Profile of reflexives in Hill Mari

2.1 Paradigm

Reflexive pronouns in Hill Mari are represented by the following items derived from the same lexical root: (i) the agreeing škə̈(m)- forms, which bear possessive and case morphology and are used in most of the argumental positions, and (ii) the non-agreeing invariant ške, often described as the nominative form (Alhoniemi 2010; Savatkova 2002), which appears with postpositions and as a prenominal possessive modifier (3).^[5]

The paradigm of the reflexive forms is given in Table 1.

A few remarks should be made regarding the paradigm. First, as identified in the table above, the nominative form ške is always unmarked.^[7] Thus, when it is used as a complement of a PP, the possessive marker appears on the postposition (compare, for instance, [3b] and [4]).

Second, škə̈m- is the morphologically marked non-nominative stem that is immediately followed by a possessive marker. A possessive suffix on reflexives obligatorily precedes accusative and genitive case markers and follows the dative marker; in this respect, reflexives resemble personal pronouns and possessed nominal phrases although the latter allow more variation in the dative (5).^[8]

Third, similarly to personal pronouns and animate nouns, reflexives prohibit locative and caritive markers, which can be used with referential inanimate nominal phrases (compare [6a] and [6b] to [6c]).^[9]

2.2 Reflexive pronouns vs. DPs

Building upon their (2002) discussion of various sub-types of personal pronouns distinguished by their structural size, Déchaine and Wiltschko (2017) propose that reflexive markers in the world’s languages also form a heterogeneous group differing in terms of their morphosyntactic behavior and structural properties. The two most widespread types of reflexives are so called D-reflexives, whose distribution is similar to that of referential nominal phrases (DPs), and clitic-like φ-reflexives. DP-like reflexives can be found in Germanic languages (for instance, English self anaphors), while anaphoric clitics are common, for example, in Romance languages (for instance, se in French and Spanish and si in Italian).

Reflexive pronouns in Hill Mari pattern with English self reflexives in that their distributional properties are parallel to those of referential nominal phrases.^[10] Firstly, reflexives in Hill Mari can both saturate various arguments and function as predicates; compare (7a) to a similar example in (7b), where the predicate is a deictic pronoun.^[11]

Secondly, reflexives allow various kinds of modifiers, including postpositive adjuncts, appositive constructions, adjectival and nominal modifiers, as seen in (8).

Thirdly, a constituent headed by a reflexive can be independently negated (9); parallel examples with a sentential negation are given in (10). Note that, although agə̑l can appear in a broad range of contexts, it can only be used as a constituent negator and never accompanies finite verbs (except for desideratives; see Kirillova 2017 for discussion). Thus, the examples in (9) support the idea that anaphoric pronouns in Hill Mari do not form a morphosyntactic complex with the main predicate.

This property also unites reflexive pronouns and referential possessed DPs. In this, the anaphor škə̈- differs from reflexive clitics common in other languages: anaphor clitics in Spanish, for instance, form a single morphophonological unit with the predicate and cannot scope independently under negation, as shown in (11).

Further evidence that reflexive pronouns in Hill Mari are not clitics is provided by the fact that, similarly to other nominal arguments, they can be separated from the predicate and the rest of the clause in focus constructions, as in (12).

Finally, reflexives and referential nominal phrases in Hill Mari can be coordinated (13a). This provides additional support for the claim that škə̈- anaphors are full DPs since clitics and structurally smaller elements cannot coordinate with larger phrases (Kayne 1975); unsuccessful attempts to construct a parallel example in Spanish are provided for comparison in (13b) and (13c).

To summarize, ške/škə̈- anaphors are similar in their behavior to referential DPs and personal pronouns and are morphosyntactically independent from the predicate unlike reflexive clitics and affixes in many languages.^[12] Although from a typological perspective it is rather common for reflexives that are derived using the pattern REFL + POSS to behave as full DPs, there is no strict one-to-one correspondence. For instance, the Russian reflexive pronoun sebja also patterns with referential DPs in its behavior while morphologically it does not contain a possessive agreement marker (Testelets 2001). At the same time, POSS-i’ anaphors in Kaqchikel (a Mayan language spoken in Guatemala) resemble English and Uralic pronouns on the surface but are similar to Spanish reflexive clitics in their syntactic distribution (Burukina 2019). In the next section I focus on the binding properties of ške/škə̈- reflexives, that is the antecedent and locality restrictions. This will allow us to determine the status of Hill Mari reflexives according to the typology of reflexive pronouns developed by Kiparsky (2002) (discussed in detail in Section 3.3).

3.1 Antecedents

In many genetically and geographically unrelated languages, including, for example, Inuit (Eskimo–Aleut; Bittner 1994), Shona (Atlantic-Congo; Storoshenko 2009), and Russian (Slavic; Testelets 2001), reflexives are subject-oriented, that is, they can be bound only by a subject antecedent and are often restricted to the direct/indirect object position (see, for example, the reflexive affix in Shona [Bantu] and reflexive clitics in French [Déchaine and Wiltschko 2017]). As will be demonstrated in this section, Hill Mari škə̈- reflexives can occur in all structural positions suitable for ordinary nominal phrases, except for the clausal subject position, and normally can be bound by any c-commanding co-argument. In exempt positions^[13] (i.e., where there is no c-commanding co-argument, as is the case within postpositional phrases or nominal phrases) reflexives allow co-reference with almost any locally available pragmatically suitable member in a sentence, the only restriction being the word order: the antecedent must linearly precede the anaphor.^[14] In what follows I will consider the reflexive-antecedent combinations in more detail one by one, providing examples.

3.2 Locality

The binding domain for škə̈- reflexives and their antecedents is restricted to a minimal clausal constituent with an overt subject. In Hill Mari, several types of clause-like constituents can be embedded: (i) finite complement clauses with overt complementizers, (ii) infinitival complement clauses, (iii) nominalized clausal constituents with genitive or nominative embedded subjects, (iv) non-finite and finite adjunct clauses. Among them, argumental infinitival constituents with a covert subject are transparent, that is a reflexive in an embedded clause can have its antecedent in the matrix clause. Constituents that have an overt subject – embedded finite clauses, nominalizations, and adjunct clauses – are opaque. This is illustrated below.

A reflexive pronoun embedded in a non-finite clause can be bound either by the local (implicit) subject, denoted here as Ø, or the matrix (explicit) one (24).

Nominalized and adverbial clauses with an overt subject are opaque for anaphor binding; thus, in (25) the matrix subjects cannot be antecedents for the embedded reflexives and the sentences receive unambiguous interpretations.

The same is true for all finite clauses: arguments, adjuncts, and relative clauses, which are also opaque for binding, see (26).

It may appear that occasionally Hill Mari reflexives, similarly to self reflexives in English, may allow co-reference with a non-local overt subject; consider, for instance, (27), where the matrix subject binds the embedded pronoun despite the presence of the local overt subject.

Note however, that in sentences parallel to (27), the pronoun škə̈- should be analyzed as emphatic (an intensifier) rather, and not as anaphoric. In (28) the emphatic pronoun škə̈m-em-ə̈m is a focused constituent and can be accompanied by the particle vele ‘only’. If another constituent is focused, the pronoun škə̈- can no longer be used; for instance, in (28), the constituent ‘only Petja’ is emphasized as the only focus and, thus, škə̈m-em-ə̈m cannot be interpreted as an intensifier; the anaphoric reading is also unavailable in this case.

3.3 ške/škə̈- in the typology of pronouns

To address the Hill Mari data from a typological perspective, I adopt the typology of pronouns based on their syntactic properties developed by Kiparsky (2002). Kiparsky follows Faltz (1977) and suggests that the binding properties of reflexives and reciprocals vary along at least the following two dimensions: (i) the size of the domain within which they must be bound, and (ii) the nature of the antecedent in the clausal domain. Importantly, these two characteristics are lexical properties of individual anaphors, not a syntactic parameter of the language as a whole, and thus may differ for distinct groups of pronouns.

As for the first parameter, Kiparsky identifies a hierarchy of five successively more inclusive antecedent domains; the category of a pronoun is determined by the maximum domain in which its antecedent may be found. Firstly, there are referentially independent vs. referentially dependent pronouns. The former can introduce a new entity into the discourse, for instance via deictic use (as in It is me!), while the latter must have at least a discourse antecedent. Secondly, referentially dependent pronouns are divided into reflexives and non-reflexives, depending on whether they require a syntactic antecedent. Thus, non-reflexives are allowed with a context antecedent in such examples as John _i is here. I saw him _i . Thirdly, reflexive pronouns may be finite-bound, that is, requiring an antecedent within the same finite clause, or non-finite-bound, that is, allowing an antecedent beyond the finite clause they are located in. Finally, among the finite-bound pronouns locally vs. long-distance bound are distinguished. The former must be bound by a constituent in the first accessible subject domain while the latter are not subject to this requirement.

In addition to this, the second parameter – the nature of the antecedent – splits each of the above mentioned categories in two, based on the obviation property [+/− Obviative] ([+/− O]). The Obviation principle is formulated as follows: “coarguments have disjoint reference” (Kiparsky 2002: 180); see also similar definitions in Hellan 1983, 1988; Sells 1986; Farmer and Harnish 1987). For instance, English personal pronouns are [+Obviative] since in examples similar to John _k hit him _i/*k , the pronoun and the co-argumental DP cannot share the referent. In contrast, English reflexives are [-Obviative]: co-reference between arguments of a predicate is enforced if one of them is a reflexive pronoun (John _k hit himself _*i/k ). This yields altogether 10 types of pronominal elements (29).

(29)

The typology of pronouns (Kiparsky 2002: 205)

Within this typology, Hill Mari škə̈- reflexives occupy the same position as the Russian reflexive sebja. On the one hand, they are referentially dependent, reflexive, finite-bound but long-distance, in a sense that non-finite clauses with an obligatorily implicit subject are transparent for binding. On the other hand, Hill Mari agreeing škə̈- anaphors are non-obviative, i.e., they allow co-reference between co-arguments, for instance, a Causee and a Theme in ditransitive constructions (see [18] above). The invariant ške reflexive is almost impossible to characterize in terms of obviation since it is used mostly in exempt positions, without co-arguments; see, however, an example in footnote. 16, reproduced in (30), where a possessive ške is bound by another dependent of the same head noun.

Having established this, I will proceed by comparing the properties of škə̈- reflexives to those of anaphors in other Uralic languages, expanding the cross-linguistic comparison of reflexive strategies presented in Volkova (2014).

4.1 Morphosyntax

Agreeing reflexives are common in Uralic languages. Consider, for example, es’-POSS reflexive forms in Erzya and Moksha (the two Mordvin languages), maga-POSS reflexive pronouns in Hungarian, itse-POSS reflexives in Finnish, xər°q-POSS anaphoric items in Nenets, ač’/aš’/as-POSS reflexives in Udmurt.^[18] A common belief is that reflexives in different Uralic languages are often derived from similar lexical roots that can be traced back to a proto-Uralic word meaning ‘shadow, soul’ (Majtinskaja 1964). This pattern is not unique among the world’s languages: as pointed out by König and Siemund (1999), most lexical reflexives stem from words denoting different body parts and related notions. A similar observation has also been made by Schladt (2000), whose typology distinguishes between eight main lexical sources for reflexives, including body-part nouns, nouns denoting a person or personality and nouns meaning ‘soul, spirit’. In Table 2, I provide partial paradigms of agreeing reflexives in several Uralic languages belonging to different branches to bring out the general pattern and to capture the cross-linguistic microvariation within the language family.

As evident from the data presented above, Hill Mari anaphors closely follow the general pattern: REFL + POSS + CASE. One easily perceived language-specific parameter is the number of cases and availability of number marking (consider, for instance, the absence of dative and partitive in Sami).

Further microvariation is attested with regard to morpheme ordering. As discussed in Section 2 of this paper, Hill Mari anaphors require a strict order of morphemes: agreement markers must precede accusative/genitive case markers but obligatorily follow the dative marker. At the same time, as shown in the table above, in Meadow Mari (the Sernur-Morkin variety) reflexive pronouns are reported to allow some variation in morphemic order in the dative (Volkova 2014). I do not have an immediate explanation for this variation, but it is important to acknowledge it. It should also be noted that the microvariation in the ordering of morphemes in anaphors matches that of the ordering of morphemes in personal pronouns. As was shown in (5d), 1pl and 2pl pronouns in Hill Mari do not allow the poss-dat/dat-poss alternation. However, in Meadow Mari such pairs of forms as mə-lan-na we-dat-poss.1pl / mem-na-lan we-poss.1pl-dat and tə-lan-da you.pl-dat-poss.2pl / ten-da-lan you.pl-poss.2pl-dat are attested.^[24] The three languages that stand out from the general system are Estonian and two Ugric languages – Mansi and Khanty. Estonian makes no person distinction and exploits no possessive markers for reflexives; for instance, the forms enese gen.sg, ennast part.sg, eneste gen.pl, and endid part.pl can be used with any singular / plural antecedents, respectively, regardless of their person characteristic. As to Mansi and Khanty, neither has dedicated lexical reflexives. In Mansi, intensified forms of personal pronouns can be used as anaphors bound by a local antecedent; consider the following paradigms as an example (Table 3).

Khanty also allows locally bound pronominals; in contrast to Mansi, no emphatic marker is present (Nikolaeva 1995; Rombandeeva 1973; Volkova 2014).

The exact reasons behind this variation remain to be adequately addressed in future work. For the purposes of the present research it suffices to emphasize that the reflexive pronouns in Hill Mari fall in with the general Uralic pattern and to mention the exceptions among Uralic languages.

4.2 Binding properties

In Section 3.2 I discussed syntactic properties of Hill Mari škə̈- reflexives and demonstrated that they are long-distance non-obviative anaphors requiring a syntactic antecedent within the minimal clause containing an overt subject. Since most traditional grammars of Uralic languages do not consider reflexives in detail, it is hard to provide a comprehensive comparison of anaphors in Hill Mari with their counterparts in terms of their syntactic properties. Fortunately, there exists a study by Volkova (2014), which compares reflexives in five Uralic languages within the formal generative framework; those are Khanty (the Shuryshkary dialect), Komy-Zyrian (the Pechora dialect and the Izhma dialect), Udmurt (the Besermyan Variety), Meadow Mari (the Sernur-Morkin dialect), and Erzya (the Shoksha dialect). Volkova’s findings alongside with information on the Hill Mari data are summarized in Table 4.

As can be seen in the table, Hill Mari mostly patterns with other Uralic languages with respect to the binding properties of its lexical reflexives, with two differences. Firstly, in (32) and (33) I reproduce Volkova’s examples for Meadow Mari and Besermyan Udmurt showing that reflexives in these languages, unlike in Hill Mari, can only have subject antecedents.

Secondly, recall that škə̈- reflexives in Hill Mari are finite-bound but non-local. This is also true for reflexives in Meadow Mari (34) and Komi-Zyrian while anaphors in Besermyan Udmurt (35), a language closely related to Komi-Zyrian, and in Erzya have a smaller binding domain – that is a minimal clause with any (overt or covert) subject.

Thus, we have observed that Hill Mari, on the one hand, follows the pattern of binding behavior most common within the Uralic language family; on the other, the behavior of reflexive pronouns in Hill Mari still differs from that of lexical anaphors in several other Uralic languages, including Meadow Mari (the sister language), which suggests an important direction for future research.^[26] In the next section I continue comparing Hill Mari to other Uralic languages, focusing on the two reflexive strategies reported as being common for the family: complex reflexives and reflexive detransitivization.

5.1 Complex reflexives

According to Volkova (2014), Meadow Mari (the closest relative of Hill Mari) allows complex (reduplicative) reflexive pronouns, derived using the pattern ške(n)- ške (36).^[27]

The complex anaphor ške(n)- ške is subject-oriented; unlike simple anaphors (škenže) it is strictly local and must always be bound by a co-argument (37). In addition to this, the reflexive ške(n)- ške cannot be used in non-co-argument position, for instance in a postpositional phrase.

At the same time, speakers of Hill Mari always interpret sentences with the equivalent sequence škə̈(m)- ške as a combination of an agreeing reflexive and an invariable ške intensifier (38).

Note that the reverse sequence ške škə̈(m)- is allowed and should also be analyzed as an occasional combination of the emphatic pronoun and an agreeing reflexive, each of them interpreted separately. This is shown by ambiguous readings of examples similar to (39): here, an invariant intensifier can be interpreted either as adverbial, related to the matrix subject, or as adnominal with the anaphor as an antecedent.

Volkova demonstrates that, in the case of Meadow Mari, ške(n)- ške is indeed a single anaphoric unit without an additional emphatic meaning; this is further confirmed by the fixed order and obligatory adjacency of the parts of an item. The word order issue has already been addressed above; both škə̈(m)- ške and ške škə̈(m)- receive similar interpretations: reflexive + intensifier. As for the adjacency, both sequences can be discontinuous (40) with no change in meaning.

The only context where ške škə̈(m)- is opaque for intervention is when the intensifier is interpreted as adnominal since in such cases it generally serves as a modifier within a nominal phrase while itcannot be separated from its head by a matrix constituent, as shown in (41).

Taking these data into account, I argue that, unlike in Meadow Mari, in Hill Mari neither the sequence škə̈(m)- ške nor ške škə̈(m)- can be analyzed as a complex reduplicative anaphor; both sequences should be decomposed into a reflexive and an intensifier, each pertaining its normal distribution and contributing its meaning. Thus, only simple reflexive pronouns are available in this language.

5.2 Reflexive detransitivization

Another reflexive strategy common among Uralic languages is to detransitivize a verb to create a reflexive predicate; that is, to assign two of the thematic roles of the verb to one of its arguments. This technique is exploited, for instance, in Estonian, Khanty, Mansi, Komi Zyrian, and Besermyan Udmurt;^[28] several examples are given in (42).

As reported by Volkova (2014), this detransitivizing strategy is also used in Meadow Mari; the suffixes -əlt- and -alt- serve as morphological exponents of reflexivity. Compare (43a) and (43b): unlike in the first case, where the Agent and the Theme are distinct, in the second case, both roles are assigned to the same argument, ‘Ivan’.

The question arises whether a similar syntactic strategy is productive in Hill Mari. At first glance, the answer seems to be positive: similar pairs of transitive – reflexive predicates can be found in Hill Mari, as seen in (44).

Despite the surface similarity between the Hill Mari and Meadow Mari examples above, I follow Belova and Dyachkov (2019) and contend that syntactic reflexive detransitivization is not operative in Hill Mari and that verbs similar to those in (44) should be considered inherently reflexive (i.e., their intransitive nature is determined in the lexicon and is not a product of a syntactic derivation).

Several facts support this claim. First, note that the marker -alt- is ambiguous between the following interpretations: middle/decausative, reflexive, and reciprocal (Galkin 1996; Salo 2015; Savatkova 2002). Only the middle (decausative), as illustrated in (45), can be considered a productive derivation.

Using a wide range of examples similar to those in (45), Belova and Dyachkov (2019) demonstrate that the central meaning of -alt- is decausative. Reflexive and reciprocal verbs, in turn, are not numerous and can be derived only from a limited number of (Hill Mari) stems. For instance, although (44) are grammatical, -alt- variants as given in (46) are totally unacceptable even though the so-called verbs of grooming are generally considered to be prototypically reflexivizible (Kemmer 1993).

Other verbs that we would expect to have reflexive counterparts but which, in fact, prohibit such a derivation include ə̑də̑raš ‘scratch’ (*ə̑də̑r- alt -aš ‘scratch oneself’), pə̈čkedäš ‘cut’ (*pə̈čked- ält -äš ‘cut oneself’), i.a.

Productivity is assumed to be the central property of syntactic reflexivization, attested, for instance, in the Romance languages including Spanish and French (Labelle 2008). The fact that the number of reflexive predicates in Hill Mari is so limited strongly suggests that they are not derived in syntax but come ‘premade’ from the lexicon (see Reinhart and Siloni (2005), who discus the differences between syntactic vs. lexical reflexivization). With this consideration in mind, I maintain that, unlike in many other Uralic languages, the only productive reflexive strategy available in Hill Mari is to use a simple anaphoric pronoun.^[29]

It might be suggested that the lexicalization of reflexivized predicates with -əlt/alt in Hill Mari happens under the influence of Russian. As shown in Belova and Dyachkov (2019), Hill Mari closely follows the Russian pattern. In the case of lexical borrowing, for instance, the suffix -alt appears in Hill Mari predicates whenever there is the suffix -sja in a Russian translation equivalent: zan’imaj alt aš/zanimat’ sja ‘to occupy oneself with something’, ubiraj alt aš/ubirat’ sja ‘to clean’, etc. Reflexive -sja verbs in Russian are argued to be derived in the lexicon and not in syntax (Pesetsky 1995; Say 2005). Unlike, e.g., -sja (medio-)passivization, -sja reflexivization is non-productive and many of such predicates have acquired non-compositional meanings (cf. strič’sja cut.hair-sja ‘to have a haircut’, not ‘to cut one’s own hair’; ubirat’sja ‘to clean’, not ‘to clean oneself’; sadit’sja ‘to sit down’, without a counterpart *sadit’, etc.).