Amadou millet [sack Loc] put-Pfv-3SgSubj

‘Amadou (man’s name) put (the) millet in a/the sack.’ (Nanga)

There is no morphological case marking for subject NPs (‘Amadou’).

1.2. Reflexives

1.2.1. Nanga third-person reflexives.

Nanga lacks transpersonal reflexives, and has no special reflexive forms coindexed to 1st-2nd person antecedents. In (2a-b), the object ‘my cow’ has the same form whether the 1Sg possessor is disjoint to the clausemate subject (2a) or coindexed with it (2b). Likewise, in (2c-d) the object ‘me’ has the same form whether it is disjoint to the clausemate subject (2c) or coindexed with it (2d). The optional accusative morpheme is added to objects but not subjects.[3]

(2) 1Sg possessor-of-object (a-b) and object (c-d)

a. [nàŋá yɛ᷈:](-ŋ̀) tɔ́rɔ́-só-Æ

[cow 1SgPoss.An](-Acc) sell-Pfv-3SgSubj

‘He/She sold my cow.’ (Nanga)

b. [nàŋá yɛ᷈:](-ŋ̀) tɔ́rɔ́-só-y

[cow 1SgPoss.An](-Acc) sell-Pfv-1SgSubj

‘I sold my (own) cow.’ (Nanga)

c. ŋ̀gí(-ŋ́) kɛ́sɛ́-só-Æ

1SgObj(-Acc) cut-Pfv-3SgSubj

‘He/She cut-Past me.’ (Nanga)

d. ŋ̀gí(-ŋ́) kɛ́sɛ́-só-y

1SgObj(-Acc) cut-Pfv-1SgSubj

‘I cut-Past myself.’ (Nanga)

1Pl, 2Sg, and 2Pl behave like 1Sg in lacking dedicated reflexive forms for use in nonsubjects. For example, (3a-b) with 2Sg subject and coindexed (but not reflexive-marked) object are parallel to the 1Sg examples (2c-d) above.

(3) 2Sg object

a. ú(-ŋ) kɛ́sɛ́-só-Æ

2SgObj(-Acc) cut-Pfv-3SgSubj

‘He/She cut you-Sg.’ (Nanga)

b. ú(-ŋ) kɛ́sɛ́-só-w

2SgObj(-Acc) cut-Pfv-2SgSubj

‘You-Sg cut-Past yourself.’ (Nanga)

For third person, however, coindexation of the object or the possessor of a nonsubject with the clausemate subject is expressed by anaphoric (reflexive) pronouns, labeled 3Refl (plus Sg or Pl) in interlinears. For example, a 3Sg possessor in an object NP appears as nɔ̀ in the absence of a coindexed antecedent (4a‑b), but as 3ReflSg á when so coindexed (4c). Likewise, a 3Sg object is expressed as ńnɛ́(-ŋ́) without a coindexed antecedent (4d‑e), but as 3ReflSg á(-ŋ́) with one (4f).

(4) 3Sg possessor-of-object (a-c) and object (d-f)

a. [nàŋá nɔ̀](-ŋ̀) tɔ́rɔ́-só-y

[cow 3SgPoss](-Acc) sell-Pfv-1SgSubj

‘I sold his/her cow.’ (Nanga)

b. [nàŋá nɔ̀](-ŋ̀) tɔ́rɔ́-só-Æ

[cow 3SgPoss](-Acc) sell-Pfv-3SgSubj

‘He/She_x sold her/his_y (someone else’s) cow.’ (Nanga)

c. [nàŋá [á yɛ̂]](-ŋ̀) tɔ́rɔ́-só-Æ

[cow [3ReflSg Poss.An](-Acc) sell-Pfv-3SgSubj

‘He/She_x sold his/her_x (own) cow.’ (Nanga)

d. ńnɛ́(-ŋ́) kɛ́sɛ́-só-y

3SgObj(-Acc) cut-Pfv-1SgSubj

‘I cut him/her.’ (Nanga)

e. ńnɛ́(-ŋ́) kɛ́sɛ́-só-Æ

3SgObj(-Acc) cut-Pfv-3SgSubj

‘He/She_x cut her/him_y (someone else).’ (Nanga)

f. á(-ŋ́) kɛ́sɛ́-só-Æ

3ReflSgObj(-Acc) cut-Pfv-3SgSubj

‘He/She_x cut himself/herself_x.’ (Nanga)

The 3Refl pronouns in (4c,f) indirectly support the syntactic validity of the subject category, since a clause-mate subject is the typologically normal antecedent for such anaphors.

To sum up so far, diagnostics for subject NPs in simple Nanga main clauses are the features in (5).

(5) Subjecthood tests in Nanga main clauses

a) linear position preceding nonsubject NPs and PPs;

b) lack of accusative -ŋ or other case-marking;

c) pronominal-subject agreement by suffixes on predicates;

d) third-person clausemate subjects that bind reflexives.

Object NPs differ from subjects on all these counts. Datives, instrumentals, purposives, and spatiotemporals are expressed as PPs and are therefore also clearly distinct from subjects.

1.2.2. Transpersonal reflexives.

1.2.2.1. Transpersonals in Dogon

Certain other Dogon languages have fully productive transpersonal reflexive morphemes. They are not marked for person in the fashion of the Nanga third-person reflexive (3Refl), and are compatible with antecedents of any pronominal person. Transpersonal reflexives also occur in Russian: pronoun seb’ya, suffixal or enclitic ‑s’a ‘self’, and possessive svoy ‘own’. Such reflexives are pure anaphors. They are more reliable for research on anaphora than typologically widespread morphologically possessive reflexives of the type ‘my/your/his/her self/head/body’, whose literal possessive reading is perhaps never entirely lost.[4]

Two Dogon languages with transpersonal reflexives are Togo Kan and Tomo Kan, whose respective abbreviations TgK and TmK will be used since they are less likely to be confused than the full names. In TgK, sǎⁿ is the transpersonal reflexive, used as objects and nonsubject possessors when they are coindexed with a clausemate subject. In (6), the antecedent person varies from second (6a) to first (6b) to third (6c), but the reflexive object form is invariant.

(6) a. ú sǎⁿ kɛ́j-ɛ̀

2SgSubj Refl cut-Pfv

‘You-Sg cut-Past yourself.’ (TgK)

b. íⁿ sǎⁿ kɛ́j-ɛ̀

1SgSubj Refl cut-Pfv

‘I cut-Past myself.’ (TgK)

c. sè:dú sǎⁿ kɛ́j-ɛ̀

Seydou Refl cut-Pfv

‘Seydou_x cut-Past himself_x.’ (TgK)

TgK (6a-c) above has the TmK counterparts in (6¢a-c) below. The TmK transpersonal reflexive singular morpheme is hà (varying with sà). TmK hà is more clearly pronominal morphologically than its TgK counterpart sǎⁿ. Like other TmK pronouns, hà has a suffixed accusative form (hà‑lì). Tone alternations in TmK like hà‑lì versus há‑lì are due to rhythmic tonal processes and may be disregarded here.

(6¢) a. à há-lì cɛ̀ʔɛ́

2SgSubj ReflSg-Acc cut.Pfv

‘You-Sg cut-Past yourself.’ (TmK)

b. ìⁿ há-lì cɛ̀ʔɛ́

1SgSubj ReflSg-Acc cut.Pfv

‘I cut-Past myself.’ (TmK)

c. sé:dù hà-lì cɛ̀ʔɛ́

Seydou ReflSg-Acc cut.Pfv

‘Seydou_x cut-Past himself_x.’ (TmK)

In both TgK and TmK, the transpersonal reflexive clearly occupies an argument position in examples like (6a-c) and (6¢a-c). This is more transparent in TmK than in TgK, thanks to its obligatory accusative marking of the reflexive in object function (6’a-c). That these reflexives are syntactic NPs rather than valency-changing derivational morphemes is also shown by the fact that both TgK sǎⁿ and TmK hà take plural forms when the antecedent is nonsingular. In TgK the plural is sǎⁿ bè, with nominal plural particle bè, exemplified in (20b) below. In TmK the plural of reflexive hà is hè, formed by a vocalic mutation also seen in other pronouns (2Sg à versus 2Pl è, logophoric singular là versus plural lè); an example is in (20’b) below.

1.3. Subjects in relative clauses.

Returning to Dogon, we now move from simple main clauses to selected subordinated clauses (relative, switch-reference, quotative), accumulating more evidence for subjecthood that we will use as subjecthood tests later on.

Relative clauses in most Dogon languages are “internally headed,” i.e. circumnominal. The overt head, maximally Poss-N-Adj-Num, occurs in the relativization site internal to the relative clause. However, this overt head has the morphological and tonosyntactic form it should get as the corresponding portion of the “upstairs” NP containing the relative clause. This requires a simple analysis involving downward (or lateral) movement after morphological and tonosyntactic processing, or a more complex analysis that mimics this (Heath & McPherson 2013).

As illustrated above, in Nanga main clauses the pronominal category of the subject is expressed by an agreement suffix on the verb. In relative clauses, this finite verb is replaced by a verbal participle with (almost) no pronominal-subject agreement.[5]

Nanga relative clauses do, however, reserve a special treatment for pronominal subjects that provides us with another subjecthood test. In subject relatives (‘the goat that ran away’), the subject is also the head NP and must be nonpronominal. (‘We who are about to die’ is phrased appositionally as ‘we, the people who are about to die’.) However, nonsubject relatives frequently have a pronominal subject. Since (with rare exceptions) the usual pronominal-suffix slot on the verb is not available in relatives, a pronominal subject is expressed by an independent pronoun proclitic to the relative verb.

Let us walk through the structure of the Nanga relative clause (7).

(7) [àrⁿà ^L nɛ́](-ŋ́) wàgàdì ^L súyɔ́ bû: tì-sɛ̀ gà

[man^L Def.AnSg](-Acc) time^L hit 3PlSubj Pfv-Ppl Loc

‘at the time (= when) they hit the man.’ (Nanga)

The head NP is ‘time’ and the subject is ‘they’, so this is a nonsubject relative. The head ‘time’ is tone-dropped by overlaying {L} melody, as usual for overt heads of relatives. The resulting wàgàdì ^L is clearly internal to the clause, though it can either follow ‘the man’ as shown in (7), or precede it. The object ‘the man’ has the same form here as it has in main clauses (the definite morpheme controls tone-dropping on ‘man’). The predicate is expressed as a chain consisting of ‘hit’ plus a perfective auxiliary verb. Since no nonpronominal subject NP is present, the pronominal subject is obligatorily expressed by 3Pl pronoun bû: proclitic to the final verb in the chain, i.e., the participialized (“Ppl”) perfective auxiliary verb tì-.

The corresponding main clause is (8). It has a 3Pl-subject suffix on the verb, and there is no proclitic subject pronoun such as bû: in (7).

(8) [àrⁿà ^L nɛ́](-ŋ) súyɔ́-tì-yà

[man^L Def.AnSg](-Acc) hit-Pfv-3PlSubj

‘They hit the man.’ (Nanga)

Subject also plays another role in relative clauses. The Nanga 3Refl pronouns (singular á, plural â:) mentioned in §1.2 above are actually a more general anaphoric category that can also be used logophorically and, more relevantly here, to index identity between main-clause subject and relative-clause subject. Therefore the bracketed relative clause in (9) has á rather than the regular 3Sg ńnɛ́ as its proclitic subject pronoun.

(9) á:mádù [dàyⁿ ^L á gɔ́rⁿɔ̀‑mì] kárí‑ŋ̀

Amadou [limit^L 3ReflSgSubj be.able-Ppl.Impf] do-Impf.3SgSubj

‘Amadou_x will do as much as he_x can.’ (Nanga)

The facts about subject pronominals in main and relative clauses are summarized in (10).

(10) Nanga pronominal subjects

a) main clauses: pronominal-subject suffixes on verbs, agreeing with (overt or covert) subject NP.

b) relative clauses: in the absence of an overt nonpronominal subject NP, a pronominal subject is expressed by a pronoun immediately proclitic to the verb.

c) The proclitic subject pronoun in (b) is 3Refl (Sg or Pl) if coindexed to the main-clause subject.

The precise modeling of Dogon internally-headed relative clauses is not crucial here. What is relevant is the fact that main and relative clauses, in different way, provide additional indicators of both morphological and referential subjecthood.

1.4. Verb-chaining and switch-reference.

Dogon languages allow chaining (serialization) of two or more verbs or VPs under some conditions. In direct chains, those with no subordinating morphology on the nonfinal verbs/VPs, the nonfinal verb occurs in bare-stem form, or in a specialized chaining form (perhaps with a final-vowel mutation), depending on the language. The final verb in the chain is inflected for whatever inflectional category (e.g. perfective negative, imperfective positive, imperative) the speech context requires.

In most Dogon languages, direct chains are restricted to cases where the two verbs denote conceptually integrated co-events, e.g. Nanga tómbó sígé‑ (~ súgó-) ‘fall down’, consisting of invariant bare stem tómbó ‘fall’ and an inflected form of sígé‑ ‘go down, descend’. In such contexts the two verbs share an actor/subject, as in (11).

(11) tómbó sígé‑só‑Æ

fall descend-Perf2-3SgS

‘He/She fell (all the way) down.’

However, the restriction of direct chains to integrated co-events means that the combinations are usually lexicalized and base-generated, analogously to compounds for other stem-classes. They are pronounced accordingly, with no pauses or other prosodic interruptions. In most Dogon languages, the only completely productive direct-chain construction that can begin with an open-ended choice of VP is the ‘can VP’ construction which involves addition of an inflected form of the ‘get, obtain’ verb (e.g. Nanga bɛ̀rɛ́‑) to a preceding VP. But as with English can VP, the same-subject condition is a trivial consequence of the semantics and there is no corresponding different-subject construction.

More loosely chained verbs/VPs, such as those denoting sequences of discrete events or denoting causal (‘if …, then …’) relations, require overt subordinating machinery and tend to be less than rigorous in distinguishing same- from different-subject combinations. Nanga, for example, does have clause-linking subordinators that usually specify either same- or different-subject (abbreviations SS, DS) relations across clauses. As usual in Dogon, in Nanga the switch-reference values are cross-cut by a perfective/imperfective split at the level of the entire two-clause construction. If the event sequence has already occurred, the same-subject subordinator is ŋ (12a). When it is not, as in future, hypothetical, and deontic contexts, the same-subject subordinator is ndé (12b) or variant.

(12) a. [ńné ŋ́] yě:-rɛ̀-y

[go Pfv.and.SS] come-Pfv-1SgSubj

‘I went and (then) (I) came (back).’ (Nanga)

b. [yè:(-y) ndé] bíndé-m-ì

[come(-1SgSubj) Impf.and.SS] go.back-Impf-1SgSubj

‘I will come and (then) go back’ (Nanga)

Since these subordinators are productive and do not require tight event integration, there is no lexicalization or base-generation. To the extent that the SS/DS opposition is rigorous, it could be used as a test for subjecthood.

However, Dogon switch reference is often more shifty than the preceding remarks suggest. In those languages for which we have adequate textual corpora, we sometimes find “same-subject” subordinators where actual coindexation of subjects is doubtful. In Ben Tey, for example, lò=náyⁿ is the nonpast anterior same-subject form of ló ‘go’, and is used in contexts like ‘I will go and come (back)’ with a clearly understood shared subject. However, lò=náyⁿ is also used in narrative to connect one scenario (perhaps prolonged) to a second scenario beginning some time later, regardless of any subject coindexation. For example, textual passage (13) follows a description of a woman’s post-partum seclusion. The forty days are introduced into the discourse in this excerpt and cannot be coindexed with any previous discourse referent.

(13) [lò=náyⁿ]

[go=Nonpast.then.SS]

hâl ló [[ùsú pɛ́‑nì:y] dɔ̀‑ý]

until go [[day ten-four] arrive-3Hort]

‘This goes on until forty days eventually arrive (=elapse).’ (Ben Tey)

There are a few similar examples in the Jamsay textual corpus, where mèyⁿ is the primary anterior same-subject subordinator, but occasionally shows up in texts in the absence of subject coindexation.

Although a full study will require a larger textual corpus on several languages, it appears that in Dogon languages that oppose same- and different-subject subordinators, the latter is marked. This allows same-subject subordinators to spread into contexts like (13), where ‘forty days’ has only marginal status as a discourse referent, in comparison to the specific human referents that abound in discourse. Switch reference is sharpest in connection with such human referents.

For this reason, same-subject marking is less reliable than some other diagnostics for (referential) subjecthood. Different-subject marking is actually better evidence, since it highlights a referential switch.

1.5. Quotative particles in quoted indicative clauses.

Quotations in most Dogon languages are marked by uninflected quotative particles. An extended quotation may be explicitly framed by an inflected ‘say’ verb, but in actual texts particles are much more usual. In an extended quotation, an inflected ‘say’ verb may occur once, while particles recur through the entire quoted passage.

In several Dogon languages, quotative particles occur at two distinct break points in clauses. The first is after the clause-initial subject. The second follows the verb or other predicate, and also follows an interrogative particle (abbreviation Q) if present, but it precedes a clause-final emphatic particle.[6] Quotative particles regularly occur at the two break points, but do not also appear after clause-final emphatics. The schema is therefore (14), where underlined gaps indicate the break points.

(14) [subject NP] ___ [nonsubject constituents, predicate] (Q) ___ [emphatic]

Whether both quotative particles that appear in the two slots have the same phonological form depends on the language. In languages like Ben Tey where they diverge in form, the abbreviations are QuotSubj for post-subject and QuotPred for post-predicate particles, respectively. In languages with just one particle form, even if present in both slots, we can use Quot in interlinears.

In Ben Tey, QuotSubj particle ma: is phonologically distinct from the QuotPred particle wa, which becomes ba after a nasal (15). Both particles acquire tones secondarily.

(15) [[á ^HLbɔ̂:] mà:]

[[3LogoSgPoss ^HLfather] QuotSubj]

[[wóŋgóró wárá‑m̀] bà]

[[farming do.farm.work-Impf.3SgSubj] QuotPred]

‘He_x says that his_x father is farming.’ (Ben Tey)

When an inflected ‘say’ verb follows the quotation, the now redundant QuotPred wa is omitted, but QuotSubj ma: remains overt (16).

(16) [[â: mà:] wóŋgóró wárá‑yɛ̀]

[[3LogoPlPoss QuotSubj] farming do.farm.work.Impf-3PlS]

gìyⁿ‑bɔ́

say.Pfv-3PlSubj

‘They_x say they_x will farm.’ (Ben Tey)

Under some conditions, pronominal-subject suffixation on the verb is suspended, or neutralized as “3Sg” subject. With or without such suffixal reduction, the pronoun must be expressed clause-initially. For example, Ben Tey wàrá:-rà-y\ ‘we are doing farm work’ in (17a) shows the usual main-clause structure with suffixally expressed pronominal subject (1Pl -y\) and no clause-initial pronominal subject. When quoted (17b), the verb is neutralized to “3Sg” subject, and the 1Pl pronoun now occurs obligatorily, followed by QuotSubj ma:, in the clause-initial position that is regular for nonpronominal subject NPs in all clause types.

(17) a. wóŋgóró wàrá:-rà-y\

farming(n) do.farm.work-Impf1-1PlSubj

‘We are farming (=working in the fields).’ (Ben Tey)

b. [î: mà:] [wóŋgóró wàrá:-rà-w]

[1Pl QuotSubj] [farming(n.) do.farm.work-Impf1-3SgSubj]

gìyⁿ-Æ

say.Pfv-1SgSubj

‘He/She said that we are farming.’ (Ben Tey)

The QuotSubj particle in languages like Ben Tey not only provides a useful test for subjecthood, it also shows that there is a major syntactic break between subject and everything else, i.e. subject versus a classically defined VP (including aspect-negation inflection).

There is, however, one limitation of the QuotSubj test for subjecthood. This is that when the subject of the original quoted sentence was ‘you’, as in [He_x said to me_y “you_y saw me_x there”], there is the possibility that the QuotSubj phrase represents an original vocative or the combination of a vocative and a subject, as in [He_x said to me_y “hey you_y, you saw me_x there”]. This point will come up in §2.3 below in connection with quoted imperatives.

We have now seen overwhelming evidence from linear position, case marking, verbal agreement, antecedent/anaphor circumscription, quotative particles, and (to a lesser extent) switch-reference, that a typologically conventional subject NP is an important morphosyntactic category in Nanga and Ben Tey. Although the inventory of reliable tests differs from one language to another, all Dogon languages we have studied (more than a dozen) have a robust subject category of this type in ordinary intransitive and transitive clauses.

2. Imperatives and hortatives.

2.1. Imperatives

The absence of overt second person pronominals in imperatives is universal or nearly so (König & Siegmund 2007, Gusev 2013, Aikhenvald 2010). Often the only sign of an imperative is just this absence of otherwise obligatory pronominal subject marking, as in English, Kirundi (Ntahokaja 1994) and Koromfe (Rennison 1997).

A standard chestnut of Linguistics 101 is that this is a surface phenomenon that is disregarded by the syntax of agreement and anaphora. The key evidence for a covert but referentially active second-person subject in imperatives is the requirement of reflexive objects in (18) (# = ungrammatical).

(18) a. (Æ_x) Kill yourself_x (#you)!

b. (Æ_x) Kill yourselves_x (#you)!

In generative circles, we are aware of no fundamental disagreement of this basic schema; indeed, a profusion of phonologically empty categories is the hallmark of today’s generative syntax. There are many ways the null imperative subject has been accounted for, ranging from surface deletion of a subject you (after it does its work) as in early transformation grammar, to a base-generated but phonologically zero addressee-related pronominal (Schmerling 1982), to a null subject coindexed in part to a higher NP (Downing 1969), to a null subject controlled by a functional projection (e.g. Zanuttini et al. (2012). Pragmatic interpretation, by which the covert addressee agency of subjectless imperative clauses is inferred from illocutionary force, has also been suggested (Downes (1977). The effect, however, is the same: the referent to whom the imperative is addressed always behaves like a referential subject NP and can bind reflexives. Dogon languages are a glaring counterexample to this.

Dogon imperative verbs occur in two forms. For singular “subject,” the imperative stem is used, with no further affixation. The imperative stem is often distinguishable by tone and/or vocalism from other unsuffixed forms of the same verb stems, such as the perfective stem or the bare stem that is used in nonfinal position in verb chains. For plural “subject,” a special suffix, in most Dogon languages distinct from the regular 2Pl suffix used with indicative verbs, is added to the same imperative stem.

Since transpersonal reflexives are important in this discussion, Togo Kan (TgK) and Tomo Kan (TmK) will be the focal languages here. (19a-b) are simple TgK perfective indicatives with clause-initial 2Sg subject pronouns. In (19b), repeated from (6a) above, the coindexation of subject and object requires the transpersonal reflexive morpheme sǎⁿ. Imperative forms of the same ‘cut’ verb are in (19c-d); here the imperative stem differs from the perfective in its final high tone. In (19c), we see that there is no overt 2Sg subject pronoun like ú in (19a-b), and that the 1Sg object form is identical to that in indicative (19a). The surprise is that (19d) does not follow the English (and supposedly universal) pattern with covert 2Sg subject and overt reflexive object sǎⁿ. Instead, we see a regular 2Sg object form, so the literal translation is ‘Cut you!’ rather than ‘Cut self!’ The same 2Sg object form is illustrated in a nonreflexive indicative in (19e).

(19) a. ú má kɛ́j-ɛ̀

2SgSubj 1SgObj cut-Pfv

‘You-Sg (have) cut me.’ (TgK)

b. ú sǎⁿ kɛ́j-ɛ̀

2SgSubj ReflObj cut-Pfv

‘You-Sg (have) cut yourself.’ (TgK)

c. má kɛ́jɛ́

1SgObj cut.Imprt

‘Cut-2Sg me!’ (TgK)

d. ú / #sǎⁿ kɛ́jɛ́

2SgObj / #Refl cut.Imprt

‘Cut-2Sg yourself!’, lit. “Cut-2Sg you-Sg!” (TgK)

e. bé ú kɛ́j-ɛ́-sɛ̂ⁿ

3PlSubj 2SgObj cut-Pfv-PlSubj

‘They (have) cut you-Sg.’ (TgK)

The TmK counterparts are exactly parallel (19¢a-e). Again, the key point is that the 2Sg imperative “subject” in (19d) does not bind a reflexive, so the TmK construction is again of the type “Cut you!” rather than “Cut self!”[7]

(19¢) a. à mí-rⁿì cɛ̀ʔɛ́

2SgSubj 1Sg-Acc cut-Pfv

‘You-Sg (have) cut me.’ (TmK)

b. à há-lì cɛ̀ʔɛ́

2SgSubj Refl-Acc cut-Pfv

‘You-Sg (have) cut yourself.’ (TmK)

c. mí-rⁿì cɛ̀ʔá

1Sg-Acc cut.Imprt

‘Cut-2Sg me!’ (TmK)

d. à-lì / #hà-lì cɛ̀ʔá

2Sg-Acc / #Refl-Acc cut.Imprt

‘Cut-2Sg yourself!’, lit. “Cut-2Sg you-Sg!” (TmK)

e. à-lì cɛ́ʔɛ̀:-y

2Sg-Acc cut-Pfv-3PlSubj

‘They (have) cut you-Sg.’ (TmK)

Since an imperative “subject” does not have full subject properties in these languages, we will refer to it for now by the noncommital term actor as we explore its properties. The data show that both TgK and TmK diverge from the predominant typological pattern in withholding normal referential subject status from the covert second person actor. The surface omission of the virtual 2Sg subject in languages like English is trumped by the deeper syntactic absence of this element in the Dogon languages.

However, the actor is not entirely covert and inert in TgK and TmK imperatives. When the actor is nonsingular, in both languages a suffix ‑ỳ is added to the imperative stem. This suffix does not occur in regular pronominal-subject suffixation on indicative verbs. In TgK this suffixation has been reduced to a binary singular- versus plural-subject distinction, the morphology depending on the aspect-negation category, e.g. zero (19a-b) versus plural-subject -sɛ̂ⁿ (19e) in perfective positive indicatives. The 2Pl equivalents of 2Sg (19a-e) are in (20a-e). Plural-actor -ỳ occurs in (20c-d). As with the 2Sg, the 2Pl actor fails to bind a reflexive in (20d).

(20) a. é má kɛ́j-ɛ́-sɛ̂ⁿ

2PlSubj 1SgObj cut-Pfv-PlSubj

‘You-Pl (have) cut me.’ (TgK)

b. é [sǎⁿ bè] kɛ́j-ɛ́-sɛ̂ⁿ

2PlSubj [Refl Pl] cut-Pfv-PlSubj

‘You-Pl (have) cut yourselves.’ (TgK)

c. má kɛ́jɛ́-ỳ

1SgObj cut.Imprt-PlAddr

‘Cut-2Pl me!’ (TgK)

d. é / #[sǎⁿ bè ] kɛ́jɛ́-ỳ

2PlObj / #[Refl Pl] cut.Imprt-PlAddr

‘Cut-2Pl yourselves!’, lit. “Cut-2Pl you-Pl!” (TgK)

e. bé é kɛ́j-ɛ́-sɛ̂ⁿ

3PlSubj 2PlObj cut-Pfv-PlSubj

‘They (have) cut you-Pl.’ (TgK)

The TmK counterparts are exactly parallel (20¢a-e) except for minor morphological features.

(20¢) a. è mí-rⁿì cɛ̀ʔɛ́

2PlSubj 1Sg-Acc cut.Pfv

‘You-Pl (have) cut me.’ (TmK)

b. è hé-lì cɛ̀ʔɛ́

2PlSubj Refl.Pl-Acc cut.Pfv

‘You-Pl (have) cut yourselves.’ (TmK)

c. mì-rⁿì càʔà:-ỳ

1Sg-Acc cut.Imprt-PlAddr

‘Cut-2Pl me!’ (TmK)

d. è-lì / #hè-lì càʔà:-ỳ

2Pl-Acc / #ReflPl-Acc cut.Imprt-PlAddr

‘Cut-2Pl yourselves!’, lit. “Cut-2Pl you-Pl!” (TmK)

e. è-lì cɛ́ʔɛ̀:-ỳ

2Pl-Acc cut.Pfv-3PlS

‘They (have) cut you-Pl.’ (TmK)

We argue later that what -ỳ marks is specifically plural addressee (not plural subject), hence “-PlAddr” in the interlinears of (20c-d) and (20¢c-d). In other words, TgK and TmK imperatives index addressee number, though they lack fully referential subjects capable of binding reflexives.

In most other Dogon languages, the absence of transpersonal reflexives, and syntactic ambiguities in the analysis of possessed body-part reflexives of the type ‘(my/your/his) head/body/soul’ (see note 3), make it more difficult to prove that referential properties of imperative actors are suppressed. On the other hand there is no evidence that they are not suppressed. Suppression may therefore be a family-wide or even regional feature, but one that is only clearly visible in the few languages with transpersonal reflexives.

The suppression of referential properties does not mean that the subject position in Dogon imperatives is syntactically vacant. Examples like (20c) and (20¢c) with non-2nd person accusative-marked objects show that imperative objects are not promoted to subject position. This contrasts with the situation in certain nominative-accusative languages in which object NPs are morphologically accusative (or otherwise object-marked) in indicative clauses but nominative (or unmarked) in imperatives: well-known cases include Southern Paiute (Sapir 1930:179-81) and several other Uto-Aztecan languages, and Finnish (e.g. Moreau 1972, Kiparsky 2001). In those languages, the object behaves as though the subject isn’t there. This is not the case in Dogon, indicating that some non-object NP occupies the subject position (for purposes of case-marking), and there is no other candidate than the 2nd-person actor.

Before returning to the analysis of plural-actor marking, we round out the coverage of deontic modality by considering hortatives.

2.2. Hortatives.

Dogon inclusive hortatives (‘let’s go!’) have much in common with imperatives. The hortative has vocalism and tonal features similar to those of the imperative stem, but the hortative also has its own suffix (e.g. TgK ‑má). Imperatives and hortatives share a special negative suffix (‘don’t go!’, ‘let’s not go!’) that is not used with indicative verbs. In TgK this suffix is ‑lé, as in kɛ́jɛ́‑lé ‘don’t-2Sg cut!’ and kɛ́jɛ́‑m‑lé ‘let’s not cut!’ (with ‑má syncopated).

Imperatives and hortatives also share plural-actor suffixes, such as TgK -ỳ and TmK -ỳⁿ (nasalized from /-ỳ/) However, “plurality” in the hortative case refers exclusively to the number of addressees, disregarding the speaker. Therefore kɛ́jɛ́‑má ‘let’s-2Sg cut!’ is the form used with a single addressee, in what would normally be labeled as first person inclusive dual hortatives, while kɛ́jɛ́‑má‑ỳ ‘let’s-2Pl cut!’ is used when there are two or more addressees. In elicitation based on cues in languages like English or French that make no number distinction in hortatives, the TgK plural form ‑má‑ỳ is the first form volunteered by informants, perhaps because Dogon are inclined to think in terms of groups rather than pairs. By contrast, imperatives are most often directed at individuals, and singular-actor imperatives are more easily elicited than plural-actor counterparts.

A more crucial difference between imperatives and hortatives is that hortatives, unlike imperatives, have overt clausal subjects. In TgK, hortative clauses normally begin with an overt 1Pl subject pronoun ɛ́mɛ́, though like other such clause-initial pronouns it is often unstressed, is often set off prosodically, and may be elided. The speaker is included in this subject pronoun, which is always 1Pl, never 2Sg or 2Pl. Furthermore, like other overt subjects, ɛ́mɛ́ in hortatives is treated syntactically as fully referential. In particular, if the object of a transitive hortative is coindexed with the subject, we get a (transpersonal) reflexive object, always plural in this case. Observe TgK reflexive plural [sǎⁿ bè] in both (21b) with singular addressee and (21d) with plural addressee. All of (21a-d) show 1Pl subject ɛ́mɛ́ in clause-initial subject position.

(21) a. ɛ́mɛ́ súgó-má

1PlS go.down-Hort

‘Let’s (two) go down!’ (TgK)

b. ɛ́mɛ́ [sǎⁿ bè] kɛ́jɛ́-má

1PlS [ReflObj Pl] cut-Hort

‘Let’s (two) cut ourselves!’ (TgK)

c. ɛ́mɛ́ súgó-má-ỳ

1PlS go.down-Hort-PlAddr

‘Let’s (three or more) go down!’ (TgK)

d. ɛ́mɛ́ [sǎⁿ bè] kɛ́jɛ́-má-ỳ

1PlS [ReflObj Pl] cut-Hort-PlAddr

‘Let’s (three or more) cut ourselves!’ (TgK)

The TmK counterparts are again exactly parallel except for language-specific morphological details (21¢a-d). 1Pl subject bè is obligatory.[8] Reflexive plural pronoun hè, with accusative suffix -lì, is required in (21¢b,d) to index the referential identity of the 1Pl subject (not just that of the singular or plural addressee) with the object.

(21¢) a. bè sùgó-mà

1PlS go.down-Hort

‘Let’s (two) go down!’ (TmK)

b. bè hé-lì càʔá-mà

1PlS Refl.Pl-Acc cut-Hort

‘Let’s (two) cut ourselves!’ (TmK)

c. bè sùgò-mà:-yⁿ

1PlS go.down-Hort-PlAddr

‘Let’s (three or more) go down!’ (TmK)

d. bè hé-lì càʔà-mà:-yⁿ

1PlS Refl.Pl-Acc cut-Hort-PlAddr

‘Let’s (three or more) cut ourselves!’ (TmK)

The analysis of hortatives in other Dogon languages that lack transpersonal reflexives, such as Ben Tey and Nanga, is less sharp since those languages lack reflexive-object marking for non-third-person subject NPs. However, the distinction between addressee (excluding the speaker) expressed by a verbal suffix, and referential actor/subject (including the speaker), is regular in Dogon hortatives. For example, Nanga has singular-addressee hortative gǒ:‑má ‘let’s-2Sg go out!’ and plural-addressee gǒ:‑mà-yⁿ ‘let’s-2Pl go out!’, making the same addressee-number distinction as in TgK and TmK. This suggests that the difference is simply that TgK and TmK make overt a pattern that is present, though partially masked, in the other languages.

2.3. Subjecthood in chained imperative and hortative clauses.

Imperatives and hortatives cannot be relativized. However, both imperative and hortative utterances may include same-subject clauses or verb chains. Subjecthood tests based on these constructions can therefore be applied.

In §1.4 we showed that Nanga (for example) has both compound-like direct verb-chains and more complex forms of clausal subordination, both of which can require shared subjects in the two clauses. We warned however, that direct chains are semantically restricted and typically lexicalized, rather than being created by combining independent clauses. We also warned that the marked same-subject subordinators can spill into different-subject territory.

For what it’s worth, the Nanga direct chain (‘fall’ + ‘descend’ ® ‘fall down’) in indicative (11) can be directly converted into imperatives, with singular (22a) or plural addressee (22b), and into hortatives like (22c) for plural addressee. As in the indicative, the nonfinal verb is in bare-stem form and the final verb is inflected.

(22) a. tómbó sígô

fall descend.Imprt

‘Fall-2Sg down!’

b. tómbó sígô-ndì

fall descend.Imprt-PlAddr

‘Fall-2Pl down!’

c. tómbó sígé‑mà‑y

fall descend-Hort-PlAddr

‘Let’s-2Pl fall down!’

Similarly, the imperfective same-subject subordinator ndé in (12b) above recurs in (23), this time with an imperative final verb rather than the imperfective indicative verb in (12b).

(23) [yè:-w ndé] bíndò

[come-2SgS Nonpast.and.SS] go.back.Imprt

‘Come-2Sg and (then) go back!’ (Nanga)

Since direct chains and (theoretically) same-subject subordinators arguably presuppose that both verbs have referential subjects that make coindexation possible, (22a-b) and (23) can be counted as (weak) evidence for referential imperative subjects, countering the (stronger) evidence to the contrary from reflexive imperatives.

2.4. Subjecthood in quoted imperative and hortative clauses.

Quoted clauses of the Ben Tey type, with both QuotSubj and QuotPred markers as in (15) above, are potentially more interesting. They strongly suggest that imperatives are conceptualized as having a second-person actor.

The Ben Tey main-clause imperative (24a), with no overt second-person subject, can be quoted as (24b) or as (24c) among other possibilities, depending on who the original addressee was (from the perspective of the current speech event). If the original addressee is also the current addressee, we get (24b); if the original addressee was a third party such as Seydou (man’s name), we get (24c).

(24) a. pɛ̀rɛ̌‑m sɛ́wⁿá

sheep-AnSg slaughter.Imprt

‘Slaughter-2Sg (a/the) sheep-Sg!’ (Ben Tey)

b. [[ú má:] pɛ̀rɛ̌‑m sɛ́wⁿɛ́‑ý]

[[2Sg QuotSubj] sheep-AnSg slaughter-IndirImprt]

gìyⁿ‑ỳ

say.Pfv-1SgSubj

‘I told you-Sg to slaughter (a/the) sheep-Sg.’ (Ben Tey)

c. [[sěydù mà:] pɛ̀rɛ̌‑m sɛ́wⁿɛ́‑ý]

[[Seydou QuotSubj] sheep-AnSg slaughter-IndirImprt]

gìyⁿ‑ỳ

say.Pfv-1SgSubj

‘I told Seydou to slaughter (a/the) sheep-Sg.’ (Ben Tey)

Both (24b) and (24c) show overt subject NPs, followed by QuotSubj ma: in the appropriate tonal form. However, the imperative verb in (24a) has been replaced by a special verb form which, for present purposes, we will label indirect imperative. The variants of the suffix are ‑ý and ‑ỳ depending on verb class (sɛ́wⁿɛ́‑ý ‘slaughter!’, éw‑yè‑ỳ ‘sit!’). Although ‑ý ~ ‑ỳ is not used in main (non-quoted) imperatives, its affinity to deontic modality is brought out by its negative form ‑rɛ́‑ý, which contains a morpheme found elsewhere only in prohibitive (negative imperative) ‑rɛ́ ~ -lɛ́ and in negative hortative ‑rɛ̀‑ḿ ~ ‑lɛ̀‑ḿ.

Indirect imperative ‑ý ~ ‑ỳ is also used in imprecations, for example those of the type ‘may God VERB you’, as in (25).

(25) jǐnjɛ̀ ú bárì‑ỳ

God 2SgO help-IndirImprt

‘May God help you-Sg!’ (Ben Tey)

Since an imperative as such is not normally directable at God, such imprecations lead to the suspicion that ‑ý ~ ‑ỳ is more like a Romance-language subjunctive, rather than a straightforward quoted imperative.[9] Confirmation of this is the absence in (25) and similar predications of quotative markers, both QuotSubj and QuotPred, except when quoting another speaker who uttered the imprecation.

Still, (24b-c) do show an overt subject NP followed by the QuotSubj morpheme. That quoted imperatives have a referential subject is also indicated by TmK reflexives. But before we return to TmK, we must consider another analytic possibility, viz., that the “subjects” in quoted imperatives like (24b-c), unlike those in other quoted sentences, are really quoted vocatives of the ‘hey you!’ type, so that what (24b-c) really quotes is not “slaughter (a/the) sheep!” but “hey you, slaughter (a/the) sheep!” or “hey Seydou, slaughter (a/the) sheep!”

That overt imperative “subjects” might be vocatives seemingly receives typological support from afar in the form of Semele (Mon-Khmer, Malaysia).[10] Semele differs from Dogon languages in that objects as well as subjects appear to be treated as non-arguments in imperative sentences:

“In the imperative construction, pronominals are in the free form, indicating in the transitive clause that they are not attached to the verb, but function as vocatives … All vocatives, regardless of their form, are optional.” (Kruspe 2004:331)

“In the transitive imperative clause the case of the core grammatical relations O and IO … is not overtly coded, and it occurs, like the addressee, as a bare NP…” (ibid: 333)

However, the vocative status of the free pronominals in Semele imperatives is in doubt. Hortatives are a subtype of imperative, and the optional free-pronominal “vocative” of a hortative is first inclusive (“1&2”), not second person, which excludes a true vocative analysis :

“In the hortative construction … , used where the addressee is invited to perform an action which the speaker is also performing, the vocative is the first person inclusive hɛ ‘1&2’, or hɛ=ʔen ‘1&2’: (ibid: 331).

Semele imperatives, then, are more likely topicalized or otherwise external(ized) subjects rather than true vocatives.[11]

Returning to Dogon, there are also objections to the vocative interpretation of the BenTey QuotSubj phrases in (24b-c). First, a QuotSubj phrase is equally common in both quoted imperatives and quoted indicatives, whereas main-clause imperatives are only occasionally preceded by an overt ‘hey you!’ vocative. Second, the form of the NP or pronoun in the QuotSubj phrase reflects the deictic and conceptual perspective of the current speaker, not that of the original speaker. It may, for example, be first person as in (26), which could not represent an original first-person vocative:

(26) sěydù [[í má:] pɛ̀rɛ̌‑m sɛ́wⁿɛ́‑ý]

Seydou [[1Sg QuotSubj] sheep-AnSg slaughter-IndirImprt]

gìyⁿ‑Æ

say.Pfv-1SgSubj

‘Seydou told me to slaughter (a/the) sheep-Sg.’ (Ben Tey)

Third, consider quoted hortatives. We have seen that main-clause hortatives have both a 1Pl subject and a second person addressee. In quoted hortatives, there is just one QuotSubj phrase, and it is 1Pl, i.e. subject rather than addressee (or vocative).

(27) [[î: ^Lmà:] sèwá:rà lò‑ḿ] gìyⁿ‑Æ

[[1Pl ^LQuotS] Sevare go-Hort.Du] say.Pfv-3SgS

'He said (to me), let's (=he and I) go to Sevare.' (Ben Tey)

So we have no alternative but to consider the QuotSubj phrases in quoted imperatives to be subjects.

This does not, however, necessarily mean that covert main-clause imperative actors are subjects. In many languages, quoted imperatives have little resemblance to main-clause imperatives, appearing as e.g. English-type infinitives (He told me [to come]) and Spanish-type subjunctives (Me dijo [que venga]. The danger of extrapolating from quoted to main-clause imperatives is brought out by the treatment of reflexives in quoted imperatives in TmK, one of the languages which (unlike Ben Tey) has a transpersonal reflexive (hà) as shown in §2.1 above. TmK quoted clauses either begin with ká ‘say’ or end with a hò (or variant) ‘say’. In either case there is a conceptualized quoted speaker-author that can bind a logophoric pronoun (singular là, plural lè). If the subject of the quoted clause is pronominal and not logophoric, it appears as a locative pronoun (e.g. 1Sg m‑á:), not as a reduced subject pronominal (e.g. 1Sg ŋ̀) as in ordinary main clauses. (28a) is a main clause with L-toned subject pronominal, (28b) is its quoted equivalent with H-toned independent pronoun, and (28c) is a version of (28b) with a logophoric subject coindexed to the quoted author.

(28) a. ŋ̀ / à wé-lì cɛ̀ʔɛ́

1SgSubj/2SgSubj 3Sg-Acc cut.Pfv

‘I/You-Sg cut him.’ (TmK)

b. ká m-á: / á-: là-lì cɛ̀ʔɛ́

Quot 1Sg-Loc / 2Sg-Loc LogoSg-Acc cut.Pfv

‘He_x/She_x said that I/you cut him_x/her_x.’ (TmK)

c. ká là à-lì cɛ̀ʔɛ́

Quot LogoSgS 2Sg-Acc cut.Pfv

‘He_x/She_x said that he_x/she_x cut you.’ (TmK)

In TmK quoted imperatives, a pronominal actor (other than logophoric) appears as a locative pronoun, e.g. 1Sg m-á: (compare 1Sg subject ŋ̀). This has affinities to the QuotSubj phrase type in Ben Tey, insofar as the actor of a quotation is set off from the rest of the clause. Unlike the case with main-clause imperatives, which cannot bind the transpersonal reflexive hà, the subject of a quoted imperative does bind it. (29a) and (29b) are respectively indicative and imperative clauses with coindexed 2Sg subject and object, repeated from §2.1. They are followed by quoted imperatives with (29c) logophoric object, and with reflexive objects coindexed to either 1Sg (29d) or 2Sg (29e) clausemate subject. Unlike Ben Bey, TmK has is no special indirect imperative verb form; the main-clause imperative stem appears in quotations in its usual form.[12] The crucial point is that reflexive object hà-lì is required in quoted imperatives (29d-e) but is not allowed in main-clause imperatives (29b).[13]

(29) a. à há-lì cɛ̀ʔɛ́

2SgSubj Refl-Acc cut-Pfv

‘You-Sg (have) cut yourself.’ (=19b) (TmK)

b. à-lì / #hà-lì cɛ̀ʔá

2Sg-Acc / #Refl-Acc cut.Imprt

‘Cut-2Sg yourself!’, lit. “Cut-2Sg you-Sg!” (=19d) (TmK)

c. ká m-á: là-lì cɛ̀ʔá

Quot 1SgS-Loc LogoSg-Acc cut.Imprt

‘(He_x/She_x) told me to cut him_x/her_x.’ (TmK)

d. ká m-á: hà-lì / #mì-nì cɛ̀ʔá

Quot 1SgS-Loc Refl-Acc / #1Sg-Acc cut.Imprt

‘(He_x/She_x) told me to cut myself.’ (TmK)

e. ká á-: hà-lì / #à-lì cɛ̀ʔá

Quot 2SgS-Loc Refl-Acc / #2Sg-Acc cut.Imprt

‘(He_x/She_x) told you to cut yourself.’ (TmK)

TmK hortatives are main-clause (30a), repeated from §2.2, and quoted intransitive (30b) and reflexive transitive (30c). The 1Pl pronoun in (30a) is obligatory. (30b-c) each include both an optional locative pronoun denoting the addressee (phrased from the perspective of the current speaker) and a regular logophoric subject marker. The latter is always plural since it encompasses the original speaker and at least one original addressee. The double nature of hortatives as both addressee-directed commands and first-plural actions is well brought out here. Also of interest is that only logophoric can behave morphosyntactically in quoted clauses like a regular subject pronoun in main clauses.

(30) a. bè / #Æ sùgó-mà

1PlS / #Æ go.down-Hort

‘Let’s (two) go down!’ (=20’a) (TmK)

b. ká (m-á: / á-:) lè sùgó-mà

Quot (1Sg-Loc / 2Sg-Loc) LogoPlS go-Hort

‘(He/She) said (to me/you-Sg), let’s go!” (TmK)

c. ká (m-á:) lè hè-lì cɛ̀ʔá-mà

Quot (1Sg-Loc) LogoPlS ReflPl-Acc cut-Hort

‘(He/She) said (to me), let’s cut ourselves!’ (TmK)

To summarize the main points: Dogon quoted imperatives have overt subjects that bind reflexives (in those languages that have reflexives); quoted hortatives have overt subjects, and in TmK also an optional overt addressee; main-clause imperatives are conceptually associated with actors that never bind reflexives and that appear overtly only in the form of plural-addressee marking on the verb; main-clause hortatives have both overt 1Pl subjects and addressee-number marking on the verb.

We are not aware of a non-Dogon language that distinguishes second person indicative versus imperative objects in the TgK/TmK fashion, i.e. reflexive indicative ‘You cut self’ versus nonreflexive imperative ‘Cut you!’. The issue of imperative-actor referential properties is not addressed in the typological literature on imperatives and other deontics, to our knowledge. For example, we can find no reference to interactions between imperatives and reflexivization, either in an exhaustive and bibliographically omnivorous compilation on the typology of imperatives (Aikhenvald 2010) or in a typologically informed book on the syntax of imperatives (Alcázar & Saltarelli 2014). We suspect that had a Dogon-like case been previously known it would have found its way into this literature. So it may be that Dogon has now complicated the structural typology of deontic clauses and will force us to add “deontic addressee number” to the typological checklist of morphologically expressable grammatical categories.

2.5. Transpersonal reflexives, imperatives, and hortatives in Russian

The typologically rare and possibily unique structure of Dogon imperatives leads us to seek parallels elsewhere. Russian is promising since it has transpersonal reflexives, used in direct object and other functions, similar to those of TmK and TgK, allowing direct comparison of them in the syntax of imperatives and hortatives. The Russian forms are full reflexive sebja, cliticized variant‑sja (~ -sj), adverb sam ‘on one’s own’, and possessive svoj ‘own’. There is no singular/plural distinction. Polish and some other Slavic languages have similar sets of forms but differ in detail (Knyazev & Nedjalkov 1985; Medová 2009; Drogosz 2005, 2012; Tabakowska 2003).

As in Dogon, Russian reflexives bring out a distinction between two types of “subject,” but the Russian distinction has nothing to do with address or with imperatives. Rather, it involves competition between sebja and ‑sja, both of which are nonsubjects that are coindexed to an antecedent (usually the clausemate subject. Mining the Russian National Corpus for data, Toldova (2011) showed that sebja is primarily used in indirect object function (benefactive, recipient) in ditransitives, and as the complement of prepositions, while ‑sja is regular as direct object. In unelaborated canonical transitives, sebja is unacceptable as direct object, indicated by # in (31a). However, it becomes possible in the presence of sam ‘on one’s own’, which emphasizes agentivity (31b).

(31) a. po-brej-Æ -sja / # sebja

Pfv-shave-Imprt (-)Refl

‘Shave!’, ‘Get a shave!’ (Russian)

b. po-brej-Æ sebja sam

Pfv-shave-Imprt Refl on.one’s.own

‘Shave yourself!’ (Russian)

Some transitive verbs allow either ‑sja or sebja as direct object. For some verbs the two are interchangeable, as in beregi sebja and beregi-sj ‘take care of yourself!’ With other verbs, the choice is determined by imputed agentivity. In (32a) the addressee is being tortured by something external; in (32b) the pain is self-inflicted.

(32) a. ne much-a-j-sja

Neg torture-Æ-Imprt-Refl

‘Don’t torture yourself!’, ‘Don’t let yourself be tortured!’ (Russian)

b. ne much-a-j sebja

Neg torture-Æ-Imprt Refl

‘Don’t torture yourself!’ (Russian)

As shown by Khokhlova (1998), in passive sentences with a syntactically demoted but semantically agentive subject, the antecedent of sebja is the demoted subject (33).

(33) [etot fakt] byl interpretirovan im [dlja sebja]

[this fact] was interpreted by.him_x [for Refl_x]

sovsem ne tak

absolutely not like.that

‘This fact was interpreted by him (=He interpreted this fact) for himself (=in his own mind) entirely differently.’

By contrast, the antecedent of svoj ‘own’ is the NP that has been promoted from object to subject, e.g. the unfortunate banker in (34a). It cannot be a demoted subject, even if agentive, so in (34b) only a nonreflexive 3Sg possessor jevo is allowed.

(34) a. [molodoj bankir ] byl zastrelen [najomnim ubijtsej ]

[young banker]_x was shot [by.hired killer]_y

[v svojom offise]

[in own_x office]

‘A young broker_x was shot by a hired killer_j in his_x own office.’

b. [proshenie ob otstavke] bylo napisano

[letter about resignation] was written

premjerom [v jevo / # svojom kabinete]

by.prime.minister _x [in his_x / # own_x office]

‘The letter of resignation was written by the prime minister_x in his_x cabinet.’

So Russian reflexives bring out two distinct kinds of “subject”: a syntactic subject sensitive to passive-type processes, and an agent. Since imperatives presuppose agency by the addressee, the two types of subject converge in the (covert) second-person imperative subject, which binds sebja (35a) and possessive svoj (35b).

(35) a. ube-j sebja

kill-Imprt Refl

‘Kill yourself!’

b. ube-j [svoj-u žen-u]

kill-Imprt [own-Acc wife-Acc]

‘Kill your wife!’

Russian hortatives express the same dualism between addressee-oriented imperative and 1Pl agency seen in Dogon, English, and elsewhere. The typical construction is dava-j ‘give!’ (formal or plural addressee dava-j-te) plus either a 1Pl perfective verb or an imperfective infinitive. dava-j(-te) is optional, and is usually omitted is in ‘let’s go!’ (36). An overt 1Pl pronoun is not allowed, following the typologically widespread pattern of subject suppression documented by Dobrushina (2003).

(36) (# my) po-jdj-om

(# 1Pl) Pfv-go-1Pl

‘Let’s go!’

In spite of its (partial or full) suppression, the 1Pl subject binds reflexives in hortatives as in indicatives. In (37), sebe is a case form of sebja.

(37) dava-j po-govori-m [o sebe]

give-Imprt Pfv-speak-1Pl [about Refl]

‘Let’s talk about ourselves!’

To summarize, Russian transpersonal reflexives are variably sensitive to agency. They are bound by hortative subjects as in Dogon and English. They are also bound by imperative subjects, as in English and most languages, but unlike Dogon. Russian hortatives are expressed by combining an imperative (with second person subject) with either a conjugated verb (with 1Pl subject) or an infinitive. Russian hortatives reflect the double nature of hortatives (as in Dogon and English), but because of their two-part form they do not challenge the validity of “subject.”

2.6. Addressees and allocutives.

The closest structural parallels we know to Dogon imperative addressee-marking of are the family of ethical (or personal) dati