DIALECTAL VARIATION OF UDMURT DISCOURSE CLITICS

. Udmurt has numerous discourse clitics. Often multiple clitics attach to the same host word, forming clusters. Clitic clusters in European languages have been successfully modeled with clitic templates. In this paper, I am considering ordering of discourse clitics across Udmurt dialects. I demonstrate that there is significant variation between dialects in this respect, as well as some free variation within varieties. While the template approach, with certain caveats, can be applied to each individual Udmurt variety, it cannot be used to model the entire language. I show that the order in clitic clusters is mostly idiosyncratic. In particular, it does not depend on the semantics of the particles or the order of their appearance in the language. Apart from that, I touch upon the peculiar situation whereby certain discourse functions are expressed by particles of completely different origin in different Udmurt dialects. I argue that this is due to a Jespersen’s cycle-style transformation, when an existing particle is gradually replaced by a new one through an intermediate stage, when both are used simultaneously.


Introduction
Although my investigation has one subject, it has a variety of objectives.In terms of its subject, this paper is dedicated to certain enclitics and their combinations (clusters) in the dialects of Udmurt.An example of such a cluster can be seen in ( 1).
leave-PST-3SG miss-FUT.1SG=ENIM=already=REPnow 'Džakyapaj, the old lady from Uddjadi, left the concert with tears in her eyes, [saying] she is going to miss [the visitors from Finland she met] already now.' There are several caveats that have to be made right away.First, as you will see from the list in Table 1, the clitics in question are rather heterogeneous, and not all of them can be unambiguously called discourse (or modal, or pragmatic) particles in the usual sense (Zimmermann 2011).Nevertheless, each of them has some clearly pragmatic functions in at least one of its uses.Uniting them under that umbrella term is thus not outright incorrect and has an advantage of making it clear right from the start that the combinations I am referring to are not pronominal clitic clusters.While the latter are common in European languages and have been extensively researched (see e.g.Bonet 1991;Cardinaletti 1999;Hellan, Platzack 1999;Bošković 2004), far less is known regarding the former.The only item on my list that probably cannot be described as a discourse clitic at all is the question marker a.I am nevertheless including it because it interacts with discourse clitic in clusters much in the same way those clitics interact with each other.
Another caveat is that I remain agnostic as to the part-of-speech characterization of the words in focus, which have been traditionally treated as adverbs, conjunctions and/or particles.When calling them discourse particles, I do not imply that "particle" is their part of speech, or indeed that there are good reasons to have "particles" on the part-of-speech list at all (cf. Zwicky 1985).Neither do I imply, by calling them simply "clitics", that clitics might be treated as a category of its own (Schachter, Shopen 2007 : 52-53) or considered acategorial (Zwicky 1985).
Finally, there has been a lot of discussion in the theoretical literature, both formal and functional, as to what should count as a clitic and whether "clitic" is a good cross-linguistic category or comparative concept (Zwicky 1977;1985;Anderson 2005;Spencer, Luís 2012;Haspelmath 2015;2022).However, Udmurt elements seem to conform to any definition of clitic ever given, including the most recent proposal by Haspelmath (2022) ("a clitic is a bound morph that is neither an affix nor a root").On the one hand, they cannot be used in isolation and always form one phonological word with their host.On the other hand, they do not have selectional preferences like affixes do, and can attach to words of any part of speech.Therefore, calling them clitics is uncontroversial, regardless of the definition one adopts.While I have no preference for any specific definition, what is important to me in their clitichood status is that the way they are positioned inside a cluster can be a priori expected to differ from both the way autonomous words are ordered in a sentence and the way affixes are ordered inside a word.
As to the objectives, there are several.First, I am going to find out which enclitic particles and particle clusters are available across Udmurt dialects.In a sense, this study is aligned in its goals and methods with the recently emerged field of variational pragmatics (Schneider, Barron 2008;Schneider 2010;2021;Aijmer 2013).The objective of this approach is, according to Schneider (2010 : 239), to add "the pragmatic level to the other language levels overwhelmingly analyzed in dialectology".Specifically, the first objective of my investigation is subsumed under the study of dialectal variation on the "formal" level of pragmatic analysis (Schneider 2008 : 20).The target variables in my case are clitics (or absence thereof) that are employed in certain pragmatic functions.Out of the main macro-social factors whose influence on the target variable is usually studied in variational pragmatics, I am focusing on the region (or dialect) of the speaker.Although I did not study the other usual factors (social class, ethnicity, gender, age) systematically, they seem to play a much smaller role, if any at all.Inter-speaker variability in my data is only witnessed in the settle-ments located on borders between contiguous dialectal areas.As to the methods, my investigation satisfies the principles of empiricity (I study language use in corpora and through fieldwork), contrastivity (I compare multiple varieties of the same language) and, although to a lesser extent, comparability (Schneider 2021 : 672-673).
However, my primary focus is not on the semantics of discourse particles, and thus not on the pragmatic level in the strict sense.Instead, I am mainly interested in the syntactic properties of the clitics, namely their position in the sentence and their position relative to each other when multiple clitics attach to the same host word.The second objective of my research is to establish the factors that may or may not affect clitic placement, namely, influence of contact languages, order of their appearance in the language and their meaning.I will only discuss the semantic side when it is relevant for this objective.
Finally, I am going to discuss why different dialects have clitics of different origin that apparently have the same non-trivial meaning.I will demonstrate how dialectal data can shed light on the historical developments that might have resulted in the observed distribution.
Udmurt is a language spoken by around 300,000 people who mostly live in the Russian region of Udmurtia and certain (mostly adjacent) parts of the neighboring regions.Udmurt was only standardized in the 1930s, and the traditional dialects are alive and spoken in Udmurt villages on a daily basis.Traditionally, Udmurt has been divided into four supradialects: Beserman, Northern, Central, Southern.The latter presents a rather sharp internal divide between Southern proper, spoken mainly inside Udmurtia (further referred to as Southern) and Southern Peripheral, spoken mainly in Tatarstan and Bashkortostan (Kelxmakov 1998).Although Beserman has been recently recognized as a separate language in Russia,1 it is actually rather close to, and easily mutually intelligible with, the Northern varieties, which is why treating Beserman on a par with Udmurt dialects as parts of an overarching languoid (Cysouw, Good 2013) makes sense.Dialectometric analysis (Arhangelxskij 2021) shows that most other supradialectal borders, unlike the ones between smaller units, are indeed rather strong and obvious.Although standard registers (standard written Udmurt and the common urban vernacular) have exerted some influence on the dialects through education, books and mass media, most Udmurt speakers never fully master them (Edygarova 2014).Udmurt has long been in contact with Tatar (Turkic > Kipchak), which has had a significant influence on Udmurt (Agyagási 2012), with southern varieties affected much more than the northern ones.Recently, Russian has replaced Tatar as the dominant language.
My investigation is based on the data I collected during two field trips in 2021-2022, as well as on several corpora and published dialectal texts.The field data consists of questionnaire sentences orally translated from Russian into dialectal Udmurt, as well as some monologue and dialogue recordings.During the 2021 field trip, I collected data in 10 settlements in Udmurtia, Tatarstan and Bashkortostan, working with multiple speakers in each place.In 2022, I worked with the members of the Estonian Udmurt community in Tartu and Tallinn, who represented 13 local varieties in total (mostly one speaker per settlement).This data was supplemented with relevant examples from the transcribed texts published by KelÍ makov (Kelxmakov 1981) and Nasibullin (Nasibullin 1981), as well as from the Corpus of Standard Udmurt,2 the Beserman Multimedia Corpus3 and the Corpus of Tatyshly Udmurt. 4he paper is structured as follows.In Section 2, I introduce the subset of Udmurt discourse particles that are the subject of my investigation.In Section 3, I demonstrate how the factors of scope and position inside the sentence can affect clitic placement inside a cluster.In Section 4, the dialectal distribution of clitic templates and variation is covered.Section 5 discusses historical processes that might have led to the observed distribution of discourse particles in Udmurt dialects.This is followed by a conclusion.

Clitics to be investigated
Udmurt counts a couple dozen clitics that can be subsumed under a broad notion of discourse particles in at least some of their uses.In this paper, I am initially focusing on the eleven particles listed in Table 1, although only a subset of them will be discussed in detail.All of them are enclitics, i.e. they phonologically attach to the right of their host word.In what follows, I am going to use the terms discourse particles and discourse clitics interchangeably when referring to the eleven items in question.With the exception of na 'yet/still' (see details in Section 4), they do not affect the stress placement in the phonological word. 5As is often the case with pragmatic elements, the translations provided are loose and do not cover the whole range of meanings the particles can have.In order to be more precise, I will translate some of the particles with labels under which their rough equivalents have been described in the typological literature.Some explanations regarding the ranges of meanings available for each particle will be offered next to the examples (2)-( 9) that introduce them.
A large number of items in question have been borrowed from Turkic languages throughout the long history of their coexistence with Udmurt in the Volga-Kama area.Therefore, it would be reasonable to investigate to which extent their syntactic properties in Udmurt could be explained by their properties in the donor language.The "Source" column contains the borrowing status."Native" in this context means 'apparently not borrowed through a relatively recent contact, i.e. from any Turkic language or Russian'.
Before moving further, a couple of short comments has to be made as to the etymology of some of the clitics.
First, it has been claimed by Tarakanov (Tarakanov 1993 : 162), who, in turn, was referring to the opinion of Munkácsi, that Udmurt (i)ńi 'already' is a borrowing of the Tatar inde 'id.'.Csúcs (1990 : 180) marks this etymology as dubious.Its areal distribution (it exists and is very frequent everywhere, including Northern areas where many other Tatar borrowings did not spread) and the existence of a very similar item in Komi, which is closely related to Udmurt, also suggest that it was not borrowed from Tatar.
However, both theories have serious flaws, as there are significant mismatches in phonology, syntax and semantics between Udmurt uk and both of its proposed sources.I argue (Arkhangelskiy 2023a) that uk is actually a native item, which has grammaticalized from a reverse-polarity tag question construction.

Functions of discourse clitics
In this section, I explain the primary functions of each of the discourse clitics under consideration, without going into much detail.Whenever semantics of a clitic has been described in the literature, I provide a reference; claims not supported by references are based on my own observations.Examples (2)-( 9) below and the corresponding comments only illustrate typical usages of the clitics.
The first clitic, no, is the standard coordinating conjunction.Apart from that, it has a range of other uses.All uses of no fall under the scope of my investigation.For example, in (2), it is used as an additive particle.Its main function here is to introduce the additive presupposition (Karttunen, Peters 1979), i.e. to signal that there is at least one other focus alternative.
this.year that-PL-GEN series-ILL-3PL.POSS Alnash-PL=ADD come.out-PST-3PL'[Volumes of cattle farming have been decreasing in several districts for years.]This year, the Alnash district also joined their numbers.' The clitic ik can be primarily described as a particle of emphatic identity, in terms of König (1986).It marks an argument in a proposition as co-referent with an argument of another proposition present in the discourse.Further information on its properties can be found in (Zakirova 2019; Zakirova, Kozlov 2022; Zubova 2016).
garden-LOC M. work:PRS.3SGchild-PL-INS=ADD M.=ID sit-PRS.3SG'Masha works in the garden, and she is also the one who does the babysitting.' The clitics na 'still/yet' and ńi 'already/anymore' ((4)-( 5), ( 8)-( 9)) can be described as time adverbials in their primary sense.The same concerns aj/alÍ i (7) and inde, which are synonymous with na and ńi, respectively, in most (although not all) contexts.However, they have other important senses, which cannot be directly reduced to their primary functions.For example, na and aj/alÍ i can mean 'apart from that' and 'else' in certain contexts, just like German noch and its other European counterparts.aj/alÍ i (but not na) also functions as a hortative particle.
we.EXCL:NOM eat-PST-1PL=already 'These tomatoes are still green, we have already eaten the red ones.' A, which marks question focus (5), and pe, which marks reported speech (6), are rather standard representatives of the respective cross-linguistic categories.A attaches to the head of the focused phrase.Pe tends to occupy the second position inside the reported clause, but can be placed elsewhere as well.Although a and pe are not discourse particles in their main function, pe can also be used without any reported speech in a pragmatic sense roughly equivalent to the English 'kind of' (Arkhangelskiy 2023b).

Frequency properties of clitics
As a further step in getting acquainted with the object of my investigation, consider the frequencies of the clitics provided in Table 2.The figures in the three columns represent the ipm (items per million) values in three corpora mentioned above: Corpus of Standard Udmurt (contemporary written texts, mostly newspapers; 9.57M words), Beserman multimedia corpus (spoken; 114K words at the time of writing) and Corpus of Tatyshly Udmurt (spoken; 10K words at the time of writing).In a sense, the two spoken corpora represent the geographical extreme points of the Udmurt-speaking area: Beserman is close to its North-Western edge, while Tatyshly (Southern Peripheral supradialect) is at its South-Eastern edge.One thing which is immediately clear from Table 2 is that the clitics in question are extremely frequent (some only in certain dialects though).Their frequency is comparable to, or higher than, that of their most frequent European counterparts.For reference, noch 'yet/still' and schon 'already; indeed' in written German have the ipm values of 3153 and 1544, respectively (DeReWo 2014).One reason for this is that the Udmurt particles tend to be available in a wider range of contexts; see the discussion of the examples (36)-(38) in Section 5, as well as the description of meanings available to ńi by Zubova (Zubova 2016).However, according to my preliminary observations, they also have a higher probability of appearance than their German or Russian equivalents in contexts where the latter would be felicitous as well.Further investigation is needed to find out why this is happening.
Another obvious conclusion is that there are clitics that exist only in some of the varieties.Out of the three varieties listed, Tatar borrowings inde 'already' and eś 'enimitive' only exist in the Tatyshly dialect, uk is absent there, and aj/ alÍi 'yet/still' does not exist in Beserman.Also, out of the pair of quasi-synonyms na and aj/alÍi 'yet/still', the former, i.e. the native one, is preferred in the standard written language, but dispreferred in Tatyshly.Although d r 'probably' exists in Beserman, it is far less common there than in the South (where it was borrowed from Tatar).
The analysis of a subset of these clitics by Zubova (Zubova 2019) helps make sense of the rest of the discrepancies.Her research reveals that genre and register may dramatically affect their frequencies.While ik 'emphatic identity marker' is much more widespread in written texts, na 'yet/still', ńi 'already' and uk 'enimitive' exhibit the opposite trend, being more common in less codified texts.The differences in the genre composition of the corpora (especially monologues/ dialogues ratio) must also explain the diverging figures for a 'question marker' and pe 'reportative marker'.To the best of my knowledge, these two are used rather uniformly across Udmurt dialects, being obligatory (a) or nearly obligatory (pe) in the same sets of contexts.

Clitic ordering in a cluster
Given the abundance and high overall frequency of discourse clitics in Udmurt, it comes as no surprise that many sentences contain more than one.When multiple clitics choose the same host, all of them form one phonological word, as in (1) or ( 10 Relative frequencies of discourse particles in three corpora 6 6 The occurrences oaf the particle uʔ a 'isn't it' in Tatyshly, which originates from a combination of a negative verb with a question marker, were excluded from the count for uk.Since it is often impossible to tell apart no as a coordinating conjunction and as an additive focus particle, all occurrences were counted.In other cases, homonymous entities were disambiguated through random sampling.AlÍ i 'yet/else' is homonymous (and cognate) with alÍ i 'now'; a 'question particle' is homonymous with the conjunction a 'and, but' borrowed from Russian; d r 'probably' is homonymous with d r 'time'.This poses a problem since the corpora I used contain ambiguous analyses.When calculating the ipm values for these items in large corpora, I manually disambiguated a sample of at least 100 search hits.The total number of hits was then multiplied by the proportion of the item in question in this sample to obtain an estimate.Since the search hits were randomized, the samples were representative of the respective corpora.For the figures obtained through sampling, Wilson's 90% confidence intervals are indicated in brackets.
hen=ADD=Q=already lay.eggs:PRS.3SG'Is the hen laying eggs already as well?' The ability to form chains, or clusters, is a well-known property of clitics observed in many languages.Voluminous research on (mostly pronominal) clitic clusters in Romance, Slavic and Greek languages allowed linguists to arrive at certain generalizations.It turns out that if several clitics meet within one phonological word, they tend to be rigidly ordered (Spencer, Luís 2012 : 47), i.e. out of several theoretically possible orderings only one is usually acceptable.
European clitic clusters have been successfully modeled with the help of clitic templates (Perlmutter 1971;Bonet 1991).Templatic organization means that all clitics in a language are split into a number of sets, and the sets can be linearly ordered from left to right, forming a template sequence.Clitics within a set cannot co-occur, and whenever several clitics from different sets form a cluster, their mutual order is the same as the order of their respective sets in the template.The set of clusters accepted by a template can be further narrowed down by additional constraints which disallow certain combinations of clitics.The most widely known example of such a constraint is the "*me-lui constraint", or Person Case Constraint in Romance languages.According to it, direct object clitics of any person other than 3 (e.g.French me 'me') cannot co-occur with dative clitics (e.g.French lui 'to him/her').
This constraint has been later shown by Sheehan ( 2020) not to be restricted to clitic clusters.
As we will see below, the template approach mostly works for each individual variety of Udmurt.Even if my insufficient data does not allow for complete reconstruction of a template in each of the varieties surveyed, it is in principle compatible with the idea of a template.Two potential contradictions to templatic organization can be resolved by stipulating two different items for no 'and / additive particle' and taking the syntactic domain into account.However, before we get into that, two general features of Udmurt clitic clusters have to be mentioned, which make Udmurt somewhat different from the well-known Romance and Slavic cases.
First, although the ordering of clitics in a given variety is often rigid, sometimes it is not completely strict.An example of an entirely rigid ordering is (10), where all alternative orders of the three clitics are considered unacceptable by Beserman speakers.However, consider two further examples from the same variety ( 11)-( 12).( 11 These orders, a=ńi and uk=ńi, are clearly considered default by the speakers.However, opposite orders (ńi=a and ńi=uk) are judged as acceptable by most of them, even if most respondents accept them with reservations, adding that "they sound somewhat worse" or "that's how they would speak in another village".There are also four examples of ńi=a and three examples of uk=ńi in the Beserman corpus, produced by different speakers.Although the "default order" is two orders of magnitude more frequent (133 and 182 occurrences in the corpus, respectively), it is nevertheless not absolutely strict.
Moreover, there are varieties where certain pairs of clitics do not have a preferred order, exhibiting both interpersonal and intrapersonal variation; these cases will be discussed in Section 4.
As noted by Heap (2005 : 86), "[t]he fixed-order generalization is so close to being categorically true that it is hardly surprising so many linguists have [---] dealt with only invariably fixed-order clitic sequences without taking into account variable sequences".Absence of variation, inferred from the strict application of the template model, is indeed sometimes presupposed: e.g.Zimmerling and Kosta (2013 : 179) define clitic clusters as "strings of clitics arranged in a rigid order according to language-specific rules".Udmurt data demonstrates that variation is actually part and parcel of clitic clustering.
Second, different varieties of Udmurt have different ordering templates.Just as phonological features or lexical items, the templates clearly demonstrate areal distribution, i.e. settlements that are close to each other are likely to have the same order of clitics in a cluster.Free variation within a template is observed at the border between areas with different templates.Interestingly, some parts of the template are invariable across the whole Udmurt-speaking area, while others fluctuate a lot.This means that a template model -even if one allows for some variation within a template -is only applicable to individual varieties of Udmurt, but not to the language as a whole.
In what follows, I will mostly deal with this dialectal distribution of templates.

Dealing with apparent counterexamples
As mentioned earlier, there are two potential obstacles for the applicability of the template model to Udmurt data -apart from the ubiquitous variation.The first one concerns an apparent violation of transitivity by the no 'and / additive particle' clitic and manifests itself uniformly across dialects.The second one involves an apparently semantically motivated variation in position of the reportative clitic pe.
In both cases, there is an interplay between the scope of the clitic and its position.

Violation of transitivity by no: narrow vs. wide scope
A binary relation "<" is called transitive if for any a, b, and c, the statements "a < b" and "b < c" imply "a < c".This is exactly the sort of relation imposed by a clitic template, since it arranges clitics or sets of clitics along a single axis.In fact, it has long been known that there are cases where such a property does not hold in the strong sense.Consider the "*me-lui constraint" in French.While, in terms of the ordering relation for French pronominal clitics proposed by Perlmutter (1971 : 57), me '1 st person clitic (object/dative)' < le '3 rd person clitic (object)' and le '3 rd person clitic (object)' < lui '3 rd person clitic (dative)', one cannot deduce that me < lui since the two clitics do not combine and the relation between them is therefore undefined.This is, however, a minor issue that can be remedied by introducing constraints which would disallow certain combinations of clitics.In that case, it is still possible to employ templates for describing clitic clustering.Restrictions of such kind are sometimes considered the only obstacle to transitivity, as e.g. by Heggie and Ordóñez (2005 : 15-16).
Nevertheless, there is another potential scenario: if any two of the three clitics a, b, and c can be combined in a cluster, it might happen that a < b, b < c and c < a. Unlike with the "*me-lui"-kind constraints, such a situation would be irreconcilable with the idea of a template.However, at the first glance, this is exactly what happens in Udmurt.We will consider Beserman variety in this subsection, however relevant properties of no hold in every Udmurt variety I examined.
Consider Beserman examples ( 13)-( 15), which include three clitics, no 'and / scalar focus particle', a 'question marker' and ńi 'already'.( 13 Assuming the template model is correct, how would the three clitics be ordered in the template?The examples suggest that in clusters no is followed by a (13), a is followed by ńi (14), and ńi is followed by no (15).However, this creates a non-transitive, circular pattern (no < a, a < ńi and ńi < no), exactly like in the hypothetical scenario described above.In addition, all three clitics can appear at once, as in (9), so the situation cannot be remedied with the help of co-occurrence constraints.Importantly, the order of clitics in each of the examples is rather rigid, with the alternatives judged as unacceptable or, in the case of ( 14), at least degraded by the Beserman speakers.Therefore, this conundrum could not be written off as a mere consequence of free variation.
There is, however, a solution to this apparent contradiction.It turns out that the position of no in clitic clusters is strictly determined by its semantics and scope.In ( 10) and ( 13), where no precedes the other two clitics, it acts as an additive particle, translated as 'as well' or 'even'; the same happens if it is used as a normal coordinating conjunction.In this case, no can have variable scope, e.g. over a NP in (10).The meaning of the first no in (15) is rather different.There, it has a scope over a clause, attaches to its head and roughly means 'although the speaker commits to the truthfulness of the proposition expressed in the clause, some natural implications do not hold or are not guaranteed to hold by the speaker'.Among its other functions, this latter no is used in concessive clauses (usually in combination with ke 'if').However, its area of applicability is broader than just concessive contexts.For example, it is often used in monoclausal sentences, where the implications that do not hold are not spelled out explicitly in a sepa-rate clause.The development of this sense could have been caused or reinforced by the etymologically unrelated Russian conjunction no 'but' (see Kaysina 2015 : 225). 7y splitting no into two distinct lexical items, 'additive' and 'contrastive', we save the template approach.The two items receive distinct slots in the templates: the additive one, close to their left border, the contrastive one, to the right one.They also have different compatibility constraints.While 'additive' no is compatible with pretty much any other clitic, the 'contrastive' one is apparently incompatible with those that contradict the speaker's commitment to the information conveyed in the clause, such as a 'question marker' or d r 'probably'.This generalization holds across dialects, so that mutual orders in examples like ( 15) and ( 16) are the same in all varieties I examined.
( run-ITER-PRS.3SG=ADD=already'He (their child) did not walk last year, but this year he already walks and runs.' With this apparent contradiction resolved, Beserman clitics can be aligned in a template, at least in terms of their default order (Table 3).Only those clitics that exist in Beserman and for which I have enough data are listed here.The additive no is labeled as no a , and the contrastive no is labeled as no c .

Beserman clitic template (based on corpus data additionally checked with elicitation)
Looking at the rest of the clitics, we observe the same trend: obligatorily narrow scope puts the clitic close to the host word, while obligatorily clausal scope places it at the end of the cluster.
One example of the narrow-scope clitic is ik 'emphatic identity marker'.When it is used outside of a handful of lexicalised combinations, it always has a narrow scope, marking an argument, an adjunct or (rarely) a predicate as focal and identical to the one in the preceding discourse.It rarely combines with other particles, but the one it easily co-occurs with, a 'question marker', is placed to the right of it in every dialect (17).
( I:ACC=ADD=Q=REP maybe let-FUT-2PL mouse say-PRS.3SG' "Maybe you will let me in as well?" says the mouse.'Corpus data that I examined and elicited examples with pe that I collected in the field reveal that, in all dialects, pe is in principle grammatical in any position inside the whole sentence, including the framing clause.It seems that its only positional constraint is that it cannot appear inside PostPs or NPs.9However, normally pe appears inside the reported clause.Its exact position within the clause can differ.In most varieties, including Standard Udmurt and Beserman, pe prefers to attach to the first phrase (e.g.NP or PostP) of the reported utterance ( 18), which makes it a second-position clitic (Bošković 2016, Haspelmath 2022 : 12).This position is however non-rigid, since even in these varieties pe occasionally appears in other positions or several times within the reported clause.In some varieties, pe may prefer other positions.For example, it seems that in Tatyshly Udmurt, it is more often encountered at the end of the reported clause; further research is needed to clarify its properties across dialects.In any case, pe always has a scope over an entire clause, marking it as reported.In clitic clusters, pe cross-dialectally almost invariably occupies the rightmost slot (19).
There are however problematic examples like (20), where pe is followed by another clitic.Although examples like (20) are rare, I have attested them in different dialects, as well as in the standard language.What's more, this does not happen due to limited free variation, as was the case with some Beserman clitics in Subsection 3.1.Comparing such examples with those where the enimitive clitic (uk or eś, depending on the dialect) precedes the reportative pe, we can conclude that the 'outer' enimitive belongs to another syntactic domain than the 'inner' one.In (20), the speaker uses jeś to mark the proposition 'she says they are studying in Moscow' as an uncontroversial piece of information that the addressee should accept as an argument in their discussion.The same happens in (21) with uk.The 'outer' particles thus describe the point of view of the current speaker and could be said to belong to the 'matrix' of the reported utterance (in terms of Spronck, Nikitina 2019) and thus to a different syntactic domain than pe.In ( 22), on the other hand, the enimitive, probably used here to signal surprise on the part of the reported speaker, who did not expect to see potatoes that big, belongs to the reported clause.two fist the.size.ofpotato-PL-2PL.POSS=ENIM=REP '[S/he says,] "Your potatoes are double the size of a fist, aren't they" ' In order to account for this scope sensitivity of pe, the template model must be further restricted to the clitics that belong to the same syntactic domain (e.g.originate from the same clause). 10With this additional constraint, pe is always the rightmost clitic in the cluster.

Position of clause-level clitics in clusters
The fact that, barring the situation described above, pe in Udmurt is always cluster-final, is probably connected to the fact that it is the only clitic that can freely move around the clause and, in most dialects, prefers to occupy the second position.It is generally assumed in syntactic theories that the placement of clause-level second-position clitics is a result of movement to a higher position in the syntactic tree.If that is correct, we could assume that the clitics that undergo such movement are placed further from the host word in clitic clusters than clause-level clitics that do not undergo movement, let alone narrow-scope clitics.This hypothesis is corroborated by the data from standard Komi-Zyrian, where not only p≠ e 'reportative marker', but also ńin 'already' (cognates of the Udmurt pe and ńi, respectively) can move to the second position -unlike in Udmurt, where clause-level ńi is normally attached to the predicate.In Komi, these two clitics are attested in the order opposite to the one Udmurt prescribes (23).Note that it clear from the context of that sentence that ńin 'already' originates in the reported clause, in contrast to the Udmurt examples with clitics that follow the reportative pe.Since both clitics undergo movement in Komi-Zyrian, the aforementioned ordering rule does not apply to them, which is why ńin 'already' can follow the reportative clitic in the cluster.
Dialectal Variation of Udmurt Discourse Clitics 113 10 Impressionistically, I would say that the 'outer' uk moves to the reported clause from the main clause after the clitic clusters in the latter have been formed.This, however, would require stipulating elliptic main clauses in examples like ( 21)-( 22), as well as adopting one of the syntactic formalisms and conducting rigorous testing.As this goes beyond both the scope of this paper and my qualification, I use an intentionally vague term 'syntactic domain' here and leave the rest to further research.
The rest of the clitics under consideration, in terms of their scope and placement, occupy an intermediate position between the narrow-scope ik 'emphatic identity marker' and the clausal-scope no 'contrastive' and pe 'reportative marker'.Semantically, they can introduce focus alternatives for constituents of various size.In terms of their placement, a 'question marker' always attaches to the focused phrase of an alternative question, while the rest of the clitics usually attach to the predicate, even if they actually modify a particular focused argument or adjunct in the clause.The scope and position, which were demonstrated above to be an important factor in clitic placement, do not play a role when ordering those.So in a sense, the contest over who gets a place closer to the host word does not have obvious winner candidates.Indeed, their placement is subject to a lot of variation, as we will see below.In what follows, I will be only concerned with the clitics other than no (both 'additive' and 'contrastive'), pe 'reportative' and ik 'emphatic identity'.

Dialectal templates
If we assume that the problems outlined in Subsection 3.2 are indeed resolved, and free variation can be allowed, then the data I have at hand allows me to come up with a clitic template for each individual variety.However, it turns our that clitics are ordered differently in different varieties, so each variety has its own separate template.Templates reconstructed for different varieties of Udmurt can be so different that it is impossible to produce a general template that would describe the entire language.
The observed variation is, first and foremost, geographic (and, therefore, dialectal) in nature.Settlements located next to each other tend to have same templates, and settlements with same templates form contiguous areas.Whenever I worked with multiple speakers who were born and raised in the same settlement inside one of such areas, no other macrosocial factor (such as gender, age or level of education) played any visible role.However, varieties of settlements located at the borders between different areas sometimes allow for variation.Unfortunately, I do not have enough data to determine if factors like age affect the preference for one of the possible orders in those settlements.One other parameter that could theoretically confound the data is the influence of the standard written register, which most respondents learned in the school and some actively use.While I cannot reliably exclude this possibility, I nevertheless have serious doubts that this would be an important factor.Unlike heavily regulated vocabulary and morphology, things like clitic ordering are seemingly considered "ephemeral" by both Udmurt speakers and language authorities.As a consequence, the latter do not try to standardize them, while the former do not feel the urge to comply with the (apparently non-existent) norm in this respect.
If we compare templates for individual varieties, we will see some stable elements common to all of them, as well as some that are subject to significant variation.Clitics with both stable and variable positions can be seen in Table 4, which presents a unified template schema for all varieties.
The structure of this schema is the following.Each cell corresponds to one clitic slot in a cluster.For adjacent cells, the left slot precedes the right one.Items without parentheses correspond to clitics that can be analyzed as holding invariable positions across varieties.Items in parentheses indicate clitics that appear in the respective slot in at least one dialect, but not in all of them.

Unified template schema for Udmurt clitics
As the enimitive particles are incompatible with the question marker, I represent the part of the schema between the extreme points na 'yet/still' and aj/alÍi 'yet/still' as consisting of two branches, rendered as two non-adjacent table rows, for convenience.The upper one contains a 'question marker', and the lower one, uk/eś 'enimitive'.Nevertheless, it can be linearized in each of the dialects, provided certain co-occurrence constraints are added.D r 'probably' is supposedly incompatible with a 'question marker', which is why it only appears in the lower row.
Items that are altogether absent in a minority of the varieties examined are not taken into parentheses.The item inde 'already/indeed', which exists in only one of the varieties I have a reasonable amount of data on (Tatyshly Udmurt), also always occurs at the rightmost end of the cluster.However, I have no data regarding its compatibility and potential mutual order with aj/alÍ i 'yet/still' and did not include it in the template.
The stable part of the template consists of the clitic na 'yet/still', which always comes first, its quasi-synonym aj/alÍ i, which always comes last, and particles a 'question marker' and uk or eś 'enimitive', which always come in between and do not combine with one another.On the other hand, ńi 'already' and d r 'probably' can occupy any place between na and aj/alÍ i.Moreover, the position of ńi in the upper row does not predict its position in the lower row: attested dialect templates include those where it precedes a in the upper line, but follows uk in the lower one, as well as vice versa.How variable the position of ńi can be even within one settlement, can be seen from the template in Table 5 based on the data from Bagrash-Bigra variety (Central/Southern; a point just north of Agryz on the maps).

Dialectal distribution of pairwise orders
The pairwise orders are clearly subject to areal distribution.In the following maps (see Fig. 1-3), the distributions of three pairs, ńi/a, ńi/uk and ńi/d r, are plotted.11 Grey background represents the area where Udmurt is spoken 12 (based on the 2010 census data).Each circle corresponds to a settlement for which the corresponding order has been attested.White circles (○) denote orders with ńi 'already' on the left, black circles (•) denote opposite orders, and mixed circles denote varieties for which both orders have been attested.The distributions mapped have to be taken with a grain of salt for two reasons.First, the amount of data behind each data point ranges from one example taken from a published speech sample to dozens of examples produced by and discussed with multiple speakers.In the former case (which, however, subsumes a minority of data points), it is possible that the variety is actually a mixed one, i.e. allows both orders, one of which has not been attested because of the insufficient data.Second, there is question of comparability of data.During elicitation, which was the primary means of data collection for varieties representing a total of 23 settlements, I used the same questionnaire to control for factors like semantics and scope.The examples I found in corpora or published collections, on the other hand, are rather heterogeneous.Neverteless, I believe them to be by and large comparable to the data I collected in the field.Comparison of corpus data for Beserman and Tatyshly Udmurt with the respective elicitation responses shows that the two types of datasets yield the same results, and pairwise orders of clitics tend to be the same regardless of the context -with the exception of two cases addressed in Subsection 3.2.

Factors that may have influenced the dialectal distribution
Even if the sample the maps are based upon is not nearly as geographically dense as the one in the Dialectological atlas (Nasibullin, Maksimov, Semenov, Otstavnova 2009), and aforementioned precautions apply, this data still provides a good first approximation of the dialectal distribution of Udmurt clitic templates.What conclusions can we draw from it?
First, let's look at the data from a purely dialectological perspective.The isoglosses one may draw based on these maps do not coincide with the supradialect borders in traditional dialectal classifications (Kelxmakov 1998;Maksimov 1999 : 171).This is hardly surprising for several reasons.First, it is known that isoglosses of different kinds do not necessarily correlate with each other.Specifically, syntactic isoglosses can exhibit low correlation with phonological, lexical and   morphological ones, as demonstrated e.g. by Scherrer and Stoeckle (2016 : 110) for Swiss German.Since traditional Udmurt classifications are based on phonological isoglosses and, to a lesser extent, on vocabulary and morphology, one should not a priori expect clitic ordering isoglosses to align with them well.Second, a dialectometric analysis by Arkhangelskiy (Arhangelxskij 2021) shows that although most of the supradialect borders are rather strong, some are fuzzy, even if one only looks at lexical data.This especially concerns the Eastern part of the border between Northern and Central supradialects.
Although the position of ńi 'already' with respect to any of the three other clitics does not automatically define its position with respect to the others, there is visible correlation.There is a large area where ńi 'already' occupies the right slot in all three pairs.In terms of traditional classifications, it consists of the western parts of the Northern, Central and Southern Peripheral supradialects, as well as, apparently, the southern and eastern outskirts of the latter.Similarly, in the central and eastern parts of the Central supradialect (the triangle between Igra, Sharkan and Izhevsk) and the central part of the Southern dialect (immediately to the West of Agryz), ńi 'already' tends to occupy the left slot in all pairs.The borders between the two areas do not coincide completely for the three clitics, but, nevertheless, seem to lie not very far from one another.Such a distribution would be compatible with a diachronic process whereby ńi 'already' started moving either from the inside outwards or from the outside inwards in clitic clusters, outflanking one other clitic at a time (different ones in different varieties).Also, the borders are more vague than the isoglosses one normally sees in dialectal maps.This indicates, apparently, that the transition from one template to another proceeded gradually rather then abruptly.However, further research, prefereably involving diachronic data, is needed to support or refute this hypothesis.Second, the maps allow us to rule out language contact effects.Russian could hardly influence Udmurt clitic templates anyway because Russian equivalents (or particles perceived as equivalents by Udmurt speakers) follow rather different prosodic and syntactic patterns.Unlike in Udmurt, Russian discourse particles are often used as proclitics and are more free to move around the sentence.Tatar, on the other hand, has a more similar system of discourse particles, many of which have actually been borrowed into Udmurt and participate in its clitic clusters.However, comparing the areal distributions with the information on the intensity of language contact, we can conclude that the variation in the mutual order of ńi 'already', a 'question marker', d r 'probably' and uk 'enimitive' is unlikely to have resulted from Tatar influence.The intensity of Tatar influence decreases if one moves from the South (many speakers of the Southern Peripheral dialect know at least some Tatar, and some are bilingual in it) to the North.The position of ńi 'already', on the other hand, produces more of an East-West divide and involves both areas moderately influenced by Tatar and those where the influence has been minimal.The only way in which Tatar probably has affected Udmurt clitic ordering is the mutual order of the clitics borrowed from Tatar.In the Tatyshly variety, where the largest number of Tatar-borrowed discourse particle has been attested, their order maps that of their Tatar sources (24).
(24) Tatar: -d r < i ́ < äle; inde Udmurt: d r < eś < aj/alÍi; inde Finally, one takeaway point relevant for the cross-linguistic study of clitic clusters is that semantics clearly does not play any role in ordering discourse clitics.
First, consider the two quasi-synonyms na and aj/alÍ i 'yet/still'.Although the sets of contexts where they are used can be slightly different, they have a significant overlap in each dialect, and are actually used interchangeably or simultaneously in most dialects (see next section).Nevertheless, na invariably occupies the leftmost slot, and aj/alÍ i, the rightmost one.
Second, na 'yet/still' and ńi 'already' in their core uses can be seen as "semantic twins", just like German noch and schon, French encore and déjà, or Russian eщe and uwe.One member of the pair normally replaces the other if the sentence containing it is negated.Nevertheless, both Udmurt items for 'yet/still' have crossdialectally fixed positions in the template, while the position of ńi 'already' varies both between dialects and within some dialects.
So basically, the positions of discourse clitics within a template, provided there are no scope precedence relations between them, can be better modeled by stochastic diachronic processes rather than explained by their semantics, contact influence or the order of their appearance in the language.But if that is true, then why are the positions of na and aj/alÍ i 'yet/still' fixed across dialects?Although I do not have a definitive answer to this, a couple of observations can be made.
In some dialects, na seems to be more integrated in the morphology than the rest of the clitics.This tight integration manifests itself in the finite negative verbal complex, which in Udmurt -like in many other Uralic languages -consists of a negative verb followed by a connegative form of the main verb.In principle, all clitics in question (probably with the exception of inde 'already') can appear in the position between the negative and the main verbs (Vilkuna 1998 : 212; see Georgieva, Salzmann, Weisser 2021 for an analysis). 13owever, in certain dialects na has a number of additional properties when placed inside the negative verbal complex than the rest.First, na can appear twice inside the complex, attaching to both the negative verb and the connegative (25).In my data, such doubling is attested mostly in the parts of the Southern (Loloshur-Vozzhi, Kadikovo) and Central (Sjumsi, Ludorvaj, Buranovo) supradialects.
this sheep small NEG.EXIST=already=ADD lamb NEG.PST.3=yetbring:CNG.SG=yet 'Although this ewe is not young already, it has not yet given birth.' Second, in combinations like that, at least in some varieties, na draws the stress that otherwise would fall on the first syllable of the connegative and/or on the negative verb. 14No other discourse particle affects stress placement in any way, as far as I know.Third, certain Southern varieties spoken in the Grakhovo district (the aforementioned village of Loloshur-Vozzhi is situated there) feature a similar construction where the future-tense negative verbal complex interspersed with the double na is used with past reference, so (26) is interpreted in the same way as (25).The exact semantic contribution of na in (26), beyond its usual meaning 'yet', is unclear.However, most other Udmurt varieties would use the past-tense copula here.
We may conclude therefore, that there are dialects in which na 'yet/still' in the negative construction is gradually turning into a morpheme.The fact that no other clitic can appear between na and its host word aligns well with this hypothesis.Aj/alÍ i, on the other hand, has more freedom of movement than the rest of the particles discussed in this section -although, apparently, also only in a subset of the varieties where it exists.In the varieties of Porez (Southern Peripheral), Verkhnjaja Igra, Jenaberdino, Kadikovo (Southern), Buranovo (Central), Stengurt (Northern), aj/alÍ i can move to the sentence-initial position (ceasing to be a clitic) or to the second position ( 27), ( 28).
then=ENIM=yet car NEG.PST be:PST 'There were no cars yet back then, you know.' This property can probably be explained by the origin of alÍ i.Just like its Tatar source, äle, the Udmurt word is polysemous.Apart from being polyfunctional discourse particles, both alÍi (in most Udmurt dialects and in standard Udmurt) and äle are used as a time adverbial meaning 'now' (which, apparently, is its original sense that gave rise to the discourse uses in Tatar).In this sense, it can be placed anywhere in the sentence, and is often found close to its beginning.Many varieties have both kinds of alÍi.In all such dialects that I have surveyed, 'lexical' and 'discourse' alÍ i can clearly be described as two separate lexical items.For example, the 'discourse' one is compatible with the past tense, as in ( 27)-( 28), the 'lexical' one is compatible with ńi 'already' (29), and both can be used in the same sentence simultaneously, as in (30).( 29 It happens rather often that the same lexical meaning is present everywhere, but expressed by distinct words in different dialects.For example, a lot of vegetables and fruits split the Udmurt-speaking area in two parts, where the Northern part has a native or a Russian name for it, e.g.jablok 'apple', and the Southern part has a Tatar one, e.g.ulmo 'id.' (Nasibullin, Maksimov, Semenov, Otstavnova 2009 : 216).There is nothing mysterious in such a situation: since apple trees are cultivated in every settlement where the Udmurts live, there will obviously be a word for 'apple' in each dialect.
Discourse particles are a very different matter though.Unlike apples, they do not refer to objects or concepts that the Udmurt speakers deal with for extralinguistic reasons anyway.A language or a dialect could do without an enimitive or a word with the polysemy of the Udmurt na or ńi just fine.Many languages, like English, do not have grammaticalized enimitives.And while most languages do have temporal/aspectual items equivalent to English yet, still, anymore and already, they can group them differently.For example, most languages in the 41-language sample by Vandeweghe (2006 : 181-185) have separate items for yet and still, in contrast to Udmurt. 16So if a certain discourse particle does not exist in a certain dialect, there is no reason for its meaning to be represented by another one or for its polysemy to be replicated elsewhere.Indeed, the enimitive apparently does not exist at all in the Bavly variety (Southern Peripheral).Yet in almost all other varieties all meanings listed in Table 1 can be expressed by one particle or another.How did that come to be?
The areal distribution of the pair na and aj/alÍ i provides a likely answer.While there are areas where the latter does not exist (most Northern dialects) or the former is very infrequent and probably limited to a small set of contexts (parts of Southern Peripheral dialects), there is a large area where both are employed either interchangeably ( 33)-(34) or simultaneously (35).In the contexts where both na and aj/alÍi are acceptable, the speakers claim there is no difference in meaning whatsoever.Their claim is supported by the data: in dialects where both clitics exist, each of them, as well as their combination, regularly appears in translations of same questionnaire sentences.
(33) Udmurt (Bagrash-Bigra, Central/Southern) Peripheral varieties).Even if the newly borrowed particle used to be only a partial semantic match for the old one and originally did not cover all senses expressed by the latter, it probably aligned with it at the stage where both were used at once.As a result, the new particle becomes available in all contexts where the old one was used and eventually replaces it.The process thus resembles the Jespersen's cycle whereby one construction is gradually replacing another without apparent change in meaning.It has to be noted however, that I am referring to the formal stages of a process that Jespersen (1917) illustrated with gradual diachronic replacement of negative elements, such as French ne V > ne V pas > (V pas).The functional explanation Jespersen offered for the initial stage of the process was that the speakers find the original item, such as the French negative particle ne, (semantically) weakened and insufficient and therefore "strengthen" it by adding another element (in that case, the postverbal element pas).While I find this explanation unconvincing even when applied to French (what does a "weakened and insufficient negation" even mean?), it definitely does not work for Udmurt.Udmurt speakers start doubling their discourse particles not out of necessity, but because these particles are available in other languages they speak, and discourse particles are very easily borrowed (Matras 2010 : 81).
The same kind of semantic alignment can be seen in the Russian L2 speech of the Udmurt and Beserman speakers.Certain Russian words, which are perceived as translation equivalents of native discourse particles, acquire new meanings and are used in contexts in which they are hardly ever used by monolingual Russian speakers.For example, Russian uwe (uže) 'already', which is considered equivalent to Beserman ńi, is almost as frequent in the corpus of L2 Russian produced by native Beserman speakers (15,061 ipm) as ńi is in the Beserman corpus (17,043 ipm).This presents a stark contrast to standard Russian usage of uwe: its frequency in the subcorpus of non-public spoken conversations of the Russian National Corpus (which contains texts of approximately the same type and register as the Beserman ones) is four times lower at 4175 ipm.This staggering discrepancy is in part due to usages like ( 36 There are parallel pairs of items in Udmurt and Komi, which are suspiciously similar in their meaning, and not cross-linguistically widespread, but clearly distinct from the etymological point of view.Such are, for example, the attenuative/comparative marker (-ges/-gem in Udmurt, Äeremisinova 2019; -ǯ k in Komi, Todesk 2015) or the self-reportative clitic (p j in Udmurt and miśa in Komi, Teptiuk 2021).In light of the situation with Udmurt discourse clitics, it would make sense to investigate if a similar process could have led to the parallel existence of such items.

Possible outcomes of Jespersen's cycle
One possible outcome of the Jespersen's style is the disappearance of the older element because it becomes redundant.There is, however, another option.If the two elements always occur next to each other (unlike French ne and pas, which are separated by a verb), they can be lexicalized as a single item.For example, na and aj could stick together at the intermediate stage, producing a joint clitic naj.
Indeed, na=aj is a frequent combination in many varieties, and sometimes it shows signs of phonetic reduction, being pronounced as naj.Nevertheless, different preferred positions of na and aj/alÍi in the negative verbal complex and their diametrically opposed slots in the clitic template ensure that they are frequently interrupted by other material, as in ( 39)-( 41).This precludes them from collapsing into a single item.( 39 Otherwise, however, lexicalization of a clitic combination would be a realistic scenario.Consider ajko, which is one of the regional variants of aj/alÍ i.Unlike all other forms of this particle, ajko could not have been derived from the borrowed alÍi through regular phonological processes.In all likelihood, it is historically a combination of aj 'yet/still' and the hortative Russian clitic ko.17 Udmurt aj/alÍi can also be used as a hortative clitic, attaching to imperatives (42), a function also available to its Tatar source, äle.When ko was borrowed into the Central dialects that had already borrowed aj/alÍi, the two were apparently used at once in this context.Since there were no intervening elements in this case, this resulted in a single clitic ajko.The equivalence relation between aj and ajko spread to other senses as well, so that nowadays ajko can be used in the rest of the contexts available for aj/alÍ i, in varieties where it exists18 (43)-( 44).The original meaning of ko thus ceased to play any role.we:EXCL.NOM eat-PST-1PL=already 'These tomatoes are still green, we have already eaten the ripe ones.'

Conclusion
Studying the dialectal distribution of Udmurt discourse particles and clitic clusters allows us to arrive at cross-linguistically relevant conclusions.First, clitic placement in clusters can be determined by hierarchies of scope (ik 'emphatic identity', two flavors of no 'and; additive particle') and syntactic constraints (pe 'reportative').Still, these factors are only relevant for those three Udmurt clitics, while the rest can be ordered by templates that can apparently be arbitrary.Semantics, influence of contact languages and the order of appearance of clitics in a language do not have any straightforward effect on their place inside a template.
Second, clitic ordering templates can be subject to variation within a single language.That variation is dialectal in nature: geographically close varieties tend to have same templates.Additionally, there is considerable free variation, both interpersonal and intrapersonal, at the borders between areas that have different templates.That means that the perception of clitic chains as always conforming to a rigid order is not entirely correct.What that also means is that when researching clitic placement in a language with a weak norm and live dialects, one should not expect the data collected from speakers of different varieties to instantiate a single underlying grammatical structure.
Finally, looking at combination of discourse particles provides a potential solution for the 'one meaning, different particles' riddle.The fact that different Udmurt dialects have completely unrelated particles that apparently have same non-trivial semantics could be the result of a gradual, Jespersen's cycle-style replacement.In the course of this process, the meanings of two items are aligned while they are used simultaneously.
The properties of Udmurt discourse clitics and dialectal distribution of clitic templates raises a number of questions.Are same particles always available in the same contexts across dialects?How does the preferred position of clauselevel reportative marker pe differ across dialects and what can affect its placement?Can we model the diachronic processes that led to the observed distribution of templates, perhaps using diachronic data?Can we find additional evidence in the diachronic data that would shed light on how exactly Jespersen's cycle progresses in the case of discourse particles?Further research is needed to answer these questions.TIMOFEЙ АРХАНГЕЛЬСКИЙ (Gamburg)

Variable position of the reportative pe
The reportative clitic pe 8 provides an opposite example.Its core function is to mark reported speech or thought, so it is normally encountered in sentences that contain a reported clause and, possibly (althought not obligatorily), a framing/main clause, such as š r šue 'says the mouse' in (19).

Table 5 Clitic template for Bagrash-Bigra (Central/Southern) na
same questionnaire sentences.In Tatyshly (Southern Peripheral), the standard Udmurt enimitive uk apparently does not exist, but the same meaning is expressed by eś, absent in most other dialects.Similarly, aj/alÍi is absent in Beserman and the Northern dialects, while na only occurs very rarely in some Since aj/alÍ i appeared in Udmurt later than na and, like other Tatar borrowings, gradually spread northwards, different latitudinal sections of the Udmurt dialectal continuum presumably represent different stages of the diachronic process whereby na is gradually replaced with aj/alÍ i.At the intermediate stages of this process, both clitics are conceptualized as full equivalents and are used simultaneously.In my data, respondents who predominantly used both clitics at once come exclusively from the Central dialectal zone (Sjumsi, Vavozh, Ludorvaj, Buranovo, Iljinskoe).Speakers from the North did not use aj/alÍ i at all.Speakers of the Southern and Southern Peripheral areas, on the other hand, tend to use the two synonyms interchangeably, but not simultaneously (and clearly prefer aj/alÍ i in Southern ), where Russian uwe is used by the Beserman speakers in contexts where standard Russian uwe would be infelicitous, but Beserman ńi Although borrowed uže is not nearly as frequent in Beserman as ńi, it cooccurs with ńi in about half of all its occurrences in Beserman (38) (see alsoKaysina 2014).Given that the equivalence relation has been established, this might prove to be an incipient stage of a shift from ńi to uže.
and this-3SG.POSS here yarn=Q what=Q say-DETR-PRS.3SG=ENIM=already'And I guess this thing (yarn) is called short [in Beserman].'I am not able to explain why ńi/uže is used in these examples, and its meaning here is clearly not temporal or aspectual.However, sentences that mean 'X is called Y' are one of the contexts where ńi frequently occurs.