If you cannot agree, move on! On labels and non-nominative subjects

This paper examines the predictions of Chomsky’s (2013; 2015) Labeling Algorithm for structures with non-nominative, non-agreeing elements in what looks like a subject position: since such non-nominative subjects do not agree with the verb, the resulting structure cannot be labeled through feature sharing. Thus, the only option for such subjects is to move from [Spec, TP] or never land there. We argue for the latter option, motivating it for PPs in Polish locative inversion, instrumental DPs in Polish predicate inversion, dative subjects in Russian and Polish, genitive subjects in Lithuanian Inferential Evidentials, and accusative DPs in Russian and Polish Adversity Impersonals. We conclude by showing that under certain very restricted circumstances (i.e. when partial Agree takes place or Agree for structural case holds but Agree for phi-features is attempted and fails for independent reasons), the non-nominative elements can remain in [Spec, TP].


Introduction
Recent work on labels, inspired by Chomsky's (2013; Labeling Algorithm, has led to new insights on the nature of successive cyclic movement, criterial positions and locality (Blümel 2013;Rizzi 2014;Cecchetto & Donati 2015;Bošković 2016, among others). In this paper, we examine the behavior of non-nominative, non-agreeing XPs from a labeling perspective. Our empirical focus is on Polish, Russian and Lithuanian, languages that allow a variety of non-nominative elements in preverbal positions. We focus on the following five types: (i) PPs in Polish locative inversion (1a) (see Babyonyshev 1996;Harves 2002;Bailyn 2004, among others, for Russian), (ii) instrumentals in Polish predicate inversion structures (1b) (Moro 1997; 2000 on Italian and English; Bondaruk 2013a on Polish, among others), (iii) dative subjects in Russian and Polish (1c) (Dziwirek 1994;Franks 1995;Moore & Perlmutter 2000;Sigurđsson 2002;Bailyn 2004, among others), (iv) genitive subjects in Lithuanian evidential constructions (1d) (Lavine 2000;, and (v) accusatives in Russian and Polish Adversity Impersonals (1e) (Babby 1994;Harves 2002;Lavine & Freidin 2002;Bailyn 2004). 1 (1) a. Polish Do pokoju wszedł Jan. to room entered Jan.nom 'Into the room walked John.' b. Polish Moim najlepszym przyjacielem był Jan. my.instr best.instr friend.instr was Jan.nom 'My best friend was Jan.' c. Polish Janowi podoba się ta książka. Jan.dat pleases refl this.nom book.nom 'Jan likes this book.' d. Lithuanian (Aušra Valančiauskienė, p.c.) Mokytojo ištaisyta klaidos. teacher.gen corrected mistakes.nom 'The teacher apparently corrected the mistakes.' e. Polish Łódkę wyrzuciło na brzeg. boat.acc threw.out onto shore 'The boat got thrown onto the shore.' A popular analysis for such constructions is the so-called Generalized Inversion analysis (in Bailyn's 2004 terms), in which the non-nominative element moves to [Spec, TP] for EPP related reasons (see Babyonyshev 1996;Lavine 2000;Harves 2002;Lavine & Freidin 2002;Bailyn 2004; among many others): 2 We show that such an analysis is incompatible with the Labeling Algorithm and argue for an alternative in which these XPs do occupy a higher position instead, as shown in (3). 3 (3) We proceed as follows. In the next section (Section 2), we review Chomsky's (2013; Labeling Algorithm and the predictions it makes for non-nominative subjects. In Section 3, we turn to empirical considerations; in Section 3.1, we review subjecthood diagnostics; 2 Bailyn (as well as the others cited here) focuses on Russian not Polish. However, the same idea has been extended to Polish (see, for example, Witkoś 2008). 3 In principle, there are (at least) two other possibilities that are consistent with the Labeling Algorithm. The non-agreeing XP could move first to [Spec, TP] and next to a higher position, where labeling could take place: Movement could also be direct, and [Spec, TP] could be occupied by an empty expletive element: However, we take the status of null expletives as elements that are both phonologically and semantically vacuous to be somewhat suspect from a theoretical perspective (see Germain 2015 for further discussion), so we will not pursue this option any further. Another reason not to pursue the null expletive option, suggested by one of the reviewers, concerns intervention; the null expletive in [Spec, TP] would block A-movement of the non-nominative argument to [Spec, FP]. This is not a problem for locative and predicate inversion, since we argue that these involve A-bar movement but it is a problem for dative, genitive, and accusative "subjects" since these for us involve A movement.
Labeling Algorithm (LA) simply selects the closest head as the label. 4 If the syntactic object (SO) to be labeled consists of a head and a phrase (as in (4a)), the LA selects the head as the label, resulting in (4b). If, however, it consists of two phrases (as in (5a)), the LA locates two potential heads (i.e. the head of XP and the head of YP) and is unable to assign an unambiguous label.
There are two ways to label the structure in (5b): either one of the phrases moves ('labeling by evacuation'), leaving only one head as a potential label, or some feature (e.g. the interrogative feature or phi-features) has to be shared ("labeling by feature sharing"). 5 Moro's (1997; analysis of small clauses provides a good illustration of the first option. The result of merging a subject and a predicate DP cannot be labeled (since two phrases are merged), as shown in (6b). Only after one DP moves, the other one can provide the label, as shown in (6c). Labeling is also an issue for subjects in [Spec, TP]; since the subject DP has merged with TP, the result cannot be labeled. Chomsky (2013: 45) suggests that "perhaps that [labeling] can be achieved by the device suggested for embedded interrogatives". DP and TP share prominent features, namely phi-features. 6, 7 This (5) a. Citko,Germain,& Witkoś 4 Labeling Algorithm (LA) simply selects the closest head as the label. 4 If the syntactic object (SO) to be labeled consists of a head and a phrase (as in (4a)), the LA selects the head as the label, resulting in (4b). If, however, it consists of two phrases (as in (5a)), the LA locates two potential heads (i.e. the head of XP and the head of YP) and is unable to assign an unambiguous label.
There are two ways to label the structure in (5b): either one of the phrases moves ('labeling by evacuation'), leaving only one head as a potential label, or some feature (e.g. the interrogative feature or phi-features) has to be shared ("labeling by feature sharing"). 5 Moro's (1997; analysis of small clauses provides a good illustration of the first option. The result of merging a subject and a predicate DP cannot be labeled (since two phrases are merged), as shown in (6b). Only after one DP moves, the other one can provide the label, as shown in (6c). Labeling is also an issue for subjects in [Spec, TP]; since the subject DP has merged with TP, the result cannot be labeled. Chomsky (2013: 45) suggests that "perhaps that [labeling] can be achieved by the device suggested for embedded interrogatives". DP and TP share prominent features, namely phi-features. 6, 7 This b. Citko,Germain,& Witkoś 4 Labeling Algorithm (LA) simply selects the closest head as the label. 4 If the syntactic object (SO) to be labeled consists of a head and a phrase (as in (4a)), the LA selects the head as the label, resulting in (4b). If, however, it consists of two phrases (as in (5a)), the LA locates two potential heads (i.e. the head of XP and the head of YP) and is unable to assign an unambiguous label.
There are two ways to label the structure in (5b): either one of the phrases moves ('labeling by evacuation'), leaving only one head as a potential label, or some feature (e.g. the interrogative feature or phi-features) has to be shared ("labeling by feature sharing"). 5 Moro's (1997; analysis of small clauses provides a good illustration of the first option. The result of merging a subject and a predicate DP cannot be labeled (since two phrases are merged), as shown in (6b). Only after one DP moves, the other one can provide the label, as shown in (6c). Labeling is also an issue for subjects in [Spec, TP]; since the subject DP has merged with TP, the result cannot be labeled. Chomsky (2013: 45) suggests that "perhaps that [labeling] can be achieved by the device suggested for embedded interrogatives". DP and TP share prominent features, namely phi-features. 6,7 This There are two ways to label the structure in (5b): either one of the phrases moves ('labeling by evacuation'), leaving only one head as a potential label, or some feature (e.g. the interrogative feature or phi-features) has to be shared ("labeling by feature sharing"). 5 Moro's (1997; analysis of small clauses provides a good illustration of the first option. The result of merging a subject and a predicate DP cannot be labeled (since two phrases are merged), as shown in (6b). Only after one DP moves, the other one can provide the label, as shown in (6c). Labeling is also an issue for subjects in [Spec, TP]; since the subject DP has merged with TP, the result cannot be labeled. Chomsky (2013: 45) suggests that "perhaps that [labeling] can be achieved by the device suggested for embedded interrogatives". DP and TP share prominent features, namely phi-features. 6,7 This makes a clear prediction for constructions with non-agreeing XPs in [Spec, TP]; since there is no agreement, labeling through phi-feature sharing is not an option. We thus argue that non-nominative subjects typically do not occupy [Spec, TP] (see Section 4 though for two principled exceptions to this generalization). To make this argument, we rely on the following assumptions. First, we assume a version of the Split CP Hypothesis of Rizzi (1997), in which different types of uninterpretable features originate on different heads. Phi-features start on the Finiteness head and are inherited by T, and wh-features start on the Force head and are inherited by the Focus head, as shown in (7a-b). 8,9 More generally, we assume that Force head is also the locus of discourse-related uninterpretable features ([uTop] and [uFoc] features, each associated with its 'own' EPP property). These are the δ-features of Miyagawa (2010;. In cases considered in this paper, the Topic head inherits the [uTop] feature from the Force head, and since we are not dealing with interrogative or focused constructions, FocP is inactive or absent altogether: We further assume that feature inheritance is optional, following Legate (2011), Ouali (2008, among others. This opens up the possibility that phi-features can remain on the Finiteness head, as shown in (9a). The novel option, due to Germain (2015), is the option illustrated in (9b), where phi-features are inherited by T but EPP features remain on Fin. Following Germain (2015), we refer to this option as Split Feature Inheritance. With these assumptions in place, we can motivate two distinct derivations for non-nominative subjects, schematized in (10a-b). In (10a), the non-nominative XP moves 6 This suggests that the result is a Phi Phrase/Person Phrase or Subject Phrase (as proposed by Rizzi and Shlonsky 2006). However, Pesetsky and Torrego's (2001) view of nominative case as a [uT] feature provides an intriguing alternative that the shared feature is simply the T feature. We thank Edith Aldridge for the suggestion. 7 This raises the question of how TP is labeled in pro-drop languages. For Chomsky (2015), in languages with rich agreement, T is strong enough to label, in languages with weak agreement, T is not and must be "strengthened" by an agreeing specifier. However, not all Ts we consider are agreeing Ts (some show default agreement, for example), and not all languages we consider here are pro-drop languages; while Polish and Lithuanian are, Russian is not (at least not in the canonical sense). 8 This division captures the A versus A-bar distinction; specifiers of Fin and T are A-positions (since these heads are hosts of phi-features) whereas specifiers of Force, Focus and Topic heads are A-bar positions (since these heads are the locus of quantificational and/or discourse features). 9 A question brought to our attention by one of the reviewers is whether (and if so, how) our proposal works in a non-split system such as that of Miyagawa (2010;, where both agreement features and discourse features enter the derivation on C and get inherited by T, subject to crosslinguistic variation. The split system we adopt here allows more options and is independently motivated (most notably by Rizzi 1997 With these assumptions in place, we can motivate two distinct derivations for non-nominative subjects, schematized in (10a-b). In (10a), the non-nominative XP moves to [Spec, TopP] and the [Spec, TP] position remains unprojected. This is what happens, we argue, in locative and predicate inversion structures, which are the focus of Section 3.2. The second option is given in (10b); the non-nominative "subject" does not move to [Spec, TP] either, but, instead, it moves to a different position (i.e. [Spec, FinP]), the option afforded to it by Split Feature Inheritance and the ability of EPP to stay on the Finiteness head. This is what happens in dative experiencer constructions, genitive evidentials, and accusative Adversity Impersonals, which we discuss in Section 3.3. 10 (10) a. 10 Another possibility, alluded to in footnote 3 above, would be for the XP to first move to [Spec, TP] and next to a higher position. We do not pursue this possibility as it would incorrectly predict that the subject can reconstruct to this position (see though Section 3.2 for discussion of movement through embedded [Spec, TP] in raising constructions). b.
If you cannot agree, move on! On labels and non-nominative subjects 7 b.
In both (10a) and (10b), the moved XP shares a feature with the head whose specifier it moves to. In (10a), it is the topic feature, and we show in Section 3.2 that the moved element is interpreted as a topic. The answer to what feature is shared in (10b) is less obvious. In this respect, we follow Rizzi and Shlonsky (2006)  In this section, we apply standard subjecthood diagnostics to the five constructions introduced above. These diagnostics have been developed primarily based on the properties of nominative subjects (henceforth canonical subjects) (see Zaenen, Maling & Thráinsson 1985;Sigurđsson 1989; and the references therein). First, canonical subjects trigger (verbal) agreement (Mary runs) and can bind anaphors (Mary saw herself in the mirror). Second, they can undergo raising to subject and to object/ECM (Mary seems to run, I consider Mary to be a runner). And third, canonical subjects license silent DPs in coordinate structures (Mary can run and will win) and can act as controllers 11 In their analysis, the -i morpheme in the French que/qui alternation is an overt realization of the In both (10a) and (10b), the moved XP shares a feature with the head whose specifier it moves to. In (10a), it is the topic feature, and we show in Section 3.2 that the moved element is interpreted as a topic. The answer to what feature is shared in (10b) is less obvious. In this respect, we follow Rizzi and Shlonsky (2006)

Subjecthood diagnostics
In this section, we apply standard subjecthood diagnostics to the five constructions introduced above. These diagnostics have been developed primarily based on the properties of nominative subjects (henceforth canonical subjects) (see Zaenen, Maling & Thráinsson 1985;Sigurđsson 1989; and the references therein). First, canonical subjects trigger (verbal) agreement (Mary runs) and can bind anaphors (Mary saw herself in the mirror). Second, they can undergo raising to subject and to object/ECM (Mary seems to run, I consider Mary to be a runner). And third, canonical subjects license silent DPs in coordinate structures (Mary can run and will win) and can act as controllers (Mary wants PRO to run). 12,13 All of these properties have been linked to [Spec, TP] position, which is the view we take as well. We will also look at scope, as the ability to exhibit inverse scope is something that can potentially distinguish nominative subjects from non-nominative ones. Other diagnostics are more language specific; in pro-drop languages, the ability to be dropped is a useful diagnostic, and in languages that allow resumptive pronouns, subject resumptive pronouns are quite restricted (a restriction known as the Highest Subject Restriction). In Russian, a language that allows so-called Main Clause Infinitives, nominative subjects are replaced by dative subjects in this construction (see Fleisher 2006 and the references therein).
In the next two sections, we show that the five constructions that are our focus in this paper form two classes. While the two classes pattern together with respect to many diagnostics (and differ from nominative subjects), they differ from each other in ways that lead us to suggest that they land in different positions. We identify the two positions as the [Spec, TopP] (an A-bar position) and [Spec, FinP], a lower A-position.

Locative and predicate inversion
The properties of both locative inversion and predicate inversion cross-linguistically have received a lot of attention in the literature (see Stowell 1981;Branigan 1992;Den Dikken & Naess 1993;Bresnan 1994;Babyonyshev 1996;Collins 1997;Culicover & Levine 2001;Harves 2002;Bailyn 2004;Postal 2004;Rizzi & Shlonsky 2006;Bruening 2010, among others, on locative inversion; and Partee 1999;Matushansky 2000;Harves 2002;Citko 2006;Pereltsvaig 2007a;Bondaruk 2013a, among others, on predicate inversion), and many of the properties we discuss below have been noted before for other languages. 14 We focus here on Polish due to its transparent case and agreement morphology. The properties of inverted locatives and predicates that distinguish them from nominative subjects, together with the labeling considerations, lead us to pursue a topicalization account of the kind developed by Stowell (1981) for English. 15 In our account, both the locative PP and the instrumental predicate move directly to [Spec, TopP], where they share a topic feature with TopP, as shown in (11b) and (12b). 16 (11) Polish -locative inversion a. Do pokoju wszedł Jan. to room entered Jan.nom 'Into the room walked John.' 12 In what follows, we do not include conjunction reduction (with the exception of a brief discussion in Section 3.3), as, in many cases, the relevant examples can be reanalyzed as coordination of smaller conjuncts. 13 There is a fair amount of cross-linguistic variation, so the applicability of a given diagnostic needs to be determined for each language. For example, the agreement diagnostic might not be very useful for languages with both subject and object agreement. Likewise, being the antecedent of an anaphor will work best for languages with subject-oriented anaphors. 14 With the exception of Citko (2006; and Bondaruk (2013a), the focus of these works is on other languages. 15 Such an account has also been proposed for non-labeling related reasons by Postal (2004), Rizzi and Shlonsky (2006), Wu (2008), and Bruening (2010)
In the rest of this section, we examine the predictions stemming from these derivations. First, since the PP and instrumental DP are never in a canonical subject position, they are predicted to be insensitive to the properties associated with canonical subjects. Second, as they end up in a topic position, movement is predicted to have consequences for information structure and scope. And third, the fact that the entire structure is a TopP (not a TP) will also have consequences for embeddability; constructions where only TPs can be embedded are predicted to be ungrammatical. 17  In both cases, T enters into an Agree relationship with the postverbal nominative 17 We focus primarily on data from Polish and refer the interested reader to Postal (2004) and Bruening (2010) for analogous examples for English locative inversion and to Babyonyshev (1996) for Russian locative inversion. For predicate inversion in English, see, for example, Moro In the rest of this section, we examine the predictions stemming from these derivations. First, since the PP and the instrumental DP are never in a canonical subject position, they are predicted to be insensitive to the properties associated with canonical subjects. Second, as they end up in a topic position, movement is predicted to have consequences for information structure and scope. And third, the fact that the entire structure is a TopP (not a TP) will also have consequences for embeddability; constructions where only TPs can be embedded are predicted to be ungrammatical. 17 First, neither inverted PPs nor inverted predicate DPs can agree with the verb: Polish a. Do Jana przyszła/*przyszedł Maria. to Jan.3sg.masc arrived.3sg.fem/*3sg.masc Maria.3sg.fem.nom 'Maria arrived at Jan's.' b. Przyczyną protestów były/*była te rysunki. cause.instr protests.gen were/*was these.nom drawings.nom 'The cause of the protests were these drawings.' In both cases, T enters into an Agree relationship with the postverbal nominative DP; this is expected in a system where spec-head configurations play no role in determining agreement: Second, the inverted PPs and DPs cannot bind anaphors, as shown in (15a-b). This is also expected given the derivations in (15a'-b'), since the inverted elements do not occupy the subject position (see, however, the discussion of the relationship between binding and subjecthood in the next section Next, we consider raising and ECM. The research on English locative inversion, which we build on here, has established that while PPs can undergo subject-to-subject raising, they do not undergo raising-to-object/ECM (see Postal 1977;Den Dikken & Nass 1993;Bresnan 1994, for example): 20 (16) a. Bresnan (1994: 96) On that hill appears to be located a cathedral.
b. Den Dikken & Naess (1993: 309) *I have never seen down the hill rolling a baby carriage.
Predicate inversion in English is also impossible in ECM contexts:  18 In the plural, Polish has two genders (masculine personal versus everything else), glossed as vir versus nonvir (virile versus nonvirile). 19 The anaphors in Polish are subject-oriented (see Section 4.3 for details of how we derive this property). 20 The examples in (16) become grammatical if the PP undergoes further movement, as also noted by Bresnan (1994): (i) Den Dikken & Naess (1993: 309) ?Down this hill I have never seen rolling a baby carriage.
Returning to Polish, the grammaticality of (18a-b) indicates that inverted locatives and instrumentals can also undergo raising to subject: 21 (18) Polish a. Na wykładzie wydawała się być jedynie Maria. at lecture seemed refl be.inf only Maria.nom 'Maria seemed to attend the lecture.' b. Przyczyną protestów wydawały się być te rysunki. cause.instr protests.gen seemed refl be.inf these.nom drawings.nom 'The cause of the protests seemed to be these drawings.' On the labeling analysis, this is not a problem since the inverted XP can move through the embedded [Spec, TP] position, as shown in (19a-b b. ?*Wydawały się przyczyną protestów być te rysunki. seemed refl cause.instr protests.gen be.inf these.nom drawings.nom 'The cause of the protests seemed to be these drawings.' Raising to object/ECM is restricted in Polish to begin with (see Brecht 1974;Lasnik 1998;Dziwirek 2000;Stepanov 2007). However, the contrast between the grammatical example in (21a), involving a non-inverted ECM construction, and the ungrammatical one in (21b), involving an inverted one, shows that raising to object (marked by accusative case on the raised element) is only possible with the non-inverted variant. 23 21 The nominative DPs can raise as well: (i) Jan wydawał się być moim najlepszym przyjacielem. Jan.nom seemed refl be.inf my.instr best.instr friend.instr 'Jan seemed to be my best friend.' (ii) Te rysunki wydawały się być przyczyną protestów. these drawings.nom seemed refl be.inf cause.instr protests.gen 'These drawings seemed to be the cause of the protests.' 22 One of the reviewers points out that non-inverted variants are degraded as well (with example (ii) provided by the reviewer). This could either mean that the complement of the raising verb has to be even smaller than a TP or that the matrix [Spec, TP] has to be filled.
(i) ??Wydawała się Maria być na wykładzie. seemed.3sg.fem refl Maria.nom be.inf at lecture 'Maria seemed to attend the lecture.' (ii) ?*Wydawały się te rysunki być przyczyną protestów. seemed.3pl.nonvir refl these.nom drawings.nom be.inf cause.instr protests.gen 'These drawings seemed to be the cause of the protests.' 23 This particular construction involves raising to object around the linker za 'as'. It is independently ruled out for locative inversion, since the complement of the linker cannot be a PP.
(21) Polish a. Uważam te rysunki za przyczynę protestów. consider these.acc drawings.acc as/for cause.acc protests.gen 'I consider these drawings as the cause of the protests.' b. *Uważam przyczynę protestów za te rysunki. consider cause.acc protests.gen as/for these.acc drawings.acc 'I consider these drawings as the cause of the protests.' The ungrammaticality of ECM/raising to object follows from the fact that TopP is simply too big to be embedded under an ECM verb. We follow Bailyn and Citko (1999) and Bondaruk (2013b), among others, and take za to be the head of Predication Phrase (PrP)/small clause. This PrP is the complement of uważać 'consider', as shown in (22) The next property we consider involves scope. While the ability to have wide or narrow scope is not a property of subjects per se, scopal possibilities do distinguish constructions with inverted XPs from their non-inverted counterparts. Kuno (1971) notes that locative inversion in English disallows inverse scope, and Moro (1991) makes a similar observation for predicate inversion: (24) Kuno (1971: 365-366) a. Many girls are in every class. (∃ > ∀, ∀ > ∃) b. In every class are many girls.
Polish patterns similarly in this respect, with an added complication that inverse scope is harder to get to begin with. 24 The ungrammaticality of (i) indicates that ECM verbs in Polish select complements that are smaller than TPs: Każdy profesor jest promotorem dwóch studentów. every.nom professor.nom is advisor.instr two.gen students.gen 'Every professor is an advisor of two students.' (∀ > ∃, ?? ∃ > ∀) b. Promotorem dwóch studentów jest każdy profesor. advisor.instr two.gen students.gen is every.nom professor.nom 'Two students' advisor is every professor.' On the topicalization account, the lack of inverse scope can be attributed to the inability of the nominative argument to undergo Quantifier Raising above the topicalized XP. 26 The fronted XPs in locative and predicate inversion also exhibit a number of other properties that are unexpected of subjects, and are indicative of movement to a higher A-bar position, [Spec, TopP] in our analysis. As noted by Bondaruk (2013a) for Polish and Babyonyshev (1996) for Russian, they undergo reconstruction. This is shown in (28a-b) for anaphor binding, in (29a-b) for variable binding, and in (30a-b) for Principle C reconstruction.

(28)
Polish ( 26 We assume this is impossible because there is a closer landing spot for the QR operation (see also Nevins and Anand 2003, who discuss this kind of explanation but ultimately discount it). We also take reconstruction in this case to be impossible; if it were, inverse scope should be available. This could be attributed to the Scope Transparency Principle (ScoT) of Bobaljik and Wurmbrand (2012: 373): If the order of two elements at LF is A > > B, the order at PF is A > > B. ScoT, seen as a soft constraint, expresses the observation that freedom of word order in scrambling languages restricts the availability of inverse scope. A similar point is captured through Ionin's (2002: 87) Preservation of Discourse Function: Topics cannot undergo (scope) reconstruction at LF. Consequently, undoing the movements in examples (26-27) would destroy the scopal relations that these movements created in the first place.
(30) Polish a. *Do pokoju Jana i wszedł on i to room Jan.gen entered he.nom 'Jan walked into his room.' b. *Najlepszym przyjacielem Jana i był on i . best.instr friend.instr Jan.gen was he.nom Lit. 'Jan's best friend was he.' The reconstruction effects are accounted for by movement to an A-bar position [Spec, TopP], as proposed by Bondaruk (2013a) for Polish predicate inversion.
Polish is a pro-drop language; however, neither the inverted PP nor the inverted predicate can be dropped, as shown in (31a-b). Irrespective of the context, the missing pro cannot be interpreted as the locative PP or the predicate DP.
Polish a. *pro wszedł Jan. entered Jan b. *pro był Jan. was Jan The last property we consider is discourse status of inverted variants. b. #Przyczyną protestów były te rysunki. cause.instr protests.gen were these drawings.nom 'The cause of the protests were these drawings.' In our account, the postverbal nominative argument in both locative and predicate inversion remains low (inside vP). It agrees with the verb (as shown in (13) above), controls PRO (as shown in (23)), binds anaphors (as shown in (28-29)), which raises the question of whether it could not be in [Spec, TP] instead, with the inverted locative (or the inverted predicate DP) occupying an even higher position, as shown in (33).
This representation, however, yields incorrect word order (note the position of the verb); the verb would have to raise to the Top head. This in turn is incompatible with negation and manner adverb placement facts: 28 27 Discourse status is also one of the diagnostics Bondaruk uses in her analysis of predicate inversion. In both the italicized inverted XP is interpreted as a topic/old information. (32a) is an answer to "Who entered the room?" and (32b) is an answer to "What was the cause of the protests?" 28 Remnant movement accounts, in which the entire VP moves to a left peripheral position (e.g. Tajsner   However, the properties of nominative subjects are not incompatible with them remaining low (inside VP for unaccusatives and in [Spec, vP] for transitives and unergatives) if agreement is a consequence of Agree, T is involved in subject-oriented binding under Agree, and T can act as a proxy controller for PRO under Agree (as in Landau 2000;. This can also handle the lack of inverse scope illustrated in (26-27) above; if TP is the target position for QR (May 1977;, the quantified subject moves there and cannot outscope the (non-reconstructed) locative PP or inverted predicate phrase. 29 The low position of the subject can also capture the low scope of the subject with respect to negation; the raised subject in (36a) is scopally ambiguous, whereas the non-raised one in (36b)  Whichever option we choose to account for (i), it should be applicable also to the position of the subject for the constructions we consider in this section. We either have to allow for the PP/?? to be labelable without the evacuation of the subject or we posit a feature [F] that makes the subject leave the PP and move to a projection above it that also has this feature (e.g. Belletti's 2004 low/clause internal Focus projection). We thank an anonymous reviewer for bringing Belletti's work to our attention. 30 This does not necessarily mean that the verb raises to T. In this respect, we depart from Alexiadou and Anagnostopoulou (1998), who take the verb raised to T in pro-drop languages to check the EPP feature. position) or that the EPP is not universal and that Polish does not have the EPP feature (as proposed by Babby 1989 for Russian). 31 Yet another logical possibility would be to resort to the presence of a null expletive in [Spec, TP], as proposed by Postal (1977; and Bruening (2010), among others, for English locative inversion. However, we take the status of null expletives as elements that are both phonologically and semantically empty to be suspect (see also footnote 3).

Dative, genitive, and accusative DPs
In this section, we show that non-agreeing, non-nominative subject DPs in Russian, Lithuanian, and Polish are not in [Spec TP], following the implications of the Labeling Algorithm. We argue that these DPs move to [Spec, FinP] directly, without stopping in [Spec, TP]. 32 We focus on dative subjects of the kind illustrated in (37-38) below, in which the subject bears the thematic role of Experiencer and can occur with a Theme or Source DP. 33 (37) Polish a. Janowi jest żal Marii. Jan.dat is sorry Maria.gen 'Jan feels sorry for Maria.' b. Janowi jest smutno. Jan.dat is sad 'Jan is sad.' 31 As pointed out by the reviewer, the issue with the EPP would not arise if the NOM argument were in [Spec, TP]. This would be compatible with our proposal that the inverted non-NOM XP is in [Spec, TopP]; however, it would leave the position of the verb (or the verbal projection) somewhat mysterious (see also the discussion in footnote 28). 32 The idea that Slavic quirky subjects occupy a higher position than nominative ones has also been proposed by Williams (2006) and Baker (2008) for independent (non-labeling) reasons. 33 Dative subjects in Slavic come in two main types: Dative experiencers (a.k.a inversion nominals of Moore & Perlmutter 2000) and subjects of infinitival clauses (a.k.a. Main Clause Infinitivals, MCI, of Fleisher 2006). Main Clause Infinitivals, illustrated in (i), are permitted in Russian. However, they occur in Polish to a very limited degree; (ii) is an example (provided by a reviewer) of a Polish MCI in a negative polarity context. They are also allowed in wh-exclamatives (e.g. "What are we to do!?"). We refer the reader to Greenberg and Franks (1991) and Franks (1995) for further discussion of this construction.

(ii) Polish
Nie nam wydawać wyroki. not we.dat give.inf verdicts.acc (Lit. 'Not (for) us to give verdicts.') 'It is not for us to judge.' Here, we focus primarily on inversion nominals whose subjects are Experiencers (illustrated in (37-38). We do not discuss dative subjects in so-called productive inversion constructions, illustrated in (iiia) for Polish (see Dziwirek 1994 for a comprehensive account). As the reviewer points out, these subjects do not have a pure Experiencer thematic role. This suggests that the construction is derived in some way from the agreeing predicate with a nominative subject in (iiib), perhaps similar to the PRO-control structure that Marušič and Žaucer (2006) propose for the Slovenian "feel-like" construction (see also Slobodchikoff 2008 for Russian).
(iii) Polish (Dziwirek 1994: 57) a The genitive subjects we discuss here come from the Lithuanian Inferential Evidential (term due to Lavine 2000; see also Gronemeyer 1997;Wiemer 2006;Lavine 2010). This construction conveys a reportative or inferential reading, the verb bears the non-agreeing neuter perfective participial suffix -ma/-ta (glossed as NONAGR in the Lithuanian examples that follow), and the object is nominative, as shown in (39a). Its agreeing counterpart has a nominative subject and an accusative object, as shown in (39b).
As the examples in (40) illustrate, these constructions are characterized by adverse events with an implied external causing event.
(40) a. Russian (Babby 1994: 25) Lodku vybrosilo na skaly (volnoj). boat.fem.acc threw.3sg.neut on rocks wave.instr 'The boat was thrown on the rocks (by a wave).' b. Polish (Kibort 2008: 254) Wyrzuciło łódkę na brzeg. threw.out.3sg.neut boat.fem.acc onto shore 'The boat got thrown onto the shore.' While Babby (1994: 58, fn. 22) argues that "bad health" verbs like tošnit' 'nauseate' are different from Adversity Impersonals in that they are "lexically impersonal" (i.e. external arguments of AIs are suppressed), we include these predicates because their accusative We propose that while these three subjects are assigned case in different locations, they all raise to [Spec, FinP] to satisfy the EPP feature on the Finiteness head, as shown in (42). Dative subjects of the kind illustrated in (37) and (38) raise from [Spec, vP], where they are assigned inherent case (Woolford 2006), leaving T free to Probe past the dative subject. The phi-features on T either remain unvalued (as in (37a-b) and (38a-b)), resulting in default non-agreeing morphology (see Preminger 2011 for a more general discussion of the so-called failed agreement), or get valued by the nominative Theme, as in (37c) and (38c). 35 The idea that non-nominative "subjects" considered in this section do not move 34 A reviewer asks whether the following constructions are also considered "bad health" verbs and therefore to be included as Adversity Impersonals.

(ii) Polish
Janka rwie w nodze. Janek.acc shoots.3sg in leg (Lit. '(To) John pain-shoots in the leg.') 'John has a shooting pain in the leg.' These predicates are not precisely of the same category as tošnit' 'be nauseous' in (41), because the Stimulus in (i) is permitted as a nominative DP. Compare the (iii-b) to (i).
(iii) Russian (Babby 1994: 27) a. Ego tošnilo ot zapaxa. he.acc nauseated.3sg.neut from smell b. *Ego tošnil zapax. he.acc nauseated.3sg.masc smell.nom 'The smell made him nauseous.' Therefore, we do not include predicates like (i) and (ii) in the group of Adversity Impersonals. We do note their similarity to the full psychological verbs discussed in this section (e.g. the example in (38c)) in that their argument structures both include a non-nominative Experiencer and a nominative Stimulus/Theme, and assume that the analysis proposed in this paper can also account for them (see also Bondaruk & Szymanek 2007 for a discussion of the parallels between accusative and dative Experiencers). 35 The question, brought to our attention by one of the reviewers, is why the experiencer does not intervene and block the relationship between T and the nominative argument in (37c) and (38c). This would be a case of Defective Intervention. Note, however, that the dative argument has moved, and on the assumption that all operations take place simultaneously at the phase level, only one link would intervene (not the entire chain). However, we do not see intervention when the Dative does not move: (i) Polish Wczoraj spodobała się Janowi Maria. yesterday pleased.3sg.fem refl Jan.dat Maria.nom 'Yesterday, Jan liked Maria.' This suggests that being inherent case marked is at issue (see Citko 2011 and the references therein for similar effects involving double objects). Crucially, we are not assuming that all datives are inherent. For example, the dative subjects of MCI (see footnote 33) are structural (as argued by Greenberg and Franks 1991;Perlmutter & Moore 2002;Nikolaeva 2014, among others). to [Spec, TP] is not unique to our proposal; Livitz (2006) takes them to move to what she dubs "Major Subject" Projection, located between CP and TP (later identified by Wood and Livitz 2012 as AboutnessP). 36 (42) Citko,Germain,& Witkoś 24 The idea that non-nominative "subjects" considered in this section do not move to [Spec, TP] is not unique to our proposal; Livitz (2006) takes them to move to what she dubs "Major Subject" Projection, located between CP and TP (later identified by Wood and Livitz 2012 as AboutnessP). 36 (42) We assume, following Rizzi and Shlonsky (2006), that Fin bears a [+N] feature. Since all three of the XPs that we argue to move to [Spec, FinP] are DPs, the resulting syntactic object is able to be labeled FinP via the sharing of the [+N] feature. Since these DPs do not occupy [Spec, TP] position, we do not expect them to have the canonical subject properties associated with this position, either because the DP does not stop in [Spec, TP], the locus of these properties, or because the phrase involved in the process is not a TP, but a FinP. In addition, the [Spec, FinP] position is not associated with any Information Structure effects, This would be a case of Defective Intervention. Note, however, that the dative argument has moved, and on the assumption that all operations take place simultaneously at the phase level, only one link would intervene (not the entire chain). However, we do not see intervention when the Dative does not move: We assume, following Rizzi and Shlonsky (2006), that Fin bears a [+N] feature. Since all three of the XPs that we argue to move to [Spec, FinP] are DPs, the resulting syntactic object is able to be labeled as FinP via the sharing of the [+N] feature. Since these DPs do not occupy [Spec, TP] position, we do not expect them to have the canonical subject properties associated with this position, either because the DP does not stop in [Spec, TP], the locus of these properties, or because the phrase involved in the process is not a TP, but a FinP. In addition, the [Spec, FinP] position is not associated with any Information Structure effects, accounting for their discourse neutral status. 37 First, similarly to the PPs and Instrumentals discussed above, these non-nominative DPs do not agree with the verb, as shown by the ungrammaticality of the (b)  In addition, the Polish AI in (47) does not allow an instrument argument, as in (ii). We leave this puzzle for future work, and thank the reviewer for encouraging us to be more specific about these cross-linguistic differences. In the case of the raising predicate 'seem' in Polish (wydawać się) and Russian (kazat'sja), a dative DP cannot be raised with the interpretation that it is the subject of the embedded clause. As one of the reviewers points out, the only reading possible in (53) is the one in which Jan perceives something else (i.e. null 'it') to be sad.
Janowi i wydawało się [ t i być smutno/wesoło]. Jan.dat seemed.3sg.neut refl be.inf sad/merry *Intended: 'Jan seemed to feel sad/happy.' 'It seemed to Jan that it was sad.' Bondaruk and Szymanek (2007) show that an additional dative Experiencer in the matrix clause is unacceptable, taking it to indicate that the matrix dative DP receives its Experiencer thematic role from the predicate wydawać się 'seem'.
(54) Polish (Bondaruk and Szymanek 2007: 21) Markowi wydawało się (*nam) [być smutno/żal, Mark.dat seemed.3sg.neut refl (*we.dat) be.inf sad/sorry że przegrał]. that lost.3sg.masc 'Mark seemed to us to feel sad/sorry that he had lost.' In the case of Russian, our consultants report that the sentence in (55)  We take the facts in (48-55) to indicate that non-nominative DPs do not undergo raising. This follows from our analysis because these subjects are in [Spec, FinP], and as argued above for TopP, FinP is too large to be embedded under raising predicates that select a non-finite TP.
Next, the ungrammaticality of (56a) and (57a)  It is impossible to test whether Adversity Impersonals behave analogously with respect to ECM, as object raising with ECM verbs like sčitat'/uważać 'consider' is only possible with subjects of small clauses.
Furthermore, these non-nominative subjects cannot be omitted in coordinate structures when the antecedent is a nominative subject, even if they share the same thematic role (e.g. Experiencer).

(58)
Polish -DAT (Dziwirek 1994 Next, we highlight other canonical subject properties that non-nominative subjects lack. First, in contrast with nominative subjects, neither dative subjects in Polish nor genitive subjects in Lithuanian can be dropped, even though Polish and Lithuanian are pro-drop languages, as shown by Dziwirek (1994) for Polish. This is exemplified in (70)  Dziwirek (1994) also observes that dative subjects require resumptive pronouns in subject relative clauses, as shown in (73a). By contrast, resumptive pronouns are ungrammatical in relative clauses with nominative subjects, as shown in (73b).

(73)
Polish -DAT (Dziwirek 1994: 140) a. Ta kobieta, co *(jej) jest zimno. the woman comp she.dat is cold 'The woman who is cold.' b. Ta kobieta co (*ona) przyszła. this woman comp she.nom arrived.3sg.fem 'The woman that arrived.' Like the dative subject, the accusative argument in an Adversity Impersonal Construction requires a resumptive pronoun in Polish when it is relativized: Polish -ACC Pacjentka co *(ją) mdli/dusi. patient comp she.acc nauseates/chokes 'The patient that is nauseated/choking.' Since Rappaport (1986), anaphor binding has been a traditional subjecthood diagnostic in Slavic. 41,42 If this is right, the pattern of anaphor binding by these non-nominative DPs would seem to indicate that (at least some of them) would have to occupy [Spec, TP]. However, recent work by Nikolaeva (2014) calls into question the conclusion that the ability to bind anaphors correlates with [Spec, TP] position.

b. Russian
Devuške i ne nravitsja eë i /*?svoj i sviter. girl.dat neg please.3sg.refl her/self's sweater.nom 'The girl doesn't like her sweater.' Dative subjects of Russian infinitival constructions, on the other hand, are able to bind reflexives (Franks 1995;Moore & Perlmutter 2000, among others). 41 To be more precise, Rappaport's (1986: 102) list of licit binders in Russian includes the following: (i) AGR, a "nominal element" sister to VP (a T head or INFL), (ii) the subject of a finite clause, (iii) a PRO-subject of an infinitival clause, and (iv) the specifier of a lexical category (e.g. NP). 42 Franks (1995), observing that possessive PPs can bind anaphors, as in (i), objects to the blanket use of anaphor binding as a subjecthood diagnostic. Siloussar (2011) counters this by arguing that svoj 'self's' is used here not as simply a possessive pronoun, but as a lexical item closer to the adjective sobstvennyj 'one's own', and that the ability to bind svoj should be used as a diagnostic for being located in [Spec, TP].
at Petrovs.gen self's house.nom 'The Petrovs have their own house.' 43 Results from Germain (2017) indicate that anaphor binding is dispreferred for dative subjects of psychological verbs compared to the binding of pronouns by these subjects (sentences with reflexive svoj received an average rating of 3.8 out of 7 while sentences with pronominal ego were rated 6.5 out of 7 on average; participants (n = 40) each saw 3 sentences of each type). Slioussar (2011Slioussar ( : 2067 also reports mixed judgements on anaphor binding by these subjects. Nikolaeva (2014: 63) does not focus on the grammaticality of anaphor binding, but rather on the complete lack of Principle B effects in the presence of a pronominal determiner.
(76) Russian Emu i ne opublikovat' svoej i stat'ji. he.dat neg publish.inf self's paper.acc 'It's not in the cards for him to publish his paper.' Furthermore, dative subjects of non-verbal psych predicates like žal' 'sorry' can bind either the reflexive or pronominal determiner: (77) a. Russian Emu i žal' svoju i /ego i sem'ju. he.dat sorry self's/his family.acc 'He feels sorry for his family.' b. Polish (Dziwirek 1994: 135) Było mu i wstyd za swoją i /jego i siostrę. was.3sg.neut he.dat shame for self's/his sister.acc 'He was ashamed for his sister.' Anaphor binding by the preverbal accusative DP of the Adversity Impersonal is ungrammatical (Slioussar 2011;Nikolaeva 2014 And finally, Lavine (2000; notes that the genitive subject in Lithuanian can bind anaphors: Lithuanian (Lavine 2010: 126) Motinos i sudeginta savo i namas. mother.gen burned-down.part.nonagr self's house.nom 'Mother apparently burned down her own house.' The fact that these subjects can bind pronouns is what distinguishes them from nominative subjects. In Slavic and Baltic, Principle B effect is triggered when a nominative subject is co-indexed with a pronominal possessive determiner. Crucially, dative subjects do not show this "anti-subject" effect, as shown by the contrast between the a and b examples in (80)(81). 45,46 44 The theory of binding proposed in Nikolaeva (2014) and the structure in (82) below predicts that arguments which receive structural case but their case do not c-command v will not be able to bind reflexive pronouns/reflexive possessives. In the concrete case of the accusative argument in adversity predicates we assume that the phase head v hands down its [uphi] and [EPP] features to its complement head V (in line with Chomsky 2008;. As a result, the case position for the (scrambled/fronted) accusative-marked argument is [Spec, VP], which does not c-command v and the index adjoined to it. Hence we predict that anaphoric binding from this position is impossible. The same mechanism explains why regular accusativemarked nominal objects do not bind reflexives. 45 The anti-subject orientation of pronominals in languages with subject-oriented anaphors was first discussed in Vikner (1985) for Danish anaphors. 46 Results from Germain (2017) indicate that binding of the pronominal determiner by the dative subject of an infinitival clause may be similarly acceptable to the binding of an anaphor (sentences with reflexive svoj received an average rating of 5.9 out of 7 while sentences with pronominal ego were rated 5.3 out of 7 on average; participants (n = 40) each saw 3 sentences of each type).  Nikolaeva (2014) proposes that the ability to bind anaphors depends on the configuration of the DP antecedent and the pronoun/anaphor, wherein pronouns and anaphors are indexes that covertly raise to certain positions. As the tree below shows, in her system, an index can first tuck in as an inner specifier of VP, and then optionally undergo head movement first to v and then to T.   Nikolaeva (2014) proposes that the ability to bind anaphors depends on the configuration of the DP antecedent and the pronoun/anaphor, wherein pronouns and anaphors are indexes that covertly raise to certain positions. As the tree below shows, in her system, an index can first tuck in as an inner specifier of VP, and then optionally undergo head movement first to v and then to T.
(82) Nikolaeva (2014: 57) In (83) below, the pronoun covertly raises to T past the dative experiencer in [Spec, vP] Nikolaeva's Principle II states that indexes adjoined to a projection headed by D, v, or T are realized as reflexives when they are c-commanded by a co-indexed specifier (Nikolaeva 2014: 68). If an index's position at Spell-out does not meet the conditions that trigger reflexivization, it is pronounced as a pronoun (see also footnote 49). The example in (84a) is ungrammatical because, as (84b) shows, the index will only be realized as svoj 'self's', and never as ego 'his', by Principle II.

(84)
Russian ( The optionality with the dative experiencers in (77) is accounted for in her system by assuming that the raising index (i.e. the un-spelled out pronoun or reflexive) can headmove from its position in VP to v and then to T or remain adjoined to v. 48 If the index is adjoined to T as in (83) above, the index will be spelled out as ego 'his'; if it stays adjoined to v, it will be spelled out as the reflexive svoj. 49 Under this proposal, the accusative in the Adversity Impersonal in (78) above is unable to bind an anaphor simply because, by the time Principle A/B is evaluated, the index will have moved beyond the accusative in the VP. 50 Nikolaeva (2014) follows Moore and Perlmutter (2000) in every node dominating x includes x and y, and x does not dominate y (where x includes y iff y is dominated by every segment of x, as proposed by May 1985)." Such a definition leaves the c-command domain of the adjunct undefined, as the node dominating the adjunct at the adjunction site does not include it. This may not be the best step for both X o and XP adjunction, so alternatively, we can invoke the Word Interpretation notion from Chomsky (1995: 322): "at LF, X o is submitted to independent word interpretation processes WI, where WI ignores principles of the computational system within X o ." 48 If the dative Experiencer of a psychological verb is also in [Spec, vP], we expect to see a similar optionality in anaphor binding with these constructions. As the examples in (75) show, this is not the case, and the dative argument of nravit'sja has different properties from the dative argument of žal. We note this and observe that one promising route for resolving this would be to adopt a different argument structure for psychological verbs and non-verbal psychological predicates, wherein the Theme is merged in different positions. Cuervo's (2003) proposal for Spanish psychological verbs has Experiencers merged in at [Spec, ApplP] and Themes at [Spec, vP].
With the nominative Theme in [Spec, vP] the index is above v forcing it to move to T without stopping in v, thereby depriving it of a position where it would is c-commanded by its antecedent. (See Germain in progress for such an analysis and discussion). 49 The appearance of the possessive pronoun and the possessive reflexive happens in accordance with Nikolaeva's (2014: 68) Reflexivization Principles IV and V: (i) Reflexivization at spell-out: when the sentence is sent to spell-out, if an index is coindexed with the specifier of the projection to which it is adjoined, the index has to be realized as a reflexive.
(ii) Pronominal is an elsewhere condition: if an index has not been realized as reflexive, it is realized as pronominal. asserting that the dative subject of the infinitival construction is assigned dative case by a non-finite T. She relies on this to account for the grammaticality of anaphor binding by these subjects, under the assumption that DPs that receive case from T are eligible binders for reflexives. We modify this assumption and propose that the dative subject in infinitival constructions of the kind given in (85) actually receives case from the Fin head. 51 (85) Russian Emu i ne opublikovat' svoej i stat'ji. he.dat neg publish.inf self's paper.gen 'It's not in the cards for him to publish his paper.' Thus, any subject assigned structural case sufficiently high in the clause (of which the dative in the infinitive construction is one example) is eligible to bind anaphors. We extend this explanation to genitive subjects of the Lithuanian Inferential Evidential, assimilating them to gerunds and treating the genitive as structural case assigned by D. This analysis of -ta/-ma as the phonological form of a "little" n follows from its historical role as a deverbalizing morpheme and is an update of Nuñes's (1994) proposal that the We would like to account for this fact by pointing to two related properties that distinguish (i) from the examples in (75) in the text: the level of embedding the reflexive and its case. It turns out that once the reflexive pronoun (or a reflexive possessive) is embedded in a larger constituent it is more liable to binding by a c-commanding antecedent in an A-position other than [Spec,TP]. On the other hand, at the observational level, the reflexive possessive shows strong reluctance to being bound when in nominative case, thus (75a) is excluded. The two properties mentioned above are intertwined, as the reflexive is typically casemarked within a larger constituent that contains it. But (75a) is also excluded on technical grounds in the account developed in Nikolaeva (2014). Assuming that the nominative Theme argument is first-merged in [Spec, vP] and the dative Experiencer in [Spec, ApplP], and that the index can attach via head movement only to v, T and D, but to no other heads, including Appl, the index can only move up and adjoin to T, too high to be bound by the dative experiencer. Crucially, the index cannot lower to v to find itself in the c-domain of [Spec, ApplP] (see footnote 48). In a way, both the structural and case-related aspects of reflexive binding come to the fore in the contrast between (ii) and (iii), also brought to our attention by the reviewer: Marek i i jego i /*swoja i żona wyjechali z miasta. Mark.nom and his.gen/*self's wife.nom left.3pl.vir from city 'Mark and his wife left the city.'

(iii) Polish
Marek i z(e) *jego i / swoją i żoną wyjechał z miasta. Mark.nom with his.gen/*self's wife.instr left.3sg.masc from city 'Mark with his wife left the city.' In (ii) the reflexive shares the nominative case with NP it modifies, which is one complicating factor. Additionally, it has to move out of a coordinate structure. The pronominal possessive (marked genitive) at least avoids the complication with case. By contrast, (iii) involves a comitative construction, where the preposition provides case for the NP including the reflexive possessive, which is more prone to being bound. Pending further analysis, we assume that the comitative construction does not involve coordination, thus it is not an island and a barrier to index/pronoun raising. 51 This follows several proposals that link the source of structural dative to a non-finite C head (Landau 2008;Livitz 2012). Germain (in progress) argues for a raising account of these constructions where the subject of the embedded clause is assigned dative case from an embedded Fin head (see Fleisher 2006 for arguments for a bi-clausal construction). If you cannot agree, move on! On labels and non-nominative subjects 41 b. Lavine (2000) argues that these constructions are not passive but assumes that the source of the genitive case on the subject is not structural, but ergative due to the similarity between this construction and the perfective constructions found in split-ergative languages like Hindi. 53 There are, however, three reasons to think that these subjects are assigned genitive case structurally by D. For one, first and second person pronouns in the Inferential Evidential construction have the same form as possessive pronominal pronouns, as shown in (87) Lavine (2010) updates this account to argue that a little v headed by the -ma/-ta morpheme assigns genitive "lexically" to its specifier or to the complement of V if the predicate is unaccusative. 54 Genitive pronouns in all other persons and numbers (along with non-pronominal nouns) are homophonous with possessors (e.g. mūsų 'us' and mūsų 'our'). Lavine (2000) argues that these constructions are not passive but assumes that the source of the genitive case on the subject is not structural, but ergative due to the similarity between this construction and the perfective constructions found in split-ergative languages like Hindi. 53 There are, however, three reasons to think that these subjects are assigned genitive case structurally by D. For one, first and second person pronouns in the Inferential Evidential construction have the same form as possessive pronominal pronouns, as shown in (87) and (88). 54 (87) Lithuanian -Inferential Evidential a. Gronemeyer (1997: 107) Mano sergma!? I.gen sick.part.nonagr 'Evidently I am sick!?' 52 An interesting question, brought to our attention by one of the reviewers, is what yields the interpretation of this construction (paraphrasable as "Given what evidence I have, it must be the case that event X happened"). Lavine (2010) analyzes the source of the evidential semantics as an Evid head in the CP layer. This could also be ascribed to a (missing) functional layer in a Speech Act Phrase like the one proposed in Speas (2004), as one reviewer suggests. Since the choice between these two options does not affect our argument here, we leave it open. The same reviewer raises the question of T directly selecting a DP in (86b). While we take this to be possible when DP denotes a complex event, nothing in our analysis precludes a null verbal projection between T and DP. 53 Lavine (2010) updates this account to argue that a little v headed by the -ma/-ta morpheme assigns genitive "lexically" to its specifier or to the complement of V if the predicate is unaccusative. 54 Genitive pronouns in all other persons and numbers (along with non-pronominal nouns) are homophonous with possessors (e.g. mūsų 'us' and mūsų 'our'). Second, unlike inherent cases, the genitive assigned in this construction can be overridden by the Genitive of Negation (GoN), the case assigned to objects and unaccusative subjects in the scope of negation in Baltic and Slavic languages (see Pesetsky 1982;Bailyn 1997;Borschev & Partee 2002;Harves 2002;Irwin 2012;Kagan 2012;Anderson 2013, among others). 55 First and second person genitive object pronouns have the form manęs/tavęs 'me/you', as shown in (89a) for the intensional genitive (see Kagan 2012 and the references therein for discussion of intensional genitives). This is also the form that first and second person subject pronouns take when they are assigned the Genitive of Negation, as shown in (89b, c). The Inferential Evidential construction can participate in the Genitive of Negation alternation, as shown in (90), which is not expected if the genitive were inherent. 55 The Genitive of Negation with subjects is limited to subjects of copular sentences in Lithuanian. Unlike in Russian, GEN is not assigned to subjects of unaccusative verbs in the presence of negation. See also Anderson (2013) for discussion of Genitive of Negation as a diagnostic for structural case in Lithuanian in general, although she does not provide a discussion of the source of GEN on subjects of Inferential Evidentials. And finally, Inferential Evidentials are possible with unaccusatives, with the subjects also marked genitive, which shows that this genitive is not correlated with a specific thematic role (see Germain 2017 for more discussion in favor of this analysis): 56 (91) Lithuanian (Lavine 2010: 124) a. Čia turbūt ir gribų esama. here evidently even mushrooms.gen be.part.nonagr 'There must be mushrooms here.' b. Ledo staiga ištirpta. ice.gen suddenly melted.part.nonagr 'The ice must have suddenly melted.' To account for the fact that the genitive subject has the same form as possessors, can be assigned Genitive of Negation, and shows a full spectrum of thematic roles, we argue that the subject of an Inferential Evidential is assigned structural genitive in [Spec, DP]. This also explains why it can bind reflexives (see example (79) above), as in this position the genitive can c-command an index which has tucked in below it; by Nikolaeva's (2014) Principle II, this position is a reflexivization site and the index is spelled out as a reflexive. Dative subjects of infinitives are similar in this respect (see example (85)); they are assigned structural case in a high position in the clausal structure (i.e. [Spec, FinP]). This is what allows them to bind reflexives in contrast to other DPs (e.g. dative experiencers and accusatives of Adversity Impersonals) that just move to such a position for other non-case related reasons. In other words, these non-nominative DPs vary in their status as eligible binders of anaphors, but this is not a reflection of their position in [Spec, TP]. Rather, it is a reflection of their case position with respect to the position of the raised pronoun/reflexive. 57 56 An additional question that one reviewer points to is the source of NOM on the Theme in the Inferential Evidential. We follow Lavine (2000; in assuming that this NOM is a default case. Evidence for this comes from the fact that some speakers do not accept the Genitive of Negation on the NOM Theme, as shown in (i), and that there seem to be PCC effects barring the NOM from being a first or second person pronoun, as shown in (ii).
(i) Lithuanian (Lavine 2010: 136) Ingos nenuraminta vaikas/*vaiko. Inga.gen not.calm.down.part.nonagr child.nom/*gen 'Inga must not have calmed down the child.' (ii) Lithuanian (Lavine 2010: 137) Ingos nuraminta *aš/*tu. Inga.gen calm.down.part.nonagr *I/*you.nom 'Inga must not have calmed down me/you.' If it is the case that [person] features must be checked via Agree with T, then we can account for (ii) by arguing that Agree here has failed to take place (see discussion in Section 4.1). Under Lavine's (2010) analysis this follows from the clause being non-finite and the T unable to assign NOM. Under the analysis given here, this follows from the NOM DP being embedded in a DP which T cannot probe into. 57 One of the reviewers points out that the minimalist perspective based on the notion of the Labelling Algorithm adopted here deserves reliance on a strictly minimalist theory of anaphora, such as Reuland (2011), rather than Hestvik (1992) and Nikolaeva (2014). It is true that Nikolaeva's approach is fairly traditional, as she assumes binding as a relation between an antecedent and the bindee based on a shared index, while Reuland's index-free proposal for binding is based chiefly on Agree and feature sharing, thus in principle it is preferred on conceptual grounds. However, we take Nikolaeva's view on binding theory to have a This concludes our survey of the properties of the three types of non-nominative "subjects" (dative, accusative, and genitive ones) that distinguish them from canonical nominative subjects. We have shown that, with respect to verbal agreement, anaphor binding, subject-to-subject raising, ECM, conjunction reduction and control, these preverbal DPs do not pattern with nominative subjects. We have also shown that despite the fact that some of them can bind reflexives, a fine-grained theory of reflexive and anaphoric binding in Nikolaeva (2014) accounts for these data without placing the non-nominative "subjects" in [Spec, TP]. We have also discussed the differences between the two types of subjects in their ability to be dropped and occur without a resumptive pronoun. We have linked all these differences to a derivation in which the non-nominative subject skips the [Spec, TP] position and moves directly to [Spec, FinP].
Our discussion may lead one to believe that there should be no cases of non-nominative non-agreeing DPs in [Spec, TP] exhibiting subject properties. In the next section, we examine two types of non-nominative non-agreeing subjects that have been argued to occupy [Spec, TP] position and show how that can be derived in our system.

Icelandic
Our system leaves open the possibility that non-nominative subjects which become involved in Agree with some probe on T can move to its specifier position. In languages in which T is "strong" enough to label (perhaps via partial agreement the quirky subject), the quirky subject may remain [Spec, TP]. This seems plausible for languages like Icelandic, whose non-nominative subjects have been documented to be quite different from the Slavic subjects (see Andrews 1976;Thráinsson 1979;Zaenen, Maling & Thráinsson 1985;Sigurðsson 1989;2002, among others hjálpað. I.nom had much to do and I.dat was nonetheless neg helped 'I had much to do, and was nonetheless not helped.' much wider descriptive coverage. Reuland's proposal is still quite programmatic, as the author admits himself (Reuland: 2011: 146): "Recall that my main goal is to show that s syntactic encoding of interpretive dependencies obeying the inclusiveness condition is in principle possible. I will therefore limit discussion as much as possible to environments and subcases needed for this goal." This index-free program for binding theory requires further meticulous application to the range of constructions we are concerned with, specifically (a) it straightforwardly covers only constructions in which both the binder and the bindee bear structural cases, (b) thus it does not easily lend itself to applications where either the binder or the bindee bear inherent/quirky cases, (c) it typically places the binder in the subject position of [Spec, TP], with little discussion of cases where it is elsewhere (for instance in one of the object positions of ditransitive verbs or an Applicative Phrase), a point which is very much necessary for our analysis, (d) it does not contain a comprehensive discussion of possessive reflexives, and (e) it leads one to expect that possessive reflexives and possessive pronouns should remain in complementary distribution, contrary to the data in (77). These reservations notwithstanding, developing applications of Reuland's index-free minimalist theory of binding to the constructions mentioned in this paper is an intriguing and challenging research task, worth pursuing independently of this contribution. Third, they can bind anaphors: Icelandic (Zaenen, Maling & Thráinsson 1985: 450) Henni i þykir [bróðir sinn i ] leiðinlegur. she.dat thinks brother.nom self's boring 'She i thinks her i brother boring.' And fourth, they are fine in ECM constructions: Icelandic (Williams 2002: 82;citing Andrews 1982) Hann telur barninu (i barnaskap sinum) hafa He believes the-child.dat (in his foolishness) have.inf batnað veikin. recovered-from the-disease.nom 'He believes the child (in his foolishness) to have recovered from the disease.' This has led many researchers to conclude that dative subjects in Icelandic occupy [Spec, TP] (Zaenen, Maling & Thráinsson 1985;Poole 2015;Wood 2015, for example). We maintain this view, and take Icelandic subjects to occupy a lower position than the Slavic ones, following in this respect Williams (2002) and Baker (2008). 58,59 The questions for our analysis is thus what allows Icelandic quirky subjects (unlike their Polish, Russian, and Lithuanian counterparts) to remain in [Spec, TP], and how the resulting syntactic object is labeled. To address these questions, we would like to point to another property of Icelandic that distinguishes it from the languages we consider, discussed by Baker (2008) 58 Baker remains somewhat agnostic about the nature of this position. For him, these subjects "sit in some higher position, such as Spec, TopicP, or they could be the specifier of of some decomposed functional category that has some of the properties of T in English and Icelandic but not others." (Baker 2008: 91). He derives the contrast from SCOPA (Structural Condition on Person Agreement), which, crucially, requires the 1 and 2 person objects to move to [Spec, TP]. Since there does not seem to be independent evidence that the nominative Theme of the psych verbs in (75), for example, is in [Spec, TP], we do not adopt this part of the proposal. 59 Williams (2002) implements the analysis in different terms (in his Representation Theory), arguing that subjects in different languages are embedded at different levels; the level of case structure versus the level of surface structure. 60 This is not the canonical PCC configuration (since it does not involve a double object construction). The structural configuration (and the account), however, is the same (see Boeckx 2000 for relevant discussion). 61 This is related to the fact that in Polish T can probe past the dative argument and agree with a nominative argument, as in (i).
(i) Polish Jankowi podobała/*podobało się Marysia. Janek.dat pleased.3sg.fem/*3sg.neut refl Marysia.nom 'Janek liked Marysia.' In footnote 35, we attribute the lack of intervention to the fact that the dative has moved. The reason why the dative continues to intervene in (96) could be related to the possibility, brought to our attention by one of the reviewers, that T is a split Probe in Icelandic and not in Polish. Kučerova (2007) shows that dative in Icelandic also does not intervene when it moves (i.e. when it undergoes object shift). PCC effects have commonly been analyzed to involve special patterns of agreement between T and the DPs in its probing domain (see Anagnostopoulou 2003;Rezac 2003;Richards 2008a). 62 In these accounts, T agrees in person with the quirky subject, and probes again for number. Since T has already agreed with the quirky subject for person, it cannot agree again with the object for the same feature (i.e. person feature). We take the first instance of Agree (Agree between the quirky subject and T) to be the crucial difference between Icelandic and Slavic-type quirky subjects, and hypothesize that this Agree makes Icelandic T strong enough to label the TP.

(Accusative) numeral subjects in Polish
The second case study involves Polish numeral subject constructions in which the numeral is 5 and/or higher (see Rutkowski 2007;Miechowicz-Mathiasen 2012 The evidence that the numeral is accusative, discussed by Franks (1994; and Przepiórkowski (1999;, comes from morphological considerations; the case form on the agreeing demonstrative is accusative not nominative, as shown in (98): Polish Tych/*ci pięciu mężczyzn stało. these.acc/*nom five.acc men.gen stood.3sg.neut 'These five men were standing.' An additional argument for the subject being accusative comes from the fact that it can be coordinated with the accusative argument of adversity predicates (Lavine & Franks 2008): Polish Tych/ci* pięciu harcerzy i tamtą harcerkę these.acc/*nom five.acc boy-scouts.gen and that.acc girl-scout.acc zemdliło po śniadaniu. nauseated.3sg.neut after breakfast. 'These five scouts and that girl scout felt nausea after breakfast.' What is relevant (and somewhat unexpected from our perspective) is that these high numeral subjects meet all the subjecthood diagnostics, other than being marked nominative and agreeing with the verb. First, they can bind anaphors: Next, just like nominative subjects, they do not allow resumptive pronouns in relative clauses introduced by the complementizer co: Te siedem studentek i co (*one i ) poszło these.acc seven.acc students.gen comp they.nom/acc went.3sg.neut do dziekana miało dobre oceny. to dean had.3sg.neut good marks 'The seven students that went to the dean had good marks.' Furthermore, both are marked with structural cases, as they both can be assigned genitive under negation in the existential construction, as shown in (106a-b) for nominative subjects and in (107a-b) for accusative ones (Franks 1994;Witkoś 1998;Przepiórkowski 1999;Błaszczak 2001 Again, in this respect, they contrast with dative subjects, which remain dative under negation: (108) Polish a. Studentom było zimno w akademikach. students.dat was.3sg.neut cold in dormitories 'The students were cold in their dormitories.' b. Studentom/*studentów nie było zimno w akademikach. students.dat/*gen neg was.3sg.neut cold in dormitories 'The students were/were not cold in their dormitories.' We are convinced that the properties of the ANS reviewed above warrant the following conclusion: although the ANS is not a nominative subject, it functions like one. 65 In principle, there are two theoretical options open to us to account for this fact, both compatible with the idea that the ANS is involved in an Agree relation with T, unlike the other subject-like elements. The first option is to follow Klockmann (2015) or Willim (2015), where T and the ANS are involved in an attempted but failed Agree. For Klockmann the ANS bears either default nominative (for non-masculine referents) or genitive case (for masculine human referents) and, crucially, it participates in an imperfect Agree with T. 66 Willim (2015: 243-246) submits that in the context of structural cases T (and v) cannot Agree for case with the ANS in narrow syntax, whose internal structure is [ FP QP [ FP F NP]] due to the fact that F (a nominal functional head) and QP bear conflicting features: F has no case feature and QP bears an unvalued case feature. As the features are not identical, with F and QP being both equidistant and accessible to T(v), the probe cannot deal with this ambiguity and Agree does not obtain. Both authors follow Preminger (2011) and Bošković (2009) and assume that the attempted but failed Agree should not terminate the derivation. Willim places the burden of spelling out the unvalued case and phi-features 65 Space limitations prevent us from discussing further properties of the ANS. Suffice it to say that the ANS shows interpretive properties of its nominative equivalent in Numeral Noun Constructions in Russian, discussed in depth in Pereltsvaig (2006) (see Witkoś et al. forthcoming for details). 66 This Agree is imperfect, as the incomplete feature specification of the numeral, with the gender feature missing, cannot fully value the phi-features of T. Klockmann's nominative/genitive hypothesis faces a problem in the case of constructions with adversity predicates (see example in 99), where the subject-like argument must appear in accusative according to Lavine and Franks (2008) and the ANS fits in this position, quite unexpectedly, if its case were different from accusative. on the PF component; in Polish, this component can interpret QP and F as the ANS, with genitive case on the NP-complement of the numeral. 67 The second option open to us is to assume that the ANS and T are involved in Agree which does not fail completely, although it does malfunction to some degree. Małecki (1863), Łoś (1927), Szober (1928), Franks (1994, and Przepiórkowski (1999; credit the behavior of the ANS in the context of structural case to the gap in its morphological paradigm: the higher numeral has no nominative form, which is why the accusative (crucially, also a structural case) "steps in" to fulfil the role of the subject. 68,69 To explain why the accusative does not agree with the verb and triggers default agreement instead, we take T to be able to copy phi-features from its Goal (as part of Agree) only when it can access the DP/NP through the minimum number of Case Projections, presumably only the projection of the nominative. 70 Projections of other cases are too opaque for phi-features to be copied onto T; so when phi-feature copying is attempted in the course of Agree, it fails and T defaults to third, singular, neuter. In sum, T and the ANS become involved in an Agree relation and they share the feature [T] ([uT] on the ANS), taken to stand for structural case. 71 It is the sharing of this feature that allows TP to project its label when the accusative subject moves to [Spec, TP]. This accounts for their subject-like properties documented in this section. Crucially, such feature sharing does 67 This solution leads to at least two potential technical problems. First, not only does it leave the ANS with unvalued case features in the narrow syntax but also lets this derivational fault spill over into LF, unaffected by the PF rescue strategy in the form of default spell-out (Bošković 2009 andPreminger 2009 postulate that only phi-features but not case features can be ignored when matched but not valued). Second, as our examples (106-107) show, the ANS shows the Genitive of Negation, a structural case, which is licensed in a configuration identical to that of the other structural cases, with Neg operating in tandem with v and jointly acting upon the goal as a Neg/v complex probe (see Witkoś 1998;Błaszczak 2001). However, the complex nature of the probe has no impact on the internal constitution of the goal; thus a consistent application of the strategy of ambiguous equidistant goals within FP would lead one to expect that the GoN should not show on [ FP QP [ FP F NP]] either, so all the three structural cases (nominative, accusative and the GoN) should be realized morphologically in the same manner. 68 This idiosyncrasy of the numeral is probably due to a twist in the diachronic development of the numeral, which shifted grammatical categories from the fully nominal form in Old Polish to an adjectival/functional class in Present Day Polish (see Rutkowski 2007;see Babby 1987 andPereltsvaig 2007b for a corresponding process in Russian). 69 This option seems to find more empirical support, for instance not only the GoN data in (106-107), the morphological data in (98) but also the fact that the ANS can function as the (accusative) argument of adversity predicates in (99) testifies to its genuine nature. The issue of a case feature unvalued in narrow syntax and thus sneaking into the LF representation dissolves immediately. We assume that nominal phrases (including Numeral Noun Constructions) are insulated with a set of Case Projections, such that nominative is the smallest case while accusative and oblique cases are "larger" (cf. Caha 2009: comitative > instrumental > dative > genitive > accusative > nominative [noun]). It must be admitted that the ANS shows hybrid agreement with the passive participle/predicative adjective, whereby they either show accusative or genitive, which is an obvious problem for the accusative subject hypothesis, as Willim (2015) observes: (i) Polish Pięć studentek było wybrane/wybranych. five.acc students.gen was.3sg.neut selected.pl.nonvir.acc/pl.nonvir.gen 'Five students were selected.' However, Witkoś and Dziubała-Szrejbrowska (2016) show how this problem can be solved if the representation of the ANS includes a functional domain consisting of one set of case projections (in the sense of Caha 2009) which needs to express the case of both the NP-complement (genitive) and the numeral itself (accusative). Witkoś et al. (forthcoming), includes a detailed discussion of competing approaches to the issue of Numeral Phrases in the subject position. 70 Jakobson (1936), Andrews (1982), and Babby (1990) take nominative nominals to be without a case value. Bittner and Hale (1996) also consider nominative and absolutive caseless forms. 71 The notion of [uT] as a feature of structural case stems from the conclusions reached in Pesetsky and Torrego (2001;. The latter source also treats the accusative case feature as [T o ], to be read as "object" Tense. One of the reviewers suggests that alternatively T and the ANS could match for the structural case feature [uT] in narrow syntax, irrespective of its content (value).
not hold of T and any other non-nominative element fronted to the clause-initial position and examined in this paper. 72

Conclusion
We began this paper by noting that the Labeling Algorithm as put forth in Chomsky (2013; creates a puzzle for XPs that appear to move to [Spec, TP] but do not agree with T; the Labeling Algorithm prevents such non-agreeing XPs from occupying the [Spec, TP] position. This is prohibited because during the derivation, two phrases are being merged: a TP and an XP (i.e. DP or PP in locative inversion cases). Since this XP does not share any features with TP, the Labeling Algorithm "finds" two heads (X and T) and is unable to label the syntactic object, rendering it uninterpretable. We have identified two possible solutions to this problem. In both, the XP moves directly to a higher position never stopping in [Spec, TP]. In locative and predicate inversion cases, the position in question is [Spec, TopP], where the moved element shares a Topic feature with TopP. In Polish and Russian dative subject constructions, Lithuanian Inferential Evidentials, and Russian and Polish Adversity Impersonals, the position in question is [Spec, FinP]. The [Spec, TP] position remains un-projected, due to Split Feature Inheritance allowing phi-features to be inherited by T without the EPP feature being inherited as well.
If you cannot agree, move on! On labels and non-nominative subjects 55 (109) a.

b.
We also showed that under some circumstances (i.e. when partial Agree happens or when Agree takes places but phi-feature sharing fails leaving only a [T] feature to be shared), the non-nominative XP can move to (and remain in) [Spec, TP]. This is what happens with Icelandic dative subjects and Polish Accusative Numeral Subjects.
If you cannot agree, move on! On labels and non-nominative subjects 55 (109) a.

b.
We also showed that under some circumstances (i.e. when partial Agree happens or when Agree takes places but phi-feature sharing fails leaving only a [T] feature to be shared), the non-nominative XP can move to (and remain in) [Spec, TP]. This is what happens with Icelandic dative subjects and Polish Accusative Numeral Subjects.
Abbreviations 1 = first person, 2 = second person, 3 = third person, ACC = accusative, AI = Adversity Impersonal, ANS = Accusative Numeral Subject, ApplP = Applicative 72 The solution that we propose to account for the properties of the Accusative Numeral Subjects rests on the assumption that their case paradigm is impoverished; nominative is absent, and accusative takes over its morpho-syntactic function as the only structural case. Clearly, such an expansion of the role of the accusative does not happen with nouns whose case paradigm contains nominative case. Therefore, as pointed out by a reviewer, the following example is still expected to be ungrammatical: (i) Polish *Zosia i harcerzy poszli na spacer. Zosia.nom and boy-scouts.acc went.3pl.vir for walk 'Zosia and the boy-scouts went for a walk.' As harcerze 'boy-scouts' in Polish has a nominative form in its paradigm, this form is always used whenever it appears in the subject position (in line with the Subset Condition of Starke 2009 andCaha 2009), either independently or in coordination with other nominals.