The Passive of Reflexive Verbs and its Implications for Theories of Binding and Case∗

Version 05.07.2012

Florian Schäfer 1. Introduction In most languages, reflexive and reciprocal verbs do not passivize. However, German (1) and Icelandic (2) do allow the formation of passives of reflexive and reciprocal verbs. (I will use the shorthand PoRs for ‘Passives of Reflexive/Reciprocal Verbs’.)1/2 (1) Zuerst wird sich geküsst, später dann geheiratet. first becomes REFL kissed, later then married ‘First people kiss each other, then they marry.’ (2) (??)Það ar baðað sig á laugardögum. expl was bathed REFL.ACC on saturdays ‘People took a bath on Saturdays.’ (Sigurðsson 1989:355, fn. 60)3 In order to identify the basic properties of PoRs and the theoretical questions posed by their existence, consider the German active-passive pair in (3) involving the reflexive use of the verb ‘waschen’ (to wash). As in ordinary passives, the lexical verb in (3b) appears as passive

∗ Earlier versions of this paper have been presented at the Universities of Lund, Cologne, HU Berlin and Stuttgart as well as at the GGS 2010 (FU Berlin), the workshop ‘Morphological Voice and its Grammatical Interfaces’ (2010, University of Vienna), Nels 41 (2010, University of Pennsylvania), the workshop ‘Variation and Change in Argument Realization’ (2010, University of Naples), the workshop ‘Verb Meaning, Event Semantics and Argument Structure’ (2010, Centre de Lingüística Teòrica, Barcelona), the workshop ‘Approaches to the Lexicon’ (2011, Hebrew University of Jerusalem) and CGSW 26 (2011, University of Amsterdam). I thank the audiences for their comments and suggestions. I would also like to thank Tor Åfarli, Artemis Alexiadou, Kirsti Koch Christensen, Martin Everaert, Thórhallur Eythórsson, Arild Hestvik, Jóhannes Jónsson, Hans Kamp, Terje Lohndal, Marcel Pitteroff, Halldor Sigurðsson, Torgrim Solstad and Jan-Wouter Zwart for their comments and help. I am also very grateful to three anonymous JCGL reviewers and the editors for their insightful comments and suggestions. All shortcomings and errors are mine. This research was supported by a DFG grant to the project B6 ‘Underspecification in Voice systems and the syntax-morphology interface’ of the Collaborative Research Center 732 ‘Incremental Specification in Context’ at the Universität Stuttgart. 1 I use the following abbreviations: NOM - nominative, GEN - genitive, DAT - dative, ACC - accusative, REFL - simple reflexive pronoun, REFL-SELF - intensified reflexive pronoun, expl - expletive, NEU - neuter, FEM - feminine, MSC - masculine, SG - singular, PL - plural, PAST - Past Tense, PRTL - verbal particle, 1/2/3 - 1st/2nd/3rd person. 2 Lithuanian is one further language for which PoRs are attested (Geniušienė 1987, Wiemer 2006). Examples such as in (i) are sometimes cited to show that English allows PoRs (Baker et al. 1989, Collins 2005). However, it has been argued that the pronoun-self element in examples like (i) is not an anaphor but an intensified pronoun (Baker 1995, Reed 2011). In any case, the phenomenon is much more restricted in English than in German, Icelandic or Lithuanian. Note in this connection, that the example in (i) would be an instance of a ‘personal PoR’ involving a nominative DP and that the predicate ‘x keeps sth. for oneself’ is inherently reflexive. See section 8 for relevant discussion. (i) Such privileges should be kept to oneself. 3 Sigurðsson (1989) is one of the first who mentions the existence of PoRs in Icelandic. Nevertheless, he marks the example in (2) with two question marks. Eythórsson (2008, ex. 43b) mentions that most speakers totally accept this example. See section 3 for a detailed discussion of Icelandic PoRs.

2

participle and the passive auxiliary werden (to become) is used.4 The external argument disappears (but it can reappear in a by-phrase as will be shown later).5 The PoRs in (1), (2) and (3b) lack a nominative DP triggering verbal agreement. In fact, most PoRs are instances of ‘impersonal passives’. Related to this is the observation that the reflexive pronoun does not change its shape under passivization: it does not shift to nominative as referential objects would but it keeps its accusative. While this is not easy to see in German, Icelandic, which has a case-inflected paradigm of reflexive pronouns, shows this clearly (cf. (2); see section 3 for a more detailed discussion).6 (3) a. Hier haben die Römer sich gewaschen. Here have the.NOM Romans REFL.ACC washed b. Hier wurde sich (von den Römern) gewaschen. Here became REFL (by the Romans) washed ‘Here, the Romans washed.’ PoRs immediately pose two questions concerning Binding Theory and Case Theory:

Q1 Binding Theory: How is Principle A of the Binding Theory satisfied in PoRs, i.e. what is the antecedent of the reflexive pronoun? (Note that the reflexive in Icelandic PoRs does not qualify as a logophor and that German lacks logophors altogether (e.g. Kiss 2001.) Q2 Case Theory: How is accusative case on the reflexive pronoun licensed in PoRs? It stands in contradiction to Burzio’s Generalization, which predicts that passives, due to the absorption of the external argument, cannot license structural accusative case. In the course of the present paper, I will propose answers to these questions. The paper is

organized as follows: In section 2 and 3, I will investigate which verb classes allow the formation of PoRs in German and Icelandic respectively. As it turns out, PoRs are subject to a semantic or conceptual restriction on the underlying reflexive verb in both languages. In section 4, I will illustrate that PoRs are not available in other Germanic languages such as Dutch or Norwegian although these languages have a very similar reflexive system and although these languages do, in principle, form impersonal passives. In section 5, I will turn to the theoretical challenges posed by PoRs including Q1 and Q2 above. As we will see, present theories of reflexivity cannot account for all properties identified for PoRs. In section 6, I will turn to my own analysis of PoRs. There I will offer an answer to Q1 which builds on the idea that ordinary anaphoric binding depends on a syntactic AGREE-relation between a c-commanding DP-antecedent and the anaphor. In the absence of such a DP-antecedent (as in impersonal PoRs) some languages allow Default φ-feature Agreement to value the anaphor. This purely formal repair strategy will be related to the semantic or conceptual restriction on the reflexive predicates underlying PoRs identified in section 2 and 3. To answer Q2, I will 4 Crucially, the reflexive element itself does not reflect the passivization process. PoRs are, therefore, not “reflexive passives” of the Romance or Slavic type as discussed, for example, in Cinque (1988). 5 When no by-phrase is present, I paraphrase PoRs as active sentences with the subject ‘people'. This is not the only possible interpretation for the implicit external argument of PoRs, that is, PoRs do not necessarily have a generic flavor. More concretely, the possible interpretations of the implicit external argument of PoRs do not differ from the interpretations of the implicit external argument in ordinary impersonal passives. For some discussion of the latter in Icelandic, see Sigurðsson & Egerland (2009). Their findings carry over to German. 6 I concentrate on passives involving reflexive pronouns such as German ‘sich’ which can have a reflexive and a reciprocal reading. PoRs can also be found with reciprocal pronouns such as German ‘einander’ (each other). While I have not studied the latter in detail, it seems to me that the same lexical semantic restrictions hold for them as identified in section 2 for PoRs involving the reflexive pronoun ‘sich’.

3

develop a specific version of a dependent case approach where dependent case can be triggered not only by the presence of a nominative DP but also by Default Agreement as we find it in impersonal PoRs. In section 7, I will update some technical aspects of this analysis to explain the historical relation between PoRs and the so-called ‘New Passive’ (also called ‘New Construction’ or ‘New Impersonal’) in Icelandic. In section 8, I will investigate the rarer cases of personal PoRs, i.e. PoRs that involve a nominative DP which, however, does not act as antecedent for the reflexive pronoun. Section 9 is dedicated to the discussion of some open issues related to the analysis of PoRs developed here. Section 10 concludes.

Before I turn to a more detailed investigation of PoRs, let me quickly provide some further motivation that the formation of PoRs is a productive phenomenon in German (on Icelandic, see section 3). In fact, PoRs were often judged as ungrammatical in the theoretical literature on German (e.g. Reis 1982:20f, Haider 1985, Kiss 2003:fn 15, Bierwisch 2006). However, other authors recognized that this view is not generally correct, and, more concretely, acknowledged the formation of PoRs as a productive option provided by the grammar of German. First examples and preliminary discussions of PoRs can be found in Wunderlich (1985:222), Abraham (1986), Fanselow (1987, 1991), Sells, Zaenen & Zec (1987), Grewendorf (1988), Frey (1993) or Müller & Sternefeld (1993). Plank (1993) and Vater (1995) are the first who investigate PoRs in more depth and they show that PoRs are, in principle, accepted among speakers of German (see also Ágel 1997, Müller 1999 and Hundt 2002 for more detailed discussions).

Three further observations suggest that PoRs are a common phenomenon in German. First, PoRs should not be classified as substandard. The examples in (4) and (5) are from the evening news of the German public-law television (Tagesschau, ARD) whose anchormen are known for their high language standard. Second, while PoRs are probably more frequent in spoken language, they can also be found in written texts. The example in (6) is from the online version of a weekly German magazine (Spiegel online). Lastly, PoRs are not a recent invention. Behagel (1924, II:214) provides some Middle High German examples. (4) Während sich heute über Sonne gefreut werden konnte, while REFL today about sun rejoiced become could, muss morgen mit Regen gerechnet werden. must tomorrow with rain calculated become ‘While today the sun could be enjoyed, rain is expected for tomorrow.’ (5) Bei der Kieler Koalitionskrise wird sich schon at the Kieler.ADJ coalition-crisis becomes REFL already gar nicht mehr bemüht, die taktischen Mätzchen zu verstecken. really no longer bestirred the tactical tricks to hide ‘Politicians involved in the coalition crisis in the regional capital Kiel do not even try any longer to hide the tactical tricks.’ (6) Bei der ARD wurde sich eiligst für den Fauxpas entschuldigt. at the ARD became REFL hastily for the faux-pas apologized ‘The people responsible at the ARD hastily apologized for the faux pas.’ 2. A semantic/conceptual restriction on the formation of PoRs In section 1, I showed that German allows the formation of PoRs. However, a closer examination reveals that PoRs are restricted by a semantic or conceptual parameter. To introduce this parameter, it is useful to make a short detour to a language with a so-called ‘two-form reflexive system’ such as Dutch in (7-9). As is well known, Dutch has two

4

reflexive pronouns, the simple reflexive pronoun zich and the complex reflexive pronoun zichzelf. Furthermore, Dutch makes a morpho-syntactic distinction between three classes of reflexive verbs, i.e. verbs where the subject binds a reflexive pronoun in the direct object position. Crucially, the morphological distinction reflects a semantic or conceptual difference (see Kemmer (1993) and references there). The following verb-classes can be identified: Inherently reflexive (or inherently reciprocal) verbs: These verbs do not allow replacing the reflexive pronoun with a referential DP. Furthermore, only the simple reflexive pronoun is allowed: (7) Jan schaamt zich/*zichzelf/*Marie. John shames REFL/REFL-SELF/Mary

‘John is ashamed.’ Naturally reflexive (or naturally reciprocal) verbs: With these verbs, the reflexive pronoun can be replaced by a referential DP. In out-of-the-blue contexts, the simple reflexive is strongly preferred. (The complex reflexive becomes acceptable under strong focus.) (8) a. Jan waste zich/??zichzelf/Marie. John washed REFL/ REFL-SELF/Mary ‘John washed (Mary).’

b. Jan scheerde zich/??zichzelf/Peter. John shaved REFL/REFL-SELF/Peter

‘John shaved (Peter).’ Naturally reflexive verbs come from a number of semantic subclasses which all represent events that carry “… inherent in their meaning [...] the lack of expectation that the two semantic roles they make reference to will refer to distinct entities …” (Kemmer 1993:58). So-called “grooming verbs” such as ‘shave’, ‘wash’ or ‘dress’ form one main subgroup of naturally reflexive verbs. Naturally reciprocal verbs involve, for example, verbs of social (‘meet’) or affectionate (‘kiss’) events but also verbs of antagonistic events (‘fight’). The complete list of naturally reflexive/reciprocal verb classes proposed by Kemmer (1993) is given and exemplified in appendix 1. Naturally disjoint verbs (called other directed verbs in Koenig & Vezzosi 2004): Again, a referential DP can replace the reflexive pronoun but the complex reflexive is strongly preferred to express binding. In opposition to naturally reflexive verbs, we can say that these verbs express events which carry the expectation that the two semantic roles they make reference to will refer to distinct entities (e.g. ‘hate’, ‘accuse’, ‘kill’, …). (9) Zij haat ??zich/zichzelf/Peter. She hates REFL/REFL-SELF/Peter

‘John hates himself/Peter.’ German differs from Dutch in that the simple reflexive pronoun ‘sich’ can be used with all three of the verb classes above (10a-c). The addition of the intensifier ‘selbst’ (self), while often possible, is hardly ever obligatory in German. Therefore, German does not make a (obligatory) morphological distinction between inherently/naturally reflexive verbs and naturally disjoint verbs.

5

(10) a. Hans hasst sich/Maria. (naturally disjoint) John hates REFL/Maria ‘John hates himself/Mary.’ b. Hans wäscht sich/Maria. (naturally reflexive) John washes REFL/Maria ‘John washes himself/Mary.’ c. Hans schämt sich/*Maria. (inherently reflexive) John shames REFL/Maria ‘John is ashamed/ashames Mary.’ While German differs from Dutch in not making a morphological distinction between inherently/naturally reflexive verbs and naturally disjoint verbs, it turns out that the difference between these verb classes is, nevertheless, relevant in the grammar of German: The same lexical-semantic or conceptual aspects that determine the distribution of simple and complex reflexives in Dutch determine the formation of PoRs in German: PoRs are overwhelmingly formed with inherently and naturally reflexive verbs; naturally disjoint verbs are very rarely found in PoRs. Such a semantic restriction on the formation of PoRs has already been suggested in Abraham (1986), Sells & al. (1987), Fanselow (1991), Kaufmann (2001) and especially Ágel (1997), but it has never been empirically corroborated. Here, I present the results of two empirical studies that I undertook to confirm this proposal.

First, I searched in Google for PoRs involving verbs of the three different verb classes. Inherently reflexive verbs and naturally reflexive verbs led to a huge number of matches involving PoRs. To identify naturally reflexive verbs, I used the classification in Kemmer (1993: chapter 3, 4), which involves 13 subclasses of naturally reflexive verbs (see appendix 1). For each of these subclasses, I selected one verb and checked it on Google. The result was positive; i.e. for each subclass, already the very first verb that I had randomly selected lead to a number of matches where the verb was used in a PoR. A possible search query for the verb ‘waschen’ (to wash) would be “sich gewaschen wird” (REFL washed becomes) which brings about 12 hits (16.01.2012). As a comparison, I checked for a number of naturally disjoint verbs whether they form PoRs. It turned out that these verbs only very rarely form such passives. Below, I illustrate these findings with some examples. The full list of verbs tested and the number of PoRs found with them via Google queries are given in appendix 1.

In A, we see the results of two inherently reflexive verbs. In B.1 and B.2, we find the results from two verbs taken from two of the thirteen subclasses of naturally reflexive/reciprocal verbs identified by Kemmer (1993). In C, we see the results for two naturally disjoint verbs. Note that the numbers of matches given below for PoRs of inherently and naturally reflexive verbs make reference to only one search string/word order, either ‘auxiliary REFL participle’ or ‘REFL participle auxiliary’ and one time specification (present or past tense) on the auxiliary. That is, alternative search strings differing in word order or tense would have lead to many more hits. Importantly, the numbers of hits for naturally disjoint verbs involve much more search string. Specifically, for each of the verbs in C, I searched for (i) PoRs involving a reflexive pronoun (sich/REFL), an intensified reflexive pronoun (sich selbst/REFL-SELF), and a reciprocal pronoun (einander/each other), (ii) PoRs in two different word orders (for matrix and embedded sentences), and (iii) PoRs in the present as well as in the past tense. Despite this 12:1 ratio of search strings I found significantly fewer PoRs with naturally disjoint verbs than with naturally and inherently reflexive/reciprocal verbs.

A. inherently reflexive/reciprocal verb ‘sich benehmen’ (to behave): search string “sich benommen wird” - 54 matches

6

(11) Es ist schlimm wie sich benommen wird. It is bad how REFL behaved becomes ‘It is bad how people behave.’ ‘sich streiten’ (to quarrel): search string “sich gestritten wird” - more than 100 matches (12) Ich verstehe echt nicht, wieso sich gestritten wird.

I understand really not why REFL quarreled becomes ‘I really do not understand why people quarrel with each other.’

B.1 grooming verbs ‘waschen’ (to wash): search string “wird sich gewaschen” - more than 100 matches (13) Dann wird sich gewaschen, schön gemacht, umgezogen Then becomes REFL washed, nice made, clothes-changed und die Zähne geputzt. and the teeth brushed ‘Then people wash, make themselves up, change their clothes and brush their teeth.’ B.2 verbs of translational motion ‘bewegen’ (to move): search string “wird sich bewegt” - more than 100 matches (14) im Musikunterricht wird sich bewegt, im Kunstunterricht wird in.the music-class becomes REFL moved, in.the art-class becomes der Pinsel geschwungen. the brush swung ‘In the music class, people move, in the art class, people swing the brush.’ C. naturally disjoint (transitive) verbs ‘schneiden’ (to cut): search string for example “sich geschnitten wurde” - 3 matches (15) Dazu lagen auch noch Glasscherben im Hauptpool, In-addition lay also even glas-fragments in.the main-pool, an denen sich geschnitten wurde! at which REFL cut became ‘In addition, there was broken glas in the main pool at which poeple cut themselves.’ ‘zerstören’ (to destroy): Search string for example “sich (selbst) zerstört wird” - 0 matches These findings (see appendix 1 for more verbs tested) further confirm the existence of PoRs in German. Furthermore, they seem to support the claim by Ágel (1997) that basically only inherently reflexive and naturally reflexive verbs form PoRs. While I found some PoRs involving naturally disjoint verbs, their number was much lower than the number of PoRs with inherently/naturally reflexive verbs.7 7 Hundt (2002) provides a list of 95 modern and 26 historical examples of PoRs. 7 of his examples involve verbs that might be best classified as naturally disjoint. Interestingly, such counterexamples to the above generalization very often contain the adverb ‘gegenseitig’ (mutually) as in (i) which necessarily triggers a reciprocal interpretation, i.e. the addition of this adverb to a naturally disjoint verb results in an inherently reciprocal context (see Geurts 2004:4 for a similar effect in Dutch). Other counterexamples turn out to be from specialized communities (see fn. 60 in appendix 1 for an example) or to involve irony (as if the event were naturally reflexive).

7

However, the results of a corpus study like the one above might be misleading because it is not clear whether we encounter a difference in grammaticality or acceptability or rather one in frequency. The problem is that, in general, passives are less frequent than active sentences and, more importantly, naturally reflexive verbs show up in a reflexive construal more often than naturally disjoint verbs (see Haspelmath 2008, Hendriks et al. 2008, Bouma & Spenader 2009). The combination of these two frequency effects might be responsible for the absence (or very low number) of PoRs with naturally disjoint verbs. PoRs of naturally disjoint verbs might, therefore, not be ungrammatical or unacceptable but just very infrequent. Therefore, we need to back up the above findings and test whether there is a difference in the grammaticality/acceptability between the reflexive active use and the reflexive passive use of the three verb classes under consideration. To do this, I ran a questionnaire study with first-year students at the University of Stuttgart. This study contained 48 target sentences all of which had the reflexive pronoun ‘sich’ in the direct object position. 24 of these target sentences were reflexive active sentences, the other 24 sentences were the impersonal passive counterparts of these active sentences (i.e., PoRs). The subject of the active sentences was either the 3rd person plural pronoun ‘sie’ (they) or the singular impersonal pronoun ‘man’ (one). The passive sentences had no by-phrase. The 24 active-passive pairs included four inherently reflexive verbs (i-ref), four inherently reciprocal verbs (i-rec), four naturally reflexive verbs (n-ref) and four naturally reciprocal verbs (n-rec) as well as eight naturally disjoint verbs, four of them used in a reflexive construal (nd-ref) and four of them used in a reciprocal construal (nd-rec). The test sentences are given in appendix 2. These 48 target sentences were arranged in two questionnaires with 24 sentences each so that no verb occurred twice in one questionnaire (i.e., each verb occurred either in the active or in the passive in one questionnaire). Both questionnaires were filled with the same 48 distractors so that the ratio of test sentences and distractors was 1:2.8 Furthermore, both questionnaires were presented in two randomized orders each. Each sentence was read by 24 speakers, who judged it on a scale from 1 (totally acceptable) to 7 (totally unacceptable). Table 1 below shows the mean results of the six verb classes in the active and in the passive.

TABLE 1

The results in the above table lead to the following conclusions:

(i) Hier wird sich nicht gegenseitig umgebracht. Here becomes REFL not mutually killed ‘People do not kill each other here.’ 8 16 of the 48 distractors were grammatical, 16 were marked (marked word order) and 16 were plainly ungrammatical. 36 of the distractors were not reflexive; of these, 18 were active and 18 were passive sentences. 12 distractor sentences were reflexive: three of them were active, grammatical sentences, three of them were active and marked and three of them were active but ungrammatical sentences. Finally, three distractor sentences were plainly ungrammatical PoRs. For the calculation of the results, only subjects were considered who did not wrongly classify more than two of the 16 plainly ungrammatical control sentences as acceptable (i.e. 4 or better).

i-ref i-rec n-ref n-rec nd-ref nd-rec active 1,56 1,87 1,68 1,66 2,26 1,79 passive 2,67 3,21 3,1 3,19 5,21 4,66

1 2 3 4 5 6 7

8

(i) First of all, we see that it does not matter for the acceptability of active as well as passive sentences (PoRs) whether the reflexive pronoun gets a reflexive or a reciprocal interpretation (that is the acceptance rates in the columns 1 and 2, 3 and 4 and 5 and 6 are approximately the same). (ii) In the active, the different verb classes were judged equally acceptable. That is, inherently reflexive/reciprocal verbs receive the same acceptance as naturally reflexive/reciprocal verbs and naturally disjoint verbs under a reflexive/reciprocal use.9 (iii) PoRs are generally less acceptable than the corresponding active counterparts. I take this general result to be (at least partly)10 independent from reflexivization. It is known that ordinary passives (with by-phrases) are judged as less acceptable than their active counterparts in out-of-the-blue contexts (at least if both arguments are animate; see Greenbaum 1977 for English, see Murphy 2007 for English and German). Furthermore, since the PoRs in the above questionnaire study were truncated and lacked by-phrase, they were less informative than their active counterparts and this probably also influenced their out-of-the-blue acceptability. (iv) The most important observation for our purposes is that PoRs of naturally disjoint verbs are much less acceptable than PoRs involving inherently and naturally reflexive/reciprocal verbs. PoRs of the latter two verb classes, on the other hand, receive approximately the same acceptance. These results are in accordance with the results of the corpus study above. The experiment also adds in an important way to such corpus data as it shows that the distribution of PoRs across verb classes cannot be explained away in terms of the frequency of reflexive uses. To conclude then, both studies confirm the claim by Ágel (1997) that inherently and naturally reflexive/reciprocal verbs enter the formation of PoRs in German without problem while naturally disjoint verbs are basically out. Note, however, that in my analysis of PoRs in section 6, I argue that PoRs of naturally disjoint verbs are not formally ungrammatical but just hard to interpret (i.e. unacceptable).11

With this background, we can turn to a more detailed discussion of PoRs in Icelandic. 3. PoRs in Icelandic Sigurðsson (1989:355, fn. 60) is one of the earliest sources for PoRs in Icelandic. He provides the active-passive pairs in (16) and (17) ((17b) repeats example (2) above). While Sigurðsson marks (16b) with one and (17b) with two question marks, Eythórsson (2008) judges these examples as totally acceptable (no question mark) and adds that most speakers he consulted agreed with him. (Maling & Sigurjónsdóttir (2002:121, fn. 12) also report a study where only 1 out of 24 speakers judged (16b) as ungrammatical. It is interesting to note that all 24 speakers were from Inner Reykjavík (see below for discussion why this is relevant)). Note that the verb in (16) is inherently reflexive, and the verb in (17) is naturally reflexive. Note furthermore, that Icelandic reflexive pronouns have a case-inflected paradigm (sig-ACC, sér-DAT, sín-GEN). Icelandic, therefore, provides clear morphological indication that the

9 Note that all test sentences involved the simple reflexive pronoun ‘sich’ without the intensivier ‘selbst’ (self). This supports the claim that this intensifier is not obligatory in German. 10 Árnadóttir et al. (2011) suggest that Icelandic PoRs, while in principle grammatical, receive less acceptance than impersonal passives without a reflexive pronoun. I have not checked this for German. 11 Thereby the analysis presented in section 6 will allow it that PoRs of basically naturally disjoint verbs become acceptable if a specific context guides a successful interpretation of PoRs because it strongly suggests that the implicit external argument of the passive and the anaphor in object position should be covalued. This then provides an explanation why we find such verbs very rarely in corpora (see also fn. 7 and fn. 60 in appendix 1).

9

reflexive pronoun in PoRs keeps the case that it has in the active. In (16), this is lexical dative, but in (17) this is structural accusative.12 (16) a. Börnin leika sér allan daginn. the.children play REFL.DAT all the.day ‘The children are playing all day.’ b. (?)Það var leikið sér allan daginn. expl was played REFL.DAT all the.day (17) a. Fólkið baðaði sig á laugardögum. the.people bathed REFL.ACC on Saturdays ‘The people took a bath on Saturdays.’ b. (??)Það var baðað sig á laugardögum. expl was bathed REFL.ACC on Saturdays Besides the fact that examples of Icelandic PoRs discussed in the literature typically involve inherently or naturally reflexive verbs, there is at least some further indication that Icelandic PoRs are restricted by the same semantic/conceptual parameter identified for German in the last section. The relevant data suggesting this are taken from a questionnaire study by Maling & Sigurjónsdóttir (2002). Before we can take a closer look at their examples, some comments are necessary. Maling & Sigurjónsdóttir are only indirectly concerned with PoRs but investigate a broader phenomenon of Icelandic syntax, which they call the ‘New Construction’ or ‘New Impersonal’. Other authors use the term ‘New Passive’, a terminology that I will adopt in this article (see the next footnote). This construction, which is a recent innovation in Icelandic, shares with the canonical passive that the same auxiliary and the same participle are used and that the external argument is absorbed. It differs, however, from the canonical passive in that no DP-movement of the internal argument takes place and the internal argument keeps its structural accusative case.13 (A more detailed discussion of the New Passive will be provided in section 7). The New Passive is subject to quite some speaker variation. As Maling & Sigurjónsdóttir (2002) show most adolescents (15-16 year old students) living outside the capital Reykjavík accept it, while adolescents living inside of Reykjavík accept it less and adults typically reject it. All speakers of Icelandic, on the other hand, accept the canonical passive. These differences between the canonical passive and the New Passive are exemplified below where (18a) is a canonical passive and (18b) is a New Passive. The numbers to the right of these examples express how many speakers within a group accepted this sentence. Elsewhere refers to adolescents living outside of inner Reykjavík, inner Rvik refers to adolescents living inside of Reykjavík and the third group represents the judgements of adults across the whole country (only binary judgements were possible in this study (yes, this is something one can say, no this is something one cannot say)). 12 One might suspect that the accusative form of the Icelandic reflexive pronoun is actually ambiguous between nominative and accusative. This hypothesis, is, however, not tenable. It has been shown that - independently of Principle A (i.e. locality) considerations - Icelandic reflexive pronouns are incompatible with nominative positions (Maling (1984), Everaert (1990); see Rizzi (1990), Woolford (1999) or Tucker (2010) on this so-called ‘anaphor agreement effect’ in other languages. See also section 9. We must, therefore, conclude that the reflexive in (17b) gets assigned structural accusative case. 13 The question whether the New Passive is a real passive or an impersonal construction with a covert subject (pro) as proposed by Maling & Sigurjónsdóttir (2002), is still a topic of discussion. I will follow here without further discussion the argumentation in some recent literature (Barðal & Molnár 2003, Eyþórsson 2008, Jónsson 2009, Sigurðsson 2011, Árnadóttir et al. 2011) that the New Passive is a real passive in that the external argument is absorbed in the same way as it is in canonical passives. Therefore, I use the term ‘New Passive’.

10

(18) a. Ólafur var rekinn úr skólanum. 00 Elsewhere Inner Rvík Adults a. * Olaf.NOM was driven from the.school 99% 96% 99% a. * ‘Olaf was expelled from school.’ b. Það var beðið mig að vaska upp. 0 Elsewhere Inner Rvík Adults a. expl was asked me.ACC to wash up 74% 47% 8% a. ‘I was asked to do the dishes.’ PoRs share with the New Passive that the object does not move to subject position and keeps its structural accusative case. However, the literature also pointed out that PoRs are not genuine instances of the New Passive (Sigurðsson 1989, Eythórsson 2008, Árnadóttir et al. 2011, Jónsson 2011). On the one hand, PoRs can be traced further back in history than the New Passive (Árnadóttir et al. 2011) and, on the other hand, many speakers who do not accept the New Passive with non-reflexive verbs accept PoRs (Eythórsson 2008, Árnadóttir et al. 2011). The latter conclusion can also be drawn from the questionnaire study by Maling & Sigurjónsdóttir (2002), which shows that PoRs receive much better acceptance than New Passives with a referential internal argument. (To see this compare the judgements in (18b) with those in (19a-c), (20a) and (21a)) This holds most significantly for adolescents from within Reykjavík and adults. Nevertheless, many authors have proposed that PoRs might have been the first step of a grammatical development that lead to the New Passive (e.g. Maling & Sigurjónsdóttir 2002, Árnadóttir et al. 2011). I will take up and develop this hypothesis in section 7. With this background, we can turn to some of the examples from Maling & Sigurjónsdóttir (2002) involving PoRs. The passive sentences in (19) all involve inherently reflexive verbs with the simple reflexive pronoun ‘sig’ in object position. The three groups of speakers show quite the same acceptance rate for each of the three sentences. (Sentence (19c) is for some reason slightly less acceptable than the other two examples.) Since all three examples are PoRs with inherently reflexive verbs, I take their acceptance rates as a baseline. The question is then how acceptable PoRs involving naturally reflexive and naturally disjoint verbs are. (19) a. Svo var bara drifið sig á ball. then was just hurried REFL to the.dance ‘People hurried to the dance.’ (Elsewhere 78% | Inner Rvík 67% | Adults 40%) b. Það var haldið sig innan dyra út af óveðrinu. it was kept REFL in doors due to bad.weather ‘People remained in the house due to the bad weather.’ (Elsewhere 82% | Inner Rvík 65% | Adults 37%) c. Það var skoðað sig um á svæðinu. it was looked REFL around in the.area ‘People took a look-around in the area.’ (Elsewhere 72% | Inner Rvík 43% | Adults 31%) The examples in (20) both involve a complex reflexive pronoun embedded in a prepositional phrase. Although both examples are syntactically identical, the first example receives much more acceptance than the second one in all three groups of speakers. I want to suggest that the reason for this difference is that the predicate “X looks at Y in the mirror” is naturally reflexive (i.e., there is a strong expectation that X=Y), while the predicate “X points at Y in the picture” is not. Typically, people look at (a representation of) themselves when they look into a mirror. But there is no expectation that someone points at (a representation of) herself/himself when s/he points at a person in a picture. This conceptual difference, I claim, is reflected in the acceptance rates of the two examples.

11

(20) a. Það var horft á sjálfan sig í speglinum. it was looked at SELF REFL in the.mirror ‘People looked at themselves in the mirror.’ (Elsewhere 58% | Inner Rvík 48% | Adults 34%) b. Það var bent á sjálfan sig á myndinni. it was pointed to SELF REFL in the.picture ‘People pointed at themselves in the picture.’ (Elsewhere 19% | Inner Rvík 11% | Adults 13%) The same conceptual parameter seems to be at play in the examples in (21) involving possessive reflexive pronouns. These examples show some syntactic differences. In (21a), the possessed noun phrase is embedded in a prepositional phrase selected by the verb, in (21b) and (21c), the possessed noun phrase is the internal argument of the verb and in (21d), the possessed NP is embedded in a prepositional phrase modifying the internal argument of the verb.14 These syntactic differences make it hard to draw strong conclusions. In any case, we see that (21a) is accepted by many more speakers in all three groups than (21b, c, d). Once again, I suggest that the reason is that “X supports Y’s team” is naturally reflexive (people typically support their own team) while this is not the case in the other three examples. For example, it is not more natural to push one’s own sister off the bike than it is to push someone else’s sister off the bike.15 (21) a. Það var haldið með sínu liði. it was held with SELF’s team ‘People supported their own team.’ (Elsewhere 63% | Inner Rvík 49% | Adults 36%) b. Í morgun var hrint systur sinni af hjólinu. this morning was pushed sister SELF’s off the.bike ‘This morning, people pushed their own sister from the bike.’ (Elsewhere 13% | Inner Rvík 7% | Adults 2%) c. Það var oft kaffært bróður sinn í sundlauginni. it was often dunked brother SELF in the.pool ‘People often dunked their own brother in the pool.’ (Elsewhere 5% | Inner Rvík 3% | Adults 1%) d. Það var klippt hárið á dúkkunni sinni. it was cut the.hair on doll SELF’s ‘People cut the hair of their own doll.’ (Elsewhere 5% | Inner Rvík 2% | Adults 2%) The examples in (21), if analyzed in the correct way, make an important point about the nature of the concept ‘natural reflexivity’. I proposed that (21a) is acceptable because there holds a naturally reflexive relation between an argument of the verb (its implicit external argument) and the possessor of an NP embedded in a PP selected by the verb. If this

14 A further difference is that the reflexive possessive in (21a) precedes the head noun while it follows the head noun in (21b-d). 15 Maling & Sigurjónsdóttir (2002:124) suggest that the examples in (21b, c, d) would also be judged as unnatural if they were presented as active sentences with an overt impersonal subject. I predict, on the other hand, that such counterparts, even if unnatural, should be much more acceptable in the active than in the passive. This was at least one of the results of the German questionnaire study reported above. However, an ultimate answer about the restrictions on Icelandic PoRs would require a similar experiment, an undertaking which I have to leave for future research.

12

explanation is correct, it follows that ‘natural reflexivity’ cannot always be determined within a verb’s co-argument domain. More concretely, since possessors are not part of a verb’s lexical representation, ‘natural reflexivity’ cannot (always) be established in the lexicon but must be computed at a conceptual level following syntactic computation. It is therefore more correct to talk about ‘naturally reflexive situations’ (see also the term ‘middle situation type’ in Kemmer 1993). To summarize this section, there is at least some indication that Icelandic PoRs are constrained by the same semantic/conceptual parameter as German PoRs. Inherently reflexive and naturally reflexive expressions receive better acceptance in PoRs than naturally disjoint expressions. Furthermore, Icelandic shows that ‘natural reflexivity’ is not necessarily a relation between co-arguments of a verb. (Below, I will argue that the same is the case with inherent reflexivity). Finally, Icelandic provides clear evidence that the reflexive pronoun in PoRs indeed keeps the case it has in the corresponding active counterpart. That is, structural accusative survives in PoRs.16 4. PoRs in other Germanic languages: Dutch and Norwegian As mentioned in the introduction, PoRs seem to be a rare phenomenon. In fact, German, Icelandic (and Lithuanian, see fn. 2) are the only languages that I know of which form PoRs productively. In this section, I exemplify the lack of PoRs in two other Germanic languages, Dutch and Norwegian. (Here, I discuss the absence of impersonal PoRs in these languages. My discussion of personal PoRs in section 8 will lead to some modification of the claim that these languages lack PoRs).

As the Dutch example in (22) and the Norwegian examples in (23) show, PoRs are judged as ungrammatical in both languages. Note that these examples are ungrammatical even though (22) involves a naturally reflexive verb and (23a) involves an inherently reflexive verb. (The verb in (23b) is naturally disjoint).17

16 Müller & Sternefeld (1993:359) try to explain away the case problem of PoRs by stipulating that German reflexives are immune to Case-absorption simply because they “do not need Case in the first place”. Although the German reflexive ‘sich’ does not carry overt case morphology, there are arguments which suggest that it has case. First, ‘sich’ can occur only in accusative or dative positions while in locally bound genitive positions, a 3rd person pronoun has to be used (e.g. Fanselow 1989, 1991, Burzio 1998). Second, if the antecedent is 1st or 2nd person, the anaphor is replaced by a 1st or 2nd person pronoun which does show overt case variation (ia-d). Third, case copying constructions (Fanselow 1991) suggest that accusative is licensed on reflexives (iia, b). Case copying is not easily applicable in PoRs (for semantic reasons, I would suggest, because such constructions destroy natural reflexivity). But if German reflexives have case in the active, there is no reason to assume that they are exempt from case assignment in PoRs. (i) a. Ich wasche mich. b. Du wäschst dich. c. Ich helfe mir. d. Du hilfst dir.

I wash me.ACC you wash you.ACC I help me.DAT you help you.DAT ‘I wash’ ‘You wash’ ‘I help myself’ ‘You help yourself’

(ii) a. weil Hans sich als einen Superhelden zeichnet. as John.NOM REFL.ACC as a.ACC superhero paints ‘because John paints (a picture of) himself as a superhero.’ b. weil Hans sich als einen Idioten ansieht. as John.NOM REFL.ACC as an.ACC idiot at-sees ‘because John regards himself as an idiot.’ 17 The ungrammaticality of (22) was confirmed by Martin Everaert, Hans Kamp, and Jan-Wouter Zwart (p.c.). The ungrammaticality of (23a, b) was confirmed by Terje Lohndal, Torgrim Solstad, and Kirsti Koch Christensen (p.c.). However, more empirical work is necessary on the availability of PoRs in these languages (see also section 8). For example, there is some indication that not all Norwegian speakers reject all impersonal PoRs to the same extent (p.c. Tor Åfarli and Arild Hestvik, see also Åfarli 1992:128 for some examples of Norwegian PoRs which he marks with two question marks; these were, however, rejected by my informants). A reviewer informs me that there are some Norwegian speakers who accept (23b) but only in the version without ‘selv’. This fits at least to some extent with my proposal that PoRs, if possible at all, must involve inherently or

13

(22) *Er werd zich gewassen. There was REFL washed ‘People washed (themselves).’ (Reinhart & Siloni 2004:169, ex. 18c) (23) a. *Det ble hygget seg. It was amused REFL ‘People amused (themselves).’ (Maling & Sigurjónsdóttir 2002:133, ex. 44b) b. *Det ble låst seg (selv) inn i fabrikken. It was locked REFL (SELF) inside in the.factory ‘People locked themselves in the factory.’ (Maling 2006:219, ex. 26b) Note that Dutch and Norwegian lack PoRs although they share two properties with German and Icelandic that might be preconditions for the availability of PoRs: First, Dutch and Norwegian have impersonal passives (recall that (most) PoRs in German and Icelandic are impersonal passives; but see section 8 on personal PoRs). Second, the reflexive system in Dutch and Norwegian is relatively similar to the system in Icelandic and German. Most importantly, both Dutch and Norwegian have a light reflexive pronoun like German sich and Icelandic sig – a SE-reflexive in the terminology of Reinhart & Reuland (1993) – which is typically used in the context of inherently and naturally reflexive verbs. 5. Theoretical evaluation Any theoretical account of PoRs should be able to answer the following questions (questions Q1 and Q2 were already formulated in the introduction, questions Q3 and Q4 grew out of the investigation in the last three sections): Q1 Binding Theory: How is Principle A of the Binding Theory satisfied in PoRs, i.e. what is the antecedent of the reflexive pronoun/anaphor?18 Q2 Case Theory: How is the accusative case on the reflexive pronoun licensed in PoRs, i.e. how can we understand/explain this counterexample to Burzio’s Generalization? Q3 Verb Class Restriction: Why are PoRs acceptable only with inherently/naturally reflexive verbs but not with naturally disjoint verbs? Recall that in the active, the three verb classes behave the same (see Table 1). Q4 Language Restriction: Why don’t we find PoRs in other (Germanic) languages, especially, if they have impersonal passives and a similar anaphoric system involving SE-reflexives? In section 6, I will provide answers to these questions. To give a very quick preview, my answer to Q1 will build on the idea that ordinary anaphoric binding depends on a syntactic AGREE-relation between a c-commanding DP-antecedent and the anaphor. In the absence of such a DP-antecedent (as in impersonal PoRs) some languages allow Default φ-feature Agreement to formally value the anaphor. Since Default Agreement is a costly operation I naturally reflexive verbs as only these two verb classes take the simple reflexive in Scandinavian languages. I would propose that these speakers could come up with a context where the concept ‘x locks y’ gets a naturally reflexive evaluation (see fn. 60 in appendix 1 for similar effects). 18 Below, I will use the term ‘reflexive (pronoun)’ to refer to an element’s reflexive morphology and the term ‘anaphor’ to refer to an element which is subject to Principle A of the Binding Theory.

14

argue that it is available only in some languages (Q4). The answer to Q3 will follow from the fact that Default Agreement can only formally value an anaphor but cannot provide a semantic antecedent for it. In this case, a sucessfull interpretation of the anaphor must depend on conceptual knowledge about the underlying predicate. Only if this is conceived as inherently or naturally reflexive, the speaker/hearer can arrive at an interpretation where the implicit external argument of the passive and the anaphor in object position are covalued (Q3). Finally, to answer Q2, I will develop a specific version of a dependent case approach where dependent case can be triggered not only by the presence of a nominative DP but also by Default Agreement as we find it in impersonal PoRs. However, before I turn in the next section to a detailed discussion of my solution for the above four questions, I want to discuss a number of theoretical proposals from the literature – either specifically on PoRs or, more generally, on reflexivity – that turn out to be insufficient to fully answer the above questions. This discussion will pave the ground for my own proposal.

As an answer to question Q1 (Binding Theory), one could suggest that the implicit argument of passives can act as the antecedent of the anaphor (e.g. Fanselow 1987, Barðdal & Molnár 2003, Sternefeld 2006, Sigurðsson 2011; see also the so-called ‘smuggling’ approach to passives in Collins 2005). However, this proposal leaves questions Q2-Q4 without answers: it does not explain why the reflexive pronoun gets accusative case (while referential DP-objects shift to nominative), it cannot explain why only inherently and naturally reflexive verbs can form PoRs and it cannot explain why only German and Icelandic form PoRs. Note in connection to the last point that the implicit argument of passives seems to have quite the same properties across languages otherwise (e.g. licensing of control, licensing of agentive adverbs, …).

In order to answer both, question Q1 about Binding Theory and question Q2 about Case Theory, one could also suggest that PoRs are hidden transitives with a covert external argument (for example pro, cf. Maling & Sigurjónsdóttir 2002 for Icelandic). If PoRs have a syntactically projected external argument they should behave like active clauses concerning Binding Theory and Case Theory. The covert external argument would act as the antecedent of the anaphor and Burzio’s Generalization would predict that accusative is available. However, this proposal cannot answer question Q3: Why should the covert external argument in PoRs be able to antecede an anaphor only if the verb is inherently or naturally reflexive but not if the verb is naturally disjoint? This is unexpected because in the active this difference does not exist (cf. Table 1). (As far as I know Romance pro-subjects can antecede reflexives in all types of reflexive predicates). Furthermore, it remains unclear why only German and Icelandic can have this type of passive (question Q4).19 German provides a further counter-argument against the idea that the anaphor in PoRs has a referential antecedent in the syntax. German impersonal passives combine with agentive by-phrases which can even introduce 1st and 2nd person pronouns (24).20 (24) a. Von mir wurde nicht gelacht. by me was not laughed

19 Maling & Sigurjónsdóttir (2002) propose a silent pro for Icelandic because this language has not only PoRs with accusative reflexive pronouns but also the New Passive where a referential internal argument keeps its strcutural accusative case. But, as discussed above, Icelandic PoRs are not genuine instances of the New Passive in that they are accepted by many speakers who reject the New Passive with referential internal arguments. Similarily, a pro-analysis would not work for German as this language lacks the counterpart of the New Passive alltogether. That is, for both standard Icelandic and German it is hard to understand why this pro should occur only if the structure involves a reflexive pronoun inside the VP. 20 This argument cannot be reproduced for Icelandic because Icelandic does not really accept by-phrases in any kind of impersonal passive (see e.g. Eythórsson 2008 or Jónsson 2009).

15

‘I did not laugh.’ b. weil von dir zu viel gelabert wird.

because by you too much babbled becomes ‘because you babble too much.’ 1st and 2nd person by-phrases are possible with PoRs, too. However, their effect on a reflexive pronoun differs from the effect that a 1st or 2nd person subject has in active clauses. In the active, the bound element agrees with the subject-antecedent in person and number (25a). Crucially and unequivocally, no agreement between the by-phrase and the bound element is possible in PoRs (25b); instead, the 3rd person reflexive pronoun is obligatory, as Plank (1993) has observed. I conclude from these data that the anaphor in PoRs does not have any argument as its syntactic antecedent at all. Instead, the anaphor gets default realization (as a 3rd person reflexive pronoun).21 (25) a. Nur wir waschen uns / *sich hier täglich. only we wash us.ACC / REFL here daily b. Nur von uns wird sich / *uns hier täglich gewaschen. only by us becomes REFL / us.ACC here daily washed

‘Only we wash ourselves here everyday.’ The existence of PoRs can also not be captured by approaches that treat reflexive verbs as

being detransitivized, be it in the lexicon or in the syntax.22 A case in point is the so-called ‘Bundling-approach’ developed by Reinhart (2000) and Reinhart & Siloni (2005). These authors assume that reflexive verbs are derived by a process which bundles a verb’s internal θ-role (<theme>) with its external θ-role (<agent>). As a result, only one argument with a complex θ-role (<agent, theme>) is merged in the external argument position. (26) illustrates the effect of Bundling:

(26) Reflexivization Bundling (Reinhart & Siloni 2005: 400) [θi] [θj] --> [θi - θj], where θi is an external θ-role.

Crucially then, the reflexive element is not an argument of the verb in this account. It is

either merged as a lexical sign that indicates that the Bundling-operation has taken place or it acts as a case-reducer which absorbs the accusative case of the basically transitive verb. These two options are correlated with the module of grammar where Bundling is assumed to take place. In some languages, Bundling is assumed to take place in the lexicon (e.g. English), in other languages, Bundling takes place in the syntax (e.g. French, German). The two options are illustrated below in (27) and (28) (Reinhart & Siloni 2005: 401, 404):

21 An alternative, not implausible idea would be that the existentially bound implicit argument of passives is always 3rd person (or impersonal) and that by-phrases do not have to match the implicit argument in φ-features. But this would leave unexplained why naturally disjoint verbs are typically judged as deviant in PoRs and why most languages lack PoRs although, across languages, implicit arguments of passives show the same properties otherwise. 22 I concentrate here on a version of the intransitivity-account of reflexive verbs that assumes that these verbs are unergative. Some researchers proposed that reflexive verbs are unaccusative (e.g. Grimshaw 1981). PoRs are problematic for such an account simply because unaccusatives are not expected to passivize. McGinnis (1998, 2000, 2004) and Embick (2004) propose that reflexive verbs are transitive but involve an unaccusative derivation. The reflexive pronoun is located in the external argument position and the internal argument moves across the external argument position to bind the reflexive pronoun. Such an account, besides having technical problems, is incompatible with PoRs because the process of passivization should absorb the external argument (i.e., the reflexive pronoun).

16

(27) Bundling in the lexicon (The case feature is reduced before syntax) a. Example: Max washes. b. Verb entry: washacc[Agent] [Theme] c. Reflexivization output: wash[Agent-Theme] d. Syntactic output: Max[Agent-Theme] washed e. Interpretation: ∃e [wash(e) & [Agent-Theme](e, Max)] -->23 ∃e [wash (e) & Agent(e, Max) & Theme(e, Max)] (28) Bundling in the syntax (The case feature is absorbed by the reflexive element) a. Example: Jean se lave. Jean REFL washes ‘John washes.’ b. VP: [se laveθi-Agent, θk-Theme] c. IP: [Jean<θi, θk> [se lavej [VP tj ]]] d. Interpretation: ∃e [wash(e) & [Agent-Theme](e, Jean)] --> ∃e [wash (e) & Agent(e, Jean) & Theme(e, Jean)]

Within such an approach, one could give the following answer to question Q1 (Binding

Theory): If the reflexive pronoun is not an argument of the verb, then it is not an anaphor either and, therefore, not subject to Binding Principle A.24

However, the Bundling-approach fails to answer all other questions identified above. It cannot explain why the reflexive element gets accusative case (Q2). In fact, if the reflexive element is actually a case absorber (as proposed by Reinhart and Siloni (2005) for German sich), its presence in PoRs is unexpected. Since the passivization process eliminates accusative case, there should be no need to insert a case absorber in PoRs. Furthermore, since the account does not make any difference between types of reflexive verbs (inherently and naturally reflexive vs. naturally disjoint) and since Bundling is claimed to hold for all languages involving SE-reflexives, the questions Q3 and Q4 remain open, too.

A potential further problem concerning the Bundling approach is related to locality. The syntactic version of Bundling is not restricted to co-arguments but can relate the embedded subject and the matrix subject in ECM-constructions as in the French example in (29).25 However, we have seen examples of PoRs where the reflexive pronoun is not the argument of a verb but the possessor of a DP and it seems impossible to broaden the concept of syntactic Bundling to such cases. The relevant Icelandic example is repeated in (30).

(29) Jean se voit [laver Marie]. Jean REFL sees wash Marie

‘Jean sees himself wash Marie.’

(30) Það var haldið með sínu liði. it was held with SELF’s team ‘People supported their own team.’ Doron & Rappaport Hovav (2007) provide a collection of arguments against the Bundling 23 The authors assume that a Bundle of θ-roles is interpreted as a disjunctive conjunction of θ-roles. 24 Some authors explicitly argued that such an intransitivity approach to reflexive verbs (i.e., that reflexive verbs are unergative) would account for the binding problem that we identified for PoRs (e.g. Sells et al. 1987, Ágel 1997, Hundt 2002, Eythórsson 2008, Árnadóttir et al. 2011). The arguments that I present below are problematic for all of these proposals. 25 However, if such examples involve complex predicate formation, even syntactic Bundling might be restricted to the co-argument domain (see Reinhart & Siloni 2005:406, fn.14).

17

approach for so-called ‘syntax-languages’, and, especially, against the claim that reflexive verbs are (necessarily) intransitive in these languages. I will give here only one of their arguments, which is originally from Labelle (2008). These authors observe that the Bundling approach predicts that it should be impossible to focus only one of the two θ-roles that are involved in the Bundling process. Recall that the Bundling approach proposes that a verb’s internal θ-role (<theme>) is not assigned to the object position but is bundled together with the verb’s external θ-role (<agent>) to a complex role (<agent, theme>). This complex role is then assigned to the verb’s external argument position. As a consequence, it should be impossible to focus only one of the two θ-roles independently of the other. But as the authors observe this is empirically not correct. Both the agent and the theme can be focused independently in a reflexive context as their French example in (31) shows. The German example in (32) illustrates the same point. (31) Jean-Pierre s’est dénoncé lui-même. Jean-Pierre REFL is denounced himself (i) ‘Jean-Pierre denounced himself, it was not others who denounced him.’ (ii) ‘Jean-Pierre denounced himself, he did not denounce others.’ (32) Morgens wäscht sie sich immer/erst mal selber. at.morning washes she REFL always/first-of-all self (i) agent focus: She washes herself, no-one else washes her. (context: She is a disabled patient.) (ii) theme focus: She washes herself, she washes no-one else. (context: She is a nurse.) Doron & Rappaport Hovav (2007) conclude that reflexive verbs as in French and German have a transitive derivation where the reflexive clitic/pronoun bears the internal θ-role, i.e. it is acts as an anaphor in argument position.26 However, while Doron & Rappaport Hovav argue against ‘syntactic Bundling’ and propose to derive its effects by ordinary anaphoric binding in syntax, they nevertheless propose that some reflexive verbs can be intransitive. Specifically, they assume that there exists, in addition to syntactic anaphoric binding, the lexical operation “Reflexivization” (and “Reciprocalization”) which identifies a verb’s internal θ-role with its external θ-role. The result is an intransitive lexical entry, which projects one complex θ-role. When this intransitive entry is projected in syntax (as a subject), a reflexive pronoun is added as a lexical marker of reflexivization. That is, these authors explicitly assume that in all languages with SE-reflexive there exists a syncretism between the anaphor and the marker of reflexivization (see Kiparsky 2002 for the same view). Furthermore, these two ways to produce a reflexive interpretation, syntactic binding and lexical reflexivization, are correlated with reflexive verb classes. The reflexive use of naturally disjoint verbs is derived in the syntax by anaphoric binding of the reflexive object. Inherently reflexive verbs, on the other hand, are always derived by lexical reflexivization. Naturally reflexive verbs, finally, have both a syntactic and a lexical derivation. While Q2 (Case Theory) and Q4 (Language Restriction) remain open, this proposal seems to answer Q2 (Binding Theory) and Q3 (Verb Class Restriction): PoRs are possible exactly if a reflexive verb is built with a marker of lexical reflexivization (which

26 See also Alencar & Kelling (2005) and Labelle (2008). Such a conclusion makes it necessary to find an alternative explanation for as to why reflexive verbs behave like intransitives with respect to a number of tests (Kayne 1975, Grimshaw 1981). See Alencar & Kelling (2005), Doron & Rappaport Hovav (2007) or Labelle (2008) for different suggestions.

18

is not subject to Principle A) but not, if it involves an anaphor (which is subject to Principle A). However, this proposal is also insufficient to derive the thematic properties of PoRs. First of all, if PoRs could only be derived on top of lexical reflexivization, it should be impossible to focus the two θ-roles independently in PoRs. But this prediction is not borne out. As in active reflexive sentences, the agent role and the theme role in PoRs can be focused independently via the addition of an intensifier.27 This shows that the reflexive pronoun bears a θ-role in PoRs and, therefore, is an anaphor in argument position. (33) Morgens wird sich immer/erst mal selber gewaschen. at.morning becomes REFL always/first-of-all self washed ‘In the morning (first) people (always) wash themselves.’ (√ agent focus, √ theme focus) Furthermore, Doron & Rappaport Hovav (2007) take lexical Reflexivization to semantically identify the verb’s external with its internal θ-role. But, as we have seen above in (21a/30), the domain of natural reflexivity exceeds the verbal coargument domain in that it can involve possessors. Since lexical processes cannot build any relation between a subject and a possessor, lexical reflexivization cannot be involved in such examples. I conclude, therefore, that the reflexive element in PoRs with naturally reflexive predicates is an anaphor with its own θ-role. For inherently reflexive verbs, I will also assume that they take as their internal argument an anaphor subject to Principle A (see also Jónsson 2011 for this claim within Icelandic). This is not the standard assumption. As mentioned above, many authors analyze inherently reflexive verbs as necessarily derived in the lexicon. The reasoning is typically the following: (i) Since the reflexive element in inherently reflexive verbs cannot be replaced by a referential DP, it cannot have a θ-role. (ii) All arguments must have a θ-role. (iii) Therefore the reflexive element cannot be an argument and, in turn, it must be a marker of lexical reflexivization. I think, all three assumptions can be challenged at least to some extent. Starting with point (iii), it is a fair question why, in language after language, markers of reflexivization and anaphors should show a syncretism. Furthermore, English has a number of inherently reflexive verbs formed with the heavy reflexive pronoun ‘himself’ as in (34) (e.g. Brame 1983). To my knowledge, no one has ever analyzed these heavy reflexive pronouns as markers of reflexivization;28 instead, these verbs seem to select a reflexive pronoun as their object. (34) a. John prided himself/*Mary on his past accomplishment. b. Sue availed herself/*me of the opportunity. Turning to point (i), German (like other Germanic languages) has a number of inherently reflexive constructions where the reflexive is embedded in and clearly θ-marked by a non-subcategorized PP. An example is the semi-idiomatic string “etwas von sich geben” (literally: ‘to give something from oneself’ meaning ‘to utter something’).29

27 See Hole (2006, 2008) on focusing the implicit argument of passives via an intensifier like German ‘selber’ (self). I should mention that I have not been able to reproduce this ambiguity with Icelandic speakers. I suggest that this is related to the fact that Icelandic has two reflexive pronouns, a simple and a complex one and that focusing the simple one is somehow blocked by the existence of the complex one. 28 In fact, the marker of reflexivity in English is typically assumed to be zero, deriving for example one-place grooming verbs as in ‘John washed’ or ‘John shaved’; see (27) above. 29 Two further examples involving prepositions are given in (i) and (ii). Here a preposition assigns a ‘locational goal’ role to the reflexive pronoun. Note that these expressions passivize very productively. The string “von sich

19

(35) Hans hat auf der Konferenz viel Unsinn von sich/ *von Maria gegeben. John has at the conference much nonsense from REFL/*from Mary given ‘John uttered lots of nonsense at the conference.’ Although the reflexive pronoun in the PP cannot be replaced by a referential expression, there is no doubt that the preposition ‘von’ (from) assigns the θ-role source to it. Since it is not the verb ‘geben’ (give) which subcategorizes for a source argument but the preposition, this example illustrates that even inherent reflexivity can exceed a verb’s co-argument domain. Note in this connection that although this example has a semi-idiomatic interpretation, its inherent reflexivity is not idiomatic at all. It is a lexical-conceptual necessity that the agent of a giving-event is identical with its source. Expressing this trivial aspect overtly with the help of a preposition opens the way for the semi-idiomatic meaning.30 It seems then as if there were two cases of inherent reflexivity, one where we can clearly identify a θ-role on the reflexive pronoun (35), and one where it seems that we cannot identify such a role (34 and its counterparts with SE-reflexives in Dutch, German or Icelandic). But even if the reflexive pronoun in the latter case lacks a θ-role this does not necessarily mean that it could not be a syntactic argument (cf. (ii)). As mentioned, it seems inevitable to analyze the SELF-reflexives in the English examples in (34) as complements, i.e. internal arguments in a syntactic sense. If these arguments lack θ-roles, then the reasoning in (ii), which is driven by the classical θ-criterion, cannot be upheld in generality. Once we accept this for SELF-reflexives, there is no a priori reason why SE-reflexives could not be analyzed the same way.31 An alternative way to react to (i) and (ii) would be to assume that inherent reflexive verbs, nevertheless, assign a θ-role to their internal argument and that the fact that this internal argument must be an anaphor is an idiomatic selectional property of these verbs which is not related to θ-theoretical considerations. The question why we lack intuitions about this internal θ-role (we cannot identify whether it is a theme, an undergoer, an experiencer or something else) might be related to the obligatory reflexivity of the θ-assigning predicate. Perhaps we can identify a θ-role only if the θ-assigner can assign this role to various alternative referential entities. That is, the θ-role of the reflexive pronoun in (35) is identified as source because, in other contexts, the θ-assigning P assigns this very same role to a referential argument. If, however, the reflexive pronoun is the internal argument of an inherent reflexive verb we might lack intuitions about this θ-role because it is never assigned to a referential entity. Whether this is on the right track or not, note that the concept of lexical reflexivization as defined by Reinhart & Siloni and Doron & Rappaort Hovav (Bundling of a verb’s internal

gegeben wurde” from the main text brings about more than 100 hits on Google. In (iii) we see an idiomatic inherent reflexive construction involving a resultative ECM-constellation. (i) etwas auf sich nehmen (ii) etwas zu sich nehmen something on REFL take something to REFL take ‘to take something upon oneself’ ‘to ingest something’ (iii) sich vor Lachen in die Ecke schmeissen REFL for laughing into the corner throw ‘to laugh one’s head off’ 30 Therefore such examples are not ‘idioms’ but ‘idiomatically combining expressions’ in the sense of Nunberg et al. (1994). While idioms are not interpreted compositionally, the interpretation of an idiomatically combining expression is derived compositionally from the meanings of its parts. Its constituents (in our case most importantly the PP) carry identifiable parts of the overall interpretation. It is just that such expressions have a figurative use. 31 In a similar vein, Steinbach (2002) and Schäfer (2008) argue that German anticausatives and middles which also come obligatorily with a SE-reflexive behave syntactically as transitive constructions although they involve only one θ-role, i.e. they analyze the reflexive pronoun in these constructions as a non-thematic argument.

20

with its external θ-role) also requires that inherent reflexive verbs involve an internal θ-role. While in these accounts this θ-role gets bundled in the lexicon with the external θ-role, the alternative sketched here would associate this θ-role with the reflexive object. It seems to me that the two alternatives cannot be distinguished on interpretative grounds.32 6. An analysis of PoRs 6.1 The distribution of PoRs As said, I assume that reflexive elements like German sich, Icelandic sig, Norwegian seg or Dutch zich (SE-reflexives in the terminology of Reinhart & Reuland 1993) are always anaphors.33 Being anaphors, these elements are subject to (some version of) Binding Principle A (Chomsky 1981), no matter whether they occur in an inherently reflexive, naturally reflexive or naturally disjoint context. In the last section, we saw that inherent and natural reflexivity sometimes exceed a verb’s co-argument domain. I assume, therefore, that inherent and natural reflexivity and their counterpart, natural disjointness, are not lexically specified but are post-syntactic phenomena which are determined at the Conceptual-Intentional interface. They are computed on the basis of conceptual information about the material inside vP/VoiceP, i.e. the meaning of the verb, its arguments as well as its adjuncts/modifiers (see the discussion in the last section; see also fn. 7 and fn. 60 in appendix 1).

Above, I claimed that the anaphor in PoRs lacks a syntactic antecedent. If the implicit external argument of the passive could act as syntactic antecedent, we could not explain why naturally disjoint verbs are typically excluded from PoRs. Furthermore, the agreement facts in (25b) would be problematic. As a consequence, this means that the traditional version of Principle A of the Binding Theory is not fulfilled in PoRs. To explain why PoRs are, nevertheless, available in some languages under some contexts, I make the following proposal: (36) Some languages can formally repair a violation of Principle A. However, the formally

repaired structure is acceptable (i.e., easily interpretable) only if the underlying predicate/event is conceptualized as inherently or naturally reflexive.

The following picture illustrates this idea. In most languages (e.g. Dutch or Norwegian), PoRs are formally ungrammatical because they violate Principle A of the Binding Theory (dark area). Some languages have a mechanism that allows repairing the Principle A violation that goes along with PoRs (grey area). However, only a subset of these formally rescued PoRs is judged as acceptable (white area). This is so because the repair mechanism is purely formal and does not, by itself, lead to a successful interpretation of the reflexive pronoun. A successful interpretation is only possible if conceptual knowledge supports it. This is the case in inherently and naturally reflexive contexts, which come with the conceptual expectation that the reflexive pronoun and the (implicit) external argument have the same semantic value. I will develop this proposal in the following subsections.

32 The focus test in (31)-(33) can, of course, not be applied to inherently reflexive constructions because inherently reflexive constructions do not allow for alternatives concerning the internal θ-role. 33 I leave aside here the role of SE-reflexives in anticausatives and generic middles. See section 9 and the references there for some discussion.

21

The distribution of PoRs (formally) ungrammatical (formally) grammatical (semantically) acceptable 6.2 Basic assumptions on anaphoric binding and structural case In this section, I provide some background assumptions concerning Binding Theory and Case Theory (see also Schäfer (2008, to appear)). I will only discuss local binding of SE-reflexives and their pronominal 1st and 2nd person counterparts. I will not be concerned with long-distance binding, the licensing of SELF-reflexives or Principle B effects. Note that I will most of the time leave aside aspects of phase theory (Chomsky 2001). I do so for reasons of space and simplicity of presentation, as the system developed below leads to nontrivial questions about the timing of the operation TRANSFER and which details of the syntactic derivation are available for evaluation at the interfaces (but see below for some comments). For further discussion of how the present system can be combined with phase theory, I refer the interested reader to Schäfer (2008, to appear).

I follow earlier proposals in assuming that anaphoric binding is grounded in a syntactic AGREE-relation (Chomsky 2000, 2001) between a DP-antecedent and an anaphoric variable (e.g. Fanselow 1991, Burzio 1991, 1998, Reuland 2001, 2005, Fischer 2004, 2006, Heinat 2006, Chomsky 2008, Kratzer 2009, Tucker 2010). I assume that an anaphoric variable is totally underspecified for φ-features: it is a set of a categorial D-feature and unvalued φ-features {D, uφ} (cf. Burzio 1991, Kratzer 2009, Tucker 2010 for similar proposals). Therefore, it is referentially defective. Furthermore, such a variable needs an antecedent to value its φ-features under syntactic AGREE. This AGREE-relation gets evaluated at the interfaces to compute the morphological form and the semantic value of the variable.

Since the variable has unvalued φ-features it is active, thereby qualifying as a probe (Chomsky 2000, 2001). The antecedent is a DP with valued φ-features and, therefore, it can act as a goal for φ-agreement. Specifically, I propose that the variable probes the tree upwards to get its features valued by a c-commanding antecedent. I will assume the option of upward-probing without further defending it here; for some recent literature on upward-probing (or “reverse probing”) see Baker (2008), Wurmbrand (to appear), Bjorkman (2011) or Zeijlstra (in press).

If AGREE between the variable and a c-commanding antecedent has taken place, the structure is sent to the interfaces for interpretation.34 If AGREE does not take place, the features of the variable remain unvalued and, as a consequence, the derivation will crash. Since syntactic AGREE is local (i.e., it has to take place within a phase), this setup has an 34 That is, I assume that the AGREE-relation between an anaphor and its antecedent is not mediated by a functional verbal head as, e.g. in Kratzer (2009). I do this mainly for reasons of simplicity, though see Tucker (2010) for some argumentation against mediation by functional heads.

PoRs in all languages (all types of verbs)

PoRs in German/Icelandic (all types of verbs)

PoRs with inherently and naturally reflexive verbs

Dutch

Norwegian

22

effect similar to Principle A of the standard Binding Theory (Chomsky 1981): An anaphoric variable needs a local, c-commanding antecedent. However, rephrasing Principle A as an AGREE-operation will allow me to implement a repair strategy for Principle A violations taking place in PoRs (see below).

At LF, the syntactic AGREE-relation between the variable and the antecedent is evaluated semantically as a binding relation (as expressed via coindexation in earlier stages of the theory; see Reuland 2001, 2005 for discussion). At PF, the AGREE-relation is evaluated morpho-phonologically and the Spell Out of the variable is determined. The specific Spell Out, either as a SE-reflexive or as a (locally bound) pronoun, depends on the φ-features of the antecedent and language-dependent morpho-phonological economy conditions (see Halle & Marantz 1993, Burzio 1998, Heinat 2006 for discussion). In most Germanic languages the situation is as follows: If the variable is valued by a 3rd person antecedent, it gets spelled out as a SE-reflexive. If it is valued by a 1st or 2nd person antecedent, it is spelled out as a 1st or 2nd person object pronoun.35 However, some languages use the SE-reflexive also for other than just 3rd person antecedents (e.g. Polish uses it for 1st, 2nd, and 3rd person antecedents) while other languages lack a SE-reflexive and always use a pronoun overtly matching the antecedent in φ-features (e.g. Frisian).

Full-fledged referential pronouns, on the other hand, are the combination of a D-feature and a set of valued φ-features {D, φ}. Therefore, they refer independently and they always spell out their inherent φ-feature.

Note that I must assume that both interfaces can differentiate between a pronominal element that started the derivation with all ϕ-features valued (a referential pronoun) and a pronominal element that started the derivation with unvalued ϕ-features and gets these features valued only during the derivation (an anaphor). PF needs to know this in order to derive the correct spell out as a SE-reflexive or a third person pronoun. The CI-interface needs to know which element valued the ϕ-features of the anaphor in order to compute the correct semantic binding relation (see Schäfer (to appear) for some further discussion).

Concerning Case Theory, I make two basic assumptions. First, I will assume (a specific variant of) a dependent case approach (cf. Yip et al. 1987, Woolford 1997, Haider 2000, Marantz 2000, Sigurðsson 2000 et seq., among others). Despite differences in execution, dependent case approaches derive Burzio’s Generalization (BG) by the assumption that there is a dependency relation between nominative and accusative assignment in that the latter case is available in a clause only if the former has already been assigned. While such a derivation of BG is a welcome result, PoRs seem to contradict it. To derive both, the standard cases of BG as well as accusative as the only structural case in PoRs, an update in the precise formulation of the dependency relation will be necessary.

Second, I follow recent proposals that morphological case is mainly a PF phenomenon (Marantz (2000), McFadden (2004), Sigurðsson (2000, 2003, 2005, 2006, 2009) among others). However, in accordance with these authors, I assume that PF determines morphological case on the basis of syntactic information. Specifically, I propose that PF evaluates the syntactic AGREE-relation involving T(ense) to determine structural case (this

35 This proposal, therefore, has to assume that 1st and 2nd person pronouns are ambiguous between referential pronouns and anaphors. If they are (locally) free as in (i), they are pronouns which start the derivation with a valued set of φ-features (see below), when they are locally bound as in (ii), they are variables that start the derivation with unvalued φ-features which receive their values from the local antecedent (cf. ‘fake indexicals’ in Katzer 2009). It should be mentioned that such an ambiguity has been rejected e.g. by Fanselow (1989, 1991) or Burzio (1998). Since this paper is mainly about SE-reflexives, I will not further go into this point (but see section 9 for some comment). (i) Eri sieht michk (ii) Ichi sehe michi He sees me I see me ‘He sees me.’ ‘I see myself.’

23

assumption will be slightly modified in section 7). As is the standard assumption, T enters the derivation with unvalued φ-features and probes its c-command domain for the closest valued DP to agree with. At PF, this AGREE-relation is evaluated to determine dependent case (accusative) and default case (nominative). Specifically, I propose the following three case-principles to be at work:

(37) a. Dependent case (ACC) (first version): A DP is realized at PF with dependent case if

a different DP has valued local T via AGREE.36 b. Default case (NOM): A DP which is not realized with dependent case, appears with

default case. c. Inherent/lexical case takes precedence over default and dependent case. To illustrate the above assumptions about Binding Theory and Case Theory, consider the derivation of the example in (38) involving a transitive verb with a subject-bound anaphor in object position.37 (38) dass Hans sich mag.

that John.NOM REFL.ACC likes ‘that John likes himself.’

(39) TP 3

T vP {uP, uN, uG} 3 Hans{P, N, G} v’ 3 v VP 3 sich{uP, uN, uG} V

The variable in object position enters the derivation with unvalued ϕ-features. The external argument is merged with a full set of valued φ-features. T enters the derivation with unvalued φ-features. Two AGREE-processes take place (indicated by the two arrows). One is initiated by the variable in object position. As said, I assume that variables probe the tree upwards; thereby the variable agrees with and gets valued by the subject. The second AGREE-process is initiated by T, which probes its c-command domain. The closest element with valued φ-features is the external argument in Spec,vP which therefore agrees with and values T (i.e. this is the standard Chomskian AGREE process where the probe c-commands the goal). I assume that German does not have obligatory movement to Spec,TP, but nothing hinges on this. Note that I dispense with an activity condition for the goal, here the subject DP. Since I assume that Case is not a syntactic feature, the standard assumption that unvalued Case activates the goal, is not applicable. The overall consequences of this move are beyond the

36 In Schäfer (2008, to appear), I argue that this combination of an AGREE-based with a dependent case approach is necessary to derive the ergative-like case-pattern of reflexive anticausatives and reflexive middles in Germanic and Romance languages (see also section 9). Specifically, I provide arguments that reflexive anticausatives/middles involve a nominative theme in object position and an accusative reflexive pronoun in Spec,vP. Since the reflexive has unvalued φ-features, the theme values T (and, indirectly, the reflexive). Baker & Vinokurova (2010) also propose the combination of an AGREE-based with a dependent-case approach. 37 Recall that I leave aspects of Phase Theory aside. The version of the present theory developed in the next section is compatible with the standard assumption that VoiceP/vP is a phase.

24

scope of this paper. These two AGREE-relations will be evaluated at the interfaces. At PF, the internal argument is marked with dependent ACC because there is a different DP (the subject) within the same clause which has valued the features on T via AGREE (cf. 37a). The subject itself, on the other hand, gets default case as nothing more specific is said about its case (cf. 37b). Finally, the variable is spelled out according to the φ-features of its antecedent. Since the antecedent is 3rd person, the variable gets realized as a SE-reflexive. At LF, finally, the AGREE-relation between the subject and the variable is interpreted as semantic binding. With this background, we can finally turn to a more technical discussion of PoRs. 6.3 Deriving PoRs Consider the German PoR in (40). The tree in (41) illustrates how the system sketched in the previous subsection handles this and similar examples. The following derivational steps take place: Both the variable and T enter the derivation with unvalued φ-features. T searches its c-command domain and the variable searches the tree upwards (as indicated by the dotted arrows). Since PoRs are impersonal constructions (but see section 8), there is no DP available that could value either T or the variable. But note that ‘AGREE’ and ‘valuation’ are different processes if we assume with Frampton & Gutmann (2000) or Pesetsky & Torrego (2007) that AGREE involves just feature sharing. Therefore, T and the variable agree with each other in (41), but since none of the two has valued features, no valuation can take place. (40) als sich gewaschen wurde. when REFL washed became ‘when people washed (themselves).’ (41) TP 3 T vPpassive {uP, uN, uG} ! v’ 3 vpassive VP 3 DP{uP, uN, uG} V (41) involves an unvalued two-member agreement chain < T – variable >. Therefore, the derivation should crash at the interfaces. This looks like the correct result for Dutch or Norwegian, but it isn’t the correct prediction for German and Icelandic. Note, however, that German and Icelandic (but also Dutch and Norwegian) have a construction different from PoRs which provides a very similar problem, namely the ordinary impersonal passive (or other impersonal constructions such as quirky intransitives in Icelandic) exemplified in (42). As these examples show, German and Icelandic do not allow the insertion of an expletive in the context of impersonal passives suggesting that these are really impersonal even in a formal sense. (An expletive can, of course, be base-generated in sentence inital position (Spec,CP) to fulfill the verb-second property.) (42) a. weil (*es) hier gestern lange getanzt wurde. (German) because expl here yesterday long danced was ‘Yesterday people danced for a long time.’

25

b. ĺ dag hefur (*það) verið dansað. (Icelandic) today has expl been danced ‘Today, people danced.’ The derivation of simple impersonal passives is sketched in (43). This time, only one element with unvalued ϕ-features is present, namely T. T probes its c-command domain but cannot find any other element that could value it. The derivation involves an unvalued one-member chain < T > that should lead to a crash at the PF-interface, contrary to fact. (43) TP 3 T{uP, uN, uG} vPpassive ! v’ 3 vpassive VP ! ? V Languages deal with impersonal passives in different ways. Some languages insert a nominal expletive with φ-features that can value T and check the EPP on T. In such languages, the problem depicted in (43) does not arise. Holmberg (2002) shows this to be the case for some Norwegian dialects where the expletive has a pronominal origin (i.e., it carries inherent ϕ-features). But other languages lack expletives with nominal features and the problem depicted in (43) holds. The languages discussed in this paper (German Icelandic, Dutch, Norwegian ) are of this latter type.38 I assume with Ruys (2010) that in the latter type of languages Default Agreement (DA) values T and saves the derivation of impersonal passives. In a minimalist framework which involves the syntactic operation AGREE and the concept of Full Interpretation (i.e., the idea that derivations crash at the interfaces if unvalued (and/or uninterpretable) features remain unchecked), default agreement must be a process taking place in core syntax, i.e. before TRANSFER to the interfaces. I assume the following conception of Default Agreement (taken from Ruys 2010): (44) Default agreement (DA): In the absence of any appropriate nominal category, the ϕ-features on an unvalued probe undergo default valuation [3rd person, singular]. DA on T is a way to rescue impersonal constructions such as impersonal passives (but see the next section for an update and especially fn. 48 for a important restriction on DA). It is a last resort operation, i.e., a repair strategy that avoids that a derivation crashes for purely formal reasons at the interfaces. Furthermore, I assume that DA is a costly operation and, therefore, it is not available in all languages. Languages lacking DA on T (as well as a nominal expletive) do not allow the formation of impersonal constructions. English might be a case in point (cf. Ruys 2010). With this conception of DA at hand, I return to PoRs and provide answers to the questions Q1-Q4 formulated at the beginning of section 5.

38 Holmberg (2002) shows that standard Norwegian uses an expletive in impersonal passives which has a locative origin and lacks nominal ϕ-features. The Dutch expletive er used in impersonal passives is of the same kind (see Ruys 2010). Such expletives can check the EPP on T but cannot value T.

26

Recall that I concluded that the anaphor in PoRs lacks a syntactic antecedent. Question Q1 asks, therefore, why PoRs are not filtered out as a Principle A violation. I proposed that anaphoric variables (often spelled out as SE-reflexives) are similar to T in that both enter the derivation with unvalued ϕ-features. In the case of ordinary binding, the unvalued features on the variable get valued by a DP-antecedent. In PoRs, T and the variable form an agreement chain <T – variable> which, however, remains unvalued (see the derivation in (41)). I argue that PoRs do not crash at the interfaces if this agreement chain <T – variable> can be valued by D(efault) A(greement). That is, DA formally precludes a violation of Principle A in PoRs: the variable gets its unvalued features valued although no c-commanding DP-antecedent is available. DA can, however, not provide any help concerning the semantic interpretation of the variable in the absence of a DP-antecedent. The question of how to interpret the variable in PoRs will be discussed below.

The idea that DA is a costly operation can provide an answer to question Q4. Recall that PoRs are available only in few languages and, more specifically, only in a subset of the languages with impersonal passives. Following Ruys (2010), I proposed that impersonal passives are rescued if the one member agreement chain <T> is valued via DA. In order to rescue PoRs, I proposed that the two-member agreement chain <T – variable> needs to be valued via DA. Arguably then, the DA-operation needed in simple impersonal passives is less complex than the DA-operation needed in PoRs. I propose, therefore, that languages differ in the complexity of the DA-operations they make available. Some languages lack DA. Other languages like Dutch or Norwegian make DA available only for heads of the extended verbal domain such as T. German and Icelandic, in addition, make DA available for non-homogeneous agreement-chains involving a verbal head (T) and an anaphoric variable. Next, I turn to question Q3 about the verb class restriction. Above, I argued that DA formally avoids that PoRs crash at the interfaces. The variable (as well as T) gets its features valued as if it had a DP-antecedent. But crucially, no DP-antecedent is available that could semantically bind the variable at LF, i.e. the derivation provides no clue how to interpret the variable. At this point, I argue that conceptual knowledge about the verbal event expressed by the PoR becomes crucial. The output of the syntactic derivation involving DA on a variable is comprehensible at the CI-interface only if conceptual knowledge provides some information about how to interpret the variable. With inherently reflexive verbs, it is conceptually clear that the anaphor has to depend semantically on the referent acting as external argument.39 Naturally reflexive events carry the strong conceptual expectation that the argument position realized by the semantically unbound reflexive pronoun is covalued with the external argument of the same event (see the discussion around example (8)). That is, in inherently and naturally reflexive contexts, the semantic coindexation that is usually computed from syntactic binding/AGREE between a DP-antecedent and a variable can be reconstructed on conceptual grounds beyond any doubt. Naturally disjoint verbs or contexts do not provide any such conceptual clue about how to interpret the variable, and this is the reason, I argue, why they typically do not enter the formation of PoRs.40 This leaves question Q2 about accusative case to be accounted for. Note that there is no reason to assume that the accusative case on the reflexive pronoun in PoRs is in any way

39 This argumentation assumes that inherently reflexive constructions involve a θ-role on the reflexive pronoun. While I argued in section 5 that this is the case in some cases (e.g. (35)), I left it open for other cases of inherent reflexivity (e.g. (34)). If examples of the latter type lack a θ-role on the reflexive pronoun, then all that is needed to derive PoRs from them is the availability of DA on the variable. 40 Note that I assume that PoRs involving naturally disjoint verbs are formally grammatical in German and Icelandic because DA values the variable. They are however, unacceptable, due to the interpretative problems described in the text. I predict, however, that PoRs involving naturally disjoint verbs become more acceptable, if the bigger context (i.e. the context beyond the bare passive vP) provides some expectation about how to interpret the variable (see. fn. 7 and fn. 60 in appendix 1 for some exemplification).

27

different than ordinary structural accusative case. Furthermore, there is no reason to assume that the case system active in PoRs is different from the general case system of the languages under consideration. That is, it does not make sense to stipulate that only PoRs but not other passives can assign structural accusative case. The availability of structural accusative in PoRs should, therefore, follow from an interplay between the general theory of structural accusative case and some property present only in PoRs but not in ordinary passives. Arguably, this property must somehow be related to the presence of the anaphor (a SE-reflexive). I argue, therefore, that the availability of accusative case in PoRs depends on a formal property of anaphors (SE-reflexives), namely their φ-feature deficiency. In the previous section, I formulated the post-syntactic algorithm for object case (ACC) repeated in (45) that crucially builds on a syntactic AGREE-relation between T and a DP. In order to derive accusative on the reflexive pronoun in PoRs, we need to update this rule for dependent case slightly as in (46) so that any kind of AGREE-valuation, either AGREE with a DP or Default Agreement, is relevant for the determination of dependent case:41 (45) Dependent case (ACC) (old version):

A DP is realized at PF with dependent case if a different DP has valued T via AGREE. (46) Dependent case (ACC) (updated version):

A DP is realized at PF with dependent case if something else (either a different DP or Default Agreement) has valued T via (default) AGREE.

In the next section, I will turn to a discussion of the relation between PoRs and the Icelandic New Passive; as mentioned earlier, this latter construction shows unexpected accusatives not only on reflexive pronouns in object position but also on referential object DPs. 7. The relation between PoRs and the New Passive In this section, I will discuss the relation between PoRs and the New Passive in Icelandic. In section 3, I had mentioned that Icelandic PoRs are not genuine instances of the New Passive because the latter is accepted only by a subset of the speakers who accept PoRs. However, some authors suggested that PoRs constituted a first step towards the development of the New Passive (e.g. Maling & Sigurjónsdóttir 2002, Árnadóttir et al. 2011).42 Below, I will update my analysis of PoRs in order to make a proposal about how this development might have happened. Specifically, I will argue that speakers of the New Passive variant have grammaticalized the output of the repair mechanism of Default Agreement, which I argued is at stake in PoRs. As a starting point, I will discuss in some more detail the differences between the canonical passive and the New Passive in Icelandic. For reasons of space, I will concentrate on passives of canonical transitive verbs (NOM-ACC verbs). As far as I can see, verbs with lexically case marked objects can easily be subsumed under the present proposal (see Sigurðsson 2011 for a more detailed discussion). The canonical passive in Icelandic is relatively straightforward (e.g. Zaenen, Maling & Thráinsson 1985, Sigurðsson 1989, 2011, Thráinsson 2007, Eyþórsson 2008). The external

41 Both versions of the dependent case approach free Burzio’s Generalization from any relation to θ-roles (see e.g. Sigurðsson (2006) and already Abraham (1986) for motivation). Both versions predict, furthermore, well-restricted, but slightly different, sets of counterexamples to the original version of Burzio’s Generalization. 42 Some authors hinted at the relevance of other phenomena for the development of the New Passive, such as cases of syncretisms between NOM and ACC, exceptions to the definiteness effect in canonical passives with postverbal internal arguments (Árnadóttir et al. 2011) or the frequent use of impersonal constructions per se as well as the existence of impersonal prepositional passives as in (48c) (Barðdal & Molnár 2006, Sigurðsson 2011).

28

argument of the corresponding active (47a) is demoted and the accusative object shifts to nominative case, thereby agreeing with the auxiliary ‘vera’ (be) (in number and person) and the past participle (in number and gender). Furthermore, this nominative DP usually undergoes DP-movement to the first position of the clause, as in (47b).43 However, there exist also passives of transitive verbs (47c) where the first position is filled by the expletive ‘það’ (there, it) (or by any other element suiting for the sentence initial position, e.g. adverbs) and the internal argument stays in its base position inside the vP.44 Even if the object DP stays inside vP, it still receives nominative case and triggers agreement on the auxiliary and the participle. However, in order to stay inside vP, the object must be indefinite (Sigurðsson 2000, 2011, Thráinsson 2007 and Eyþórsson 2008). Due to this definiteness effect, the example in (47d) is ungrammatical. ((47a-d) are taken from Eyþórsson 2008). (47) a. Einhver lamdi stúlkuna. (Active) someone beat girl.the.ACC ‘Someone beat the girl.’ b. Stúlkan var lamin (Canonical Passive) girl.the.NOM was beaten.FEM.SG ‘The girl was beaten.’ c. Það var lamin stúlka. (Impersonal Canonical Passive) it was beaten.FEM.SG girl.NOM ‘A girl was beaten.’ d. *Það var lamin stúlkan. it was beaten.FEM.SG girl.the.NOM ‘The girl was beaten.’ As discussed earlier, Icelandic also allows impersonal passives of unergative verbs. In this case, the expletive (or an adverb) fills the sentence initial position and default morphology shows up on the auxiliary (3rd person, singular) and the participle (neuter singular). ((48a) is from Eyþórsson 2008, (48b-c) are from Sigurðsson 2011). (48) a. Það var dansað alla nóttina. it was danced.NEU.SG all night.the ‘People danced all night.’ b. Hér er verið að vinna. here is been.NEU.SG to work ‘People are working here.’ c. Þess vegna var ekki talað við þá. that for was not talked.NEU.SG with them ‘Therefore, people did not talk to them.’ The New Passive is also built with the auxiliary ‘vera’ and the verb’s past participle and the external argument is not overtly projected. However, three properties set this construction apart from the canonical passive: (i) there is no ACC-to-NOM conversion and, in turn, the auxiliary and the participle show up with default morphology (see (49), where the percent sign indicates that only some speakers accept the construction). (ii) NP-movement of the internal argument out of the vP is impossible and in turn the expletive ‘það’ (or an adverbial) is inserted in the clause initial position, see (49 vs. 50). (iii) As first discussed by Maling &

43 This is typically analyzed as cyclic movement to Spec,CP via Spec,TP. 44 The indefinite object can also undergo short movement to the left of the participle, but this is not important for our purposes here (see Sigurðsson 2011:fn. 3 for discussion).

29

Sigurjónsdóttir (2002), there is no definiteness effect, see (49b). (The examples in (49) and (50) are taken from from Eyþórsson 2008. (50a, b) were slightly modified; p.c. Thórhallur Eythórsson). (49) a. %Það var lamið stúlku. (New Passive) it was beaten.N.SG girl.ACC ‘A girl was beaten.’ b. %Það var lamið stúlkuna. (New Passive) it was beaten.N.SG girl.the.ACC ‘The girl was beaten.’ (50) a. *Það var stúlku lamið. (New Passive) it was girl.ACC beaten.N.SG ‘A girl was beaten.’ b. *Það var stúlkuna lamið. (New Passive) it was girl.the.ACC beaten.N.SG ‘The girl was beaten.’ All three properties are prototypically associated with the internal argument of transitive verbs, but here we find them with a passive (see fn. 13). I will mainly concentrate on the case problem of the New Passive here as a full discussion of the ban on A-movement and the absence of the definiteness effect are beyond the scope of this paper. However, I will also provide some informal suggestions about the latter two properties below. As Sigurðsson (2011) stresses, accusative case preservation takes place only in the context of the New Passive, i.e. such violations of Burzio’s Generalization are not a general property of speakers of the New Passive variety. Similarly, the absence of the definiteness effect is restricted to the context of the New Passive and does not show up with other structures lacking an external argument (e.g. canonical passives, unaccusatives). Any analysis of the New Passive must explain why these three properties are strictly restricted to this specific construction. For this reason, I will update the case theory developed in the last section. Specifically, I will assume that the determination of case morphology is based on syntactic information computed inside of the vP/VoiceP (see Sigurðsson (2000), (2003), (2009) and Schäfer (2008, to appear) for a general motivation of this move). The role that T(ense) played in the earlier version above is now taken over by little v (also called Voice). This means that v is equipped with a set of unvalued φ-features which must be valued by the closest DP. In order to derive that the subject, which is base generated in Spec,vP, will be realized with nominative case, I assume that v searches its m-command domain for the closest DP and that the DP in the specifier of vP is closer to v than a DP embedded in the complement of v.45/46 In

45 A definition of AGREE that meets our requirements has been formulated by Müller (2009): AGREE: α agrees with β with respect to a feature bundle Γ iff (a), (b), and (c) hold: a. α bears a probe feature [*F*] in Γ, β bears a matching goal feature [F] in Γ (Starring a feature indicates its probe status; see Sternefeld 2006). b. α m-commands β. c. There is no δ such that (i) and (ii) hold: (i) δ is closer to α than β. (ii) δ bears a feature [F] that has not yet participated in AGREE. d. δ is closer to α than β if the path from δ to α is shorter than the path from β to α. The path from X to Y is the set of categories Z such that (i) and (ii) hold: (i) Z is reflexively dominated by the minimal XP that dominates both X and Y. (ii) Z dominates X or Y.

30

transitive vPs, v will therefore agree with the external argument as illustrated in (51), in canonical unaccusative and passive structures illustrated in (52), v will agree with the internal argument. (51) a. dass Hans Maria kämmt. that John.NOM Mary.ACC combs

‘that John combs Mary.’ b. TP 3

T vP 3 Hans{P, N, G} v’ 3 v{uP, uN, uG} VP 3 Maria{P, N, G} V

(52) a. dass Maria gekämmt wird. that Mary.NOM combed becomes ‘that Mary is (getting) combed.’

b. TP 3 T vPpassive ! v’ 3 vpass{ uP, uN, uG } VP 3 Maria{P,N, G} V To derive from these derivations morphological case, I update the principle determining dependent case as below: (53) Dependent case (ACC) (final version): A DP is realized at PF with dependent case if it

is not involved in the valuation of local v/Voice via AGREE. (This is the case if Voice/v gets valued by a different DP, by Default Agreement or if Voice/v is lexically equipped with a set of valued ϕ-features (for the latter, see below)).

If nominative case is generally negotiated inside the vP and not via T, something in addition has to be said about DP-movement to Spec,TP as well as about verbal agreement which is typically assumed to be located in T. Concerning A-movement to Spec,TP, it has been argued

The length of a path is determined by its cardinality. It follows that the specifier and the complement of a head qualify as equally close to the head; and that the specifier of a head is closer to the head than a category that is further embedded in the complement of the head. 46 As an alternative to m-command, we could follow proposals in the literature which dissociate Voice from the base position of the external argument (e.g. Collins 2005, Sigurðsson 2011). In these accounts, VoiceP dominates vP and the latter introduces the external argument. If Voice is active, an external argument is projected in Spec,vP, if Voice is passive, no Spec,vP is projected and the external argument remains implicit. If we assume that the ϕ-features relevant for case-determination are located in Voice, we can derive a nominative-accusative case system by letting Voice search its c-command domain for the closest valued DP.

31

that this is triggered by an EPP-feature on T, not by Nominative Case (Alexiadou & Anagnostopoulou 1998, 2001, Marantz 2000, Sigurðsson 2000, McFadden 2004, among others). The standard implementation of A-movement along these lines takes it that T has unvalued ϕ-features and an EPP-feature and that the DP which values T’s ϕ-features is afterwards moved to Spec, TP. However, quirky subjects in Icelandic make it clear that the element that moves to Spec,TP is not necessarily the one which triggers agreement morphology on T. This means that we should dissociate DP-movement to Spec,TP not only from nominative case but also from overt agreement morphology on T. I follow, therefore, proposals which argue that verbal agreement morphology on T should be seen as the result of cyclic AGREE between Tense and the lower Voice/v-head (Legate 2005, Marantz to appear, Schäfer to appear); since the latter has been valued by the DP that receives nominative, T will indirectly agree with this DP, too.47 In the trees below, I leave out this AGREE-relation between T and the lower Voice/v-head for simplicity, but note that this AGREE-relation allows it to uphold the crucial connection between finiteness/verbal agreement on T and nominative case in this theory. The derivation of PoRs is straightforward. Instead of T, it is now little v that agrees with the anaphor. More concretely, v probes downwards and the anaphor probes upwards and the two form an unvalued two-member agreement chain. This derivation crashes in Dutch or Norwegian, but in German and Icelandic Default Agreement rescues it.48 (54) a. dass sich gewaschen wird. that REFL washed becomes ‘that people wash.’ b. TP 3 T vPpassive ! v’ 3 vpass{ uP, uN, uG } VP 3 DP{uP, uN, uG} V To derive the preservation of structural accusative in the New Passive, I propose that speakers of this variant of Icelandic have grammaticalized the outcome of this Default Agreement operation; i.e., these speakers (mis-)interpreted the PF of PoRs as an indication that there exists a passive v-head with valued ϕ-features.49 Such a v-head does not probe and the case 47 That is, I assume that T is involved in two different AGREE-processes, one triggering dislocation to Spec,TP and the other one leading to overt agreement morphology in T. While the former process arguably takes place in the syntax, it has been proposed that the latter takes at PF (Bobalijk 2008, Sigurðsson 2011). See the next footnote for conceptual considerations why this latter process might also take place in the syntax. 48 In ordinary impersonal passives lacking any DP argument, T and v form an unvalued agreement chain which must be valued by DA. In order to derive that PoRs are available only in a subset of languages forming impersonal passives, I propose that DA on a chain between two verbal heads (T <---> v) is less costly than DA on a chain between two verbal heads and a DP (T <---> v <---> anaphor). Note that this is compatible with the idea that DA is restricted in that it can only apply if syntactic AGREE between a probe and (an unvalued) goal has taken place, i.e., DA can only value chains, not just single heads like T or v (see D’Alessandro 2007:4.3 and Lopez 2008 for this idea). This might be important to avoid overgeneration. 49 This proposal leads to the question whether ordinary impersonal passives could also lead to a New Passive. On the empirical side, I would hypothesize that this is not possible. In order to derive this result, phase theory might be relevant though I can only speculate here. If the passsive vP is a phase, DA between v and an anphor will take

32

assignment constraint in (53) predicts that the internal argument gets accusative case at PF.50 Speakers of Icelandic, therefore have (a subset of) the following inventory of v-heads ((55a) comes actually in two different versions for transitive/unergative and unaccusative verbs): (55) a. v-active {uP, uN, uG} (all speakers) b. v-passive {uP, uN, uG} (all speakers) c. v-new-passive {3rd, SG, Neuter} (New Passive variant)51 With this proposal, lets turn quickly two the other two properties of the New Passive, the ban on A-movement and the absence of the definiteness effect. Note that these two properties are typically assumed to be correlated in that a DP inside vP has to be indefinite only if its definite counterpart could A-move out of the vP. Taking this correlation for granted, we only have to account for the ban on A-movement for the internal accusative argument in the New Passive. As a starting point, we know that internal arguments of transitive constructions cannot A-move. This can be analyzed as an intervention effect triggered by the external argument (leaving phase-theoretical considerations aside). v probes for the ϕ-features of the closest DP and this DP checks afterwards the EPP feature on T via displacement to Spec,TP. Following the argumentation developed in Sigurðsson (2011), I want to propose, therefore, that A-movement of the internal argument is blocked in the New Passive because the valued ϕ-features on Voice intervene between T and the internal argument. Due to this blocking, the EPP feature has to be checked by an expletive in the New Passive. In principle, it seems to me a welcome result that all three properties of the New Passive are related to one underlying factor, i.e., valued ϕ-features on v/Voice. Note that I hereby depart form Sigurðsson’s (2011) analysis of the New Passive who dissociates the ban on A-movement from the assignment of accusative case. The problem that I see with this dissociation is that it leaves it as a coincidence that the speakers of the New Passive variety introduced two then independent properties into their grammar at the same time and in the context of the very same construction. On the other hand, Sigurðsson has good reasons to dissociate the two processes and I should briefly go into them. The point is that Icelandic has another construction which shows only the case preserving property of the New Passive but no ban on A-movement. These are the so-called ‘FATE-accusative’ verbs (or ‘stray accusative’ verbs; see Haider 2001, Svenonius 2002, 2006, Platzack 2006, Sigurðsson 2005, Schäfer 2008:chapter 7). This term makes reference to a set of verbs which undergo the causative alternation, but where, unexpectedly, the sole argument of the intransitive variant keeps the same case marking as the object of the transitive variant. An example of this alternation taken from Sigurðsson (2005) is given in (56). (Note that the accusative is structural as it disappears under passivization). Crucially, however, Fate accusatives differ from New Passives in that the accusative DP can move out of the vP (and therefore also shows a definiteness effect) as shown in (56b). (56) a. Stormurinn rak bátinn á land. (causative) the.storm.NOM drove the.boat.ACC on land ‘The storm drove the boat onto land.’ place in the lower phase, while DA between T and v will take place in the higher phase. This difference might be relevant for a grammaticalization process involving features of the lower phase head v. 50 If the internal argument is an anaphor, it probes upwards and gets formally valued by the inherently valued v-head. This means that PoRs in the New Passive variety do not involve any formal repair mechanism anymore. But the interpretation of PoRs still needs conceptual backup (inherent or natural reflexivity) as the syntax does not provide a DP which could act as semantic binder at LF. 51 Note that these ϕ-features cannot be equated with the implicit external argument.

33

b. Bátinn rak á land. (Fate accusative) the.boat.ACC drove on land ‘The boat drifted onto land.’ For Sigurðsson (2005, 2011), Fate accusatives differ form the New Passive in that they do not involve ϕ-features on v/Voice. Therefore A-movement is not blocked. Simplifying his analysis for reasons of space, the two constructions share, however, the property that their v/Voice heads (a passive v/Voice head and an unaccusative v/Voice head respectively) carry the same diacritic which leads to the realisation of accusative morphology on the theme argument at PF. The fair question is, therefore, how Fate accusatives can be integrated in my system which seems to necessarily correlate the licensing of the “unexpected accusative” with the ban on A-movement. In Schäfer (2008), I followed the proposal in Haider (2001) and Platzack (2006) that Fate accusatives are hidden transitives which involve a covert external argument pro in subject position. This covert subject is the same as the one which arguably shows up with Icelandic weather verbs. The latter never take an overt (semi-)argumental expletive in Icelandic (cf. 57 from Sigurðsson (1989:164, ex. 15a)), but if we believe that every verb must take at least one semantic argument (due to a ban on empty predication) then Icelandic weather verbs should involve a covert subject, a specific kind of weather-pro in Spec,v (Sigurðsson 1989, Haider 2001; on the (semi-)argumental status of the “expletive” in weather verbs, see e.g. Rizzi 1986 or Vikner 1995). (57) Rigndi (*það) mikið í gær ? rained it much yesterday ‘Did it rain a lot yesterday?’ If Fate accusatives involve such a covert external argument (pro) which values v/Voice (for some further empirical motivation see Schäfer (2008) and the references there), then accusative on the theme follows trivially from the case assignment rule in (53). However, we must ask why this covert subject does not intervene for A-movement? This question becomes even more urgent because Icelandic has another construction involving a covert pro-subject where A-movement of the internal argument is in fact blocked (see also Svenonius (2002:199) for raising this point). This is the case with the ‘impersonal modal’ construction (Sigurðsson & Egerland 2009). (58a) vs. (58b) shows that the internal accusative argument cannot move out of the vP in this construction, arguably because the covert subject intervenes. (58) a. Hér má ekki auglýsa bókina. here may not advertise book.the.ACC b. *Hér má bókina ekki auglýsa. here may book.the.ACC not advertise ‘One may not advertise the book here.’ If we want to keep the claim that Fate accusatives take a covert pro subject, we have to explain why this subject does not intervene for A-movement of the internal argument while the covert subject in the modal construction does. Note, in this connection, that the interpretation of the covert subjects in the two constructions clearly differs. In the modal construction, the subject refers to a human entity similar to English ‘one’, German ‘man’ or French ‘on’ (see Sigurðsson & Egerland 2009 for a detailed discussion of the interpretation of this covert subject). The covert pro is, therefore, fully specified for ϕ-features. The subject of weather verbs, and by hypothesis also of Fate accusative constructions, on the other hand, is referentially extremely reduced. (For example, it cannot be asked for.) In Schäfer (2008), I

34

proposed, therefore, that weather-pronouns only have a reduced set of φ-features (see also Szucsich 2007); specifically, they only involve a number-feature but no person-feature and no gender-feature. I would hypothesize then that the amount of feature specification on pro determines whether it acts as an intervener for A-movement or not. If it is fully specified, it intervenes, if it is only partly specified, it does not. (However, this partly specified pro must, nevertheless, be “strong” enough to trigger ACC on the internal argument). This makes the prediction that – just as the (hypothetical) subject in Fate accusatives – the subject of weather verbs should not intervene for A-movement. While weather verbs are normally intransitive, it is nevertheless possible to combine them with internal arguments as in (59).52 The example then shows that this prediction is confirmed, i.e. the internal argument leaves the vP if definite (p.c. Halldor Sigurðsson). (59) Þá mundi spurningunum rigna (*spurningunum) yfir okkur. then would questions.the.DAT rain questions.the.DAT over us ‘Then it would rain these questions on us. (We would have to face these questions.)’ 8. Personal PoRs So far, I have discussed PoRs as instances of impersonal passives, i.e., passives which lack a nominative DP triggering verbal agreement. While the majority of PoRs fit this description, sometimes PoRs do, in fact, involve such a nominative DP. (I concentrate first on German. Later on, I will turn to Icelandic, which shows an interesting quirk in the relevant contexts). Cases in point are passives of double object verbs as in (60a) or cases where the VP contains an object and a prepositional phrase as in (60b). (60) a. Anschließend hat er sich einen Hamburger gekauft. Afterwards has he REFL.DAT a.ACC hamburger bought ‘Afterwards he bought (himself) a hamburger.’ b. Auf der Konferenz hat er viel Unsinn von sich gegeben. at the conference has he much nonsense.ACC from REFL.DAT given ‘At the conference, he uttered lot’s of nonsense.’ These strings passivize quite well as in (61). In these examples, the accusative theme-DP shifts to nominative and triggers agreement on the finite auxiliary. I call such passives personal PoRs. Note that personal PoRs show the very same semantic/conceptual restriction identified for impersonal PoRs in section 2 in that the relation between the external argument and the argument expressed by the anaphor must be inherently or naturally reflexive. The semi-idiomatic example in (60b, 61b) is inherently reflexive (see the discussion about (35) above). The predicate ‘x buys y for z’ in (60a, 61a) is naturally reflexive in that x is typically identical to z (see Kemmer 1993: section 3.5.2). Double object constructions that lack this semantic property typically do not form PoRs.53 This is predicted because, as with impersonal PoRs, the syntax does not provide an antecedent for the anaphor.

52 The internal argument of Icelandic weather verbs is marked with dative (Maling 2002). Note that the corresponding German example would use an overt weather-expletive as in (i), suggesting that the Icelandic example is in fact transitive, not an unergative with a quirky subject replacing the weather-pro of the canonical weather verb. However, arguments for this conclusion from within Icelandic would be preferable. (i) Dann hat *(es) tausende Fragen auf uns geregnet. then has it thousand questions.ACC on us rained ‘Then it rained thousand questions on us. (The we had to face lots of questions.)’ 53 A Google query on 22.2.2012 received 30 hits for “wird sich gekauft” (becomes REFL bought), but zero hits for “wird sich (selbst) (zu)geschickt” (becomes REFL (SELF) (at)send). The latter predicate is naturally disjoint.

35

(61) a. Anschließend wurde sich ein Hamburger gekauft. Afterwards was REFL.DAT a.NOM hamburger bought ‘People bought themselves a hamburger.’ b. Auf der Konferenz wurde viel Unsinn von sich gegeben. At the conference was many nonsense.NOM from REFL.DAT given ‘People uttered lots of nonsense at the conference.’ Personal PoRs pose some questions for my account developed in the last sections. First, why does the nominative DP not antecede the anaphor? This is of course possible as the example below shows. There the anaphor probes the tree upwards and gets valued by the (A-moved) nominative theme leading to an ordinary binding relation at LF. (62) Die Gäste wurden sich / einander vorgestellt. the guests.NOM were REFL.DAT / each-other.DAT introduced ‘The guests were introduced to each other.’ The only thing I have to say about this question is that, independently of PoRs, any binding theory must allow for bypassing potential antecedents. Cases in point are active double object constructions where the direct object can, but does not have to be bound by the indirect object as in the English example below. (63) Johnk showed Maryi herselfi/himselfk in the mirror. The next question is how the anaphor gets valued if the nominative DP does not bind it. Default Agreement is clearly not at stake as the nominative DP values T (or within the theory developed in the last section, the passive v-head). My answer to this question builds again on the dissociation between formal valuation of an anaphor in the syntax, on the one hand, and translating this valuation into a semantic binding relation at LF, on the other hand. The derivation I want to propose is shown in (64). I assume a low applicative structure for double object constructions (Pylkkänen 2008). Passive v probes its c-command domain and agrees with the variable but no valuation takes place because both are unvalued. But this time further probing of v beyond the anaphor will meet the DP-theme. This DP values the v-head, and, indirectly, the anaphor, too, because the latter is in an AGREE-relation with v. Therefore, the anaphor does not trigger a crash at the interfaces. However, since the anaphor is not in a direct AGREE-relation with a c-commanding DP, no semantic antecedent can be computed from this derivation at LF. Instead, conceptual licensing has to guide the interpretation of the anaphor, as was the case with impersonal PoRs. (64) a. dass sich ein Hamburger gekauft wird. that REFL.DAT a.NOM hamburger bought becomes ‘that people buy a hamburger for themselves.’

36

b. TP 3 T vPpassive ! v’ 3 vpass{uP, uN, uG} V ! ApplP 3 DP{uP, uN, uG} Appl’ 3 Appl DP{P, N, G} This derivation makes two predictions. First, since the anaphor is not valued via Default Agreement, such a derivation should be available also in languages lacking this repair strategy.54 In section 4, I had discussed that Dutch and Norwegian lack impersonal PoRs. The explanation offered for this in section 6 was that these languages do not have Default Agreement involving the chain between an anaphor and T (or passive v). If personal PoRs do not built on Default Agreement, these two languages should in principle allow their formation.55 Indeed, there is very preliminary indication that personal PoRs are judged better than impersonal PoRs in Dutch and Norwegian. Jan Wouter Zwart (p.c.) judged the example in (65) as “not strikingly bad”. Tor Åfarli (p.c.) mentions that examples such as in (66) are found in colloquial Norwegian and Terje Lohndal (p.c.) agrees that this example is better than the Norwegian impersonal PoRs discussed in section 4. Note that the Dutch example involves an inherently reflexive predicate and the Norwegian one involves a naturally reflexive predicate. As said, this conceptual restriction should also hold for personal PoRs as the anaphor does not find a DP-antecedent within syntax. However, the significance of these very preliminary results should be checked with many more examples and speakers. I must leave this for future research. (65) Op de workshop moest er weer een hoop geklets on the workshop must there again a heap.NOM nonsense tot zich genomen worden. to REFL taken become ‘On the workshop one had to take in a lot of nonsense again.’ (66) Vi var på puben i går kveld, We went to the.pub in last night og det ble kjøpt seg øl over en lav sko. and it was bought REFL beer.NOM over a low shoe ‘We went to the pub last night, and one bought oneself beer all the time.’56

54 The formal aspects of this derivation (but not its interpretation at the interface) is also very similar to the one proposed in Schäfer (2008, to appear) for reflexive anticausatives and reflexive middles which exist in many languages with SE-reflexives (see section 9). 55 But see the discussion of Icelandic further below in this section. If Dutch ‘zich’ and/or Norwegian ‘seg’ have the same restriction that I propose for Icelandic ‘sig’ then these two languages are predicted not to form any PoRs at all because then even the PoRs in (65) and (66) would have to make use of DA. 56 There is actually some discussion in the literature whether non-moved internal arguments in Norwegian

37

The second prediction concerns the PF Spell Out of the anaphor. Although the nominative DP does not value the anaphor directly, it does so indirectly via the connection with the passive v-head (see the tree in (64b)). This predicts that if the nominative is a 1st or 2nd person pronoun, the anaphor should actually also show this feature make-up. To test this prediction, consider the data below which are taken from Plank (1993). (67a, b) provide some context. (67b) is an active clause which is inherently reflexive. The indirect argument is an anaphor which is obligatorily bound by the external argument, and the theme is a 2nd person plural pronoun. In (68), we see four passive versions of this active clause which differ in the following way: In (68a), the accusative theme has shifted to nominative and agrees with the finite verb and the indirect object variable is realized as the 3rd person anaphor ‘sich’. The b-example differs in that the variable is realized as the 2nd person plural pronoun (‘uns’), i.e., it has the same feature content as the nominative theme and the verbal agreement. In (68c), the theme stays in the accusative, the verb shows default agreement and the indirect object is realized as the 3rd person anaphor ‘sich’. In (68d) finally, the theme is in the accusative, the finite verb shows default agreement and the variable is realized as the 2nd person pronoun ‘uns’, i.e., it shares the ϕ-features of the theme.57 (67) a. Firma Ai will sichi Firma B einverleiben. company.NOM A wants REFL.DAT company.ACC B incorporate ‘Company A wants to incorporate/absorbe Company B.’ b. The workers of company B say: So einfach verleibt man sich uns nicht ein. So easily corporates one.NOM REFL.DAT us.ACC not in ‘One does not absorb us so easily.’ (68) a. *So einfach werden sich wir nicht einverleibt. So easily become.PL REFL.DAT we.NOM not incorporated b. *So einfach werden uns wir nicht einverleibt. So easily become.PL us.DAT we.NOM not incorporated c. ?So einfach wird sich uns nicht einverleibt. So easily become.SG REFL.DAT us.ACC not incorporated d. *So einfach wird uns uns nicht einverleibt. So easily become.SG us.DAT us.ACC not incorporated Formally, the theory proposed above predicts (68b) to be the grammatical version. However, beyond any doubt, the only acceptable version is the one in (68c). But note that the string in (68b) suggests that there is a semantic binding relation between the recipient and the theme (due to the inherent reflexivity of the verb ‘einverleiben’ (incorporate) the interpretation would be something like ‘We incorporate ourselves into ourselves’.) The reason is, of course, the indexical nature of the 2nd person pronouns ‘uns’ and ‘wir’ which necessarily refer to the same group of individuals involving the speaker. What is going on here, in my opinion, is that in order to avoid this wrong interpretation speakers adhere to the repair strategy that we know already from impersonal PoRs, namely Default Agreement. More precisely, the passive v-head and the anaphor agree with each other without valuating each other and, in order to avoid the misleading interpretation which the surface form of (68b) would bring about, the v- passives keep the accusative or shift to nominative (see e.g. Åfarli 1989). If they keep accusative, this example is an instance of an impersonal PoR of a ditransitive verb similar to the Icelandic examples discussed in (69). 57 In (68a-d), I keep the canonical order between the indirect and the direct object for reasons of readability. Note that all judgements remain the same if the order is changed by scrambling. In fact, nominative pronouns typically precede dative pronouns in the German pronoun cluster.

38

head does not further probe beyond the variable to agree with the theme DP. Instead, Default Agreement values the agreement chain between the v-head and the variable and, as a consequence, the theme appears with accusative case (cf. the case-algorithm in 53), the variable gets the Spell Out as the 3rd person reflexive ‘sich’ and the verb shows default morphology. Of course, this proposal leaves open many questions about when and how the grammar makes available the costly operation of Default Agreement. These are beyond the scope of this paper. However, it is very interesting in this connection to note that Icelandic uses Default Agreement quite regularly to form PoRs of ditransitive verbs (while German does so only in the very specific constellation in (68)). The phenomenon is discussed in Eythórsson (2007, 2008) and Árnadóttir et al. (2011) under the label ‘Impersonal Ditransitive Reflexive Construction’ (IDRC). As these authors show, speakers of Icelandic quite generally avoid personal PoRs. Instead, they keep accusative on the theme and the finite verb and the participle appear with default agreement. Examples are given in (69). The examples in (70) show that the theme cannot shift to nominative (although Eythórsson (2007) mentions some counterexamples). Note in passing that the verbs involved in IDRCs are also either inherently or naturally reflexive and that examples as in (69) are accepted by speakers who do not accept ordinary New Passives. (69) a. Það var fengið sér öllara. it was got.NEU.SG REFL.DAT beer.ACC ‘People got themselves a beer.’ b. Það var keypt sér pizzu. it was bought.NEU.SG REFL.DAT pizza.ACC ‘People bought themselves pizza.’ (70) a. *Það var fenginn sér öllari. it was got.MSC.SG REFL.DAT beer.MSC.SG.NOM b. *Öllari var fenginn sér. beer.MSC.SG.NOM was got.MSC.SG REFL.DAT ‘People got themselves a beer.’ To make clearer this difference between German and Icelandic, let us turn back to the tree in (64b), which shows the derivation of PoRs involving ditransitive verbs. The difference seems to be that in German the variable (‘sich’) allows v to further probe down the tree and to agree with the theme, while in Icelandic the variable (‘sig’) seems to block further probing of v. As a consequence, the theme will value the chain between v and the variable in German while DA will value this chain in Icelandic. Due to the case constraint in (53) this will have as a result that the theme argument surfaces with nominative in German but with accusative in Icelandic.58 A broader study of the availability of and the restrictions on Default Agreement in general and of IDRCs in particular might shed more light on whether this proposal goes into the right direction or not. 58 This proposal makes interesting predictions for reflexive anticausatives in Icelandic. In the next section (see also fn. 36), I describe the derivation of reflexive anticausatives (e.g. in German) where crucially the variable/anaphor has to allow v to bypass and to probe the theme. If ‘sig’ blocks further probing of v, then such a derivation should be impossible in Icelandic. And in fact Icelandic does not form reflexive anticausatives with the full reflxive pronoun ‘sig’ but with the affix ‘-st’. Interestingly, this affix seems to be caseless and not be involved in any agreement processes (see Wood (2012, submitted) for discussion).

39

9. Open issues: potential problems and topics for future research In this last section, I will adress some open issues, potential problems and directions for future reasearch that might grow out of my proposal. They relate to the interaction of PoRs with pronominal expletives, reflexives merged in the external argument position and the so-called ‘Anaphor Agreement Effect’. In section 6.3, I discussed two strategies how a language can make available impersonal passives. Leaving aside the question how the EPP on T is checked in impersonal passives, languages can use either Default Agreement or a nominal expletive, i.e. an expletive that is inherently equipped with ϕ-features (such as English ‘it’ which is, however, restricted to argument positions). Holmberg (2002) reports this latter strategy for some Norwegian dialects. If such expletives can value T (or passive v), they might also be able to formally value the variable/anaphor in PoRs. Since no Default Agreement would be necessary, one might predict PoRs to be grammatical. However, the Norwegian dialects discussed by Holmberg (2002) do not seem to license impersonal PoRs any more than those dialects using featureless expletives (locative adverbs such as English ‘there’). There are two ways how this could be derived. Either, passive v/Voice is a phase and the expletive is merged outside of the phase, say directly in Spec,TP, where it values the chain between T and passive v/Voice. In this case, the expletive might enter the derivation too late to formally value the anaphor inside vP. Alternatively, the expletive might enter an AGREE-relation with the anaphor and, in turn, value it. In this case, we must ask how the CI-interface would interpret this structure. Since the expletive is fully nominal and c-commands the variable, the CI-interface should try to compute a semantic binding relation between the pronominal expletive and the anaphor. But since the expletive lacks a thematic role no such interpretation can be computed. Therefore, the derivation necessarily leads to semantic gibberish. If anaphors can be valued by default agreement and can afterwards be interpreted with the help of conceptual knowledge about events and their participants, there might be the danger of overgeneration. Specifically we must ask what would happen if we merge an anaphor in a verb’s external argument position? First, an anaphor could be merged in Spec,vP of an unergative verb, get formally valued by DA and receive accusative at PF ([TP T [vP REFLACC v [VP sleeps]]] ). The problem with this derivation is that, at the CI-interface, no interpretation can be computed for the anaphor, since, at no level of representation, the sentence contains a second argument which could be used as an antecedent (either via formal binding, or via conceptual considerations as in PoRs). Second, the anaphor could be merged in the external argument position of a transitive verb. In this case, the derivation would be similar to the one proposed for personal PoRs in section 8. v would search its m-command domain for a goal. It would first agree with the anaphor in its specifier but no valuation could take place at this point. Afterwards, v would agree with the internal argument (a DP fully specified for ϕ-features), which would value v, and, indirectly, also the anaphor. At PF, the internal argument would get nominative case and the anaphor would get accusative case. Furthermore, the anaphor would be spelled out with the ϕ-features of the internal argument. Turning to the CI-interface, once again, no ordinary binding relation can be computed on the basis of this syntactic derivation since the anaphor is not in a direct AGREE-relation with a (c-commanding) DP antecedent. In Schäfer (2008, to appear), I argued that such derivations are used to form reflexive anticausatives (71a) or reflexive middles (71b) in German and other Germanic languages (as well as in Romance and Slavic; see fn. 58 for Icelandic). In these constructions, the reflexive pronoun does not carry any interpretation and is present only for purely formal reasons, namely to fill the external argument position syntactically without introducing an external argument semantically. This expletive use of the reflexive pronoun is possible exactly because it lacks a c-commanding antecendent.

40

(71) a. [CP weil [TP T [vP sichACC v [VP die TürNOM öffnet]]] because REFL the door opens ‘... because the door opens’ b. [CP weil [TP T [vP sichACC v [VP das BuchNOM gut verkauft]]] because REFL the book well sells ‘... because the book sells well’ What would happen with a derivation as in (71), if the verb involved were inherently or naturally reflexive as in (72)? While the anaphor would lack a c-commanding DP-antecedent, perhaps conceptual knowledge could lead to an ordinary transitive/active interpetation similar to PoRs. (72) a. [CP weil [TP T [vP sichACC v [VP MariaNOM wäscht]]] as REFL Mary washes ‘... because Mary washes’ I would like to suggest that this is not possible because the structure provides a Principle C violation. The reflexive pronoun (which has a thematic interpretation) would c-command a co-indexed R-expression. Of course, this presupposes that Principle C effects cannot be circumvented by conceptual considerations, while Principle A effects can. A full integration of Principle C is beyond the scope of the present paper. As a basic assumption in this paper, I followed earlier proposals in the idea to encode the relation between an anaphor and its antecedent via a syntactic AGREE-relation (see the references in section 6.2). There are different motivations for this move, but one central one is the so-called ‘Anaphor Agreement Effect’, which such theories are intended to derive. This term refers to the observation that across languages anaphors are excluded from syntactic positions that trigger verbal agreement (Maling 1984, Everaert 1990, Rizzi 1990, Woolford 1999, Tucker 2010). In NOM-ACC languages, these are basically argument positions to which nominative is assigned. The idea is that arguments in such positions have to value a functional head, say T, with ϕ-features and anaphors, which are by hypothesis devoid of inherent ϕ-features, cannot fulfill this job. As a consequence anaphors in such a position lead to a crash. Note that passives of verbs with a reflexive pronoun in structurally case marked object positions (PoRs) are exactly a case in point. As I argued, some languages can rescue the latter by making use of Default Agreement, i.e. by unburdening the anaphor from the duty to trigger verbal agreement. The question that I want to address is whether the concept of Default Agreement leads to unwanted consequences for this general approach to the Anaphor Agreement Effect. Consider Icelandic DAT-NOM constructions where the dative subject locally binds the nominative object. As we see in (73) and (74), the internal argument cannot be realized by an anaphor. Instead, a (intensified) pronoun has to be used (Everaert 1990). (Everaert shows that the pronoun cannot be analyzed as an anaphor in disguise; for example, it allows for sloppy as well as strict readings. Note also that a dative subject can antecede an anaphor if the latter is marked with genitive case or embedded in a nominative DP.) (73) a. *Honum finnst sig veikur. Him.DAT finds REFL.ACC sick b. Honum finnst ?hann / hann sjálfur (vera) veikur. Him.DAT finds he.NOM/ he.NOM SELF (be) sick ‘He considers himself sick.’

41

(74) a. *Maríu leiðist sig. Maria.DAT finds.boring REFL.ACC b. Maríu leiðist hún sjálf. Maria.DAT finds.boring her.NOM SELF ‘Maria considers herself boring.’ The question is why the a-examples cannot be rescued by Default Agreement. The derivation would look as depicted in (75): the anaphor probes upwards, agrees with the dative subject and gets valued by it. This will lead to a semantic binding relation between the two at the CI-interface. For some reason, the ϕ-features which the anaphor aquires from the dative cannot be used to trigger verbal agreement (Note that this assumption is necessary for all theories assuming that syntactic agreement is at the base of anaphoric relations. Somehow, the fact that a quirky subject itself cannot trigger verbal agreement makes it impossible for the anaphor which agrees with the quirky subject to trigger verbal agreement indirectly). Afterwards, v and T form an unvalued agreement chain. In order to let this structure survive, Default Agreement values this chain in a similar vein as in canonical impersonal passives (see fn. 48). (75) [TP T [vP DPDAT v [VP V REFLACC]]] For some reason, such a derivation is not available. Instead, the anaphor is replaced by a locally bound pronoun, in contradiction to what Principle B would suggest. Since such a pronoun enters the derivation with valued ϕ-features, it can value v without problem. Replacing an anaphor with a pronoun is arguably an alternative repair strategy to rescue structures from the Anaphor Agreement Effect (cf. Fanselow 1989, 1991, Burzio 1998). This leads to two questions. Why does DA not work in (73a) and (74a). And why are PoRs not rescued by inserting a pronoun in object position? Concerning the first question, we could suggest that DA is somehow restricted to passive structures. But this does not seem to be correct because at least some speakers of Italian allow exactly this repair strategy in similar constructions as shown in (76b) (Rizzi 1990, Woolford 1999:fn. 5, Tucker 2010). (See Woolford (1999) for further instances were Default Agreement avoids the Anaphor Agreement Effect.) I leave the question why a language uses one or the other repair strategy in this context for future research. Note that, as in Icelandic, most speakers of Italian would replace the reflexive in (76) with a nominative pronoun.

(76) a. *A loro interessano solo se stessi. to them.DAT interest.3.PL only REFL selves b. (?)A loro interessa solo se stessi. to them.DAT interest.3.SG only REFL selves ‘They only care about themselves.’ This leaves the other question, namely why PoRs are not rescued by replacing the anaphor in object position with a pronoun, i.e. why a string such as ‘He was washed’ cannot get the interpretation ‘He washed himself’. To explain this, we can follow the proposal in Kratzer (1996) that the implicit argument of passives, although it cannot bind an anaphor as I argued in section 5, nevertheless triggers Principle B and C effects. That is, the implicit argument of passives behaves like a weak implicit argument in the sense of Landau (2010); it is active for some but not all grammatical processes.

42

10. Conclusions This paper discussed Passives of Reflexive Verbs (PoRs), which exist in a subset of languages with impersonal passives and SE-reflexives, namely German and Icelandic. This construction poses questions for standard theories of case and anaphoric binding. With respect to Case Theory, the accusative case on the reflexive pronoun contradicts the predictions of Burzio’s Generalization. I argued that there is no reason to assume that an idiosyncracy in the case system is at play and that this accusative should be analyzed as ordinary structural object case. Concerning Binding Theory, I showed that the formation of PoRs is conceptually restricted to inherently and naturally reflexive contexts. I argued that lexical theories of reflexivity cannot account for the availability of PoRs because the reflexive pronoun in PoRs behaves like a syntactic and semantic argument and the reflexive relation found in PoRs is not confined to the verbal co-argument domain. However, a standard version of Principle A of the Binding Theory cannot account for PoRs either because a closer inspection of PoRs showed that the implicit argument of the passive is not the formal, i.e. syntactic antecedent of the reflexive pronoun.

I proposed that the reflexive pronoun in PoRs is an ordinary anaphor/a variable and that it should therefore be subject to some version of Principle A of the Binding Theory. I followed recent proposals which reformulate the traditional Principle A in terms of a syntactic AGREE-relation between an anaphor and its antecedent. Since in PoRs the anaphor cannot find a c-commanding DP-antecedent, PoRs lead to a Principle A violation in most languages. I argued, however, that in German and Icelandic a formal violation of Principle A can be avoided because Default Agreement (DA), a last resort repair operation, can formally value the φ-features of the variable. DA only avoids a formal crash of PoRs at the interfaces but it does not lead to a semantic interpretation of the variable. Therefore, a successful interpretation of PoRs depends on an inherently or naturally reflexive context. Finally, I proposed an AGREE-based version of a dependent case approach which allows the application of Default Agreement to trigger structural accusative on the reflexive pronoun in PoRs. I used this case theory to elaborate the hypothesis that the Icelandic New Passive developed out of PoRs due to a grammaticalization process where the outcome of DA was reinterpreted as a lexical property of a passive v-head. Finally I discussed personal PoRs where the passive vP contains an anaphor and a referential object DP (e.g., PoRs involving ditransitive verbs). The theory developed here predicts these to be more readily available than impersonal PoRs (i.e., even in Dutch and Norwegian) because they do not depend on the costly operation of DA. While there are some preliminary indications that this might be empirically correct, this topic awaits further research. Furthermore, while impersonal PoRs seem to work quite similar in German and Icelandic, this parallelism breaks partly down in the context of personal PoRs. While the referential direct object gets nominative case and triggers verbal agreement in German, it gets accusative case in Icelandic. I hypothesized that the reason for this might be that German ‘sich’ allows v to bypass and to probe for the object DP while Icelandic ‘sig’ blocks further probing of the object DP by v. As a result, only the Icelandic construction involves Default Agreement, which in turn triggers accusative case on the object. However, this proposal as well as the broader questions about when and why a language makes available the process of Default Agreement must be left for future research. In any case, it seems that even PoRs involving ditransitve verbs are subject to the same conceptual restriction as impersonal PoRs: they are acceptable only in inherently and naturally reflexive contexts.

43

Appendix 1: Results of the Internet query in section 2 Recall from section 2 that the number of matches in A and B result from a single search query while those in C result from 12 different search queries. A. inherently reflexive/reciprocal verb ‘sich benehmen’ (to behave): search string “sich benommen wird” - 54 matches (77) Es ist schlimm wie sich benommen wird It is bad how REFL behaved becomes ‘It is bad how people behave.’ ‘sich streiten’ (to quarrel): search string “sich gestritten wird” - more than 100 matches (78) Ich verstehe echt nicht, wieso sich gestritten wird

I understand really not why REFL quarreled becomes ‘I really do not understand why people querry.’

B.1 grooming verbs ‘waschen’ (to wash): search string “wird sich gewaschen” - more than 100 matches (79) Dann wird sich gewaschen, schön gemacht, umgezogen Then becomes REFL washed, nice made, clothes-changed und die Zähne geputzt and the teeth brushed ‘Then people wash, make themselves up, change their clothes and brush their teeth.’ B.2 verbs of translational motion ‘bewegen’ (to move): search string “wird sich bewegt” - more than 100 matches (80) im Musikunterricht wird sich bewegt, im Kunstunterricht wird in.the music-class becomes REFL moved, in.the art-class becomes der Pinsel geschwungen the brush swung ‘In the music class, people move, in the art class, people swing the brush.’ B.3 verbs of non-translational motion ‘verbeugen’(to bow): search string “wird sich verbeugt” - 40 matches (81) Vor den Prinzen wird sich verbeugt in.front.of the princes becomes REFL bowed ‘People bow down in front of the princes.’ B.4 verbs of change in body posture ‘hinsetzen’ (to sit down): search string “wird sich hingesetzt” - more than 100 matches (82) Danach wird sich hingesetzt und den Nachbarn ausgefragt Later becomes REFL sit-down and the neighbour asked

44

nach wohin, warum, usw.59 about where, why, etc. ‘Later, people sit down and ask the neighbour about where to go, why, etc.’ B.5 verbs of antagonistic events ‘prügeln’ (to trash): search string “wird sich geprügelt” - more than 100 matches (83) Da wird sich geprügelt, der geliebte Nachbars-Hund entführt, ... there becomes REFL beaten, the beloved neighbour-dog kidnapped ‘People beat each other, they kidnapp the beloved dog of the neighbour, ...’ B.6 verbs of affectionate actions ‘küssen’ (to kiss): search string “wird sich geküsst” - 30 matches (84) Ringe werden getauscht, es wird sich geküsst und fertig ist die Sache Rings become exchanged, it becomes REFL kissed and ready is the matter ‘People exchange rings, they kiss and then it is done.’ B.7 social/encountering actions ‘treffen’ (to meet): search string: “sich getroffen wird ” - more than 100 matches (85) Wann sich getroffen wird steht noch nicht fest when REFL met becomes stands not yet fixed ‘It is not fixed yet when to meet.’ B.8 verbs of interlocution ‘unterhalten’ (to talk to s.o.): search string “sich unterhalten wurde” - 73 matches (86) Worüber sich unterhalten wurde, haben wir nicht erfahren about-what REFL talked became have we not found.out ‘We did not figure out what they were talking about.’ B.9 verbs of emotion ‘wundern’ (to wonder): search string “wird sich gewundert” - more than 100 matches (87) Und dann wird sich gewundert, warum man keinen Schritt weiter kommt and then becomes REFL wondered why one no step further comes ‘And then people wonder why one does not make any progress.’ B.10 emotive speech actions ‘beschweren’ (to complain): search string “wurde sich beschwert” - more than 100 matches (88) In vielen Reviews wurde sich beschwert, dass ... in many reviews was REFL complained that ‘In many reviews people complained that ... .’ 59 This example involves a VP-conjunction, where the object of the first conjunct is the reflexive pronoun ‘sich' and the object of the second conjunct is the DP ‘den Nachbarn’ (the.ACC neighbour). This second DP is overtly marked for accusative case. This provides further evidence that the reflexive pronoun in PoRs is marked with accusative case. In my judgment, nominative case on this second DP would be equally good. A possible explanation might be that this example is ambiguous between a VP and a vP conjunction.

45

B.11 simple cognitive events ‘überlegen’ (to consider): search string “wird sich überlegt” - 31 matches (89) Es wird sich überlegt, wie man die Zuschauer entschädigen kann it becomes REFL considered how one the audience compensate can ‘People thought about how to pay compensation to the audience.’ B.12 verbs of perception ‘anschauen’ (look at): search string “wird sich angeschaut” - 48 matches (90) Auch das Opel-Problemforum wird sich angeschaut Also the Opel-problem-forum becomes REFL at-looked ‘People also looked at the forum for problems with Opel.’ B.13 intentive verbs ‘wünschen’ (to desire): search string “wird sich gewünscht” - 25 matches (91) Auch ein Kino wird sich gewünscht, genau wie Konzerte, ... Also a cinema becomes REFL desired exactly like concerts ‘People wanted to have a cinema, exactly like concerts, ... .’ C. naturally disjoint (transitive) verbs ‘schneiden’ (to cut): search string for example “sich geschnitten wurde” - 3 matches (92) Dazu lagen auch noch Glasscherben im Hauptpool, In-addition lay also even glas-fragments in-the main-pool, an denen sich geschnitten wurde! at which REFL cut became ‘In addition, there was broken glas in the main pool at which people cut themselves.’ ‘zerstören’ (to destroy): search string for example “sich (selbst) zerstört wird” - 0 matches ‘töten’ (to kill): Search string for example “wurde sich getötet” - 1 match (93) Doch wie kann Menschlichkeit in einem Geschäft existieren, But how can humanness in a business exist, das darauf hinausläuft das einander getötet wird? which amounts-to that each other killed becomes ‘But how can humanness exist in a business, in which people kill each other?’ ‘verletzen’ (to hurt): search string for example “wird sich verletzt” - 10 match60 (94) Es wird sich verletzt, es wird probiert, getestet, therapiert It becomes REFL hurt, it becomes tried, tested, therapy-made ‘People hurt themselves, try different things, test different things, make a therapy.’ ‘loben’ (to praise): search string for example “sich selbst gelobt wurde” - 11 matches

60 Some of the matches with ‘verletzen’ (to hurt) come from bulletin boards for borderline patients. For this community it is “natural” to hurt oneself. However, other examples have a reciprocal interpretation which does not allow such an explanation.

46

(95) wenn ... über alles Mögliche geredet, geschimpft und sich selbst gelobt wurde when ... about everything possible talked, blustered and REFL self praised became ‘when people talk about all possible things, bluster and praise themselves.’ ‘beschuldigen’ (to accuse): search string for example “wurde sich beschuldigt” - 3 matches (96) Da wurde sich beschuldigt, angepflaumt und geschmollt there was REFL accused, flamed and sulked ‘People accuse each other, flame each other and they sulk.’ ‘angreifen’ (to attack): search string for example “wird sich angegriffen” - 5 matches (97) Da wird sich angegriffen, aufgesprungen und gekniffen und gezwickt There becomes REFL attacked jumped-up and pinched and tweaked ‘People attack each other, jump up and pinch and tweak each other.’ ‘beladen’ (to load): search string for example “sich beladen wird” - 1 match (98) ... sondern mit einer Menge von Lehrern sich beladen wird but with a lot of teachers REFL loaded becomes ‘People load themselves with many different teachers.’ ‘kritisieren’ (to) critisize: search string for example “sich kritisiert wird” - 0 matches ‘einsperren’ (to lock in): search string for example “sich selbst eingesperrt wird” - 1 match (99) Vielmehr die Frage, wie dadurch sich selbst eingesperrt wird? rather the question how thereby REFL SELF in-locked becomes ‘Rather, the questions is, how people thereby lock themselves in.’ ‘aussperren’ (to lock out): search string for example “sich ausgesperrt wurde” - 1 match (100) weil der Schlüssel verloren gegangen ist oder sich ausgesperrt wurde because the key lost went is or RELF locked-out became ‘Because the key got lost or people locked themselves out.’ ‘beobachten’ (to watch): search string for example “wird einander beobachtet” - 2 matches (101) Es wird einander beobachtet, und der andere wird wie ein Feind betrachtet It becomes each-other watched and the other becomes as an enemy seen ‘People watch each other and the other one is seen as one’s emeny.’ Appendix 2: Test sentences of the questionnaire study in section 2 inherently reflexive verbs (i-ref): (I.act) Viele Jahre lang hat man sich einfach so durchgewurschtelt. many years long has one REFL simply so through-muddled ‘One simply muddled through for many years.’ (I.pas) Viele Jahre lang wurde sich einfach so durchgewurschtelt.

47

many years long became REFL simply so through-muddled (II.act) Auf dem Spielplatz toben sie sich dann richtig aus. at the playground let.steam.of they REFL then well PART ‘They let steam of at the the playground.’ (II.pas) Auf dem Spielplatz wird sich dann richtig ausgetobt. at the playground becomes REFL then well PART-let.steam.of (III.act) Erst danach erkundigten sie sich nach den Preisen. only afterwards inquired they REFL after the prices ‘They asked about the prices only afterwards.’ (III.pas) Erst danach wurde sich nach den Preisen erkundigt. only afterwards was REFL about the prices inquired (IV.act) Trotzdem weigert man sich, die Kosten zu übernehmen. Nevertheless refused one REFL the expenses to over-take ‘Nevertheless, one refused to take over the expenses.’ (IV.pas) Trotzdem wird sich geweigert, die Kosten zu übernehmen. Nevertheless becomes REFL refused the expenses to over-take inherently reciprocal verbs (i-rec): (I.act) Nach dem Streit vertrugen sie sich dann aber doch wieder. after the fight got.along they REFL then after-all again ‘After the fight, they got along with each other again.’ (I.pas) Nach dem Streit wurde sich dann aber doch wieder vertragen. after the fight became REFL then after-all again gotten.along (II.act) Über ethnische Grenzen hinweg freundet man sich kaum an. Across ethnical borders away makes.friends one REFL rarely PART. ‘People rarely make friends outside of their ethnical group.’ (II.pas) Über ethnische Grenzen hinweg wird sich kaum angefreundet. Across ethnical borders away becomes REFL rarely PART-made.friend. (III.act) Als Termin einigte man sich auf den Dienstag. as appointment.time agreed one REFL at the Tuesday ‘One agreed on Tuesday for the appointment.’ (III.pas) Als Termin wurde sich auf den Dienstag geeinigt. as appointment.time became REFL at the Tuesday agreed (IV.act) Über den Unterhalt streiten sie sich dann jahrelang. about the child.support argue they REFL then years.long ‘They argue for years about the child support.’ (IV.pas) Über den Unterhalt wird sich dann jahrelang gestritten. about the child.support becomes REFL then years.long argued naturally reflexive verbs (n-ref): (I.act) Vor der Feierstunde zogen sie sich noch schnell um. before the ceremony changed.dress they REFL yet quickly PART ‘Before the ceremony, they quickly changed their dresses.’ (I.pas) Vor der Feierstunde wurde sich noch schnell umgezogen. before the ceremony became REFL yet quickly PART-changed.dress (II.act) Mit Freude erinnert man sich an den letzten Titelgewinn. with happyness remembers one REFL at the last championship ‘One remembers the last championship with happyness.’ (II.pas) Mit Freude wird sich an den letzten Titelgewinn erinnert. with happyness becomes REFL at the last championship remembered (III.act) Allzuoft entfernen sie sich heimlich von der Unfallstelle.

48

too.often move.away they REFL silently from the scene.of.accident ‘Too often, they silently leave the scene of accident.’ (III.pas) Allzuoft wird sich heimlich von der Unfallstelle entfernt. too.often becomes REFL silently from the scene.of.accident moved.away (IV.act) Im Mittelalter rasierte man sich natürlich immer nass. in.the middle.ages shaved one REFL of.course always wet ‘In the middle ages, one shaved always wet of course.’ (IV.pas) Im Mittelalter wurde sich natürlich immer nass rasiert. in.the middle.ages became REFL of.course always wet shaved naturally reciprocal verbs (n-rec): (I.act) Auf dem Volksfest prügelten sie sich wieder heftigst. at the folk-festival beat they REFL again wildely ‘At the folk festival, they beat each other again wildely.’ (I.pas) Auf dem Volksfest wurde sich wieder heftigst geprügelt. at the folk-festival was REFL again wildely beaten (II.act) Bei der Ankunft begrüßte man sich erst einmal herzlich. at the arrival welcomed one REFL first of.all cordially ‘At the arrival, one first of all welcomed each other cordially.’ (II.pas) Bei der Ankunft wurde sich erst einmal herzlich begrüßt. at the arrival became REFL first of.all cordially welcomed (III.act) In solchen Filmen küssen sie sich nämlich andauernd. in such movies kiss they REFL to.wit permanently ‘In such movies, they kiss all the time.’ (III.pas) In solchen Filmen wird sich nämlich andauernd geküsst. in such movies becomes REFL to.wit permanently kissed (IV.act) Anschließend unterhält man sich lange über den Hauptdarsteller. afterwards chats one REFL long about the main-actor ‘Afterwards one chats for a long time about the main actor.’ (IV.pas) Anschließend wird sich lange über den Hauptdarsteller unterhalten. afterwards becomes REFL long about the main-actor chatted naturally disjoint verbs in a reflexive construal (nd-ref): (I.act) Im Kunstunterricht photographieren sie sich mit Hilfe eines Spiegels. in.the art-class photograph they REFL with help of-a mirror ‘In the art class, they photograph themselves with the help of a a mirror.’ (I.pas) Im Kunstunterricht wird sich mit Hilfe eines Spiegels photographiert. in.the art.class becomes REFL with help of-a mirror photographed (II.act) Nach dem Sturz verarztete man sich mit einem Druckverband. after the accident treated one REFL with a compression-bandage ‘After the accident, one treated oneself with a compression bandage.’ (II.pas) Nach dem Sturz wird sich mit einem Druckverband verarztet. after the accident becomes REFL with a compression-bandage treated (III.act) Wegen der Daten-CD zeigten sie sich bei der Polizei an. because of-the data-CD accused they REFL at the police PART ‘Because of the data-CD, they pressed charges against themselves at the police station.’ (III.pas) Wegen der Daten-CD wurde sich bei der Polizei angezeigt. because of-the data-CD became REFL at the police PART-accused (IV.act) Aus Angst sperrt man sich dann im Keller ein. due.to fear locks one REFL then in.the basement in

49

‘Because one is afraid, one locks oneself into the basement.’ (IV.pas) Aus Angst wird sich dann im Keller eingesperrt. due.to fear becomes REFL then in.the basement in-locked naturally disjoint verbs in a reciprocal construal (nd-rec): (I.act) Oftmals behindern sie sich durch die enge Startaufstellung. often impede they REFL due.to the tight starting-lineup ‘Often, they impede each other due to the tight starting linup.’ (I.pas) Oftmals wird sich durch die enge Startaufstellung behindert. often becomes REFL due.to the tight starting-lineup impeded (II.act) Schon vor dem Startschuss beobachtete man sich argwöhnisch. already before the starting-shot watches one REFL distrustfully ‘Even before the starting shot, they watch each other distrustfully.’ (II.pas) Schon vor dem Startschuss wurde sich argwöhnisch beobachtet. already before the starting-shot was REFL distrustful watched (III.act) Hinter verschlossenen Türen kritisierten sie sich ziemlich ungehemmt. behind closed doors critizised they REFL quite unresistedly ‘Behind closed doors, they critisized each other quite unresistedly.’ (III.pas) Hinter verschlossenen Türen wurde sich ziemlich ungehemmt kritisiert. behind closed doors became REFL quite unresistedly critizised (IV.act) Im Wartesaal starren sie sich fast immer gelangweilt an. in.the waiting-room stare they REFL almost always bored at ‘In the waiting room, they almost always stare at each other in a bored way.’ (IV.pas) Im Wartesaal wird sich fast immer gelangweilt angestarrt. in.the waiting-room becomes REFL almost always bored at-stared References Abraham, Werner. 1986. Unaccusatives in German. Groninger Arbeiten zur germanistischen

Linguistik (GAGL) 28:1–72. Åfarli, Tor A. 1992. The syntax of Norwegian passive constructions. Amsterdam: John

Benjamins. Ágel, Vilmos 1997. Reflexiv-Passiv, das (im Deutschen) keines ist. Überlegungen zu

Reflexivität, Medialität, Passiv und Subjekt. In Sprache im Fokus. Festschrift für Heinz Vater zum 65. Geburtstag, eds. Christa v. Dürscheid, Karl-Heinz Ramers, and Monika Schwarz, 147–187. Tübingen: Niemeyer.

Alencar, Leonel F. de, and Carmen Kelling. 2005. Are reflexive constructions transitive or intransitive? Evidence from German and Romance. In Proceedings of the LFG05 Conference, eds. Miriam Butt, and Tracy H. King. Stanford, CSLI Publications.

Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word Order, Verb-movement and EPP checking. Natural Language & Linguistic Theory 16.3: 491–539.

Alexiadou, Artemis, and Elena Anagnostopoulou. 2001. The subject in situ generalization, and the role of Case in driving computations. Linguistic Inquiry 32:193–231.

Árnadóttir, Hlíf, Eythórsson Thórhallur. and Einar F. Sigurðsson. 2011. The passive of reflexive verbs in Icelandic. Nordlyd 37:39–97.

Baker, Carl L. 1995. Contrast, discourse prominence, and intensification, with special reference to locally free reflexives. Language 71:63–101.

Baker, Mark. 2008. The Syntax of Agreement and Concord. Cambridge: Cambridge University Press.

50

Baker, Mark, Johnson, Kyle, and Ian Roberts. 1989. Passive arguments raised. Linguistic Inquiry 20:219–252.

Baker, Mark C., and Nadya Vinokurova. 2010. Two modalities of case assignment: case in Sakha. Natural Language and Linguistic Theory 28:593–642.

Barðdal, Jóhanna, and Valeria Molnár. 2003. Passive in Icelandic – compared to Mainland Scandinavian. In Structures of Focus and Grammatical Relations, eds. Jorunn Hetland, and Valeria Molnár, 231–260. Tübingen: Niemeyer.

Behagel, Otto. 1924. Deutsche Syntax. Eine geschichtliche Darstellung. Volume II. Heidelberg: Winter.

Bjorkman, Bronwyn M. 2011. BE-ing Default: the Morphosyntax of Auxiliaries. Dissertation, MIT.

Bierwisch, Manfred. 2006. German Reflexives as Proper and Improper Arguments. In Form, Structure, and Grammar: A Festschrift Presented to Günther Grewendorf on Occasion of his 60th Birthday, eds. Patrick Brandt, and Eric Fuß, 15–36. Berlin: Akademie-Verlag.

Bobalijk, Jonathan D. 2008. Where’s ϕ ? Agreement as a Post-Syntactic Operation. In Phi-Theory: Phi-features across interfaces and modules, eds. Daniel Harbour, David Adger, and Susana Béjar, 295–328. Oxford University Press.

Bouma, Gosse, and Jennifer K. Spenader. 2009. Strong and weak reflexives in Dutch. Proceedings of the 7th International Workshop on Treebanks and Linguistic Theories, eds. Frank Van Eynde, Anette Frank, Koenraad De Smedt, and Gerzjan van Noord, 103–114. Utrecht: LOT Publication.

Brame, Michael. 1983. Bound Anaphora is not a Relation between NPs: Evidence for Local Word Grammar (Without Trees), Linguistic Analysis 11:139–166.

Burzio, Luigi. 1991. The morphological basis of anaphora. Journal of Linguistics 27:81–105. Burzio, Luigi. 1998. Anaphora and Soft Constraints. In Is the Best Good Enough? Optimality

and Competition in Syntax, eds. Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky, 93–114. Cambridge, MA: MIT Press.

Cinque, Guglielmo. 1988. On Si constructions and the theory of Arb, Linguistic Inquiry 19:521–581.

Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 2000. Minimalist Inquiries: The Framework. In Step by Step: Essays on

Minimalist Syntax in Honor of Howard Lasnik, eds. Roger Martin, David Michaels, and Juan Uriagereka, 89–155. Cambridge, MA: MIT Press.

Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in Language, ed. Michael Kenstowicz, 1–52. Cambridge, MA: MIT Press.

Chomsky, Noam. 2008. On Phases. In Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud, eds. Robert Freidin, Carlos P. Otero, and Maria L. Zubizarreta, 133–166. Cambridge, MA: MIT Press.

Collins, Chris. 2005. A smuggling approach to the passive in English. Syntax 8: 81–120. D’Alessandro, Roberta. 2007. Impersonal “si” constructions. Berlin: Mouton de Gruyter. Doron, Edith, and Malka Rappaport Hovav. 2007. Towards a Uniform Theory of Valence-

changing Operations. Proceedings of IATL 23. Embick, David. 2004. Unaccusative Syntax and Verbal Alternations. In The unaccusativity

puzzle: explorations of the syntax-lexicon interface, eds. Aartemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert, 137–158. Oxford: Oxford University Press.

Everaert, Martin. 1990. Nominative Anaphors in Icelandic: Morphology or Syntax? In Germanic Syntax Workshop, eds. Werner Abraham, and Eric Reuland, 277–307. Amsterdam: John Benjamins.

Eythórsson, Thórhallur. 2007. Reflexive passives in Icelandic. Handout of a talk given at NORMS Workshop on Pronouns, Binding and Anaphors, University of Iceland.

51

Eythórsson, Thórhallur. 2008. The New Passive in Icelandic really is a passive. In Grammatical Change and Linguistic Theory: The Rosendal Papers, ed. Thórhallur Eythórsson, 173–219. Amsterdam: John Benjamins.

Fanselow, Gisbert. 1987. Konfigurationalität. Tübingen: Narr. Fanselow, Gisbert. 1989. Konkurrenzphänomene in der Syntax: eine nicht-pragmatische

Reduktion der Prinzipien B und C der Bindungstheorie. Linguistische Berichte 123:385–414

Fanselow, Gisbert. 1991. Minimale Syntax. Habilitationsschrift, University of Passau. Groninger Arbeiten zur germanistischen Linguistik (GAGL) 32.

Fischer, Silke. 2004. Towards an Optimal Theory of Reflexivization. PhD thesis, University of Tübingen.

Fischer, Silke. 2006. Matrix Unloaded: Binding in a Local Derivational Approach. Linguistics 44:913–935.

Frampton, John, and Sam Gutmann. 2000. Agreement is Feature Sharing. Ms., Northeastern University.

Frey, Werner. 1993. Syntaktische Bedingungen für die semantische Interpretation. Über Bindung, implizite Argumente und Skopus. Berlin: Akademie-Verlag.

Geniušienė, Emma. 1987. The typology of reflexives. Berlin: Mouton de Gruyter. Geurts, Bart. 2004. Weak and strong reflexives in Dutch. In Proceedings of the ESSLLI

workshop on semantic approaches to binding theory, Nancy, France. Greenbaum, Sidney. 1977. Judgements of syntactic acceptability and frequency. Studia

Linguistica 31:83–105. Grewendorf, Günther. 1988. Aspekte der deutschen Syntax. Tübingen: Narr. Grimshaw, Jane. 1981. On the lexical representation of Romance reflexivie clitics. In The

Mental Representation of Grammatical Relations, ed. Joan Bresnan, 87–148. Cambridge, MA: MIT Press.

Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA: MIT Press. Haider, Hubert. 1985. Von sein oder nicht sein: Zur Grammatik des Pronomens “sich”. In

Erklärende Syntax des Deutschen, ed. Werner Abraham, 223–254. Tübingen: Narr. Haider, Hubert. 2000. The license to license. In Arguments and Case. Explaining Burzio’s

Generalization, ed. Eric Reuland, 31–54. Amsterdam: John Benjamins. Haider, Hubert. 2001. How to stay accusative in insular Germanic. Working Papers in

Scandinavian Syntax 68:1–14. Halle, Morris, and Alec Marantz. 1993. Distributed Morphology and the Pieces of Inflection.

In A View From Building 20, eds. Ken Hale, and Samuel J. Keyser, 111–176. Cambridge, MA: MIT Press.

Haspelmath, Martin. 2008. A Frequentist Explanation of Some Universals of Reflexive Marking. Linguistic Discovery 6:40–63.

Heinat, Fredrik. 2006. Probes, pronouns, and binding in the Minimalist Program. PhD thesis, Lund University.

Hendriks, Petra, Jennifer Spenader, and Eric-Jan Smits. 2008. Frequency-based constraints on reflexive forms in Dutch. In Proceedings of the 5th International Workshop on Constraints and Language Processing (CSLP 2008), eds. Jorgen Villadsen, and Henning Christiansen, 33-47. Computer Science Research Reports, #122, Roskilde University, Denmark.

Hole, Daniel. 2006. Agentive selbst and other instantiations of the identity function in German. Ms., University of München.

Hole, Daniel. 2008. Focus on identity – The dark side of zìjĭ. The Linguistic Review 25:267–295.

Holmberg, Anders. 2002. Expletives and Agreement in Scandinavian Passives. Journal of Comparative Germanic Linguistics 4:85–128.

52

Hundt, Markus. 2002. Formen und Funktionen des Reflexivpassivs im Deutschen. Deutsche Sprache 30:124–166.

Jónsson, Jóhannes G. 2009. The new impersonal as a true passive. In Advances in Comparative Germanic Syntax, eds. Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger, and Florian Schäfer, 281–306. Amsterdam: John Benjamins.

Jónsson, Jóhannes G. 2011. Reflexive sig is an argument. Nordlyd 37:99–118. Kaufmann, Ingrid. 2001. Medium: Eine Studie zur Verbsemantik. Habilitationsschrift,

University of Düsseldorf. Kayne, Richard S. 1975. French Syntax. The Transformational Cycle. Cambridge, MA: MIT

Press. Kemmer, Suzanne. 1993. The middle voice. Amsterdam: John Benjamins. Kiparsky, Paul. 2002. Disjoint Reference and the Typology of Pronouns. In More than

Words: A Festschrift for Dieter Wunderlich, eds. Ingrid Kaufmann, and Barbara Stiebels, 179–226. Berlin: Akademie Verlag.

Kiss, Tibor. 2001. Anaphora and Exemptness. A comparative treatment of anaphoric binding in German and English. In The Proceedings of the 7th International Conference on Head-Driven Phrase Structure Grammar, eds. Dan Flickinger, and Andreas Kathol, 182–197. CSLI Publications.

Kiss, Tibor. 2003. Die Genese der Ausnahmeanapher. In Arbeiten zur Reflexivierung, eds. Lutz Gunkel, Gereon Müller, and Gisela Zifonun, 157–188. Tübingen: Narr.

Koenig, Ekkehard, and Letizia Vezzosi. 2004. The role of predicate meaning in the developement of reflexivity. In What makes Grammaticalization? A Look from its Fringes and its Components, eds. Walter Bisang, Nikolaus Himmelmann, and Björn Wiemer, 213–244. Berlin: Mouton de Gruyter.

Kratzer, Angelika. 1996. Severing the External Argument from its Verb. In Phrase Structure and the Lexicon, eds. Johan Rooryck, and Laurie A. Zaring, 109–137. Dordrecht: Kluwer.

Kratzer, Angelika. 2009. Making a Pronoun: Fake Indexicals as Windows into the Properties of Pronouns. Linguistic Inquiry 40:187–237.

Labelle, Marie. 2008. The French Reflexive and Reciprocal se. Natural Language and Linguistic Theory 26:833–876.

Landau, Idan. 2010. The Explicit Syntax of Implicit Arguments. Linguistic Inquiry 41:357–388.

Legate, Julie A. 2005. Phases and Cyclic Agreement. In Perspectives on Phases (MITWPL 49), eds. Martha McGinnis, and Norvin Richards. Cambridge, MA: MIT Press.

Lopez, Luis. 2008. The [person] restriction: why and why not. In Agreement Restrictions, eds. Roberta D’Alessandro, Gunnar Hrafbjargarsson, and Susann Fischer, 129–157. Berlin: Mouton de Gruyter.

Maling, Joan. 1984. Non-clause-bounded reflexives in modern Icelandic. Linguistics and Philosophy 7:211–241.

Maling, Joan. 2002. Það rignir þágufalli á Íslandi: verbs governing dative objects. Íslenskt mál 24:31–105.

Maling, Joan. 2006. From passive to active: Syntactic change in progress in Icelandic. In Demoting the agent: Passive, Middle and Other Voice Phenomena, eds. Benjamin Lyngfelt, and Torgrim Solstad, 197–223. Amsterdam: John Benjamin.

Maling, Joan. & Sigrídur Sigurjónsdóttir. 2002. The new impersonal construction in Icelandic. Journal of Comparative Germanic Linguistics 5:97–142.

Marantz, Alec. 2000. Case and Licensing. In Arguments and case: explaining Burzio’s Generalization, ed. Eric Reuland, 11–30. Amsterdam: John Benjamins.

Marantz, Alec. to appear. Phases and Words. MS, NYU. McFadden, Thomas. 2004. The Position of Morphological Case in the Derivation: a study on

the syntax-morphology interface. PhD thesis. University of Pennsylvania.

53

McGinnis, Martha. 1998. Reflexive external arguments and lethal ambiguity. In Proceedings of WCCFL 16, eds. Emily Curtis, James Lyle, and Gabriel Webster, 303–317. Stanford: CSLI Publications.

McGinnis, Martha. 2000. Reflexive clitics and the specifiers of vP. In Papers from the UPenn/MIT Round Table on the Lexicon (MITWPL 35), eds. Liina Pylkkänen, Heidi Harley, and Angeliek van Hout, 137–160. MIT Working Papers in Linguistics.

McGinnis, Martha. 2004. Lethal Ambiguity. Linguistic Inquiry 35:47–95. Müller, Gereon. 2009. Ergativity, accusativity, and the order of Merge and AGREE. In

Explorations of Phase Theory: Features and Arguments, ed. Kleanthes K. Grohmann, 269–308. Berlin: Mouton de Gruyter.

Müller, Stefan. 1999. Deutsche Syntax deklarativ. Head Driven Phrase Structure Grammar für das Deutsche. Tübingen: Niemeyer.

Müller Gereon, and Wolfgang Sternefeld. 1993. Scrambling as A-bar movement. In Studies on Scrambling. Movement and Non-Movement Approaches to Free Word Order Phenomena, eds. Norbert Corver, and Henk van Riemsdijk, 331–385. Berlin: Mouton de Gruyter.

Murphy, Brian. 2007. A Study of Notions of Participation and Discourse in Argument Structure Realisation. Doctoral Thesis, Trinity College Dublin.

Nunberg, Geoffray, Sag, Ivan A., and Thomas Wasow. 1994. Idioms. Language 70: 491–538. Pestsky, David, and Esther Torrego. 2007. The Syntax of Valuation and the Interpretability of

Features. In Phrasal and Clausal Architecture: Syntactic Derivation and Interpretation, eds. Simin Karimi, Vida Samiian, and Wendy K. Wilkins, 262–294. Amsterdam: John Benjamins.

Plank, Frans. 1993. Peculiarities of Passives of Reflexives in German. Studies in Language 17:135–167.

Platzack, Christer. 2006. Case as AGREE Marker. Working Papers in Scandinavian Syntax 77:71–99.

Pylkkänen, Liina. 2008. Introducing Arguments. Cambridge, MA: MIT Press. Reed, Lisa A. 2011. Get-passives. The Linguistic Review 28:41–78. Reinhart, Tanya. 2000. The Theta System: syntactic realization of Verbal concepts. OTS working

papers TL-00.002, Utrecht University. Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24:657–720. Reinhart, Tanya, and Tal Siloni. 2004. Against the Unaccusative Analysis of Reflexives. In

The Unaccusativity Puzzle, eds. Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert, 288–331. Oxford: Oxford University Press.

Reinhart, Tanya, and Tal Siloni. 2005. The Lexicon-Syntax Parameter: Reflexivization and Other Arity Operations. Linguistic Inquiry 36:389–436.

Reis, Marga. 1982. Reflexivierung im Deutschen. In Actes du Colloque du Centre de Recherches Germaniques de l’Université de Nancy & Journée Annuelle des Linguistes de l’Association des Germanistes de l’Enseignement Supérieur 12, ed. E. Faucher, 1–40. Nancy.

Reuland, Eric. 2001. Primitives of Binding. Linguistic Inquiry 32:439–492. Reuland Eric. 2005. Agreeing to bind. In Organizing Grammar: Linguistic Studies in Honor

of Henk van Riemsdijk, eds. Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz, and Jan Koster, 505–513, Berlin: Mouton de Gruyter.

Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17:501–558.

Rizzi, Luigi. 1990. On the anaphor-agreement effect. Rivista di Linguistica 2:27–42. Ruys, Eddy. 2010. Expletive selection and CP arguments in Dutch. Journal of Comparative

Germanic Linguistics 13:141–178.

54

Schäfer, Florian. 2008. The Syntax of (Anti-)Causatives. External arguments in change-of-state contexts. Amsterdam: John Benjamins.

Schäfer, Florian. to appear. Local Case, Cyclic AGREE and the Syntax of truly Ergative Verbs. In Local Modelling of Non-Local Dependencies in Syntax, eds. Artemis Alexiadou, Tibor Kiss, and Gereon Müller. Tübingen: Niemeyer.

Sells, Peter, Zaenen, Annie, and Draga Zec. 1987. Reflexivization variation: Relations between syntax, semantics, and lexical structure. In Working Papers in Grammatical Theory and Discourse Structure: Interactions of Morphology, Syntax, and Discourse, eds. Masayo Iida, Stephen Wechsler, and Draga Zec, 169–238. Stanford: CSLI Publications.

Sigurðsson, Halldór A. 1989. Verbal Syntax and Case in Icelandic. Doctoral dissertation, University of Lund.

Sigurðsson, Halldór A. 2000. The locus of case and agreement. The locus of case and agreement. Working Papers in Scandinavian Syntax 65:65–108.

Sigurðsson, Halldór A. 2003. Case: abstract vs. morphological. In New Perspectives on Case Theory, eds. Ellen Brandner, and Heike Zinsmeister, 223–268. Stanford: CSLI Publications.

Sigurðsson, Halldór A. 2005. Accusative and the Nom/Acc alternation in Germanic. Working Papers in Scandinavian Syntax 76:93–133.

Sigurðsson, Halldór A. 2006. The nominative puzzle and the low nominative hypothesis. Linguistic Inquiry 37:289–308.

Sigurðsson, Halldór A. 2009. The No Case generalization. In Advances in Comparative Germanic Syntax, Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger, and Florian Schäfer, 249–280. Amsterdam: John Benjamins.

Sigurðsson, Halldór A. 2011. On the new passive. Syntax 14:148–178. Sigurðsson, Halldór Á., and Verner Egerland. 2009. Impersonal null-subjects in Icelandic and

elsewhere. Studia Linguistica 63:158–185. Steinbach, Markus. 2002. Middle Voice. Amsterdam: John Benjamins. Sternefeld, Wolfgang. 2006. Syntax. Eine morphologisch motivierte generative Beschreibung

des Deutschen. Tübingen: Stauffenberg. Svenonius, Peter. 2002. Icelandic Case and the Structure of Events. Journal of Comparative

Germanic Linguistics 5:197–225 Svenonius, Peter. 2006. Case alternations and the Icelandic passive and middle. In Passives

and Impersonals in European Languages. eds. Satu Manninen, Katri Hiietam, Elsi Kaiser, and Virve Vihman. Amsterdam: John Benjamins.

Szucsich, Luka. 2007. Nothing Wrong with Finite T: Non-Agreeing Accusative Impersonal Sentences. In Formal Approaches to Slavic Linguistics 15, The Toronto Meeting 2006, eds. Magda Golędzinowska, Ulyana Savchenko, and Richard Compton, 417–435. Ann Arbor: Michigan Slavic Publications.

Thráinsson, Höskuldur. 2007. The Syntax of Icelandic. Cambridge: Cambridge University Press.

Tucker, Matthew A. 2010. Even more on the anaphor agreement effect: When binding does not agree. Ms., University of California, Santa Cruz.

Vater, Heinz. 1995. Zum Reflexiv-Passiv im Deutschen. In Deutsch als Fremdsprache: an den Quellen eines Faches. Festschrift für Gerhard Helbig zum 65. Geburtstag, ed. Heidrun Popp, 185–192. München: Iudicium.

Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: Oxford University Press.

Wiemer, Björn. 2006. Relations between Actor-demoting devices in Lithuanian. In Passivization and Typology (Form and Function), eds. Werner Abraham, and Larisa Leisiö, 274–309. Amsterdam: John Benjamins.

Wood, Jim. 2012. Icelandic Morphosyntax and Argument Structure. Dissertation, NYU.

55

Wood, Jim. submitted. Reflexive -st verbs in Icelandic. Woolford, Ellen. 1997. Four-Way Case Systems: Ergative, Nominative, Objective and

Accusative. Natural Language and Linguistic Theory 15:181–227. Woolford, Ellen. 1999. More on the Anaphor-Agreement Effect. Linguistic Inquiry 30:257–

287. Wunderlich, Dieter. 1985. Über die Argumente des Verbs. Linguistische Berichte 97:183–

227. Wurmbrand, Susi. to appear. The syntax of valuation in auxiliary–participle constructions. In

Coyote Working Papers: Proceedings of the 29th West Coast Conference on Formal Linguistics (WCCFL 29). Tucson: University of Arizona.

Yip, Moira, Maling, Joan, and Ray Jackendoff. 1987. Case in tiers. Language 63:217–250. Zaenen, Annie, Maling, Joan, and Höskuldur Thráinsson. 1985. Case and Grammatical

Functions: The Icelandic Passive. Natural Language and Linguistic Theory 3:441–483. Zeijlstra, Hedde. in press. There is only one way to agree. The Linguistic Review.