the second important cue after release. Adding it to the implicational hierarchy in (3) will give the statement in (4).

4. Release » Burst » Approach

1.3 Distribution of cues according to environments Having considered the cues to palatalized stops and their relevant importance we can look how the distribution of cues in stops differs depending on environments. I will remind the reader that the ‘full package’ of cues for a stop consists of the periods of approach, burst, and release, where release plays the primary role. The burst is the second important cue. The perceptual and acoustic descriptions are based primarily on Jones&Ward (1969) and Bolla (1981). For simplicity I will restrict the discussion to the voiceless labial stop /pj/. In gestural representations I will also ignore the gestures Glottis and Velum as well as the specification of Constriction Degree.

1.3.1 Intervocalic position The most distinctive position for the realization of the palatalized contrasts is the intervocalic environment. The gestures of plain and palatalized labial in this environment are shown in (5) and (6). The gestures involved in the articulation of the consonant in question are shaded. As I mentioned before, the plain labial involves one gesture while the palatalized one is articulated by two gestures. The alignment of consonant gestures with those of vowels gives the following sequence of acoustic events, (5b) and (6b). The main part of the vowel is followed by a V-C transition (approach), silence as a result of the stop constriction, a short period of burst , a C-V transition (release), and the following vowel itself. These acoustic events are perceived as sequences of segments [a], [p], [a] and [a], [pj], [a] correspondingly (5c) and (6c). Note that much shorter cues suffice for a plain labial, while longer release, burst, and approach are employed in the articulation of the palatalized labial. The [i]-glides arise from the overlap of the palatal gesture with the gestures of the vowels. 182 TORONTO WORKING PAPERS IN LINGUISTICS

5. [apa] a. Articulatory sequence of events Lips labial

TB pharyngeal pharyngeal

b. Acoustic sequence of events [a] [] silence noise [u] [a] |------|-|------|--|--|------| ↑↑↑↑ approach closure burst release c. Percept apa |------|------|------| Note: TB = Tongue Body

6. [apja] a. Lips labial

palatal TB pharyngeal pharyngeal

b. Acoustic sequence of events [a] [i] silence noise [i] [a] |------|---|------|----|------|------| ↑ ↑↑↑ approach closure burst release c. Percept apj a |------|------|------|

I conclude that the presence of all cues to palatality makes the intervocalic environment ideal for realization of the contrast. The prediction that follows is that if a has a palatalized contrast, it will be found first of all intervocalically. It has to be pointed out, however, that this holds only if the following vowel has an F2 value lower than that at the release of the palatalized stop: there has to be a perceptible difference between the release and the nucleus of the vowel (Bondarko 1977).

1.3.2 Initial prevocalic position Now we turn to the initial position in (7). The contrast in this environment is likely to be preserved, since the two most significant cues are present: release and burst. ALEXEI KOCHETOV 183

This environment, however is not as good as the previous one, since the approach period containing the on-glide on the consonant is absent. Still, due to the two other cues the plain and palatalized consonants are likely to be distinguished.

7. [pja] a. Lips labial

palatal TB pharyngeal

b. |------|-----|------|------| ↑↑↑ no approach closure burst release

1.3.3 Final postvocalic position Now let us look at the word-final position before a pause as in (8). As Kuznetsova (1977) notes, palatalized stops tend to increase the duration of the burst in the final position. The palatal glide at the right edge is devoiced and turned into a short component [ç], which represents an audible friction (Jones&Ward 1969). This partly compensates for the loss of the most important cue to a palatalized consonant, release. The approach to the consonant is still present adding to the distinctiveness of the contrast.

8. [apj] a. Lips labial

palatal TB pharyngeal

b. |------|----|------|------| ↑↑↑ approach closure burst no release

In general, this environment is at a serious disadvantage compared to the pervious ones: it lacks the release. This allows one to suggest that a language is more likely to preserve the contrasts in the first two environments and neutralize it in the final position. It is important to recall the fact that the coronal palatalized stop has the longest and most phonetically salient burst (among other palatalized stops). With the positional increase of the burst word-finally (positional enhancement: Flemming 1995) the coronal will be in a much more advantageous position than labial or velar palatalized stops. The bottom-line is that palatalized 184 TORONTO WORKING PAPERS IN LINGUISTICS coronals are more likely to be found word finally, even though the most important cue is missing.

1.3.4 Final postconsonantal Having another consonant before a stop will deprive it from another cue, transitions in the approach (here and below I will present only the gestural scores of palatalized consonants) (9).6 The second cue to the contrast, burst, is still present. If it is salient enough for a language, the contrast will be preserved. Again, there is a strong preference for a more salient coronal stop vs. labial and velar stops.

9. [arpj] a. Lips labial TT alveolar TB palatal pharyngeal

b. |------|------|------|-----| ↑↑ no approach closure burst no release Note: TT = Tongue Tip.

1.3.5 Medial/final preconsonantal Now consider the environment before a consonant, while the preceding vowel is retained (10). In this position the release transitions cannot be perceived since they are blocked by the closure of the following stop. The approach transitions and the burst are present.

10. /apjka/ → [apk]a: a. Lips labial

palatal TB pharyngeal velar

b. |------|---|------|----|------| ↑↑ ↑ approach closure burst no release

6A preceding sonorant, however, is a better environment than a fricative or a stop, since it has a weak formant pattern. ALEXEI KOCHETOV 185

When discussing the burst of a stop in this position, additional phonetic details should be mentioned. First, the duration of the burst here is shorter before a consonant than word-finally. Second, a consonant in Russian is not released before a homorganic stop or nasal, or before a lateral (Jones&Ward 1969:94). That is, a coronal /t/ can be released before the labials /p/, /m/, and the velar /k/ (11a), but not before /t/ and /n/, where neither burst, nor glide are possible (11b).7 No independent j j j j release (and thus no fricative off-glide) is observed for /t / in sequence [t s ] and [t s≥ ]. Before /n/ and /l/ it has nasal and lateral plosion correspondingly, or no independent release and the fricative element.

11.a. tjp, tjk, tjm, tjf Burst b. *tjt, *tjs, *tjs≥ *tjn, *tjl, *tjts No burst

Similar conditions hold for labials and velars (which, however, almost never live up to the preconsonantal environment without getting neutralized). A gestural score of a less distinct sequence, a palatalized labial followed by a homorganic nasal, is shown in (12). Note that the neutralization of the plain- palatalized contrast in this environment is not obligatory. It can be avoided if the preceding cue, approach, is adequate to be perceived. It has been noted that an [i]- glide between the vowel and the palatalized stop in this position tends to be slightly more salient than in other environments, when other, more important cues are present (Jones&Ward 1969:104).

12. /apjma/ → [apm]a a. Lips labial labial

palatal TB pharyngeal

b. |------|---|------|------| ↑↑ approach closure no burst no release

1.3.6 Initial preconsonantal A consonant articulated word-initially before another consonant presents another ‘bad’ case (13). The stop has neither release, nor approach. It can have a short burst, however, the same condition on the place and manner of the second consonant applies. The prediction is: if the contrast is possible in the environment, it is more likely that it will occur before a hetero-organic stop or a nasal.

7Consider the data from Zsiga (1998): while [d] is always released before [k] in Russian, it is release before [t] only in 10 percent of the cases. Homorganicity affects release even more in velars, with a 89 percent release in k#t and 0 precent release in k#k. Zsiga attributes this fact to different gesture overlap patterns for homorganic and hetero-organic stops. 186 TORONTO WORKING PAPERS IN LINGUISTICS

13. /pjra/ → [pra] a.

Lips labial

TT alveolar

TB palatal pharyngeal b. |------|----|------|------| ↑↑ no approach closure burst no release

Another factor that may play role here is the palatality of the following consonant. The following palatalized segment provides additional cues to the preceding one. Thus, it is more likely to find a palatalized stop before another palatalized consonant than before a plain one.

1.3.7 Interconsonantal And now, the worst scenario is to be considered. A stop is blocked between two consonants (either initially, or medially, or finally) (14).

14. /arpjka/ → a[rpk]a: a. Lips labial

palatal TT alveolar

TB velar

b. |------|------|---|------| ↑ ↑ no approach closure burst no release

This position is similar to the previous one in terms of the lack of the cues of release and approach as well as in terms of presence of burst. However, it is quite possible that burst between other consonants is shorter and thus less salient than that at the beginning of the word. As in the two previous cases, the presence of burst crucially depends on the quality of the following consonant: it is impossible before homorganic stops and nasals, or a lateral. Thus, if a language distinguishes a palatalized contrast in an environment like this, the contrast will be found in clusters like [pjk], [pjt], or [pjn]. It is highly unlikely that a contrast or even a non- contrastive palatalized consonant is found in clusters like [pjf], [pjm], or [pjl], where ALEXEI KOCHETOV 187 no independent burst is possible (15). In this case, in the absence of contextual cues a palatalized stop may make use of internal cues, such as overall segment duration. These cues are, however, less salient than the contextual ones (Steriade 1997). If the contrast in this environment is possible in a language, it has to be preserved in all other environments previously discussed.

15. /arpjma/ → a[rpm]a: a. Lips labial labial

palatal TT alveolar

b. |------|------|------| ↑ no approach closure no burst no release

It should be mentioned that a palatalized consonant is more likely occur in the environment of other palatalized consonants, e.g. Cj__Cj. The presence or absence of burst does not seem to matter here: palatality of the segment in question is conditioned (regressively or progressively). However, in this case the consonant is likely to be non-contrastive.

1.3.8 Summary Let us now summarize the phonotactic environments and phonetic cues provided by them. In Table 7 I give the discussed contexts, the cues that are present in them and those missing. I put the similar environments together and order them according to the number and salience of cues. The ordering of environments presents an implicational hierarchy with the most likely contexts for realization of the plain- palatalized contrast at the top and the least favourable ones at the bottom. Having a contrast in a less favourable environment predicts that a contrast will occur in a position better equipped with cues. 188 TORONTO WORKING PAPERS IN LINGUISTICS

Table 7. contexts presence of cues absence of cues V__V release none burst approach __V release approach burst V__# burst release approach C__# burst release approach V__C1 approach release burst (C)__C1 burst release approach V__C2 approach release burst C__C2 no contextual cues release burst approach Note: C1 = hetero-organic stop or nasal; C2 = other consonants.

The hierarchy of cued environments given in Table 7 can be recast into a perceptibility scale, or harmonic ranking of negative constraints (*Gesture: Hamilton 1996, Boersma 1997) requiring neutralization of palatality in different environments (16).8 The description of some of the constraints is given in (17). The constraints on gesture combinations that result in fewer cues are ranked higher, and those with more cues are lower.

16. *PAL/C__C2 fewer cues | *PAL/V__C2 | *PAL/C__C1 | *PAL/C__# | ⇓ *PAL/V__C1 | *PAL/V__# | *PAL/__V | *PAL/V__V more cues

8 The prevocalic environments ((V)__V) involve a sub-hierarchy: *PAL/__Vback » *PAL/__Vfront (see 1.3.1). ALEXEI KOCHETOV 189

17. *PAL/V__V: Neutralize palatality contrast between vowels. *PAL/#__V: Neutralize palatality contrast word-initially (in the absence of the preceding vowel). *PAL/V__#: Neutralize palatality contrast word-finally (in the absence of the following vowel), etc.

The language-particular faithfulness constraint to palatalized consonants, PRESPAL (18) (cf. Preserve (voice): Steriade 1997), can be ranked against the fixed hierarchy in (16). Different kinds of ranking of this constraint will yield different patterns of neutralization of palatalized contrasts.

18. PRESPAL: Preserve a palatalized segment.

The proposed fixed constraint hierarchy and the faithfulness constraint PRESPAL ranked against it make straightforward and testable predictions of what neutralization patterns can be attested and why. They also predict what types of consonant clusters are more likely to be found in a language with a certain neutralization pattern and which ones are impossible. In the next section I will put this hypothesis to the test using the phonotactic distribution of plain and palatalized consonants in Russian.

2. Distribution of palatality contrasts in Russian

In this section I will demonstrate that the predictions made by the hierarchy of constraints in (16) are supported by the actual distribution of plain-palatalized contrasts in Standard Russian. I will consider all attested clusters with plain and palatalized stops and will account for most unattested ones. The data on single consonants are from Jones&Ward (1969); two-consonant clusters are taken from Tolstaia (1968) and Dmitrienko (1985) and supplemented by examples from Avanesov (1972), Jones&Ward (1969), Lazova (1974), and some of my own ones. Due to the limit of the paper I will consider clusters with voiceless consonants, with the exception of cases when only clusters with voiced consonants are found.

2.1 Single consonants

2.1.1 Medial First, I will discuss the distribution of single consonants. All Russian single plain and palatalized consonants contrast medially between vowels (19).9

9Here and below I will use the transliteration adopted in North American literature on Russian, while j j using the IPA symbols for transcription. Note the following correspondences: ë = [] (C __) , y = [π], C j j j j j = [C ]. Sequences [-p´], [-t´], [k u], [k o], [k o], [k a], and [kπ] occur only in loans; [-pπ] and [-tπ] are found only in native vocabulary (Avanesov 1972). 190 TORONTO WORKING PAPERS IN LINGUISTICS

19. a. p pj ko[p]at’ ‘dig’ o[pj]at’ ‘again’ sa[p]og 'high boot' sa[pj]ër 'combat engineer' lo[p]ux ‘burdock‘ ku[pj]ura ‘banknote’ kana[p]e ‘canapé‘ te[pj]er’ ‘now’ u[p]yr’ ‘vampire’ u[pj]irat’ ‘to prop’ b. kkj o[k]urok ‘cigarette-butt’ mani[kj]ur‘manicure’ u[k]ol ‘picket’ li[kj]ër ‘liqueur’ re[k]a ‘river’ *-[kj]a- *-[k]e- pa[kj]et ‘package’ *-[k]y- di[kj]ij ‘wild’ c. ttj va[t]a ‘cotton’ ba[tj]a ‘dad’ po[t]ok ‘stream’ po[tj]ok ‘began to flow’ pe[t]ux ‘rooster’ u[tj]ug ‘iron’ pa[t]ent ‘patent’ po[tj]erja ‘loss‘ mo[t]yga ‘hoe‘ mo[tj]iv ‘tune’

Thus, all contrasts are preserved. This is not surprising, since the intervocalic environment is the most favourable for the perception of a palatalized sound, providing the full package of the cues: approach, burst, and release (20). It should be noted, however, that palatalized velars are still rare before back vowels (front vowels neutralize the contrast), occurring mostly in recent loans (Avanesov 1972, Farina 1981).

20. Release » Burst » Approach √√√

2.1.2 Initial Apparently, the lack of the cue of the approach in the initial position (21) does not affect the distribution of palatalized consonants: they are preserved almost to the same extent as in the previous case.10

21. a. ppj [p]at ‘stalemate’ [pj]at ‘heel’, gen, pl [p]ot ‘sweat’ [pj]ëk ‘baked’ [p]ugat’ ‘to scare’ [pj]ure ‘puree’ [p]er ‘peer’ [pj]erl ‘pearl’ [p]yl’ ‘dust’ [pj]il ‘drank’ b. kkj j [k]uvs‡in ‘pitcher’ [k ]uvet ‘cùvétte’ [k]ol ‘picket’ *[kj]ë- j [k]as‡a ‘porridge’ *[k ]a- *[k]e- [kj]edr ‘cedar’ *[k]y- [kj]ino ‘cinema’

10See Endnote 9. ALEXEI KOCHETOV 191

c. ttj [t]ak ‘so’ [tj]aga ‘drought’ [t]omnyj ‘obese’ [tj]omnyj ‘dark’ [t]ura ‘rook’ [tj]ur’a ‘tjurja (soup)' [t]est ‘test’ [tj]esto ‘dough’ [t]ykat’ ‘to poke’ [tj]ikat’ ‘to tick’

As it is seen in (21b), there are severe restrictions on initial velars. On the other hand, there is a constraint against sequences of plain velars with front vowels /e/ and /i/ (considering [π] an of /i/). Only palatalized velars occur here. To leave the constraints on velars aside, we can state that all contrasts are allowed initially. The key to this distribution is the hierarchy of cues. The lack of approach does not play any role here (22). Recall that it is the least important cue to palatalized consonants. The most important indicators of the contrast, release on the following vowel and burst, are present in the environment in question and, thus, make it contrastive.

22. Release » Burst » Approach √√

2.1.3 Final Plain and palatalized segments contrast in word-final position (23): palatalized coronals and labials can occur in this environment. However, the lack of the most important cue, release, is crucial to the neutralization of /kj/. Only plain /k/ is found in this position (23b). The prohibition against the final [kj] is shown by an asterisk.

23. a. ppj gri[p] ‘mushroom’ progi[pj] ‘caving in‘ ry[p] ‘fish‘, gen, pl sy[pj] ‘rash‘ ce[p] ‘flail’ ce[pj] ‘chain’ gra[p] ‘hornbeam‘ gra[pj] ‘rob’, imp glu[p] ‘stupid’ glu[pj] ‘depth’ uto[p] ‘has sunk’ to[pj] ‘swamp’ b. kkj zvu[k] ‘sound’ *-[kj] la[k] ‘varnish’ *-[kj] kri[k] ‘shout’ *-[kj] ve[k] ‘age’ *-[kj] ro[k] ‘horn’ *-[kj] c. ttj li[t] ‘was poured‘ li[tj] ‘to pour’ ry[t] ‘was digged’ ry[tj] ‘to dig’ ode[t] ‘dressed’ ode[tj] ‘to put on’ ma[t] ‘door-mat‘ ma[tj] ‘mother’ du[t] ‘was blown’ du[tj] ‘to blow’ komo[t] ‘drawer‘ lomo[tj] ‘chunk‘ 192 TORONTO WORKING PAPERS IN LINGUISTICS

It has, however, to be mentioned that palatalized labials are less frequent finally than palatalized coronals (Lazova 1974, Farina 1981) and are commonly neutralized in Northern and Western (Avanesov 1972). The fact that palatalized consonants, particularly velars and, to some extent, labials, are more susceptible to neutralization in this environment can be well explained by the crucial importance of the release cues and a relatively smaller salience of burst of these segments, compared to that of coronals. Thus, the data lives up to the expectations represented in the hierarchy in (24).

24. Release » Burst » Approach √√

2.2 Two-consonant clusters

2.2.1 Initial prevocalic Now I turn to two-consonant clusters. All plain and palatalized consonants widely occur as second segments of these clusters (Tolstaia 1968) (25).

25. a. ppj [sp]at’ ‘to sleep’ [spj]atit’# ‘to go crazy’ [fp]ast’# ‘to fall into’ [fpj]atero# ‘five times’ b. kkj [sk]ot ‘cattle’ [tkj]ët ‘s/he weaves’ c. ttj [st]ado ‘herd’ [stj]ag ‘flag’ [ft]oroj ‘second’ [ftj]ër# ‘rubbed in’ Note: # = an inter-morpheme cluster.

This is, again, not surprising, since the following vowel provides a good environment for the realization of phonetic cues. As it was the case word-initially, the lack of the approach does not affect the perceptibility of the segment (26).

26. Release » Burst » Approach √ √

2.2.2 Initial preconsonantal The situation is quite different with the first consonant of the initial cluster. In (27), (28), and (29) I give possible and disallowed initial clusters for labial, velar and coronal stops.11 I list the examples with the following plain (e.g. 27a) or palatalized segments (e.g. 27b) separately. Further, I indicate the number of possible clusters with a given initial consonant as well as a number of contrasting pairs (e.g. 27c). Note that palatalized labials and velars in the initial preconsonantal position

11A marginal cluster [pfj] is found in [pfj]ennig from German ‘Pfennig’; the cluster is not-contrastive: *[pjfj]-. ALEXEI KOCHETOV 193 pattern together. They surface as unmarked plain segments, even though the following segment is palatalized (27b) and (28b).

27. a. ppj [pr]avyj ‘right’ *[pjr]- [pl]avat’ ‘to swim’ *[pjl]- [pn]u ‘I will kick’ *[pjn]- [ps]y ‘dogs’ *[pjs]- j [ps≥]eno ‘millet’ *[p s≥]- j [pt]as‡ka ‘little bird’ *[p t]- b. [prj]amo ‘straight’ *[pjrj]- [plj]aska ‘dance’ *[pjlj]- [pnj]i ‘stumps’ *[pjnj]- [psj]ix ‘phycho’ *[pjs]- [ptj]ica ‘bird’ *[pjtj]- j [pê]ela ‘bee’ *[p ê]- c. Number of clusters with [p]: 12 Number of clusters with [pj]: 0 Number of contrasts [p] • [pj]: 0 Note: • = versus.

28. a. kkj [kr]asnyj ‘red’ *[kjr]- [kl]ast’ ‘to put down’ *[kjl]- [kn]ut ‘whip’ *[kjn]- [kv]as ‘kvass’ (drink) *[kjv]- [kt]o ‘who’ *[kjt]- b. [krj]ik ‘scream’ *[kjrj]- [klj]in ‘wedge‘ *[kjlj]- [knj]az’ ‘prince’ *[kjnj]- [kvj]ity ‘to be quits’ *[kjvj]- [ksj]enofob ‘xenophobe’ *[kjsj]- [gdj]e 'where' *[gjdj]- c. Number of clusters with [k]: 11 Number of clusters with [kj]: 0 Number of contrasts[k] • [kj]: 0

Coronals fare better having a few clusters (28).12

12There is an onomatopoeic word with a cluster [tjf]: [tjf]u ‘pah’, 'pshaw'; the cluster is not- contrastive: *[tf]-; an additional cluster is attested in a place-name [dbj]ilisi 'Tbilisi'; it is not-contrastive: *[djbj]-; variation exists in some CCj clusters; here I give the variants characteristic of younger speakers of Standard Russian (cf. Avanesov 1972). 194 TORONTO WORKING PAPERS IN LINGUISTICS

29. a. ttj [tr]ud ‘labour’ *[tjr]- *[tm]- [tjm]a ‘darkness’ [dn]o ‘bottom’ *[djn]- [tv]oj ‘your’, sg *[tjv]- [tk]at’ ‘to weave‘ *[tjk]- b. [trj]i ‘three’ *[tjrj]- [tlj]a ‘aphis’ *[tjlj]- [tmj]in ‘caraway’ [tjmj]e ‘darkness’, loc, sg *[dnj]- [djnj]ax ‘day’, loc, pl [tvj]ërdyj ‘hard’ *[tjvj]- j *[t ß:]- [t ß:]ëtnyj ‘futile’ [tkj]ët ‘s/he weaves‘ *[tjkj]- c. Number of clusters with [t]: 8 Number of clusters with [tj]: 5 Number of contrasts [t] • [tj]: 1

There is, however, only one contrasting pair of clusters, tjmj and tmj. The evidence for the contrast is not overwhelming if we consider the fact that there are only two words that have these clusters: tmjin ‘caraway’ and tjmje , loc, sg of tjma ‘darkness’.13 Besides, the form tjmje is rarely used on its own but rather with a preposition vo ‘in’. This provides an approach cue to the palatalized segment. Possibly, the output-output correspondence with related forms with prevocalic palatalized segment ([tj]ëmnyj 'dark', [tj]emnota, 'darkness'), supports the realization of the palatalized consonant in a less cued environment. In Old pronunciation (Avanesov 1972) [t] in tmin would agree with the palatalized [mj] in palatality, and thus the clusters in question would not be contrastive. It is worth mentioning that the clusters tjm and tjmj are among a few possible clusters predicted in (11). The presence of the labial nasal allows for the burst of the coronal stop. The palatalized [tj] is even more likely here when the following labial nasal is also palatalized. We also know that a coronal palatalized stop has the most salient burst, the fact that explains why labial and velar stops are not found in the same environment. In sum, neutralization of the contrast in preconsonantal position in initial clusters is adequately explained by the hierarchy of positions of neutralization and the phonetic details of the stop burst (30).

30. Release » Burst » Approach √

2.2.3 Final postconsonantal The next group is final two-consonant clusters (31), (32), and (33). The analysis of contrasts with respect to the second consonant of the cluster shows that this is a

13Also [dmj]itrij 'Dmitri' (name). ALEXEI KOCHETOV 195 fairly good environment for coronals and less so for labial stops.14 Velars are neutralized to the unmarked plain segment.

31. a. ppj go[rp] ‘hump’ sko[rpj] ‘grief' sto[lp] ‘pole’ *-[lpj] j s‡ta[mp] ‘stamp’ *-[mp ] i[sp] ‘cottage’, gen, pl *-[spj] j tja[s≥p]# ‘lawsuit’, gen, pl *-[s≥p ] xo[tjp] ‘walk’, gen, pl *-[tjpj] ska[ljp] 'scalp' *-[ljpj] pro[sjp]# ‘request’, gen, pl *-[sjpj] b. Number of clusters with [p]: 8 Number of clusters with [pj]: 1 Number of contrasts [p] • [pj]: 1

32. a. kkj pa[rk] ‘park’ *-[rkj] to[lk] ‘sense‘ *-[lkj] ta[ljk] ‘talc' *-[ljkj] ta[nk] ‘tank’ *-[nkj] gla[fk] ‘central board’ *-[fkj] lo[sk] ‘lustre’ *-[skj] b. Number of clusters with [k]: 6 Number of clusters with [kj]: 0 Number of contrasts [k] • [kj]: 0

Unlike the final postvocalic environment, the consonants here lack the approach cue in addition that of release (34). This obviously contributes to the neutralization. It is particularly true for velars and labials. However, the position allows for the presence of the second most important cue: fricative burst, which is usually longer in this position (Kuznetsova 1969). Coronals, having a strident release, are better equipped for this environment.

14The words gost and gostj could be considered a minimal pair, however, they also differ in the palatality of [s] which assimilates to the following segment. Under regressive palatality assimilation in Standard Russian a coronal (except /r/ and /l/) assimilates to the following palatalized coronal (except /rj/ and /l/). Morpheme boundaries may block the assimilation (Avanesov 1972). 196 TORONTO WORKING PAPERS IN LINGUISTICS

33. a. ttj j c‡ë[rt] ‘devil' sme[rt ] 'death' j bo[lt] ‘bolt’ proz‡e[lt ] 'yellow tint' ekspro[mt] ‘impromptu’ *-[mtj] komenda[nt] ‘superintendant’ *-[ntj] li[ft] ‘elevator’ ne[ftj] ‘oil’ go[st] ‘quality standard’ *-[stj] j nu[s≥t]# ‘need’, gen, pl vo[s≥t ] 'leader' fra[xt] ‘freight’ *-[xtj] rece[pt] ‘prescription’ *-[ptj] fa[kt] ‘fact’ *-[ktj] j ma[êt] ‘mast’, gen, sg *-[êt ] vo[ljt] 'volt' se[ljtj] ‘herring' *[sjt] go[sjtj] ‘guest’ b. Number of clusters with [t]: 12 Number of clusters with [tj]: 6 Number of contrasts [t] • [tj]: 5

34. Release » Burst » Approach √

2.2.4 Final preconsonantal The preconsonantal position in the final cluster, as seen in (35), (36), and (37), is not the best place for contrasting palatality. Neither palatalized labials nor velars contrast, or even occur here. The usual output are the plain [p] and [k].

35. a. ppj Dne[pr] 'the Dnieper' (river) *-[pjr] xri[pl] 'husky' *-[pjl] zum[pf] 'dibhole' *-[pjf] gi[ps] ‘gypsum’ *-[pjs] efkali[pt] ‘eucalyptus’ *-[pjt] b. ve[prj] ‘wild boar’ *-[pjrj] vo[plj] ‘yell’ *-[pjlj] c. Number of clusters with [p]: 7 Number of clusters with [pj]: 0 Number of contrasts [p] • [pj]: 0

36. a. kkj mo[kr] ‘wet’ *-[kjr] cy[kl] ‘cycle’ *-[kjl] brju[kf] ‘swede’, gen, pl *-[kjf] va[ks] ‘shoe polish’, gen, sg *-[kjs] a[kt] ‘act’ *-[kjt] Va[kx] ‘Bacchus’ *-[kjx] ALEXEI KOCHETOV 197

b. spekta[klj] 'show' *-[kjlj] c. Number of clusters with [k]: 7 Number of clusters with [kj]: 0 Number of contrasts [k] • [kj]: 0

The contrast among coronals is largely neutralized (37).15 All the three j j j possible clusters, [d m], [t p], and [t ê ] (probably found only in these three words) occur in oblique forms. The [t’ê] is homorganic in terms of palatality (obligatory regressive assimilation). It is important that [djm] and [tjp] are among a few clusters that allow burst (11): [tj] is followed by a hetero-organic nasal or stop. Note that these are the only non-coronal nasal and stop found in clusters in (37c).

37. a. ttj tea[tr] ‘theatre’ *-[tjr] më[tl] 'broom', gen, pl *-[tjl] ri[tm] ‘rythm’ ve[djm] 'witch', gen, pl bo[tf] ‘beet tops’, gen, pl *-[tjf] ve[tx] ‘decrepit’ *-[tjx] j bri[ts]≥ ‘bridge’ (game) *-[t s]≥ *-[tp] *xo[tjp] ‘walk’, gen, pl b. vnu[trj] ‘inside’ *-[tjrj] ve[tfj] ‘twig’ *-[tjfj] j *-[tê] pri[t ê] ‘parable’, gen, pl c. Number of clusters with [t]: 8 Number of clusters with [tj]: 2 Number of contrasts [t] • [tj]: 1

The neutralization patterns of stops in (37) support the proposed ranking of cues in (4): having a burst is more important in the preservation of a consonant than an approach (38).16

38. a. Release » Burst » Approach √√ b. Release » Burst » Approach √

2.2.5 Medial clusters All palatalized consonants are found in medial prevocalic clusters. The reason for this is the availability of the two most important cues, release on the following vowel and burst (39). As in other prevocalic environments the presence of velar /k/ is restricted before certain vowels. Since the clusters found in this position are similar to those found in the initial position, examples seem to be unnecessary.

15These are not fully contrastive clusters, since the stops differ in voice. 16In some of the clusters in (61)-(63) sonorants may be considered syllabic. This, however, is not crucial for our analysis, since a following syllabic sonorant deprives a preceding segment from a release to the same extent as any other consonant. 198 TORONTO WORKING PAPERS IN LINGUISTICS

39. Release » Burst » Approach √√

The second medial environment, is medial preconsonantal, V__CV. The survey of possible clusters of this type in Dmitrenko (1985) reveals that no palatalized consonants are allowed in this environment within a root. Inter-morpheme clusters (root + suffix) are allowed (40), (41) and (42). However, only palatalized labials and coronals are found here.

40. a. ppj ko[pr]a ‘copra’ *-[pjr]- so[pl]o# ‘nozzle’ *-[pjl]- o[bm]an ‘deception’ *-[bjm]- j s‡lë[pn]ut’# ‘to slap’ *-[p n]- ca[pf]a ‘pin’ *-[pjf]- le[pt]a ‘mite’ *-[pjt]- po[ps]a 'pop music' *-[pjs]- j o[px]az‡ivat’# ‘coax’ *-[p x]- to[pk]a# ‘furnace’ *-[pjk]- j o[ps≥]it’# ‘to trim’ *-[p s≥]- ku[pts]a# ‘merchant’, gen, sg *-[pjts]- b. ko[prj]e ‘copra’, loc, sg *-[pjrj]- so[plj]a ‘snivel’ *-[pjlj]- o[bmj]en# ‘exchange’ *-[bjmj]- ko[pnj]e# ‘stack’, loc, sg *-[pjnj]- ca[pfj]ennyj ‘pin’, adj. *-[pjfj]- ko[ptj]it’ ‘to smoke’ prisposo[pjtj]e# 'adapt' pu[psj]ik 'doll' ne zlo[pjsj]a#‘don’t be angry’ o[pxj]itrit’# ‘to outwit’ *-[pjx]- ru[pkj]e# ‘deckhouse’, loc, sg *-[pjkj]- j o[pß:]ij ‘general’ *-[p ß:]- j golu[pê]ik# ‘my friend’ *-[p ê]- c. Number of clusters with [p]: 22 Number of clusters with [pj]: 2 Number of contrasts [p] • [pj]: 2

The clusters with [pjtj] and [pjsj] are found only with suffixes -te (imp, pl) and sja (inf). Note that the second consonant in these clusters is palatalized. In colloquial speech labials in these forms are often depalatalized (Avanesov 1972:138-39). ALEXEI KOCHETOV 199

41. a. kkj la[km]us 'litmus' *-[kjm]- o[kn]o ‘window’ *-[kjn]- i[kr]a ‘caviar’ *-[kjr]- ste[kl]o ‘glass’ *-[kjl]- bu[kv]a ‘letter’ *-[kjv]- la[kt]oza ‘lactose’ *-[kjt]- bo[ks]a ‘boxing’. gen, sg *-[kjs]- va[kx]analija ‘Bacchanalia’ *-[kjx]- b. a[kmj]eist ‘akmeist’ *-[kjmj]- o[knj]e ‘window’, loc, sg *-[kjnj]- i[krj]e ‘caviar’, loc.sg. *-[kjrj]- ste[klj]annyj ‘glass’, adj. *-[kjlj]- so[gbj]ennyj ‘bent’ *-[kjpj]- ty[kvj]e 'pumpkin', loc, sg *-[kjvj]- pa[ktj]e ‘pakt’, loc, sg *-[kjtj]- bo[ksj]ër ‘boxer’ *-[kjsj]- j ni[kê]emnyj# ‘useless’ *-[k ê]- c. Number of clusters with [k]: 16 Number of clusters with [kj]: 0 Number of contrasts [k] • [kj]: 0

For coronals the environment is more favourable. A few clusters both homorganic and hetero-organic in terms of palatality are attested (42).17 Some of them, as [tjk], are quite common in the language.

17The cluster [tjm] occurs in the place-name Tot'ma; the clusters [tjv] and [tjv'] occur in the forms of hydronym Ut'va (Ut’ve, gen, sg); the cluster [tjnj] and some other ones below involve regressive palatality assimilation. See footnote 15. 200 TORONTO WORKING PAPERS IN LINGUISTICS

42. a. ttj e[tn]os ‘ethnos’ *-[tjn]- j o[tl]oz‡it’# ‘to put off’ *-[t l]- u[tr]o ‘morning’ *-[tjr]- po[dm]oga# ‘help’ ve[djm]a 'witch' bi[tv]a# ‘battle’ *-[tjv]- o[tp]ast’# 'to fall off’ su[djb]a# ‘fate’ re[tk]o# ‘rarely’ re[tjk]a# ‘radish’ o[tt]orgnut’# ‘to tear away’ *-[tjt]- o[ts]adit’# ‘to displant’ *-[tjs]- ve[tx]oje ‘decrepit’, neu *-[tjx]- j o[ts≥]el’nik ‘hermit’ *-[t s≥]- o[tts]epit’# ‘to unhook’ *-[tjts]- b. *-[tnj]- plo[tjnj]ik# ‘carpenter’ pe[tlj]a ‘loop’ *-[tjlj]- v nu[trj]e ‘inside’ *-[tjrj]- po[dmj]esti# ‘to sweep’ ve[djmj]e 'witch', dat, sg o[tvj]et# ‘responce’ *-[tjvj]- o[tpj]isat’# ‘write back’ sva[djbj]e# ‘wedding’, loc, sg re[tkj]ij# ‘rare’ re[tjkj]e# ‘radish’, loc, sg *-[ttj]- o[tjtj]esnit’# ‘to press back’ *-[tsj]- o[tjsj]ejat’# ‘to siff’ ve[txj]ij ‘decrepit’ *-[tjx]- j *-[tê]- o[t ê]im ‘father-in-law’ j *-[t ß:]- o[t ß:]ipyvat’#‘to nip off’ c. Number of clusters with [t]: 19 Number of clusters with [tj]: 11 Number of contrasts [t] • [tj]: 6

Again, the attested combinations of consonants in clusters (palatalized + plain consonant) are not random and are governed by the same phonetic principle as in the previous case: the burst at the end of the first consonant is possible only when the following segment is of a different place of articulation, preferably a stop or a nasal. Thus, we get the clusters in (43a), and the clusters in (43b) are not found in the language.

43. a. CC1: tjp, tjk, tjm, tjf Burst j j j j j j b. CC2: *t t, *t s, *t s≥, *t n, *t l, *t ts No burst

The clusters in (43a) have the corresponding sets of cues in (44a). Thus, the availability of a burst is crucial to whether a given cluster is phonetically viable.

44. a. Release » Burst » Approach √√ b. Release » Burst » Approach √ ALEXEI KOCHETOV 201

The constraint on clusters manifests itself in the synchronic depalatalization in (45): e.g. the nasal plosion in pu[tn]yj or lateral plosion in ko[tl]y, or the following fricative in my[ts]a do not allow for an independent burst, and lead to neutralization of the underlying palatalized coronal.

45. pu[tj] ‘way’ pu[tn]yj ‘worthwhile’ my[tj] ‘to wash’ my[ts]a# ‘to wash oneself’ ko[tj]ël ‘boiler’ ko[tl]y ‘boiler’, pl o[tj]ec ‘father’ o[ts:]a ‘father’, gen, sg

Unlike in the final preconsonantal environment ((35), (36), and (37)), more non-coronal stops and nasals are allowed in the medial preconsonantal position (due to independent sonority constraints). This gives us more possible CjC clusters. The fact explains the superficial difference between the different outputs in these environments, although both positions have the same sets of cues (43). The analysis of neutralization of palatality in two-consonant clusters is supported by the data on more complex clusters in Russian (Kuc‡era&Monroe 1968). The patterns attested in three- (46), four-, and five-consonant clusters are governed by the same principle and the presence of palatalized consonants in them is even more restricted.

46. a. ppj [spr]os# ‘demand’ *[spjr]- ra[spr]os# ‘interrogation’ *-[spjr]- b. kkj [skr]omnyj ‘modest’ *[skjr]- i[skr]a ‘spark’ *-[skjr]- c. ttj [str]ax ‘fear’ *[stjr]- lju[str]a ‘lustre’ *-[stjr]- mini[str] ‘minister’ *-[stjr]

It is important to mention that segments at word edges may benefit from more acoustic cues in connected speech than segments in word-internal clusters (Hamilton 1996:235): a word-initial consonant following a vowel-final word or a word-final consonant preceding a vowel-initial word receive additional approach (47a) and release cues (47b) correspondingly. On the other hand, segments in internal clusters are not affected and thus are at a disadvantage.

j 47. a. vo t’mu -[√t m]- ‘in the dark‘ approach b. golup’ uletel -[upju]- ‘a pigeon flew away’ release

2.2.6 Summary To sum up, reference to the articulatory gestures and acoustics of palatalized segments accounts for all cases of neutralization and preserving palatalized contrasts in Russian single consonants and clusters. It gives an insight into the question of why certain environments are ‘better’ or ‘worse’ for the contrast and why different 202 TORONTO WORKING PAPERS IN LINGUISTICS segments behave differently in the same environments. The crucial factor in determining whether to preserve or neutralize palatalized contrasts is the distribution of contextual cues, primarily, release and burst. The fewer cues an environment hosts and the less perceptually salient they are, the more likely palatality will be neutralized in favour of the unmarked plain segment. Compare the proposed hierarchy of perceptibility from (16), repeated in Table 8, with the actually attested number of palatalized contrasts across places of articulation.18 These facts can be accounted for with an Optimality Theory-style ranking of PRESPAL with respect to neutralization constraints as in (48). It is seen that the environments with fewer cues never serve to distinguish the palatalized contrast. The patterns have some constrained variation among places of articulation: coronals are the most preserved, velars are the most neutralized, and labials pattern in between.

Table 8. cues environ contrast C • Cj ment release » burst » approach labial velar coronal a. √√√V__V yes yes yes b. √√(√) #__V yes yes yes c. √√ C__V yes yes yes d. (√) √√V__# yes no yes e. √√V__C1 yes/no no yes/no (2) (7) f. (√) √ C__# yes/no no yes/no (1) (5) g. √ (√) #__C1 no no yes/no (1) h. √ C__C1 no no no i. √ V__C2 no no no j. (√) #__C2 no no no k. C__C2 no no no Note: (√) = the cue is optionally provided in connected speech.

18 Restrictions apply before front vowels (58abc), especially for velars (*PAL/__Vback » *PAL/__Vfront ). ALEXEI KOCHETOV 203

48. labials coronals velars *PAL/C__C2 neutralize | *PAL/V__C2 | *PAL/(C)__C1 | *PAL/C__C2 *PAL/C__# || *PAL/V__C2 *PAL/C__C2 *PAL/V__C1 ||| *PAL/(C)__C1 *PAL/V__C2 *PAL/V__# PRESPAL ------|------|------|------*PAL/C__# *PAL/(C)__C1 *PAL/__V ⇓ ||| *PAL/V__C1 *PAL/C__# *PAL/V__V || *PAL/V__# *PAL/V__C1 || *PAL/__V *PAL/V__# preserve || *PAL/V__V *PAL/__V | *PAL/V__V

3. Typology of licensing palatality by cue

I have argued for an analysis of that distribution of palatality in Standard Russian is based on phonotactics. However, assuming that the phonetic properties of release, burst, and approach are universal, the conclusions made for Russian should extend to other languages that make use of plain-palatalized contrast. In this section I will show that distribution of palatalized consonant and their contrastiveness in different languages readily conforms to the prediction made by Licensing by Cue. Due to limits of the paper I will present only some general conclusions. I will argue that phonotactics of all languages is governed by the same scale, or hierarchy of perceptibility, proposed in section 1. Here I will use a simplified version of the hierarchy as in (49), omitting the environment C_# (due to insufficient data) and the distinctions between C1 (a hetero-organic non-lateral) and C2 (any other consonant). 204 TORONTO WORKING PAPERS IN LINGUISTICS

49. release » burst » approach hierarchy (√)*PAL/C__C | (√) √ *PAL/V__C | √√ *PAL/V__# | √√ *PAL/#__V | √√ √ *PAL/V__V Note: (√) = burst before certain consonants.

Languages may differ in ranking the faithfulness constraint PRESPAL with regards to the given hierarchy. The major implication that emerges from the phonetic markedness hierarchy is that a language is more likely to preserve a plain- palatalized contrast in a more cued environment (i.e. lower in ()) than in less cued one (i.e. higher in (49)). In other words, if a language does contrast plain and palatalized segments in a less perceptually salient environment (e.g. V__C), it will certainly do so in a more salient context (e.g. #__V or V__V). This implication can be translated into a highly constrained set of language types ranging from the type that preserves the contrast in all environments (type 1) to the one that neutralizes the contrast in all contexts (type 6) (Table 9). Whether a language licenses palatality in a given context or neutralizes it, is marked with yes or no.

Table 9. type V__V #__V V__# V__C C__C aCja • aCa Cja • Ca aCj • aC aCjka • arCjka • aCka arCka 1 yes yes yes yes yes 2 yes yes yes yes no 3 yes yes yes no no 4 yes yes no no no 5 yes no no no no 6nonononono

Type 1 and type 6 represent two extremes. In type 1 languages, palatality is found in all environments in which a consonants occurs; in type 6 no palatalized consonants are found at all. The others fall in between. Going from type 1 down, each language type distinguishes palatalized from non-palatalized environments being eliminated in a non-systematic way. In which box that ‘no’ is written, palatalized consonant may be present non-contrastively, or absent altogether. Type 6 must be most common, since apparently not many languages tend to have this articulatorily and perceptually marked distinction.19

19In the sample of UPSID (Maddieson 1984) only 21 out of 318 languages are noted to have distinctive palatalized consonants. ALEXEI KOCHETOV 205

My survey is based primarily on Slavic, Uralic, and Celtic languages and dialects. I consulted the following sources: Comrie&Corbett 1993 (Belorussian, Ukrainian, Standard Bulgarian, and Slovene), Chekman 1970 (Belorussian and Bulgarian dialects), Tolstaia 1968 (Bulgarian), Mathiassen 1996 (Lithuanian), Abondolo 1998 (Erzya Mordva, Nenets, Saami), Redei 1984 (Erzya Mordva), Leskinen 1968 (Karelian dialects), Decsy 1966 (Nenets), De Burca 1958 (Irish), Macaulay 1992 and Ball&Fife 1993 (Scots Gaelic, Manx, and Welsh). In Table 10 I present the neutralization patterns of palatalized coronals.20

Table 10. Coronal stops type language family V__V #__V V__# V__C C__C or atja • tja • ta atj • at atjka • artjka • group ata atka artka 1 ??? yes yes yes yes yes 2 Russian Slavic yes yes yes yes no Belorussian Slavic Polish Slavic Bulgarian (NN) Slavic Erzya Mordva Uralic Ukrainian Slavic Scots Gaelic Celtic 3 Bulgarian (BC) Slavic yes yes yes no no Irish Celtic Manx Celtic Karelian (O1) Uralic 4 Bulgarian (S) Slavic yes yes no no no Karelian (O2) Uralic Lithuanian Baltic Nenets Uralic 5 Karelian (A) Uralic yes no no no no Saami Uralic 6 Slovene Slavic no no no no no Welsh Celtic

Note: Bulgarian (S) = Standard Bulgarian; Bulgarian (NN) = the Nova Nadezhda of Bulgarian; Bulgarian (BC) = the Bela Cherkva dialect of Bulgarian; Karelian (A) = the Arkhangelsk dialect of Karelian; Karelian (O1) = the Olonetsk dialect of Karelian 1; Karelian (O2) = the Olonetsk dialect of Karelian 2.

20Here I omit the initial preconsonantal contrast [tmj] • [tjmj]; in some languages (e.g. Belorussian, Polish, and Irish) palatalized coronal stops surface as alveolar or postalveolar ; the presence of the Scots Gaelic palatalized coronals and velars before plain consonants is restricted to certain grammatical forms (Ball & Fife 1993: 163). 206 TORONTO WORKING PAPERS IN LINGUISTICS

Five out of six predicted types are attested. Note that genetic affiliation of languages is of no importance here: related languages can belong to different types, and unrelated languages may be of the same type. What all of them have in common is the distribution of the plain-palatalized contrast in accordance with the predictions made by the contextual markedness hierarchy. The distribution of the contrast in labials and velars is illustrated in Tables 11 and 12.21 No languages in the sample belong to types 1 and 5.

Table 11. Labial stops type language family V__V #__V V__# V__C C__C or apja • pja • pa apj • ap apjta • arpjta • group apa apta arpta 1 ??? yes yes yes yes yes 2 Bulgarian(NN) Slavic yes yes yes yes no Russian Slavic 3 Irish Celtic yes yes yes no no 4 Bulgarian (S) Slavic yes yes no no no Polish Slavic Belorussian Slavic Nenets Uralic 5 Lithuanian Baltic yes no no no no 6 Karelian Uralic no no no no no Erzya Mordva Uralic Ukrainian Slavic Slovene Slavic Scots Gaelic Celtic Manx Celtic Welsh Celtic

One of the generalizations that is apparent from these tables is that if labials or velars maintain the plain-palatalized contrast in a given environment, coronals will certainly do so. Thus, the attested patterns are subject to the harmonic place markedness hierarchy (Prince&Smolensky 1993), translated here in terms of palatality (50).

21Palatalized labials in Russian can hardly be considered fully contrastive in the medial preconsonantal environment, since they occur before palatalized consonants and in very limited grammatical contexts with respect to labials (See 2.2.5); the sequence pjV may be allowed; the same is true for Type 6, e.g. Ukrainian and Scots Gaelic. ALEXEI KOCHETOV 207

Table 12. Velar stops type language family V__V #__V V__# V__C C__C or akja • kja • ka akj • ak akjta • arkjta • group aka akta arkta 1 ??? yes yes yes yes yes 2 Bulgarian (NN) Slavic yes yes yes yes no Scots Gaelic Celtic 3 Bulgarian (BC) Slavic yes yes yes no no Irish Celtic Manx Celtic 4 Russian Slavic yes yes no no no Belorussian Slavic Bulgarian (S) Slavic Lithuanian Baltic 5 ??? yes no no no no 6 Karelian Uralic no no no no no Erzya Mordva Uralic Nenets Uralic Ukrainian Slavic Polish Slavic Slovene Slavic Welsh Celtic

50. *Labialj, *Velarj » *Coronalj Note: Labialj = palatalized labial, Velarsj = palatalized velar, Coronalj = palatalized coronal.

Note that this hierarchy involves several possible typological patterns Table 13. I will assign integers to each pattern. These integers correspond to the types of neutralization (1-6). The first integer refers to labials, the second one represents coronals, and the fourth one characterizes velars. Labials and velars may behave similarly with respect to maintaining palatality, i.e. we have non-crucial ranking of the constraints *Labialj, *Velarj (ab). Palatalized labials and velars may show the same distribution as coronals (a), or may be more restricted in distribution of their plain counterparts than coronals (b). Labials and velars may also behave differently. In (c) velars are ranked higher than labials, since palatalized velars may contrast with plain velars in a more restricted set of environments than palatalized and plain labials. In (d) we have the reverse situation: labials are ranked higher than velars. There are no patterns where palatalized labials or velars are contrastive in more environments than coronals. 208 TORONTO WORKING PAPERS IN LINGUISTICS

Table 13. a. Ranking: *Labialj, *Velarj, *Coronalj Languages: Bulgarian (NN) (2-2-2) Irish (3-3-3) Lithuanian (4-4-4) Bulgarian (S) (4-4-4) b. Ranking: *Labialj, *Velarj » *Coronalj Languages: Belorussian (4-2-4) Erzya Mordva (6-2-6) Ukrainian (6-2-6) Karelian (O1) (6-3-6) Karelian (O2) (6-4-6) Karelian (O2) (6-5-6) c. Ranking: *Labialj » *Velarj, *Coronalj Languages: Scots Gaelic (6-2-2) Manx (6-3-3) Bulgarian (BC) (6-3-3) d. Ranking: *Velarj » *Labialj, *Coronalj Languages: Russian (2-2-4) Polish (2-2-4) Nenets (4-4-6)

I will leave other details of typological patterns for further investigation. In considering these patterns a more refined hierarchy, cluster types and their frequency, as well as other factors, should be examined. So far, however, we can conclude that all the attested language types are far from being random and they follow directly from the hierarchy based on the reference to contextual phonetic cues. More and better cues give a better preserved plain-palatalized contrast. In addition, the gaps in patterns (e.g. less common types 1 and 5) also follow from the acoustic information. The lack of the less important cue, approach, hardly affects the contrast, provided the two most important cues are present (type 5). On the other hand, the absence of both release and approach makes it difficult to preserve the contrast based solely on burst. It is even more difficult to do so in the absence of all contextual cues (type 1). In this case non-stops (affricates, fricatives, and sonorants) are better off, since they have salient internal cues. They are more likely to be found in a context like this than stops. The key to the universal preference of coronal as the most contrastive place of articulation in terms of palatality is also to be found in the phonetic detail: a coronal stop has the most long and perceptually salient strident burst. Labials and velars are not as well phonetically equipped and thus they are more commonly neutralized in the same environments. Thus, the universal markedness hierarchy (49) receives phonetic motivation rather than being a stipulative axiom.

4. Conclusion

The analysis of neutralization of plain-palatalized contrasts presented in this paper has provided evidence for the hypothesis of Licensing by Cue. It has demonstrated ALEXEI KOCHETOV 209 that palatalized consonants can be contrastive or neutralized depending not on syllabic position but rather on the availability of phonetic auditory information, and particularly the contextual cues of release, burst, and approach. The contrast is more likely to be preserved in an environment that provides more contextual cues. If a language distinguishes the contrast in a less-cued environment, it will also preserve it in all more-cued contexts of the hierarchy. The analysis is supported by the distribution of palatalized segments in Russian as well as in other related and non- related languages. The patterns attested in these languages are strikingly similar and can be legitimately considered universal. This analysis demonstrates that phonetic information plays important role in phonology, at least in the distribution of phonemic contrasts. The reference to phonetic markedness in the distribution of palatalized segments opens many directions for research. Among them are contextual enhancement of palatalized segments, types of clusters, restrictions on the contrast before certain vowels, regressive and progressive palatality assimilation patterns and their relation to language-particular gestural overlap, the role of morpheme boundaries in neutralization, etc. One of the important questions that arises from this work is the distribution of true palatals compared to palatalized consonant. This requires an investigation of acoustic and articulatory properties of these segments. Preliminary observation suggests that the distribution of palatal stops is subject to the same contextual constraints as that of palatalized stops. Consider the distribution of the (pre)palatal /c/ in Australian aboriginal languages (based on the data provided in Hamilton 1997) (Table 14).22

Table 14. Distribution of the palatal stop /c/ in Australian aboriginal languages environments cues number of languages Release » Burst » Approach V__V √√√ 82 #__V √√(√)77 C__V √√ 64 V__# (√) √√ 36 V__C √√ 31 C__# (√) √ 0 #__C √ (√) 0 Note: (√) = the cue may be available only in connected speech.

Eighty two languages in the sample have /c/ in their inventories. It occurs intervocalically in all of the languages, since all contextual cues are present. With the diminishing of cues fewer languages tend to contrast the palatal segment. The availability of release seems to be of particular importance: the lack of it causes a drop from sixty four to thirty six languages. The lack of approach is apparently less

22Final and initial clusters are absent in these languages altogether, the fact that can also be deduced from the contextual markedness hierarchy (Hamilton 1997). 210 TORONTO WORKING PAPERS IN LINGUISTICS crucial. The implicational hierarchy holds here as well: e.g. all of the languages that have the contrast finally (thirty six languages), have it initially and intervocalically as well. The presence of burst also seems to correlate with the quality of the following segment in a cluster. Only 5 mono-morpheme clusters with palatal stops are permitted in these languages (68). Note that the most frequent clusters, [cp] and [ck], are strikingly similar to the ones predicted for palatalized segments, [tjp] and [tjk] (Table 15). These are the combinations that allow for a salient burst.

Table 15. cluster number of languages [cp] 30 [ck] 14 [ct] 1 [cn] 1 [cw] 1

It should be noted that some languages tend to exhibit a different distribution of palatals: these segments are more likely to be found finally rather than initially (e.g. Komi-Permyak /c/ (Lytkin 1962) or the French or Spanish palatal nasal /µ/). It is possible that the approach transitions are more important for these segments than the release (cf. Hamilton 1997, Kuznetsova 1969). Thus the hierarchy of contexts for palatals can allow for some cross-linguistic variation. This question requires further attention and theoretical investigation.

References

Abondolo, Daniel, ed. (1998). The Uralic languages. New York, N.Y.: Routledge. Avanesov, R. I. (1972). Russkoe literaturnoe proiznoshenie. Moscow: Prosveshchenie. Ball, Martin J. & James Fife, eds.. 1993. The Celtic languages. New York, N.Y.: Routledge. Boersma, Paul (1997). Inventories in functional phonology. Manuscript, Rutgers Optimality Archive. Bolla, K. (1981). A conspectus of Russian speech sounds. Cologne: Boelau. Bondarko, L.V. (1977). Zvukovoi stroi sovremennogo russkogo iazyka. Moscow: Prosveshchenie. Browman, Catherine P. & Louis Goldstein (1989). Articulatory gestures as phonological units. Phonology 6,201-252. Chekman, V.M. (1970). Gistoryja protsipastaulenniau pa tsverdostsi-miakkastsi u belaruskai move. Minsk: Navuka i tekhnika. Comrie, Bernard & Greville G. Corbett, eds. (1993). The Slavonic languages,. London, New York, N.Y.: Routledge. De Burca, Sean (1958). The Irish of Tourkmakeady, Co. Mayo: A phonemic study . Dublin: Dublin Institute for Advanced Studies. Decsy, Gyula (1966). Yurak chrestomathy. Bloomignton, Ind.: Indiana University. ALEXEI KOCHETOV 211

Derkach, M. (1975). Acoustic cues of softness in Russian syllables and their application in automatic speech recognition. In Auditory analysis and perception of speech, edited by G. Fant & M.A.A. Tatham, 349-358. New York, N.Y.: Academic Press, 1975. Dmitrienko, S.N. (1985). Fonemy russkogo iazyka. Moscow: Nauka. Farina, Donna Marie (1991). Palatalization and jers in Modern Russian phonology: An underspecification approach. Doctoral dissertation: University of Illinois at Urbana-Champaign. Flemming, Edward Stanton (1995). Auditory representations in phonology. Doctoral dissertation: University of California at Los Angeles. Halle, Morris (1959). The sound pattern of Russian. The Hague: Mouton. Hamilton, Philip James (1996). Phonetic constraints and markedness in the phonotactics of Australian aboriginal languages. Doctoral dissertation, Toronto, University of Toronto. Jones, Daniel & Dennis Ward (1969). The phonetics of Russian. Cambridge: University Press. Kochetov, Alexei (1998). Phonology and phonetics of labial palatalization. Buffalo Papers in Linguistics, 98-101. Papers presented at the Fourth Buffalo-Toronto student conference in linguistics. State University of New York at Buffalo. Kuc‡era, Henry & George K. Monroe (1968). A comparative quantitative phonology of Russian, Czech, and German. New York: American Elsevier. Kuznetsova, A.M. (1969). Nekotorye voprosy foneticheskoi kharakteristiki iavlenia tverdosti-miagkosti soglasnykh v russkikh govorakh. In Eksperimental’no- foneticheskoe izuchenie russkikh govorov, edited by S.S. Vysotskii, 35-140. Moscow: Nauka. Ladefoged, Peter & Ian Maddieson (1996). The sounds of the world’s languages. Cambridge, Mass.: Blackwell. Lazova, M.V. et al. (1974). Obratnyi slovar’ russkogo iazyka. Moscow: Sovetskaia Entsiklopediia. Leskinen, H. (1984). Karelischen. In Studien zur phonologischen Beschreibung uralischen Sprachen, edited by P. Hajdu & Laszlo Honti, 247-257. Budapest: Akadémiai Kiadó. Lytkin, V. I., ed. (1962). Komi-Permiatskyi iazyk. Kudymkar: Komi-Permiatskoe knizhnoe izdatel’stvo. MacAulay, Donald (Ed.) (1992). The Celtic languages. New York, N.Y.: Cambridge University Press. Maddieson, Ian (1984). Patterns of sounds. Cambridge: Cambridge University Press. Mathiassen, Terje (1996). A short grammar of Lithuanian. Columbus, Ohio: Slavica. Prince, Alan & Paul Smolensky (1993). Optimality Theory: Constraint interaction in generative grammar. Manuscript, Rutgers University. Pulleyblank, Douglas (1997). Optimality Theory and features. In Optimality Theory: An overview, edited by D. Archangeli & D. Terence Langendoen, 59- 101. Oxford: Blackwell. Redei, K. (1984). Erza Mordwinisch. In Studien zur phonologischen Beschreibung uralischen Sprachen, edited by Hajdu, P. & Laszlo Honti, 247-257. Budapest: Akadémiai Kiadó. Silverman, Daniel (1997). Phasing and recoverability. New York, N.Y.: Garland. 212 TORONTO WORKING PAPERS IN LINGUISTICS

Skalozub, L.G. (1963). Palatogrammy i rentgenogrammy soglasnykh fonem russkogo literaturnogo iazyka. Kiev: Izdatel’stvo Kievskogo universiteta. Steriade, Donca. (1997). Phonetics in phonology: The case of laryngeal neutralization. Manuscript, University of California at Los Angeles. Zsiga, Elizabeth C. (1997). Features, gestures and Igbo vowels: An approach to the phonology-phonetics interface. Language 73,227-274. Zsiga, Elizabeth C. (1998). Phonetic alignment constraints. Paper presented at the Linguistic Society of America Annual Meeting, New York, N.Y.