Journal of Slavic Linguistics

1 introduction

Anaphor binding is a broadly-studied and widely-used diagnostic for syntactic structure across languages: since anaphors (which are subject to Principle A of the Binding Theory) require an antecedent that is local, c-commanding and in an A-position, the possibility of anaphor binding is commonly used as a diagnostic for both c-command (e.g. Barss & Lasnik 1986) and A- vs. A'- distinction. This use of anaphor binding as a syntactic diagnostic is not without problems, however, as evidenced by the debate surrounding the structure of OVS sentences in Russian, to be discussed in more detail below. The goals of this paper are, thus, three-fold: (a) to gain a better understanding of anaphor binding as a syntactic diagnostic, (b) to shed new light on the long-standing and hotly-debated puzzle of OVS sentences in Russian, and (c) to explore the interaction of syntax and semantics, as concerns binding and thematic roles.

The rest of the paper is organized as follows. §2 introduces the reader to the basics of anaphor binding in Russian and the puzzle of OVS sentences. §3 discusses the properties of the Russian reflexive sebja 'self' in more detail, bringing to the fore a couple of additional complications concerning its distribution and interpretation. Finally, §4 attempts a solution that relates anaphor binding to thematic roles.

2 anaphor binding in russian and the ovs puzzle

As is true cross-linguistically, Russian has a class of anaphors, that is, elements that require an antecedent that satisfies the following ABC requirements (not to be confused with Principles A, B and C of the Binding Theory; Russian anaphors, like their counterparts in English and many other languages are subject to Principle A). First, an anaphor requires an antecedent that must be "A", i.e. in an A-position (rather than in A'-position). Second, the antecedent must be "B", i.e. binding, which means the antecedent must c-command the anaphor and the two must match in phi-features (if applicable). Third, the antecedent [End Page 1] of an anaphor must be "C", i.e. "close" or "local". The phi-feature requirement is only applicable to the reciprocal anaphor drug druga 'each other', whose antecedent must be plural, not to reflexive anaphors sebja 'self' and its possessive counterpart svoj 'self's', which lack phi-features. (The possessive reflexive agrees in features with the possessed, not with the antecedent.) The remaining requirements (A-position, c-command and locality) are illustrated in (1) below. In (1a), the antecedent of sebja 'self' is in an A-position (particularly, in Spec-TP), c-commands the anaphor and is in the same local binding domain, defined for present purposes as a finite clause. (We return to the question of the appropriate definition of local binding domain for sebja 'self' in §3 below.) In (1-b), 'Vanya' cannot be the antecedent of sebja 'self' because it does not c-command the anaphor; the relation between the putative antecedent and sebja 'self' here is one of subcommand. In (1-c), 'Vanya' cannot be an antecedent for sebja 'self' because it is not in the same clause as the anaphor. Finally, in (1-d), an OSV clause, the putative antecedent 'Vanya' fails because it is fronted by A'-movement to an A'-rather than an A-position and hence it cannot bind the anaphor sebja 'self'.

(1)

With this in mind, we can now turn to the so-called OVS puzzle. As is well-known, Russian allows all six imaginable orders of subject, object and verb, with the choice among them determined largely by discourse context. Of the six options, the SVO order, as in (2-a), is the most frequent in corpora (cf. Bailyn 1995, inter alia). The OVS order, as in (2-b), is "the most frequent non-canonical word order" (Ionin & Luchkina 2018, 742) and the only other order, besides SVO, that can be used discourse-initially. It occurs in 11% of all 3-member sentences (Bivon 1971); cf. also Sirotinina (1965), Bailyn (1995), Kallestinova (2007), inter alia.¹

(2)

[End Page 2]

Given its prominent nature, it is no surprise that the OVS order has attracted a great deal of attention from scholars of Russian syntax, with a plethora of analyses proposed to date (see the references cited in (3) below).

(3)

While the position of the verb and that of the subject are by no means agreed upon by all, here we shall focus on the position of the object, as it sets the goalposts for the placement of the two elements that linearly follow it. Two approaches to the position of the O in OVS have emerged; we shall call them the A-approach and the A'-approach, based on the type of position in which the object is argued to be located. Specifically, the advocates of the A-approach place the O in OVS in an A-position, particularly, in Spec-TP, as illustrated in (3). In other words, according to this approach the O in OVS is in the same position as the S in SVO and OSV. In contrast, the advocates of the A'-approach place the O in OVS in an A'-position, particularly, in Spec-CP (although some variation exists as to which exact A'-position the O in OVS is in). Thus, for the proponents of the A'-approach, the O in OVS is not in the same position as the S in SVO but in the same position as the O in OSV. This comparison to the placement of the subject and the object in various other word orders will be important below, as we can argue for this or that approach by showing patterns of similar behavior of various elements with respect to a variety of phenomena.

While arguments have been brought up in support of both approaches, in what follows we shall outline four arguments in support of the A-approach, followed by an argument for the A'-approach based on anaphor binding. That latest argument constitutes what we dub the OVS puzzle.

The first argument in support of the A-approach comes from the WCO (Weak Cross Over) effects, or rather from the lack thereof.² As shown in (4), the derivation of the OVS order does not cause a WCO violation, in contrast to the derivation of OSV order. [End Page 3]

(4)

The second argument in favor of the A-approach comes from subject oriented adverbs, such as oxotno 'willingly', neoxotno 'unwillingly', žadno 'greedily', etc. As shown in (5-a), in SVO clauses such adverbs are placed after the subject and before the verb. (The orders marked as ungrammatical here are, in fact, grammatical, but only if the adverb is contrastively focused; such clauses are not relevant here as we are interested in the discourse-neutral position of the adverbs.) As can be seen in (5-b), in OSV clauses these adverbs are likewise placed after the subject and before the verb. In contrast, in OVS clauses, such as (5-c), such adverbs are placed not after the subject but after the O and before the V (cf. Pereltsvaig 2019). In effect, the O in OVS occupies the same position in relation to a subject-oriented adverb as the S in SVO or OSV.

(5)

The third argument in support of the A-approach comes from the that-trace effect, which, as shown by Antonuyk-Yudina (2021, 6), is induced by the O in OVS in parallel to the S in SVO. As noted by Antonuyk-Yudina, this effect is ameliorated by additional adverbials inserted between the complementizer and the gap, as in the case of the that-trace effect in English. If we assume that the O in OVS is in Spec-TP, the that-trace effect can be formulated uniformly as a prohibition against extracting the element in Spec-TP following an overt complementizer.

(6)

[End Page 4]

Click for larger view
View full resolution

Table 1.

From Zubkov (2018, 4)

The fourth and final argument to be presented here in support of the A-approach is based on reciprocal anaphor binding. This binding is getting us ever so close to the OVS binding puzzle. As shown in (7-a), the O in OVS can serve as an antecedent of the reciprocal anaphor drug druga 'each other' embedded inside the S. (Note that for some speakers this is better if the S is inanimate, as in this example, as reported by Titov (2012, 95; 2018, 2); however, I have not been able to replicate Titov's observation in my surveys.) In contrast, the O in OSV, which is located unarguably in an A'-position, namely Spec-CP, predictably, cannot serve as the antecedent for drug druga 'each other'. Thus, once again the O in OVS and the O in OSV do not behave in the same way, as would be predicted by the A'-approach.

(7)

If the A-approach is on the right track, we expect the O in OVS to be able to serve as the antecedent for a reflexive anaphor, just as it can for a reciprocal anaphor (cf. (7) above). That expectation, as we shall see below, is not met. Hence, for the rest of this paper we shall explore differences between the reciprocal and reflexive anaphors in Russian and their differing behavior in OVS contexts. This, in effect, is the OVS puzzle. But first, let's consider the evidence showing that the O in OVS cannot bind sebja 'self' in the S position.

The most obvious way to test this is by attempting to make sebja 'self' the S in OVS. As shown in (8), OVS sentences where the O binds sebja 'self' in the S position are ungrammatical. Crucially, there does not seem to be a difference between OVS and OSV sentences in this respect, an argument used by advocates of the A'-approach in support of their position. There is, however, a very simple explanation for the ungrammaticality of sentences like (8-a): there is no nominative form for sebja 'self'. The form sebja 'self' is both accusative and genitive form, and there is no nominative counterpart. Moreover, speakers of Russian, when asked what could possibly be the nominative form, come up with a variety of tentative answers but exhibit no certainty in their "solution", a hallmark of a morphological gap. [End Page 5]

(8)

Since there is no nominative form of sebja 'self', we need to try other potential binding configurations. One way to test this is by using the above-mentioned possessive reflexive svoj 'self's', which has nominative forms (for the different genders and numbers). As can be seen in (9), the O in OVS can bind svoj 'self's' inside the S. Yet, there is a potentially confounding factor here. As noted in Padučeva (1983, 7), Rappaport (1986, 114–118), among others, svoj has several non-anaphoric uses, in which svoj does not require an antecedent at all. This can be seen from the Russian proverbs in (10), where svoj is contained within the S in SVO, which is in Spec-TP. Hence, there is no element in this sentence that could be a c-commanding antecedent for svoj, and yet the sentence is grammatical. Many additional examples, both proverbs and non-idiomatic sentences, can be brought to illustrate this further; a curious reader is referred to Padučeva's and Rappaport's work for additional examples. The necessary continuation in example (9) suggests that we are dealing with a non-anaphoric use of svoj. Yet, since the anaphoric and non-anaphoric uses of svoj are not easily distinguishable, the grammaticality of svoj in a particular configuration may be a false-positive. We claim that this is indeed what happens in (9). For the purposes of this paper, we shall use only the reflexive sebja 'self' and set aside the possessive reflexive svoj 'self's'.

(9)

(10)

So if we cannot use svoj 'self's' and there is no nominative form of sebja 'self', how can we test whether the O in OVS can bind into the S? Three structural configurations that contain sebja 'self' embedded into the S in OVS are discussed below. The first such configuration involves an eventive nominalization as the S, of which sebja 'self' is the internal argument. (Eventive nominalizations in Russian are discussed in detail in Pereltsvaig 2015, 2017, 2018a,b, Pereltsvaig et al. 2018.) A relevant – and grammatical! – example is given in (11). Here, the O in OVS, namely, Mjunxauzena (the accusative form of 'Munchhausen'), appears to bind sebja 'self,' which is the internal argument of the nominalized vytaskivanie 'pulling out'.

(11)

Does this mean that the O in OVS can bind sebja 'self' in the S? Our claim is that here too we have a confounding factor: in particular, the antecedent of sebja 'self' is not the object Mjunxauzena, but the unpronounced PRO subject of the eventive nominalization. The reference of the PRO itself depends on whether the predicate selecting the nominalization [End Page 6] is one of subject control or one of object control, as shown in (12).

(12) Binding mediated by control:

Thus, the example in (11) above has the structure as in (13), where the apparent binding of sebja 'self' by the sentence-initial O is mediated by a control relation: the O controls the PRO and the PRO binds the anaphor within the eventive nominalization. (We assume that nominals and not only clauses can serve as the local binding domain.)

(13)

The second configuration where we attempt binding of sebja 'self' inside the S by the O involves placing sebja 'self' as the Theme argument of a 'picture'-nominal.³ Since a single-word Theme argument of such a nominal would be preferred as expressed by a possessive (e.g., Vasina fotografija 'Vasya's photo' rather than fotografija Vasi 'photo of Vasya'), and we are trying to avoid the possessive reflexive svoj 'self's', for reasons discussed above, we need to make the internal argument multi-word, for example, by coordinating sebja 'self' with another nominal. What the resulting 'picture'-nominal would look like is shown in (14-a), where such a nominal functions as the O in SVO. Note that such a sentence was judged grammatical by speakers in an online survey.

What if we now make such a 'picture'-nominal the S in OVS and attempt binding of sebja 'self' by the sentence-initial O? As shown in (14-b), such a sentence was judged ungrammatical. Thus, we can conclude that the O in OVS cannot bind sebja 'self' inside the S.⁴

(14) Survey III
[Context: A picture of Sherlock Holmes and Dr. Watson appeared in the Times…]

[End Page 7]

A similar, yet slightly different, configuration emerges if we use a 'rumor'-nominal (e.g., sluxi o sebe 'rumors about self', anekdoty o sebe 'jokes about self', ljubov' k sebe 'love of self', (ne)uverennost' v sebe '(un)certainty in self' etc.) instead of a 'picture'-nominal; the chief difference between the two types of nominals is that 'rumor'-nominals take oblique rather than accusative complements. Here too, as shown in (15), the O in OVS cannot bind sebja 'self' in its oblique forms (e.g. its locative form sebe) inside the S.⁵

(15)

Thus, we end up with the following conundrum: the O in OVS cannot be an antecedent for sebja 'self', and the A-approach does not explain it. If we adopt the A'-approach instead, we are left with no explanation for the four patterns brought up above in support of the A-approach, namely, the lack of WCO effects, the that-trace effect induced by the O in OVS, the placement of the subject-oriented adverbs in OVS and the possibility of the O binding the reciprocal anaphor inside the S.

3 a closer look at sebja 'self'

In the previous section, we showed that sebja 'self' behaves in a puzzling way: while four pieces of evidence, including a comparison to another anaphor, suggest that sebja 'self' inside the S in OVS should be able to take the O as its antecedent, that expectation is not met. What makes sebja 'self' different from the reciprocal anaphor?

The existing literature on sebja 'self' (Timberlake 1979, Yang 1983, Rappaport 1986, Progovac 1992, Asarina 2005, Bailyn 2007, 2012, Marelj & Matushansky 2015, Zubkov 2018, Haspelmath 2019, 14) mentions three properties that characterize it in contrast to the reciprocal drug druga 'each other'. First, sebja 'self' is said to be monomorphemic. While it clearly is not, as it contains the stem and a case ending, the important insight here is that sebja 'self' has no phi-features. The second property of sebja 'self' is that it allows long-distance (LD) binding, namely it is possible for the antecedent to be outside the minimal binding domain (clausal or nominal). This occurs when sebja 'self' appears inside an infinitive clause, as in the oft-cited example from Timberlake (1979), given [End Page 8]

Click for larger view
View full resolution

Table 2.

Russian in the typology of long-distance binding according to Manzini & Wexler (1987) (+ indicates where long-distance binding is possible)

in (16). Here, the antecedent of sebja 'self' can be local, namely, PRO referring to the yard-keeper, or the PRO of the higher clause referring to the secretary, or the matrix subject 'the general'.

(16)

Beside infinitives, such an extension of the binding domain is possible in the case of small clauses and of complex DPs. In all three types of structures, what is lacking is the expression of tense or its structural counterpart, the TP. In this respect, the Russian reflexive anaphor sebja 'self' fits very nice with the typology of LD anaphors proposed by Manzini & Wexler (1987). In particular, sebja 'self' fits right between the Italian reflexive sè 'self', which cannot be LD bound in infinitives but only in small clauses and complex DPs, and thus requires an INFL, according to Manzini & Wexler (1987), and the Icelandic reflexive sig 'self', which can be LD bound not only in infinitives, small clauses and DP but also in subjunctive clauses, and thus requires independent/referential tense, according to Manzini & Wexler's typology.⁶ While the issue of the typology of LD anaphors is interesting in and of itself, here we move to the third oft-mentioned property of the Russian sebja 'self'.

The third characteristic of sebja 'self', noted in the earlier literature, is its subject-orientation, illustrated in (17): since sebja 'self' has no phi-features, we might expect both the subject 'Katya' and the object 'Vanya' to be possible antecedents of sebja 'self' (as is the case in the English translation), but that expectation is not met. Instead, only the subject and not the object can be an antecedent for sebja 'self', even in an SVO clause.

(17)

(18)

[End Page 9]

We shall return to the subject-orientation of sebja 'self' below, but for now it should be noted that these three properties—the lack of phi-features, the possibility of LD binding (in some structures) and the subject-orientation—are often noted to go together in comparative and cross-linguistic studies (Faltz 1977, Pica 1987, Cole et al. 2001a: xiv, xxx–xxxi; Reuland 2003 inter alia on Chinese, Italian, Icelandic, Kannada, Hindi). However, the Russian sebja 'self' cannot be simply assigned to the class of "long-distance anaphors", for three reasons. First and foremost, various elements across languages that allow for LD binding do not share the same set of properties, and upon closer examination some of them turn out to be pronouns or logophors and not anaphors after all, unless they are bound locally. For example, for English Hestvik & Philip (2001, 121) argue binding into 'picture'-phrases to be logophoric. The same does not hold of the Russian sebja 'self': even when its antecedent is outside the minimal binding domain, it still exhibits all the hallmarks of an anaphor. For example, unlike Mandarin Chinese ziji 'self', sebja 'self' requires a c-commanding antecedent, even when that antecedent is not local in a narrowly defined way. Moreover, the relation of subcommand, which allows for LD binding in Mandarin Chinese, does not allow for binding of sebja 'self' in Russian. For example, in (18), sebja 'self' inside an infinitive may be bound locally by the PRO (which is coindexed with 'Boris', its controller in the matrix clause), or it can be bound LD by the matrix subject 'Katya' (similarly to the LD binding by 'the general' in (16) above). What is not possible here is LD binding of sebja 'self' by 'Vanya' because this putative antecedent does not c-command the anaphor, being too deeply embedded inside the subject DP 'Vanya's sister Katya'. Another property that characterizes sebja 'self' as an anaphor, in contrast to ziji 'self' in Mandarin Chinese or the inflected reflexive kendi-sin 'self-3sg' in Turkish, is that sebja 'self' cannot take an extra-sentential antecedent (cf. Kornfilt 2001: 200 for Turkish and Pollard & Xue 2001: 329; Huang & Liu 2001: 157 for Mandarin Chinese). The ability of the Turkish inflected reflexive to take an extra-sentential antecedent is illustrated in (19-a), and the ungrammaticality of the corresponding binding configuration in Russian is shown in (19-b).

(19)

In addition, the Russian reflexive sebja 'self' cannot take split antecedents, again in contrast to the inflected reflexive in Turkish Kornfilt (2001, 205): compare the grammaticality of the split antecedent interpretation in Turkish in (20-a) with the ungrammaticality of its Russian counterpart in (20-b).

(20)

[End Page 10]

Furthermore, the Russian reflexive sebja 'self' exhibits only a sloppy reading under VP ellipsis (cf. Cole et al. 2001a: xvii–xviii; Lidz 2001: 239-240).⁷

(21)

Besides remaining a syntactic anaphor (i.e., subject to Principle A of the Binding Theory) even when LD bound, sebja 'self' does not fit with apparent counterparts in other languages in another important way. Recall from the above discussion that sebja 'self' is subject-oriented; cross-linguistically, subject-orientation of various LD anaphors has been accounted for in term of an agreement analysis, which correlates the possibility of LD binding with the absence of morpho-syntactic subject-predicate agreement. Thus, it has been noted that LD binding is possible in languages lacking subject-predicate agreement, such as Mandarin Chinese, or in languages that have subject-predicate agreement, LD binding is possible only in constructions where subject-predicate agreement is lacking (e.g. in Icelandic). The gist of the analysis is that the anaphor in such languages/constructions is bound not by another DP but by an Agr˚; if an Agr˚ in a local binding domain is lacking, binding by a higher Agr˚ is possible and since the subject (in Spec-AgrP) is co-indexed with Agr˚, apparent subject-orientation of LD anaphors emerges. Russian, however, is a language with morpho-syntactic subject-predicate agreement and yet it allows LD binding even in constructions with subject-predicate agreement, such as ECM clauses, contrary to the predictions of the Agr-based analysis (cf. Progovac 1992, developing insights in Yang 1983 and Pica 1987, and similar to the proposal in Kornfilt 1984, 1988, 2001). As illustrated in (22), sebja 'self' inside the predicate of a (bracketed) small clause can be bound either by a local subject, 'Holmes', or by an LD subject, 'Watson'.⁸ [End Page 11]

(22)

Yet another way in which sebja 'self' differs from "LD anaphors" in other languages, such as Mandarin Chinese, is the lack of blocking effect. Thus, in Mandarin Chinese (cf. Huang & Liu 2001: 142-143, 145-146; Pollard & Xue 2001), LD binding of ziji 'self' is possible only if the local antecedent matches the LD antecedent in features. In (23-a), Zhangsan and Lisi match in number, gender and person features, so either of those phrases can serve as the antecedent of ziji 'self'. In contrast, in (23-b), Zhangsan and ni 'you' do not match in person features and hence only local binding of ziji 'self' is possible.

(23) Mandarin Chinese:

In contrast, in Russian the person, number or gender features of the local antecedent do not matter as far as the possibility of LD binding: for example, in (24-a), the LD antecedent 'Holmes' and the local antecedent PRO (controlled by the object 'Watson') match in features, whereas in (24-b), the LD antecedent 'Holmes' and the local antecedent PRO (controlled by the object 'me') do not match in person, yet the grammaticality of LD binding in both examples is judged to be the same. The only difference in the grammaticality of LD binding in case of feature mismatch is that the 1^st person LD antecedent appears to be dispreferred, for reasons not yet understood.

(24) Russian:

The same is true if a mismatch is in number or gender (the lack of number mismatch is illustrated below).⁹ [End Page 12]

(26) No number effect (and there's also no gender effect)

Finally, a curious observation has so far eluded researchers, to the best of our knowledge: the subject-orientation of sebja 'self' is not absolute. Two exceptions emerged from our survey of native speakers. First, an object (in SVO or OVS) can in some cases be an antecedent of sebja 'self' inside an oblique argument, for about 20-30% of the speakers. For example, in (27), virtually all speakers accept the subject as the antecedent of sebe 'to self' (here, an external possessor of 'room', in order to avoid svoj 'self's'). However, 20-30% of speakers accept the direct object, 'Watson', as the antecedent of sebe 'to self'. Importantly, the SVO vs. OVS order does not seem to matter, and contrary to the prediction of the A-approach, changing the order from SVO to OVS does not make the O a better antecedent for the reflexive anaphor. (If anything, the acceptance of the O as the antecedent in OVS is slightly lower than in SVO.) Another crucial observation in connection with these examples is that the matrix predicate is a causative one: SEND potentially translates as CAUSE GO. This observation will be important for the eventual analysis that we develop for these and other facts in the following section. We shall therefore call this a "causative effect".

(27) [Context < in Russian>: On the envelope there were several little brown spots. What is it: dirt, blood or gravy sauce? Holmes couldn't see it clearly with a naked eye. And so…]
Survey II

Another apparent exception to the subject-orientation of sebja 'self', which we shall dub the "psych-verb effect", is as follows: binding of sebja 'self' inside the S by the O in OVS is [End Page 13] somewhat more acceptable if the matrix predicate is a psych verb than with a non-psych verb. (A similar effect is also noted by Sells (1987, 476) for Italian proprio 'self'.) Thus, in (28), speakers judge a sentence with a psych verb such as udivit' 'surprise' better than a similar sentence with a non-psych verb such as diskreditirovat' 'discredit'. Although the average difference in grammaticality between these two sentences is not huge (2.4 vs. 1.7 on a 1-5 scale), nearly half the speakers show some preference for (28-a), with a psych verb, over (28-b) with a non-psych verb.

(28) [Context: A picture of Sherlock Holmes and Dr. Watson appeared in the Times…]

Given that subject-orientation is seen as a crucial property of LD bound elements in the previous literature, the ultimate analysis should account for these two exceptions—the causative effect and the psych-verb effect—as well as for the overall subject-orientation in examples such as (17) above.

4 the solution: where syntax and semantics meet

The solution, we believe, is in an additional requirement that sebja 'self' imposes on its antecedent; to the best of our knowledge, this observation has only been made in passing in Zubkov (2018, 49–57). Thus, we claim that the OVS puzzle and the two apparent exceptions to the subject-orientation of sebja 'self' can be understood if, in addition to the "ABC" requirements mentioned in section 1, sebja 'self' imposes yet another, "D," requirement, namely, the de se requirement. (De se translates from Latin as 'of self'.) This requirement can be formulated as follows: the antecedent of sebja 'self' must denote an individual who self-ascribes the property denoted by the predicate containing the reflexive.

The idea of the de se requirement comes from the work of Giorgi (1984) and Chierchia (1989) on the Italian possessive reflexive proprio 'self's'. They note that proprio is appropriate only if the individual denoted by the antecedent self-ascribes the relevant property. For example, consider (29) in the following context: just before a Three-Tenors concert, an assistant wheels a rack with the costumes into the performers' dressing room. Pavarotti sees that a pair of pants has caught on fire, but he does not know whose pants these are as they all look the same on their hangers. In such a context, claim Giorgi and Chierchia, the use of propri 'self's' is impossible (even though the sentence is grammatical and can be used if Pavarotti realizes that the pants in question are his own).

(29)

[End Page 14]

A similar effect has been noted with respect to reflexive anaphors in several other languages, most notably with respect to kaki 'self' in Teochew Chinese (a Southern Min language spoken in Guangdong; Cole et al. 2001b: 23) and the LD anaphor ziji 'self' in Mandarin Chinese (Pan 2001, 293).¹⁰

Our proposal is that LD bound sebja 'self' in Russian is also subject to the de se requirement, namely that its antecedent ascribes to him/herself the proposition of which sebja 'self' is an argument. This observation can be appreciated in the context of a Russian joke about Comrade Brezhnev (who, at least according to the Soviet lore, was barely in touch in reality, especially in his later years). Consider the following context: Comrade Brezhnev is having a guided tour of an art museum and is nodding at each artwork with a knowledgeable expression. Then, he stops in front of a mirror. He points at his reflection, not recognizing it as such, and asks who that is. The reflection, obviously, points back at him. Brezhnev starts to make funny faces, teasing who he thinks is another man in an art installation, and "the man," obviously, teases him back. Comrade Brezhnev gets very angry and orders to arrest "that man". In this context, the example in (30) is judged as unacceptable. (Again, the sentence is grammatical and is fine if Brezhnev recognizing the reflection as being, in some sense, "himself"; for a more detailed discussion of such "near-reflexive" readings, see Lidz 2001.)

(30)

Beside a situation of a mental illness or dementia, as in the above example, another way in which one might not ascribe a certain property to oneself without self-ascribing (i.e., without realizing that one indeed ascribes the property to oneself) is if one speaks of a future situation in which the identity of a given definite description is not yet known (to the protagonist). For example, in the context of being on the eve of a presidential election, a candidate may make a statement about whoever gets elected (which, presumably, is not known yet, if the elections are fair). Such a situation is illustrated below:

(31) [Context: On the eve of a presidential election in Ukraine, the then-candidate Zelensky suggests that Trump should invite whoever is ultimately elected Ukrainian president to Washington.]

Let's now consider how the de se requirement translates into the subject-orientation property of sebja 'self' and its exceptions. Here, we propose that in order to satisfy the de se requirement, a putative antecedent must meet the minimum threshold: in particular, in order to be able to self-ascribe a property, one must have the mental capacity to ascribe propositions at all. As noted by Zubkov (2018, 46–47), LD binding of sebja 'self' requires the LD antecedent to be animate, as can be seen from (32): 'Katya' but not èta kniga 'this book' can be a LD antecedent of sebja 'self'. However, I take animacy to be a pre-requisite for the required sentience of the antecedent.

(32)

[End Page 15]

We would like to take this one step further: in order to pass the minimum threshold of being able to ascribe properties, the individual's mental capacity should not be merely known to the speaker or the addressee but it should be known from linguistic context rather than from general encyclopedic knowledge. Here, we develop ideas from Ariel (2014), who claims that knowledge from linguistic context preempts knowledge from general encyclopedic knowledge. Our claim is that knowing that a given individual can ascribe properties from general encyclopedic knowledge is simply not good enough.

But what are the linguistic means of expressing an individual's mental capacity (or the lack/irrelevance thereof)? And since it appears that the S vs. O is more important to the binding of sebja 'self' than structural positions such as Spec-TP, how do those linguistic means of expressing an individual's mental capacity relate to grammatical functions? Here, we turn to ideas in Dowty (1991), who addresses the very important question of where theta-roles come from and how they relate to grammatical functions. According to Dowty, theta-roles and grammatical functions are tied to the semantic implications of a given predicate. In particular, the argument with the most Proto-Agent properties maps to the S, the argument with the most Proto-Patient properties maps to O. The Proto-Agent and Proto-Patient properties are listed below. As can be seen from (33), two of the Proto-Agent properties directly relate to the individual's mental capacity: its "volitional involvement in the event or state" and its "sentience and/or perception". Crucially, none of the Proto-Patient properties relate to one's mental capacity.

(33)

i. Proto-Agent properties:

(i) volitional involvement in the event or state (mental capacity)

(ii) sentience (and/or perception) (mental capacity)

(iii) causing an event or change of state in another participant

(iv) movement (relative to the position of another participant)

(v) (exists independently of the event named by the verb)

ii. ii. Proto-Patient properties:

(i) undergoes change of state

(ii) incremental theme

(iii) causally affected by another participant

(iv) stationary relative to movement of another participant

(v) (does not exist independently of the event, or not at all)

Let's now see how this works to account for the subject-orientation of sebja 'self' in a "normal" transitive clause, with a non-psych predicate, such as the example in (17), repeated below as (34). Although both Vanya and Katya are animate individuals, only one of them has a theta-role that indicates that individual's mental capacity: Katja's being an Agent (and hence the S) derives in part from Katya's volitional involvement in the event in question. Vanja, albeit denoting an animate and sentient individual as well, does not wear its mental capacity on its sleeve, so to speak. We know that Vanya has a mental capacity to ascribe properties, but we know it from our general encyclopedic knowledge (since Vanya is a human), not from linguistic context, such as Vanja's theta-role. In other words, in such a clause the S has the Proto-Agent properties including volitionality and/or sentience, i.e. properties that mark it as having the mental capacity for ascribing properties; in contrast, the O does not. Therefore, the S, Katja, passes the minimum threshold for the de se requirement. Furthermore, the grammaticality of the sentence with that coindexing indicates that Katja indeed satisfies the de se requirement, i.e., that she is aware that that her questions are about herself. [End Page 16]

(34)

To recap, the O in either OVS or SVO becomes the O in the first place because it has Proto-Patient rather than Proto-Agent properties, such as volitionality and/or sentience. Consequently, its volitionality/sentience cannot be known from its theta-role. Thus, the O does not satisfy the minimum threshold for the de se requirement and cannot be the antecedent of sebja 'self'. Importantly, the word order does not matter in this respect and neither does the structural position of the O, inside the VP or in Spec-TP. We now turn to the apparent exceptions to the subject-orientation discussed above: the causative effect and the psych-verb effect.

First, let's consider the causative effect, illustrated in (27) above. Here the object of a causative verb can be the antecedent of sebja 'self' inside an oblique argument, for about 20-30% of the speakers. Our proposed analysis is that the causee object in such sentences can be interpreted as a volitional Agent of the embedded predicate, in this example, GO. After all, if Holmes sends Watson somewhere, it is still up to Watson to decide whether he goes or not.

Second, let's consider the psych-verb effect, illustrated in (28) above. As discussed in detail in Dowty (1991), some psych verbs are exceptional to the overall pattern that volititional/sentient arguments are the S; this includes such Russian verbs as ispugat' 'frighten', udivit' 'surprise', rasstroit' 'upset', bespokoit' 'worry', etc. With the so-called object-experiencer verbs, the sentient Experiencer maps to the O. According to Dowty (1991, 579–580), this is because such predicates imply a Proto-Patient property for the Experiencer, namely a change of state in the Experiencer, which correlates with an inchoative interpretation of 'frighten' (vs. subject-experiencer psych verbs such as 'fear'); the reader is referred to Dowty's work for a more detailed discussion. In other words, with object-experiencer psych verbs, which are the ones giving rise to the psych-verb effect in Russian, the Experiencer maps to the O despite its sentience. It is the sentience, however, and not directly the grammatical function or the structural position that matters.

5 summary and further questions

In this paper, we have argued that the reflexive anaphor sebja 'self' in Russian imposes not only the three well-understood "ABC" requirements on its antecedents ("A" for A-position, "B" for binding, i.e. c-commanding position, and "C" for closeness, i.e. locality, defined for sebja 'self' as a domain defined by Tense), but also a fourth condition, namely "D" for "de se requirement". In other words, the antecedent of sebja 'self' must self-ascribe the relevant property. The de se requirement comes with a minimum threshold: the antecedent must denote an individual capable of ascribing properties, that is one with certain mental capacities, and such mental capacities must be known from linguistic contexts (specifically, from the antecedent's theta-role) rather than from general encyclopedic knowledge.

This proposal allows us to shed new light on the OVS puzzle: despite a host of A-position-related properties, the O in OVS is unable (in most cases) to be an antecedent for sebja 'self'. According to the proposed analysis, this has nothing to do with the O's structural position (which we take to be an A-position, Spec-TP). Instead, the inability of the O in OVS to bind sebja 'self' relates to the fact that although the O satisfies the "ABC" requirements of the anaphor, it fails to satisfy the "D" (de se requirement), since its volitionality/sentience cannot be known from linguistic context (i.e. theta-role). In relatively rare cases, such as objects of causative predicates and of object-experiencer psych-verbs, the O in OVS (or SVO, for that matter) can be the antecedent for sebja 'self', at least for some speakers of Russian. This is because in those exceptional cases, the O [End Page 17] bears a theta-role that is associated with volitionality/sentience implications.

Finally, we would like to note that if one presupposes the "inverted Y" model of syntax, it is unsurprising to find that anaphor binding (which occurs at LF rather than PF) interacts with other LF phenomena (such as theta-role implications) and not with PF properties such as word order.

One big question left unanswered in this regard is what happens in passives in Russian: can the internal argument that appears in Spec-TP (same as the O in OVS, according to the proposed analysis) be an antecedent of sebja 'self'? We leave this question open for future research and a different paper, however, because the two issues that need to be addressed in order to answer this question deserve their own separate papers: (i) what are the empirical facts concerning binding of sebja 'self' in Russian passives? and (ii) what is the proper analysis of Russian passives, even without binding? The first question is touched upon in the existing literature, but our preliminary survey of Russian native speakers indicates that there is more variability and complexity in the data than previously reported. As for the second issue, there are at least two possible analyses one could consider: one along the lines of Baker et al. (1989) and the "smuggling"-type analysis along the lines of Collins (2005). Discussing these issues would be central to understanding the workings of sebja 'self', so we hope to undertake it in the future.