2.1 The Erzya phoneme system

Erzya has a relatively rich consonant system (Table 1).¹ From the point of view of harmony, the most important feature of the consonant system is that unpalatalised and palatalised alveolar consonants contrast.²

On the contrary, the Erzya vowel system is relatively poor, it consists of five vowels (see Table 2, but also Footnote 4). In non-initial syllables, /i/ is rare and /u/ is extremely rare – except for suffixes, where /i/ is common in verbal inflection. Mid vowels /o/ and /e/ alternate due to harmony. All the other vowels are neutral, which means that their occurrence is not restricted by the preceding vowels or consonants; however, they are opaque, i.e. they determine which of the alternating consonants and vowels follow them.

2.2 Harmony in Erzya

Erzya harmony is usually labelled as vowel harmony in the literature. As /o/ and /e/ differ not only in their backness/frontness but also in rounding, it is not self-evident at first whether front/back or rounding harmony is involved. However, as unrounded /ɑ/ is regularly followed by rounded /o/, a front/back harmony must be assumed. Additionally, the assumption that vowels undergo front/back harmony is underpinned by the fact that VH is intertwined with consonant harmony due to palatalisedness:³ while the palatalisedness feature of consonants is evidently connected to the frontness of vowels, it has nothing to do with rounding. The interference with the palatalisedness of the consonants is also generally mentioned in the literature. However, the interference is so strong that it is more appropriate to speak about vowel-consonant harmony.

• Vowels trigger vowel alternation: /kudo-so-nzo/ ‘house-ine-poss.3sg’ : /vel^je-se-nze/ ‘village-ine-poss.3sg’.

• Consonants trigger consonant alternation: /kɑl-t/ ‘fish-pl’ : /kɑl^j-t^j/ ‘willow-pl’.

• Vowels trigger both consonant and vowel alternation: /kudo-vtomo/ ‘house-abe’ : /vel^je-vt^jeme/ ‘village-abe’.

• Consonants trigger both consonant and vowel alternation: /kɑl-do/ ‘fish-abl’ : /kɑl^j-d^je/ ‘willow-abl’.

• Vowels trigger consonant alternation: /kudo-t/ ‘house-pl’ : /vel^je-t^j/ ‘village-pl’.

• Consonants trigger vowel alternation: /kɑl-on^j/ ‘fish-gen’ : /kɑl^j-en^j/ ‘willow-gen’ or /kɑl-so/ ‘fish-ine’ : /kɑl^j-se/ ‘willow-ine’.

The general rule is that harmony proceeds from left to right: once a vowel is front or an alveolar is palatalised, the subsequent mid vowels are front and the subsequent alveolars are palatalised. However, palatalised consonants and front vowels do not only occur due to harmony in non-initial syllables:

• Palatalised alveolars can also occur after back vowels: /kɑl^j/ ‘willow’, /kɑl-on^j/ ‘fish-gen’.

• High /i/ can also occur after back vowels and unpalatalised alveolars: /udi/ ‘sleep-prs.1sg’, /vɑltne/ ‘word-def.pl.nom’.

There are also some exceptions to vowel harmony.

• Alveolar sibilants and other alveolars in a cluster with them do not undergo harmony, but they are are transparent for vowels: /vel^je-se-nze/ ‘village-ine-poss.3sg’, /kɑl^j-se/ ‘willow-ine’.

• Stem-internal unpalatalised alveolars are transparent for vowels, but not for consonants: /stɑl^jin-de/ ‘Stalin-abl’.

• Back low /ɑ/ is opaque for harmony: /vel^je-vɑ-nzo/ ‘village-prol-poss.3sg’.

• High front /i/ is opaque for harmony: /t͡ʃumot͡ʃi-n^jek/ ‘sin-poss.1pl (sinner-nmlz-poss.1pl)’ (cf. /t͡ʃumo-nok/ ‘sinner-poss.1pl’).

• High back /u/ is opaque for harmony, but only appears in some foreign words: /inst^jitut-so/ ‘institute-ine’.

Harmony is also present stem-internally as a tendency, cf. Fejes (2021a) for details.

Additionally to the vowel-consonant harmony sketched above, there is also a syllable-internal harmony in Erzya: front vowels following unpalatalised alveolars are retracted: /sij/ [si̠j] ‘pus’ (cf. /s^jij/ [s^jij] ‘louse’), /seme/ [se̠me] ‘tinder (fungus)’ (cf. /s^jeme/ [s^jeme] ‘brush’). As syllable-internal harmony does not interfere with vowel-consonant harmony, it will be ignored in the following discussion.⁴

2.3 Nominal and verbal morphology

In Erzya, inflectional morphemes occur to the right of the stem, and prefixing is extremely marginal. In Erzya declension, there are 12 cases,⁵ there are indefinite, definite and possessive forms, while singular and plural are only differentiated in definite and some possessive forms, but, except for the nominative, not in the indefinite subparadigm. Additionally, nouns and adjectives can get verbal suffixes: such forms function as predicative forms (the role of the suffixes resembles the function of the copula in other languages). Although Erzya is generally labelled as an agglutinative language, affixes are not clearly segmentable. As Table 3 illustrates, one cannot segment a consistent marker for definiteness, plural or some of the cases (e.g. illative),⁶ nor even one that alternates only due to phonological processes.

Erzya verbs agree with the subject in person and number (although many syncretic forms occur); they have indeterminative and determinative forms, the latter agreeing with the object in person and number; they occur in four tenses (present, two pasts, future) and seven moods (indicative, imperative, optative, conjunctive, conditional, conjunctive-conditional, desiderative), although most tenses and moods cannot be combined with each other. Similarly to the declensions, suffixes for different conjugational categories can hardly be segmented. Table 4 shows some forms from a single paradigm. When the forms within the same person and number are contrasted, while /i/ seems to mark past tense in the first and second person forms, it appears only in the present tense in the third person forms. When comparing forms within the same tense, the suffix of the third person singular seems to be /-i/ in the present, /-s^j/ in the past (without an explicit tense marker) and zero in the second past tense. The situation is even more complicated if other conjugation types are considered as well.

Although some markers (especially markers of rarer moods) are much better segmentable, in this study, the complex of the grammatical markers attached to the stem will be labelled as terminations. Thus every word form consists of a stem and a termination. Nonetheless, in some cases, segments which tend to regularly occur in given forms are sometimes referred to as suffixes or markers, even when the exact placement of their boundaries is debatable.

Additionally, even the boundary between the stem and the termination can be debated. As for declension, Rueter (2010, 15–16) points out that different schools tend to analyse the very same segments attested at the morpheme boundary differently: some authors attach these segments to the stem, others to the suffix, while some others consider it as a kind of linking element neither belonging to the stem nor to the suffix. In this study, following Rueter (2010, 71–72), these segments are analysed as belonging to the suffix. In addition to being a decision of principle, this solution serves practical reasons: if these segments were analysed as belonging to the stem, their behaviour in harmony would be missed by our analysis, which is restricted to the suffixes; and their analysis as independent segments would complicate the analysis without any reason or advantage.

The case of conjugation is a bit more complicated. The key data are presented in Table 5. It is clear that there are strings which are present word-initially in all forms of the verb, as /sod-/ in the case of ‘know’ or /t^jej-/ in the case of ‘do’. Similarly, there are also strings which are always present word-finally in the given form of all the verbs, e.g. /-ms/ in the infinitive, /-ɑn/ in the first or /-i/ in the third person singular in the present indicative. However, at least in the infinitive, there is a vowel between the string that is present in all forms of the verb and the termination that is present in the given form of any verb. At first glance, it is not clear where the given vowel belongs.

The fact whether the vowel is low or mid seems to be completely lexicalised (the frontness/backness of the mid vowels is determined by vowel harmony, although there are some marginal exceptions, cf. Fejes 2020, 43–44). Therefore, it is reasonable to suppose that at least either low or mid vowels (or both) are part of the stem, rather than being determined by a morpholexical feature of the stem.⁷ Especially in the case of stems with the low vowel, which is present in most of the forms, it is well-grounded to analyse the forms as if the vowel is the part of the stem, but it is deleted before suffixes beginning with a vowel, such as /-ɑn/ or /-i/.

As far as mid vowels are concerned, there are two types of forms: in one type, as in the infinitive, the mid vowel is always present before the /-ms/ string. However, in the other type, such as the third person singular past indicative forms ending in /-s^j/, the mid vowel appearing before /-ms/ in the infinitive form is present just in some cells of the paradigm. One solution could be that all stems end in a consonant, the termination of the infinitive is /-Oms/ (in which /O/ stands for the alternating mid vowel which is dropped after /ɑ/-final stems); while before the termination of ind.pst.3sg /-s^j/, the mid vowel is only present (or missing) if the stem is marked as such. Nonetheless, in such an analysis it remains unclear where the vowel appearing with some stems before the segment /-s^j/ belongs: is the stem or the suffix extended with the vowel? Another possibility is to assume similarly that the ind.pst.3sg termination is /-s^j/, and if it is preceded by a vowel, that vowel is part of the stem. When the infinitive termination /-Oms/ is attached to such a vowel-final stem, its vowel is dropped, similarly to when it is attached to an /ɑ/-final stem.

Although there is no generally accepted segmentation of verb forms (cf. Rueter 2010, 69–70), a segmentation that is similar to the one proposed here is used in Serebrennikov & Buzakova (1993), Mészáros (1998) or Rueter (2010), so it is at least not completely incompatible with the Erzya grammatical tradition.

2.4 Previous studies on Erzya suffix alternation

Evidently, the descriptions of Erzya usually deal with suffix alternation due to harmony in some way, cf. Bubrix (1953, 36), Bondarko & Pol'akova (1993, 94–95), Keresztes (1990, 37), Bartens (1999, 66–67), Keresztes (2011, 22–23), etc. These just mention the most salient regularities of Erzya harmony, and many details remain obscured. There is at least one attempt at its formal description (Koizumi 1980 – in an SPE-like framework, for its criticism, see Fejes 2022). However, there are not many studies on Erzya morphophonology from the point of view of suffix alternation.

Maticsák (2003) seems to be the only study in which suffix alternation takes centre stage. However, this research is restricted to the alternations of the case suffixes (in the indefinite subparadigm) due to the phonological characteristics of the stem; furthermore, it does not examine the type of alternation, only the fact whether the following circumstances cause suffix alternation:

• whether the stem ends in a vowel or a consonant;

• whether the stem ends in a single consonant or a consonant cluster;

• whether the last vowel of the stem is front or back;

• whether the final consonant is palatalised or not;

• whether the final consonant is voiced or voiceless.

Rueter (2010, 62–66) gives a useful overview of the range of harmony in suffixes – however, similary to many others, he treats it as two distinct phenomena, vowel harmony and palatal harmony. In the cases of vowel harmony, he examines whether affix-initial, affix-internal and affix-final harmony is attested in a given suffix. He also distinguishes another type, stand-alone harmony for the locative suffix /-o/ : /-e/. Although Rueter (2010, 92–93) considers this suffix a case marker, this categorisation is hardly tenable because of the very restricted productivity of the suffix. It appears only with stems ending in /n, m, l/ or /r/, it does not combine with plural or possessive markers, and mostly appears after adverbial and postpositional stems. The only inflectional suffix consisting of one mid vowel (/-o/ : /-e/) is the second person singular imperative ending. However, it could also be analysed as a suffix with both affix-initial and affix-final alternation. In the case of palatal harmony, he examines whether the harmony trigger can be a(n affix-)preceding vowel or consonant. The picture is quite simple here: a preceding vowel can always be a trigger, while a preceding consonant can be one only when the suffix begins with a(n alternating) consonant. Nonetheless, Rueter (2010, 65) also points out the specific behaviour that is labelled as inverse alternation below (cf. Section 5.3).

3.1 Consonants

Consonants are analysed in three groups depending on their behaviour in suffixes. Affricates, laterals and trills, discussed in Section 3.1.1, are very rare in suffixes; therefore, it is difficult to make generalisations on their behaviour. Sibilants, which form unpalatalised–palatalised pairs, explored in Section 3.1.2, could potentially undergo harmony, but they never do. Finally, stops, analysed in Section 3.1.3, potentially undergo harmony, and generally – although not always – they alternate in suffixes due to harmony. In Section 3.1.4, the findings about the consonant alternations (and their lack) are summed up.

3.2 Vowels in suffixes

As it was mentioned earlier, only mid vowels undergo harmony. Even these avoid alternation when they are preceded by opaque vowels or consonants.

Non-alternating /o/ occurs when there is /ɑ/ in the nucleus of the preceding syllable: in third person singular and first person plural in the prolative (/-vɑnzo/ : /-kɑnzo/ : /-ɡɑnzo/, /-vɑnok/ : /-kɑnok/ : /-ɡɑnok/) and comparative case (/-ʃkɑnzo/, /-ʃkɑnok/); the first and second person plural suffixes in present indicative (/-tɑno/ : /-t^jɑno/, /-tɑdo/ : /-t^jɑdo/), conditional (/-in^jd^jer^jɑtɑno/, /-in^jd^jer^jɑtɑdo/) and optative (/-ztɑno/ : /-oztɑno/ : /-eztɑno/, /-ztɑdo/ : /-oztɑdo/ : /-eztɑdo/); third person singulart and plural in the conditional conjunctive (/-in^jd^jer^jɑvol^j/, /-in^jd^jer^jɑvol^jit^j/). In all these cases, there is an unpalatalised alveolar (/t, d/ or /n/), a cluster of unpalatalised alveolars (/nz/) or a non-alveolar (/v/) consonant between /ɑ/ and the non-alternating /o/.

Non-alternating /e/ usually occurs when there is an /i/ in the nucleus of the preceding syllable: in the indeterminative subparadigm, first and second person in the past indicative (/-in^jek/, /-id^je/), second past indicative (/-il^jin^jek/, /-il^jid^je/), past conditional (/-in^jd^jer^jin^jek/, /-in^jd^jer^jid^je/), conditional conjunctive (/-in^jd^jer^jɑvl^jin^jek/, /-in^jd^jer^jɑvl^jid^je/) and in all persons and numbers in desiderative (1sg /-iksel^jin^j/, 2sg /-iksel^jit^j/, 3sg /-iksel^j/, 1pl /-iksel^jin^jek/, 2pl /-ikseil^jid^je/, 3pl /-iksel^jt^jek/). Additionally, several similar cases occur in the determinative subparadigm, e.g. prs.3sg>3sg /-siz^je/), prs.3sg>3pl /-sin^jz^je/), prs.1pl>3 /-sin^jek/), etc. As the examples show, the preceding /i/ and the non-alternating /e/ are usually separated by non-alternating palatalised consonants (/n^j, d^j/ or /z^j/), clusters of non-alternating palatalised consonants (/n^jd^j/ or /n^jz^j/), or exceptionally by the cluster /ks/, consisting of a non-alveolar consonant and an alveolar sibilant (which always avoids alternation due to harmony).

Non-alternating /e/ also occurs following an inversely alternating alveolar /n^j/ : /n/ (indf.dat /n^jen^j/ : /nen^j/) or cluster /n^jt^j/ : /nt/ (def.sg.dat/ill /n^jt^jen^j/ : /nten^j/) and /t^jn^j/ : /tn/ (e.g. def.pl.nom /t^jn^je/ : /tne/). In these cases, the termination-initial alveolar consonant(s) get(s) unpalatalised due to an immediately preceding unpalatalised alveolar. However, the vowel is front even when no preceding vowel can result in its fronting (e.g. /pɑz-nen^j/ ‘god-dat’). No other cases were found where non-alternating /e/ could occur “in its own right” or even because it is preceded by a consonant that is palatalised “in its own right” (although verbal terminations 3sg>3sg and 3sg>3pl, which end in /iz^je/ and /in^jz^je/ respectively, are worth considering). For further discussion, see Section 5.3.

The mid vowel alternates:

• termination-initially in the indefinite singular genitive ((/-n^j/ :) /-on^j/ : /-en^j/), lative ((/-v/ :) /-ov/ : /-ev/, cf. Footnote 5), the possessive suffixes in nominative singular 1sg ((/-m/ :) /-om/ : /-em/), 2sg ((/-t/ : /-t^j/ :) /-ot/ : /-et^j/), 3sg ((/-zo/ : /-ze/ :) /-ozo/ : /-eze/), etc.;

• following a termination-initial alternating alveolar in the indefinite singular ablative (/-to/ : /-t^je/ : /-do/ : /-d^je/), first person plural possessive suffix in the nominative singular (/-nok/ : /-n^jek/ (: /-onok/ : /-en^jek/));

• following a termination-initial cluster containing a non-alveolar and an alternating alveolar (/vt ː vt^j/): abessive ((/-tomo/ : /-t^jeme/ :) /-vtomo/ : /-vt^jeme/);

• following a termination-initial unpalatalised alveolar sibilant (/s, z/) in the indefinite singular inessive (/-so/ : /-se/), illative in the possessive subparadigm, e.g. first person singular (/-zon/ : /-zen/ (: /-ozon/ : /-ezen/)), the third person singular optative in the indeterminative subparadigm (/-zo/ : /-ze/ (: /-ozo/ : /-eze/)) and with a third person singular object in the determinative subparadigm (/-so/ : /-se/ (: /-oso/ : /-ese/));

• following a termination-initial cluster containing an unpalatalised alveolar sibilant (and possibly also an unpalatilised alveolar or a non-alveolar consonant) in the indefinite singular elative (/-sto/ : /-ste/), definite singular translative (/-kson^jt^j/ : /-ksen^jt^j/) and similar forms with possessive suffixes (e.g. first person singular /-kson/ : /-ksen^j/), third person singular possessive suffix in the nominative plural (/-nzo/ : /-nze/ (: /-onzo/ : /-enze/)), the syncretic elative infinitive and past gerund (/-msto/ : /-mste/ (: /-omsto/ : /-emste/));

• following a termination-initial non-alveolar consonant in the nominative infinitive (/-mo/ : /-me/ (: /-omo/ : /-eme/));

• following an alternating mid vowel with an alternating alveolar in between in the nominative singular first person plural possessive suffix and termination-finally in the oblique cases (/-nok/ : /-n^jek/ (: /-onok/ : /-en^jek/)), in the imperative second person plural termination (/-do/ : /-d^je/ (: /-odo/ : /-ed^je/));

• following an alternating mid vowel with an unpalatalised alveolar sibilant in the illative in the possessive subparadigm, e.g. first person singular ((/-zon/ : /-zen/ :) /-ozon/ : /-ezen/), the third person singular optative in the indeterminative subparadigm ((/-zo/ : /-ze/ :) /-ozo/ : /-eze/) and with a third person singular object in the determinative subparadigm ((/-so/ : /-se/ :) /-oso/ : /-ese/);

• following an alternating mid vowel with a cluster containing an unpalatalised alveolar sibilant in the third person singular possessive suffix in the nominative plural and termination-finally in the oblique cases ((/-nzo/ : /-nze/ :) /-onzo/ : /-enze/), the syncretic elative infinitive and past gerund ((/-msto/ : /-mste/ :) /-omsto/ : /-emste/);

• following an alternating mid vowel with a non-alveolar consonant in the abessive throughout the paradigm (/-tomo/ : /-t^jeme/ : /-vtomo/ : /-vt^jeme/), the conjunctive third person singular ((/-vol^j/ :) /-ovol^j/ : /-evel^j/ and plural (/-ovol^jt^j/ : /-evel^jt^j/ and in the nominative infinitive ((/-mo/ : /-me/ :) /-omo/ : /-eme/).

To sum up: vowels alternate termination-initially or when there is an alternating vowel in the preceding syllable and either an alternating alveolar or a non-alveolar consonant (or a cluster containing only non-alveolar consonants) between them. Another possibility is that there is a non-alternating unpalatalised alveolar consonant (alone or in a cluster) between the two alternating vowels: however, it occurs only if the consonant is a sibilant (or, in the case of the cluster, one of them is a sibilant). Mid vowels do not alternate if there is a non-alternating vowel (high /i/ or low /ɑ/) in the preceding syllable. Theoretically, it could happen that a termination begins with a syllable or more with alternating vowel(s), but then non-alternating front /e/ occurs because of a palatalised alveolar. However, similar cases are not attested among the inflectional terminations.

5.1 The specific behaviour of sibilants

The specific behaviour of sibilants needs an explanation irrespective of the chosen framework, but it is salient in an autosegmental description. The fact that sibilants do not undergo harmony could be easily explained if palatalised sibilants did not exist: in this case, it could be supposed that J spreads in them but is not spelled out. The fact that palatalised sibilants are attested but never as a result of harmony suggests that their palatalisedness is different. This would mean that the palatalisedness of the sibilants is represented as a feature different from J (that is, a different tier). If it were not different, it could not be explained why J does not spread to sibilants, or why the spread J is not spelled out.

Moreover, the fact that all unpalatalised alveolars (both sibilants and stops) are transparent for frontness harmony suggests completely different tiers for vowels and consonants. Although this is technically possible, the one-to-one correspondence of phonological features and phonetic (articulatory) gestures is preferred in phonological theory. Additionally, the fact that sibilants assimilate other alveolar consonants in palatalisedness (/ver^jɡiz-de/ ‘wolf-abl’ but /r^jivez^j-d^je/ ‘fox-abl’) and vowels in frontness (/pɑz-on^j/ ‘god-gen’ but /ɑrɑz^j-en^j/ ‘shortage-gen’) suggests that the very same feature is involved.

An additional possibility is to assume a difference in feature geometry. Even if it is supposed that the feature geometry does not allow the spreading of J on the consonant if the feature for the friction (narrow channel) is present – or if the feature for the stop (block) is absent – but allows the spreading of J even from such an alveolar to another or a mid vowel), it is difficult – if not impossible – to identify independent phonological facts which support such an analysis. Consequently, such an assumption would be circular reasoning. Furthermore, the supposed feature geometry could be convincing if supporting phenomena were found in other languages as well. However, it seems that there is no empirical evidence for the specific behaviour of sibilants in palatalisation cross-linguistically.¹¹

One could argue that the specific behaviour of sibilants can be explained by historical developments. It is well-known that while original palatalised consonants were usually depalatalised in a back-vowel environment, unplatalised alveolars went through palatalisation sometimes between Proto-Uralic and Proto-Mordvinic, sibilants did not go through these processes (cf. Bartens 1999, 33–37; Keresztes 2011, 50–54; Zhivlov 2023, 128–129, etc.). Nonetheless, referring to this historical change, it is also necessary to explain the specific behaviour of the sibilants. Therefore, this kind of argumentation would only shift the issue from the present to the past. Although one could argue that the opposition of palatalisedness was functionally loaded much more for sibilants than for other consonants, and exactly that blocked their merger, it would not completely explain their current behaviour. Although many instances of syncretism are attested in Erzya paradigms, their number would not be seriously increased if currently non-alternating unplalatalised sibilants underwent harmony. However, such a development is not reported from any dialect.

5.2 Regressive depalatalisation

As the examples discussed above show, J assimilation is always progressive. However, the non-alternation of /nz/ in 3sg (as in nom.pl.3sg and many others) /-(o)nzo/ : /-(e)nze/ seems to contradict this assumption. In such examples, it would be expected that J spreads on /n/, but then cannot spread forward to /z/, so the expected alternation would be /-(o)nzo/ : */-(e)n^jze/. Therefore, it must be either supposed that */-(e)n^jze/ emerges at some point in the dervation, but later */n^j/ loses its palatalisedness due to regressive assimilation in a subsequent phase; or that the sibilant blocks the spread of J on /n/ in some obscure (but clearly regressive) way. However, there are no other cases when regressive J assimilation is involved. Moreover, no regressive assimilation is attested across morpheme boundaries: /lomɑn^j-se/ (*/lomɑn-se/) ‘human-ine’.

The case of the /n^jz^j/ cluster must be discussed here. As it occurs following an /i/ (prs.3sg>3pl /-sin^jz^je/, pst.3sg>2sg /-in^jz^jit^j/, pst.3sg>3pl /-in^jz^je/, etc.), the palatalisedness of /n^j/ is not surprizing. However, the palatalisedness of /z^j/ is unexpected, as sibilants do not undergo this process. As a consequence, it must be assumed that J is lexically linked to the sibilant /z^j/, which does not block the spread of J on the preceding nasal.

5.3 Inverse alternations

In Erzya, two types of suffix alternations should be distinguished. One of these is normal alternation, when J is not linked lexically to the (initial) segment of the suffix. However, even in this type of alternation, J can be lexically linked to a farther phoneme such as in the case of the genitive /-n^j/ : /-on^j/ : /-en^j/ or the conjunctive third person singular (/-vol^j/ : /-ovol^j/ : /-evel^j/). In such cases, further phonemes of the suffix can be palatalised/fronted when J spreads on them. The other type is the inverse alternation, in which J is lexically linked to the termination-initial phoneme (always a consonant), but can be unlinked in certain circumstances: namely, when the stem-final consonant is an unpalatalised alveolar.

Table 6 demonstrates the differences between the two kinds of alternation through the example of the normally alternating abessive and the inversely alternating dative suffixes. The forms of ‘village’ do not show any differences: both suffixes begin with a palatalised alveolar. It is expected that J spreads to the suffix if the feature does not come with the termination-initial consonant underlyingly. Similarly, the forms of ‘wolf’ suggest that J cannot spread to the termination-initial consonant because of the stem-final unpalatalised alveolar. However, this cluster is transparent for the vowels, so the mid suffix vowels are expected to be front – and they are.

The forms of ‘house’ show the difference between the two suffixes. As J cannot spread from the stem, the the alveolar consonants and the mid vowels of the abessive suffix remain unpalatalised and back, respectively. However, the alveolar consonants and the mid vowel of the dative suffix are palatalised and front, respectively, which suggests that J is linked to the termination-initial consonant (and spreading forward to the right).

The forms of ‘god’ provide the problem for analysis. The abessive form is unproblematic, it can be explained like the abessive form of ‘house’. However, the dative form of ‘god’ is a real puzzle. Although the stem cannot spread J on the suffix, the termination-initial nasal should be palatalised, as J is linked to it underlyingly. However, it is unpalatalised, and a straightforward explanation is that J was unlinked due to the preceding sibilant, which does not allow an alveolar in the same cluster to be palatalised. But if J cannot spread on the mid vowel either from the stem or from the termination-initial consonant, why is it /e/ and not /o/ that is attested in the dative suffix following stems like ‘god’?

A simple answer could be that J is also associated to the vowel. However, this solution contradicts to a basic principle of Autosegmental Phonology, namely the Obligatory Contour Principle, which “prohibits consecutive or adjacent identical segment” (Goldsmith 1990, 309). This principle eliminates ambiguity in the analysis: if it were allowed to link two features to two neighbouring elements, it could not be distinguished from the case when one feature is linked to both of them. Additionally, there are internal arguments, coming from the facts of Erzya, to reject such an analysis. Non-alternating /e/ (in addition to cases when it is preceded by a phoneme which can be a source of J spreading) only occurs in inverse alternation, there are no cases when non alternating /e/ occurs termination-initially, following a termination-initial unpalatalised alveolar or a non-alveolar consonant, or following an /ɑ/ in the preceding syllable without a palatalised alveolar in between. This argument against /e/ being front “in its own right” holds independently of the framework chosen.

Another possibility is to assume that although J is delinked from the nasal but does not remain unlinked, it is linked to the following element, to the vowel. Although this solution is practically possible, this is a new element in the analysis: up to this point, it will be supposed that only those features are spreading which are linked to the same timing point both underlyingly and the surface. Additionally, there is no other phenomenon which could support this analysis, so this argument would be circular reasoning again.