This article shows how a systematic impersonalization alternation in Russian provides additional evidence for underspecification in argument structure. In the case of a large class of lexically causative verbs, the causer is realized either as a volitional Agent in the nominative case or as an oblique-marked, nonvolitional causer, depending on how the event is construed. A causative theory of accusative is advanced, according to which the mere presence of external causation is a sufficient condition for accusative licensing, including those cases that lack an external argument altogether. The analysis is extended to explain accusative preservation in the Icelandic “fate accusative” construction.
1 Lexically Causative Verbs in English
It has been observed that certain lexically causative verbs in English occur with a range of subject arguments, including volitional Agents, Instruments, and Natural Forces, as in (1).
The storm / Max / The stone broke the window.
[Natural Force / Agent / Instrument]
The heat / Max / The candle melted the ice.
[Natural Force / Agent / Instrument]
According to Reinhart (2002:233–235), the subjects in (1) minimally bear the feature [+c]—meaning they are causers, but the kind of causation is crucially not fixed in the lexicon (see Levin and Rappaport Hovav 1995:92–107, where such verbs are defined as “externally caused”). These verbs are distinguished from those that bear an additional [+m] feature in Reinhart’s system ([+c+m]), where [+m] indicates a particular “mental state,” which we take to mean volitional human participation, as indicated in (2a) and (2b).
Verbs that are [+c+m] like shave and assassinate/murder are necessarily agentive and, for this reason, do not admit the range of subjects illustrated in (1). Levin and Rappaport Hovav (1995: 106–110) observe that it is precisely verbs like those in (1) that enter into the causative alternation in English (see also Haspelmath 1993, Reinhart 2002, Folli and Harley 2005, Levin and Rappaport Hovav 2005, Alexiadou, Anagnostopoulou, and Schäfer 2006, 2015, Schäfer 2008, 2012, Martin and Schäfer 2014, and Cuervo 2015 for much additional discussion). Observe, for example, the difference between English peel and break. Peel is not entirely underspecified—it selects either an Agent or Instrument subject, but not a Natural Force (Reinhart 2002), and as a result does not form an anticausative, in contrast to break, as shown in (3)–(4).
Max / The knife peeled the apple.
*The heat peeled the apple.
*The apple peeled.
The vandals / The rocks / The storm broke the window. (cf. (1a))
The window broke.
Thus, underspecification for the source of causation in English (and elsewhere) affects a verb’s range of possible subjects ((1)–(2)) and its ability to participate in the causative alternation ((3)–(4)). We show in section 2 that variable realization of the causer in lexically causative verbs in Russian explains a type of causative alternation that is not possible in English.
2 The Russian Transitive Impersonal
In this article, we introduce a new argument for variable causer realization, which has received scant attention in the generative literature. A broad class of Russian two-place, lexically causative verbs freely project their causer argument either as a nominative agentive subject (with a sentient, volitional causer reading) or as an oblique causer (typically giving “out-of-human-control” semantics). The appearance of an oblique causer gives rise to a transitive impersonal, a construction in which accusative appears in the absence of an Agent or any nominative argument (Babby 1994, 2010, Markman 2004, Lavine 2010, 2016). Transitive impersonals alternate with transitive agentive predicates, as illustrated in (5)–(6) from Russian.
V Kieve nacionalisty razbili okna otdelenija Sberbanka
in Kiev nationalists.NOM.PL broke.PL windows.ACC of.branch of.bank
‘In Kiev nationalists broke the windows of a branch of the Bank of Russia.’
(Rossijskaja gazeta, 22 June 2014)
Šxunu razbilo volnami.
schooner.ACC broke.IMP waves.INST
‘The schooner was smashed by the waves.’
(Vasilii Golovanov, Ostrov; Russian National Corpus)
Gospod’ uslyšal moi molitvy i rastopil ledyšku v tvoej grudi.
lord heard my prayers and melted.M.SG icicle.ACC in your chest
‘The Lord heard my prayers and melted the icicle in your heart.’
(Ekaterina Koljučkina, “Ty prišel”)
Ledyšku rastopilo solncem.
icicle.ACC melted.IMP sun.INST
‘The icicle was melted by the sun. / The icicle melted due to the sun.’
This article pursues the hypothesis that Reinhart’s (2002) [+c] feature, broadly construed as realizing a causative subevent in the syntax, is a sufficient condition for accusative in some languages (regardless of the presence or absence of a higher nominative argument). The [+c] feature is a complex of θ-roles: Agent, Instrument, or unspecified cause. It represents the underspecification of a lexical entry for the source of causation, which, as we have shown, determines subject selection and argument structure alternation possibilities in English. Transitive impersonals in Russian occur with accusative on the Theme, in the absence of an Agent, so long as the predicate is causative (as indicated by the oblique Natural Force arguments in (5b) and (6b)). In informal terms, the impersonalization alternation in Russian replaces the prototypical subject (Agent) with a VP-internal argument that names a Natural Force or causative event.2 Thus, while the impersonalization alternation does not alter the basic valency of the predicate, this rearrangement of arguments determines whether or not an Agent will be projected, so it follows that the impersonalization alternation is a Voice alternation (see Alexiadou, Anagnostopoulou, and Schӓfer 2015, where the causative alternation is likewise treated as a Voice alternation).3
If the argument structure for a verbal predicate specifies that the causer must be realized as an Agent ([+c+m] in Reinhart’s terms), then the transitive impersonal is blocked. This is illustrated by Russian (7b), in contrast to (8b).
Ustav ot boli, mužčina sam amputiroval sebe
having.grown.tired from pain man.NOM.M.SG by.self amputated.M.SGSELF.DAT
nogu stolovym nožom.
leg.ACC table knife.INST
‘Having grown sick of the pain, a man amputated his leg with a kitchen knife.’
(Vesti, 15 October 2013)
*Emu amputirovalo nogu.
him.DAT amputated.IMP leg.ACC
(Intended: ‘His leg somehow became amputated (not by a human agent).’)
Bezrabotnyj avstriec otrezal sebe nogu.
unemployed Austrian.NOM.M.SG severed.M.SGSELF.DAT leg.ACC
‘An unemployed Austrian cut off his leg.’
(Argumenty i fakty, 27 March 2012)
Emu otrezalo nogu.
him.DAT severed.IMP leg.ACC
‘His leg was severed (not by a human agent).’
In the ungrammatical example (7b), the verb amputirovat’ ‘amputate’ is necessarily agentive; it has no nonvolitional usage, in contrast to otrezat’ ‘sever’ in (8). For example, (8b) is appropriate in describing a car or train accident, which are decidedly out-of-human-control events. Since the argument structure for amputirovat’ is fully specified in the lexicon (like English shave and assassinate), the impersonalization alternation is unavailable.
Note that Russian also has transitive verbs like tošnit’ ‘to experience nausea’, which are lexically specified never to occur with an external argument (or subject of any kind). Such verbs permit only a nonvolitional causer (the source of the nausea) and, as a result, they occur exclusively as transitive impersonals, as in (9).4
Russian: Transitive impersonal
Ego tošnilo ot zapaxa.
he.ACC nauseated.IMP from smell.GEN
‘He became nauseated from the smell.’
It follows that (7) and (9) do not participate in the impersonalization alternation for the same reason: their argument structure is fixed in the lexicon (amputirovat’ ‘amputate’ is fixed as [+c+m], and tošnit’ ‘to experience nausea’ is fixed as [+c−m]).5 Thus, there are three main types of externally caused transitive verbs in Russian: those that are obligatorily agentive (7); those that obligatorily select a nonvolitional causer (9); and those whose type of causation is underspecified in argument structure (8), which gives rise to the impersonalization alternation in Russian.6
It is important to note the crucial role that causation plays in our analysis with respect to the unexpected licensing of accusative in transitive impersonals. Compare, for example, “pure unaccusatives,” such as those described by Levin and Rappaport Hovav (1995:90–101) as “internally caused.” Pure unaccusatives do not form transitive impersonals and never assign accusative to their sole argument (in any language, as far as we know). The Russian verbal predicate meaning ‘freeze’, for example, comes in two varieties: one externally caused and one internally caused. Externally caused zamorozit’ in (10) occurs, as expected, as a transitive impersonal (10b). In the case of zamerznut’ in (11), the internally caused counterpart to zamorozit’, the transitive impersonal fails (11a), and nominative occurs on the verb’s sole argument (11b).
Mama zamorozila mjaso.
mom.NOM.F.SG froze.F.SG meat.ACC
‘Mom froze the meat.’
Počemu Evropu zamelo, a Rossiju zamorozilo?
why Europe.ACC covered.IMP and Russia.ACC frozen.IMP
‘Why was Europe covered with snow, whereas Russia was frozen over?’
(utro.ru, 11 January 2010)
Monadic (“pure”) unaccusative
‘The river froze up.’
Note that on a causative theory of accusative, which is discussed below, zamelo ‘covered with snow’ and zamorozilo ‘froze over’, in (10b), must then be two-place predicates, even if we cannot name the Natural Force responsible for the events they describe. In any case, we know that these verbs are dyadic (cf. (10a) for zamorozit’) and that no valency-reducing operation has applied. This illustrates the variable mapping of the causer argument in externally caused verbal predicates: it can occur as a nominative Agent, occur as an oblique VP-internal argument, or remain unstated, but nonetheless be interpreted both semantically (it introduces a causative subevent) and syntactically (it is responsible, we will claim, for the appearance of accusative on the Theme).7 We assume a parsimonious lexicon, whereby the transitive agentive and transitive impersonal alternates involve a single root verb participating in two distinct event types, each determined by different functional projections (Folli and Harley 2005).
Note that Alexiadou, Anagnostopoulou, and Schafer 2006, 2015 and Schäfer 2008 argue that anticausatives, such as English (12) and German (13), are also causative since they license causer PPs that indicate the presence of a [cause] head. Such PPs are introduced by English from and German durch (while by/von, which introduce Agents, and with/mit, which introduce Instruments, are ungrammatical with anticausatives).
(12) The window cracked *by / *with / from the pressure.
Die Vase zerbrach *von Peter / *mit dem Hammer / durch ein Erdbeben.
the vase.NOM broke by Peter with the hammer through an earthquake
‘The vase broke as a result of an earthquake.’
By way of contrast, the corresponding transitive impersonal in Russian, (14a), is two-place and transitive. Observe that plotina ‘dam’ appears in the accusative and that the causer (Natural Force) zemletrjasenie ‘earthquake’ is realized as an argument, with the syntactic effect of introducing a causative subevent, which we argue below to be responsible for the transitivity property. Note, however, that the causer argument is not necessarily Natural Force. The only requirement imposed by the transitive impersonal is that the causer be outside of human control. For example, in (14b) the causer is a moth, in (14c) it is a tram, and in (14d) it is an economic event.
(14) Russian: transitive impersonal8
Zemletrjaseniem povredilo plotinu.
earthquake.INST damaged.IMP dam.ACC
‘A dam was damaged due to an earthquake.’
(Tatyana Solomatina, Otojti v storonu i posmotret’; Russian National Corpus)
Mol’ju iz’’jelo plat’e.
moth.INST eat.through.IMP dress.ACC
‘A dress was eaten through by moths.’
Grabitelja v Taganroge razdavilo tramvaem.
robber.ACC in Taganrog crushed.IMP tram.INST
‘A robber in Taganrog was crushed by a tram.’
(Biznes gazeta, 12 October 2015)
Èstoniju zaxlestnulo volnoj nekačestvennogo benzina.
Estonia.ACC engulfed.IMP wave.INST low.quality gasoline.GEN
‘Estonia was swamped with a wave of low-quality gasoline.’
We argue below that v[cause] probes accusative. It follows that admitting a v[cause] head into the structure of anticausatives falsely predicts the appearance of accusative in the one-place predicates in (12) and (13) (see (11)). We assume, instead, following Cuervo (2015:414–415), that English from and German durch modify the “dynamic event of change,” Cuervo’s vGO (or Ramchand’s (2008) process head), rather than the head that initiates the event or sets the event in motion.9
3 v Heads and “Independent” Accusative
We assume that the Agent is not part of the lexical entry of verbal roots (Kratzer 1996); instead, it may or may not be projected by a Voice head, depending on the compatibility of an Agent with the eventuality described. Lexically causative verbs are minimally specified as [+c], as discussed above. The variability seen in the syntactic projection of the causer argument is the principal concern of this article. We take the appearance of accusative case in the absence of an active Voice head to indicate that the features [voice] and [cause] operate independently, in the form of two discrete v heads, as indicated in the “split vP” in (15), following Pylkkänen 2008.10 If an Agent is not projected, as in the case of transitive impersonals, v[voice] is inactive and absent in the syntax altogether (indicated in (15) by shading).
In the event that [voice] is not argument-projecting ([−voice]), accusative appears, so long as the event is caused ([+cause]). Voice in (15) is “unbundled” in Pylkkänen’s terms (see also Bowers 2002, Folli and Harley 2005, Alexiadou, Anagnostopoulou, and Schäfer 2006, Kallulli 2007, Harley 2013).11 In (14), v[cause] is activated by the presence of oblique nonvolitional causers, which are construed as initiating an out-of-human-control event in the absence of an active [voice] head (i.e., an Agent).12 In the case of lexically causative verbs, v[cause] is active as an accusative probe, regardless of the setting for [voice], so long as the features [voice] and [cause] operate independently. This is precisely the arrangement of v heads that gives the impersonal transitive passive in Ukrainian (Lavine 2013) and the Japanese adversity causative (Pylkkänen 2008:89–92), as well as the Russian transitive impersonal and, as suggested below, the Icelandic fate accusative construction.13
Recall that transitive impersonals are ruled out in contexts that are not causative, such as basic unaccusatives (11). In such cases, v[cause] is not activated and accusative does not appear. This predicts that the sort of independent accusative that we describe will never occur, in any language, with internally caused “pure” unaccusatives. Observe how different event structures are built via the interaction of lexical roots and functional heads. A breaking or melting event can be construed as caused by a sentient, volitional Agent, merged in Spec,v[voice] (as in (5a)/ (6a)) or as set in motion by a nonvolitional causer argument (e.g., a Natural Force or Instrument), thereby activating v[cause] (as in (5b)/(6b)) in the absence of a [voice] head, giving the transitive impersonal. Note that the approach taken here supports much recent work in the constructionist approach to verbal meaning, whereby different event construals are built via the interaction of lexical roots and functional heads. In this way, multiple meanings are generated in the syntax on the basis of a single verbal root, potentially giving rise to systematic alternations (see, e.g., Marantz 1997, Borer 2005, Folli and Harley 2005, Alexiadou, Anagnostopoulou, and Schäfer 2006, 2015, Schäfer 2008, Harley 2013, Wood 2015).
4 The Icelandic Fate Accusative
Our analysis of the transitive impersonal in Russian suggests a potential explanation of the “fate accusative” construction in Icelandic, exemplified in (16) (for additional discussion, see Zaenen and Maling 1984, Ottósson 1988, Jónsson 2003, Sigurðsson 2005, 2006, 2011, Svenonius 2005, Schäfer 2008, Lavine 2014, Wood 2017).
(16) Icelandic: Fate accusative
Bátinn rak á land.
boat.the.ACC drove to land
‘The boat drifted ashore.’
Strompinn blés af húsinu.
chimney.the.ACC blew off house.the
‘The chimney blew off the house.’
Bátinn fyllti á augabragði.
boat.the.ACC filled in flash
‘The boat swamped immediately.’
We argue here that the fate accusative is an Icelandic version of the transitive impersonal construction: a two-place predicate in which accusative is assigned to the object in the absence of an Agent subject.14 At first blush, the licit examples of the fate accusative in (16) superficially resemble the ungrammatical Russian example (11a) in occurring without a causer. However, note that the reading in (16) is not anticausative, as shown in (17), in which accusative on the sole argument is ungrammatical with the “middle” suffix -st.15
The fate accusative also resists ‘by itself ’ modification, which is acceptable in anticausatives.16
Icelandic: Fate accusative
Strompinn blés af húsinu (*af sjálfumsér).
chimney.the.ACC blew off house.the of self
(Intended: ‘The chimney blew off the house (on its own).’)
The transitive impersonal in Russian is likewise incompatible with ‘by itself ’ modification, as indicated in (19).
Russian: Transitive impersonal
Reku zamorozilo (*samu po sebe).
river.ACC froze.IMP by itself
(Intended: ‘The river froze (on its own).’)
The incompatibility of the fate accusative with anticausative morphology and ‘by itself ’ modification strongly suggests that the Icelandic construction, like its Russian counterpart, is two-place and externally caused, even if the causer argument is left unexpressed in Icelandic.17 The Russian transitive impersonal can also occur without the overt expression of its causer (as in (10b)), which is always present in the argument structure, but sometimes left unexpressed for the simple reason that it cannot be identified by the speaker, as in (20).18
(20) Russian: Transitive impersonal
(Letčik . . . sbrosil tysjaču gallonov topliva) i samolet srazu
(the pilot dumped a thousand gallons of fuel) and plane.ACC immediately
‘The pilot dumped a thousand gallons of fuel and the plane immediately rose.’
(Mašina vnezapno ostanovilas’) i menja brosilo vpered.
(the car suddenly stopped) and me.ACC threw.IMP forward
‘The car suddenly stopped and I was thrown forward.’
Again, note that the predicates in (20a–b) are not construed by native speakers as anticausative. Russian potjanulo ‘pulled’ and brosilo ‘threw’ are ordinary externally caused, transitive verbs.
Recall Reinhart’s (2002) Theta System, in which traditional θ-roles are formally encoded in terms of [±c] (cause change) and [±m] (mental state or sentience) and the lexical entry of arguments is stated in terms of how these features cluster. Note crucially that, as in any potential transitive impersonal, the causer of potjanulo ‘pulled’ and brosilo ‘threw’ is listed merely as [+c] in argument structure; it is unspecified for mental state (i.e., volition), which makes it compatible with any kind of causation, including Natural Forces, Instruments, and ambient conditions. Natural Force is [+c−m], which can also be realized as an Instrument. The exact reading is contextually determined. The Instrument reading involves an implied Agent, which manipulates the Instrument, in contrast to the Natural Force reading, which is entirely incompatible with an Agent (Reinhart 2002:279–280). The idea is that the [+c] cluster gives “greater interpretive freedom” (Reinhart 2002:232), which in turn gives rise to variable argument realization via further specification of the [m] feature. The transitive agentive alternate, as in the (a) examples of (5)–(8) in Russian and of (21)–(22) below in Icelandic, is formed by adding [+m] to the [+c] cluster, identifying the causer as sentient ([+c−m]). The Russian transitive impersonal and Icelandic fate accusative are formed by specifying [−m], giving the [+c−m] cluster, denoting, in this case, Natural (or ambient) Force.19 On our account, this greater interpretive freedom—what drives the impersonalization alternation—is realized syntactically in terms of whether [voice] is activated in conjunction with [cause] or whether [cause] operates independently of [voice] (i.e., an Agent), assuming, as described above, the potential “unbundling” of the [voice] and [cause] v heads, as motivated in Pylkkänen 2008.
The transitive impersonal and fate accusative constructions occur only with those externally caused verbs that are minimally specified as [+c], leaving the volitional status of the causer argument ([m]) unvalued. This captures Ottósson’s (1988:148) intuition regarding the Icelandic fate accusative: “[T]hese verbs are marked for accusative subject only when the agent side argument is natural force.” Sigurðsson (2006:21) echoes this intuition, in describing the term fate: “I . . . take the liberty of using the notion ‘fate’ as a cover term for forces that are not in human power.” Sigurðsson (2005:106) hints at the argument structure alternation that we are suggesting here, describing the “Fate feature” as “a voice feature of a sort, blocking or ‘turning off ’ the usual voice feature that otherwise introduces AGENT in both transitives and passives.” Sigurðsson (2011) subsequently introduces a special v head, VoiceFATE, to account for the unexpected accusative preservation (no “case-star deletion”) and out-of-human-control semantics of the construction. We follow Sigurðsson 2011 in locating the source of accusative and the construction’s out-ofhuman- control semantics within the system of v heads. Our particular implementation of Sigurðsson’s intuition capitalizes on the idea of the fate accusative construction as a kind of causative without an external argument. In this way, we provide a unified treatment for both the fate accusative and the transitive impersonal. In both cases, the causer argument is realized as Natural Force (or Instrument, in Russian), and the event is construed as externally caused, but agentless. The Natural Force causer sets the event in motion, thereby activating the accusative probe v[cause].20
We noted above that the fate accusative, like the transitive impersonal, also has a transitive agentive alternate when [m] is positively valued (giving a sentient, volitional Agent). That is, the Icelandic fate accusative, like the Russian transitive impersonal, is formed from verbs that also realize basic, transitive agentive two-place predicates.21
Bóndinn rak hestana í burtu.
farmer.the.NOM drove horses.the.ACC away
‘The farmer drove the horses away.’
Bátinn rak á land.
boat.the.ACC drove to land
‘The boat drifted ashore.’
Hún fyllti bátinn.
she.NOM filled boat.the.ACC
‘She filled the boat.’
‘The boat swamped.’
Like the transitive impersonal in Russian, the fate accusative construction does not involve a valency-reducing operation. The syntactic presence of the causer is identified by virtue of (a) the incompatibility of the construction with the anticausative, (b) its infelicity with ‘by itself ’ modification, and (c) the role it plays in licensing accusative by activating the v[cause] head. The causer argument remains part of the verb’s denotation, even though it is not overtly realized in an argument position.
Further evidence that the fate accusative is sensitive to nonvolitional causation comes from Experiencer predicates, which are causative only when they occur with a subject causer argument (Pesetsky 1995:55–60). Otherwise, the non-Experiencer argument is interpreted as a Stimulus or Target of Emotion (Pesetsky 1995). As a result, we predict that accusative should not be possible in the Icelandic fate accusative or the Russian transitive impersonal when formed from Experiencer verbs, since, as impersonals, they lack the causer subject (Spec,v[voice]P is not filled, and v[cause] is not identified). In the absence of such external causation, as in the case of impersonal (subjectless) Experiencer predicates, the special ‘fate’ reading is lost. Compare the licit appearance of accusative in (23a), which is causative, with its ungrammatical appearance in the object Experiencer predicate in (23b). Now observe that (23c) is grammatical, but only on the unremarkable, noncausative Experiencer reading: undra ‘be surprised, marvel at’ is an accusative-subject Experiencer verb (Jónsson 2003:157–158)—no ‘fate’ reading is involved. Note finally that (23d), coerced here as externally caused by an unexpressed Natural Force (i.e., the ‘fate’ interpretation), is ungrammatical.22 The ‘fate’ reading fails because the (accusative) subject Experiencer predicate is not causative.
Fiskimennina rak á land vegna mikils vinds.
fishermen.the.ACC drove to land because.of heavy winds
‘The fishermen drifted ashore due to heavy winds.’
*Fiskimennina hræddi vegna mikils vinds.
fishermen.the.ACC frightened because.of heavy winds
(Intended: ‘The fishermen were frightened because of the heavy winds.’)
(Jóhannes Gísli Jónsson, pers. comm.)
Krakkann undraði þetta.
child.the.ACC marveled this.ACC
‘The child marveled at this.’
*Krakkann undraði í svefni.
child.the.ACC marveled in sleep
(Intended: ‘The child marveled at some force of nature in its sleep.’)
The same facts hold for the Russian transitive impersonal in (24a–b). In the case of the object Experiencer verb napugat’ ‘frighten’ in (24b), the expected causer subject is not merged due to impersonalization (i.e., the sentence is subjectless). The non-Experiencer argument, igruškoj ‘toy.INST’, is realized instead as a VP-internal target stimulus, the entity toward which the emotion is directed, rather than its cause.
To review, our analysis treats the Icelandic fate accusative as a two-place predicate whose causer argument is obligatorily interpreted as a Natural Force, which sets the event in motion and introduces into the structure a v[cause] head, which in turn licenses structural accusative case. This construction, it follows, is not an Icelandic-specific quirky phenomenon, but a transitive impersonal, in the sense described for Russian.25 This explains why accusative is impossible in anticausatives (perhaps universally): these are one-place predicates that assert only the root verbal meaning in relation to the Theme argument; as in the case of Experiencer predicates, no causative subevent is asserted.
A central claim of our analysis is that all transitive impersonals, including the fate accusative, lack a Voice projection: these constructions are subjectless and impersonal, the result of an impersonalization alternation. Note, however, the alternative treatment provided by Schäfer (2008: 291–302) and Wood (2017), following, in part, Haider 2001. Schäfer and Wood treat the fate accusative as likewise two-place and caused (rather than anticausative). Their account differs from ours in positing a covert external argument in the Spec,VoiceP position with the stipulated properties of bearing nominative case and the semantic feature of a referentially reduced (though thematic) weather pronoun. According to their account, the assumption that the covert external argument bears nominative case explains the appearance of accusative on the Theme argument, as an instance of (postsyntactic) dependent case (see Marantz 1992, Sigurðsson 2005, 2006). And the weather pronoun is posited to account for the atmospheric (Natural Force/out-of-human-control) semantics of the construction.
What is unique about our approach is that it accounts for accusative preservation and the necessary out-of-human-control semantics without stipulating a null thematic nominative pronoun. First, positing a thematic covert pronoun in these two-place predicates forces the treatment of the VP-internal causer as either a third argument or an adjunct (as noted in footnote 3, the latter is suggested by an anonymous reviewer). Recall that we are proposing an alternation regarding the position in which the causer argument is generated, either as an Agent in Spec,v[voice]P or as a nonvolitional oblique causer in a VP-internal position (in which it receives either instrumental case or genitive as complement of the preposition ot ‘from’). We posit no reduction or expansion in the basic valency of the verb, but rather only a rearrangement of its arguments (cf. the sprayload and dative alternations). This approach is supported by the fact that the removal of the causer argument in Russian, barring suppletion, requires the affixation of anticausative -sja, which does not occur in the transitive impersonal (in the same way that Icelandic -st and -na are incompatible with the fate accusative).26 Indeed, the very purpose of the low merger of the causer argument (i.e., the impersonalization alternation) is to construe the event as out of human control. The fact that accusative, on our account, is linked to causation (the v[cause] probe), rather than covert material higher in the structure, is demonstrated by the inability to form the transitive impersonal on the basis of verbal roots that are inherently monadic (basic unaccusatives). And it explains the incompatibility of the transitive impersonal with anticausatives and stative Experiencer predicates. Finally, note that the Russian transitive impersonal occurs not only with Natural Forces, but also with Instruments and causing events.27 We would not expect, for example, a “weather”-like, referentially reduced pronoun to set in motion events caused by a smell (9), a moth (14b), a tram (14c), the glutting of a market with low-quality gasoline (14d), the effect of a pilot’s deliberate dumping of fuel (20a), the (unstated) force of momentum (20b), an electrical current (24a), a motorcycle (26a), or the force exerted by a crowd of people (27). We believe that a more general argument alternation of the type described here for Russian can be extended to Icelandic without the additional machinery of null elements with special case and semantic properties. Recall the key similarities between the Russian and Icelandic transitive impersonal constructions: (a) they are both formed on verbs that are otherwise two-place and causative; (b) accusative is preserved, as in the transitive agentive variant; (c) both constructions resist ‘by itself ’ modification, indicating the presence of a causer (of some variety); and (d) the interpretation in both constructions is necessarily ‘out of human control’.28 We provide a unified account for the source of external causation in both constructions as a consequence of the lower merger of the causer (and its suppression in Icelandic), whose very purpose is to provide a means of describing an event as out of human control.
5 The Derived Transitive
Our analysis of transitive impersonal sentences is supported by the following correct prediction: namely, when the Agent argument of an externally caused verb is not selected, the nonvolitional causer can actually be realized in two different ways, either as a VP-internal, obliquely marked argument in the transitive impersonal as in (25a) or as the nominative subject of a derived transitive as in (25b). (26) is an additional example.
Menja oslepilo molniej.
me.ACC blinded.IMP lightning.INST
‘I was blinded by the lightning.’
Menja oslepila molnija.
me.ACC blinded.F.SG lightning.NOM.F.SG
‘The lightning blinded me.’
Ego sbilo s nog motociklom.
him.ACC knocked.down.IMP from legs motorcycle.INST
‘He was knocked off his feet by a motorcycle.’
Ego sbil s nog motocikl.
him.ACC knocked.down.M.SG from legs motorcycle.NOM.M.SG
‘A motorcycle knocked him off his feet.’
In point of fact, the feature complex of the causer argument posited for the transitive impersonal, [+c−m], applies equally well to the same causer argument when it appears in subject position (i.e., the argument marked nominative in the (b) examples in (25)–(26)).
The alternation exemplified in (25)–(26) raises two questions: (a) assuming that both alternates are derived from the same root and have the same basic argument structure, how are they derived (i.e., at what point do their derivations diverge)? and (b) assuming that systematic alternative morphosyntactic realizations of the same root verb have different, sometimes subtle meanings (cf. the active ~ passive alternation) or are appropriate in different discourse contexts, what precisely is the systematic meaning difference between the alternative argument realizations in pairs like (25)–(26)?
First, while the derivation of both sentences in (25)–(26) has the same underlying argument structure, in which a volitional Agent has not been selected, the subject position itself is not necessarily eliminated. In the case of the derived transitive in (25b) and (26b), the nonvolitional causer (molnija ‘lightning.NOM’ and motocikl ‘motorcycle.NOM’) occupies the subject position and is assigned nominative case by agreement with T.29 Next, note that there is a systematic difference in discourse appropriateness between the derived transitive and the transitive impersonal. While the transitive impersonal is used to assert that the event is not under the control of a volitional causer, the derived transitive is unmarked for human control: it neither asserts nor denies it. As we have shown, the transitive impersonal is common in events involving Natural Forces and situations in which the event is out of human control. In (27), the event described involves human beings, but the dynamics of the enormous crowd they are in has gone out of (human) control.
Russian: Transitive impersonal
(Tolpa byla ogromma.) Postepenno ee prilivami i otlivami menja
(the crowd was immense) gradually its flows.INST and ebbs.INST me.ACC
prižalo k stene.
pressed.IMP to wall
‘The crowd was immense. Its ebb and flow gradually pinned me to the wall.’
(V. Sysoev, Xodite tixo, govorite tixo)
In contrast, where human control is asserted, as in (28), only the derived transitive is appropriate, because it is clear from the context that the source of causation (‘a strong blow’) was intentional. The transitive impersonal, given in parentheses, is reported by Mel’čuk (1974) as ungrammatical on this reading.30
Russian: Derived transitive
(Soldaty brosilis’ na Ivana) i srazu že sil’nyj udar ego
(soldiers hurled themselves at Ivan) and immediately strong blow.NOM.M.SG him.ACC
sbil s nog (*sil’nym udarom ego sbilo s
knocked.down.M.SG from legs strong blow.INST him.ACC knocked.down.IMP from
‘Soldiers hurled themselves at Ivan and a powerful blow immediately knocked him off his feet.’
Thus, the derived transitive is used when the event is under human control, as in (28), or when there is no need to assert or emphasize that the event is not under human control, as in (25b) and (26b). The (a) and (b) examples in (25)–(26) describe the same real-world event from two different points of view. The transitive impersonals in the (a) examples emphasize the event as out of human control, while the derived transitives in the (b) examples are unmarked for this feature, but consistent with it.
Argument structure, it follows, is not fixed in the lexicon, but is subject to competing event construals, each with its own pattern of argument realization in the syntax. Transitive impersonals instantiate the very underspecification in argument structure that is predicted by constructionist syntax. More specifically, while the “affected” argument is fixed as the Theme, the predicate’s causer argument is freely merged high, as an Agent, in Spec,v[voice]P and realized as nominative, or low, as a nonvolitional causer, and realized by an oblique case, if the causer is known and relevant to the discourse. The nonvolitional causer activates v[cause], an accusative probe. The nonvolitional causer can also be realized as subject, an alternate we dub the derived transitive, when the event is not necessarily construed as out of human control or when the out-of-humancontrol semantics is not relevant.
We have shown, on the basis of the following observations, that both the Russian transitive impersonal and the Icelandic fate accusative are two-place and externally caused: (a) they do not occur with basic unaccusative or anticausative constructions; (b) they resist ‘by itself ’ modification; (c) they do not occur with noncausative (stative) Experiencer predicates; and (d) they target verbs that also project transitive agentive predicates. This all suggests that these verbs are necessar- ily causative, but that the kind of causation is not fixed in the lexicon. Reinhart’s (2002) minimal specification for the causer argument as [+c] captures its variable realization described above, which we have referred to as the impersonalization alternation. Our analysis in terms of the underspecification of the causer argument offers a significantly more parsimonious explanation for the range of data surveyed here than (a) positing multiple lexical entries for every verb that participates in the alternation or (b) stipulating a null thematic subject, which we have argued is unnecessary. We have advanced a causative theory of accusative, which extends Pylkkänen’s (2008) observation that causative verbs appear in the absence of canonical external arguments. And we have demonstrated the role that external causation plays in accusative licensing. More generally, we have shown how different eventualities are built up in the syntax as a function of whether the root verb combines with an Agent in Spec,v[voice]P or with a nonvolitional oblique causer, which activates v[cause]. The syntactic presence of a [cause] head in the absence of [voice] results in a transitive, though subjectless, impersonal. We thereby provide new evidence for the independence of the Agent from the root verb’s lexical entry: a large class of lexically causative verbs (namely, those minimally specified as [+c]) occurs with or without an Agent argument, depending on the eventuality described. Finally, our elucidation of a systematic impersonalization alternation in both Russian and Icelandic provides a new argument for the underspecification of causer arguments in lexical entries.
1 Transitive impersonals in Russian (and elsewhere) are not passive, despite the English passive-like translation. They bear active morphology and do not contain an Agent or subject of any kind at any level of representation. The suffix -o, glossed as IMP(ersonal), appears in lieu of agreement morphology in subjectless sentences.
2 The oblique case marking on the causer argument indicates its VP-internal status. When the same nonvolitional causer merges higher, in a position in which it is sufficiently local to T(ense), it appears in the nominative (as in the case of the “derived transitive,” discussed in section 5).
3 An anonymous reviewer asks whether the causer in the transitive impersonal construction might best be treated as an adjunct rather than an argument. We assume that the causer is an argument since there is no overt valency-reducing morphology, such as passive -en or anticausative -sja. At the same time, we do not want to overstate this claim. Whether the causer is an argument or an adjunct, it has to perform the same function, namely, to identify a causative subevent, which we link to the presence of a discrete [cause] head in the syntax (section 3). It follows that if we treat the causer as an adjunct, it would be an obligatory (selected) adjunct, since the transitive impersonal is necessarily causative ([+c]). And unlike other syntactic adjuncts, such an adjunct would have a direct relationship with a (suppressed) argument of the verb, similar to Grimshaw’s (1990) description of “argument adjuncts.” In this sense, the causer has an “intermediate status” (Grimshaw 1990) in that it obligatorily identifies a position in event structure (Grimshaw’s aspectual tier), but fails to satisfy an argument position. Note, however, that if the causer were a simple adjunct, it would be expected to occur freely with any semantically compatible verb. Yet instrumental causers are systematically ungrammatical with anticausatives, as in (i), adapted from (6b).
*Ledyška rastopilas’ solncem.
icicle.NOM.F.SG melted.F.SG.ANTICAUS sun.INST
The fact that ledyška rastopilas’ ‘the icicle melted’ occurs felicitously with the PP ot solnca ‘from the sun’ suggests that the ot ‘from’ + GEN phrase may best be analyzed as an adjunct, in contrast with the bare instrumental causer phrase (see (9) and the discussion in footnote 4). We assume therefore that the instrumental causer is an argument and that we are describing an argument alternation that realizes the causer VP-internally rather than as an Agent argument of Voice. Alternatively, if the causer is an adjunct (e.g., the ot ‘from’+GEN phrase), then the impersonalization alternation reduces the predicate’s valency by one. Either way, the causer is necessary to identify a causative subevent.
4 For an explicit analysis of the argument structure of verbs like tošnit’ ‘to experience nausea’, see Babby 2009: 39–44. An anonymous reviewer notes that other transitive impersonals, in addition to tošnit’, occur with an ot ‘from’ + GEN causer phrase. The ot + GEN phrase indicates a Source semantically. Its principal syntactic function, like the instrumental causer, is to identify an independent [cause] head, perhaps, as suggested in footnote 3, as a “selected adjunct.” An anonymous reviewer provides the following additional example with ot ‘from’ + GEN:
Russian: Transitive impersonal
Ego trjaslo ot straxa.
he.ACC shook.IMP from fear.GEN
‘He trembled with fear.’
5 It remains to be explained why tošnit’ ‘to experience nausea’ fails to occur as a derived transitive (section 5), whereby the nonvolitional causer would be realized as a nominative subject, as in (i).
Russian: Derived transitive
*Zapax ego tošnil.
smell.NOM.M.SG he.ACC nauseated.M.SG
(Intended: ‘The smell nauseated him.’)
As noted above, tošnit’ is a rare and highly idiosyncratic verb, which does not occur with a subject of any kind. It follows that it has no reflexive, participial, or gerundive form. We do not further derive its defective paradigm.
9 It follows that anticausatives, on our analysis, contain a v head with change-of-state semantics and a stative vP, its complement, which combines with the root (Cuervo 2015). This lower v head identifies the result state (e.g., broken, melted, burned). Crucially, we assume no [cause] or [voice] head in the anticausative, following, in spirit, Ramchand 2008, where the anticausative contains a process head, with no initiator head; this is similar to standard approaches in which the structure of the anticausative is contained in the causative alternate. Other approaches, which assume a specifierless (or expletive) [voice] head (Schäfer 2008, Wood 2015), are not incompatible with the approach taken here, so long as [voice] does not introduce a thematic argument.
11Pylkkänen (2008) proposes the “Voice-bundling parameter,” whereby non-Voice-bundling languages, which represent [voice] and [cause] on two discrete syntactic heads, as in (15), are distinguished from Voice-bundling languages, like English, in which “the causative relation and the external θ-role are ‘packaged’ . . . into one syntactic head” (Pylkkänen 2008:100). An anonymous reviewer questions whether there is any evidence for non-Voice-bundling elsewhere in Slavic or Icelandic. In Slavic, we would point to the accusative-preserving Ukrainian impersonal passive (see Lavine 2010, 2013); in Icelandic, non-Voice-bundling plausibly gives rise to accusative preservation in the “new passive” (or “new impersonal”) (see Eythórsson 2008 and Jónsson 2009, among others; for an alternative account, see Maling and Sigurjónsdóttir 2002 and Legate 2014).
12 Compare the ungrammatical English sentence in (i) (on the expletive interpretation for it) with the grammatical Russian sentence (14a).
*It damaged the dam due to the earthquake.
13Alexiadou, Anagnostopoulou, and Schäfer (2015) provide further evidence for dissociating the causative relation from the Agent in their analysis of an oblique causer construction in German, illustrated in (i) (see also Schäfer 2008).
German: Oblique causer
Dem Mann ist die Vase zerbrochen.
the.DAT man is the.NOM vase broken
‘The man unintentionally caused the vase to break.’
Here, an oblique DP (marked dative), when added to an anticausative, is interpreted as the unintentional causer of the change-of-state event (see Alexiadou, Anagnostopoulou, and Schäfer 2015:44–51 for details). Note, however, that dem Mann ‘the man.DAT’ is not an argument of anticausative ‘break’ and does not introduce an accusative probe (in independent v[cause]), in contrast to the transitive impersonal in Russian.
14 The distribution of both the Russian transitive impersonal and the Icelandic fate accusative follows Wood’s (2017) Accusative Subject Generalization, which states that accusative subjects in Icelandic never occur as arguments of morphologically intransitive verbs, namely, passives, morphologically marked anticausatives, and verbs with intransitive stem morphology (in Russian (11)).
15 The -st suffix is referred to as “middle” by Jónsson (2003:131). See Wood 2015:61–73 for discussion of the full range of its usage. In (17), -st derives an anticausative and is therefore glossed as such. Accusative is likewise incompatible with the -na morpheme, which also forms anticausatives for certain verbs. See Wood 2017 for details.
16 By way of contrast, Schäfer (2008:298) notes that in genuine anticausatives in Icelandic, ‘by itself ’ modification is perfectly acceptable.
Dyrnar opnuðu-st af sjálfumsér.
door.the.NOM opened-ANTICAUS of self
‘The door opened on its own.’
The idea is that the fate accusative resists ‘by itself ’ modification because the Theme argument undergoes a change of state as a result of external causation, whereas in the anticausative the event is construed as if the Theme underwent a change of state with no such external force (see Schäfer 2008:297–299 for related discussion). If ‘by itself ’ modification denies the presence of a causer, as suggested by Alexiadou, Anagnostopoulou, and Schäfer (2015:21–22), then its incompatibility with the fate accusative is correctly predicted.
17 Note the similar case of “causer suppression” in the Sinhala involitive, as discussed by Beavers and Zubair (2013). Sinhala also admits accusative in seemingly one-place predicates that crucially entail the implicit presence of a causer argument.
18 Compare the use of the subjectless 3rd plural verb form, which is standardly used to indicate an unknown human actor, as in (i).
Ego ubili na vojne.
he.ACC killed.3PL at war
‘He was killed at war.’
Impersonal -o is the correlate to the 3rd plural form for potentially unknown Natural Forces.
20 Each functional head in the syntax is a predicate of sorts, identified with a particular argument (cf. Bowers 2010). Think of v[cause] as a predicate unsaturated until an argument merges that is capable of independently setting the event in motion. Secondary Instrumentals (discussed in Alexiadou, Anagnostopoulou, and Schäfer 2006) that require human manipulation, like ‘fork’, would thus fail to “saturate” (= license) the [cause] head. Only causative obliques, expressed overtly or merely understood, saturate the v[cause] predicate, licensing its independent presence in the structure and giving rise to its morphosyntactic reflex of licensing ( probing) accusative (cf. the licensing of Borer’s (2005) Aspect head).
21 An exception is Russian tošnit’ ‘to experience nausea’, whose causer argument is fully specified as [+c−m] and which, as a result, does not alternate with an agentive counterpart, but appears as a transitive impersonal only.
22 We thank Tolli Eythórsson for discussion of (23c–d).
23 An anonymous reviewer notes the long list of Experiencer predicates in Jónsson 2003 that occur with accusative subjects. Note that Jónsson limits his discussion of the fate accusative construction to the non-Experiencer, transitive motion verb reka ‘drive, drift’. We assume that (23c) is not a fate accusative construction, on the reading that þetta ‘this.ACC’ is interpreted as the target stimulus, rather than as Natural Force (or as some out-of-human-control causer). As discussed in the text, we posit that psych verbs, as a general rule, should be degraded in the fate accusative construction, since they lack a causer subject. This is a strong hypothesis that may have to be weakened in future work upon consideration of expected counterexamples.
24 Anton Zimmerling ( pers. comm.) suggests the improved Experiencer predicate in (i).
Russian: Experiencer predicate
?Mal’čika napugalo vspyškami molnii.
boy.ACC frightened.IMP flashes.INST of.lightning
‘The boy was frightened by lightning flashes.’
Though (i) is still degraded, lightning flashes cannot be manipulated by humans and, it follows, (i) is a better candidate for the necessary out-of-human-control causer reading for Russian. See Zimmerling 2013 for related discussion.
25 To be sure, a narrower range of verbal roots participates in the Icelandic construction; indeed, some verbs whose causer is [+c] with [m] unspecified—our stated criterion for transitive impersonal formation—resist the fate accusative construction in Icelandic (Wood (2017:262n17) notes: “It is not enough to say that F[ate]A[ccusative] verbs are unspecified for cause. . . . First of all, not all cause unspecified verbs can even form FAs”). We thank an anonymous reviewer for providing eyðileggja ‘destroy’ as one such verb with a [+c] causer ([m] unspecified) that does not form a fate accusative.
26 The absence of Voice (or of any external causation) is indicated morphologically in Russian by either anticausative -sja or the intransitive verb stem of the causative alternation. The latter is illustrated in (11). An example of anticausative ‘melt’, affixed with -sja, is repeated in (i) from footnote 3.
Ledyška (*Ledyšku) rastopilas’.
icicle.NOM.F.SG icicle.ACC melted.F.SG.ANTICAUS
‘The icicle melted.’
Neither anticausative -sja nor intransitive verb stems occur in the transitive impersonal. Recall that this is because the transitive impersonal is necessarily caused (though, we posit, not by an argument of Voice).
27Wood (2017) notes that not all fate accusative constructions involve an implied Natural Force causer. He refers to Sigurðsson’s (2006:21) formulation of “forces that are not in human power.” Likewise, according to Schäfer (2008: 291–302) the null weather pronoun, a “referentially reduced” thematic subject, does not require a strict ‘weather’ interpretation; instead, it refers to the unidentified force associated with the ‘fate’ interpretation. Note that the causer in the Russian transitive impersonal is standardly identified—not vague or abstract—and, in point of fact, is often quite concrete.
28 A full account of the Icelandic construction along these lines would require discussion of causer suppression. This is a complex problem involving the existence of elements that are semantically active but not syntactically projected. See Beavers and Zubair 2013 for discussion of similar cases of causer suppression in Sinhala.
29 We assume that nonvolitional causers are not arguments of Voice (see Alexiadou, Anagnostopoulou, and Schäfer 2015 for extensive discussion). In the case of the derived transitive, the nonvolitional causer appears in Spec,v[cause]P, from which position it is visible to the nominative probe in T.
30 Ksenia Zanon ( pers. comm.) suggests a stronger context for the derived transitive, given in (i).
Russian: Derived transitive
(Mašu atakoval razbojnnik.) Ee sbil s nog sil’nyj udar / (*ee
(A thief attacked Masha.) she.ACC knocked.down.M.SG from legs strong blow.NOM.M.SG she.ACC
sbilo s nog sil’nym udarom).
knocked.down.IMP from legs strong blow.INST
‘A thief attacked Masha. A powerful blow knocked her off her feet.’
We gratefully acknowledge Thórhallur Eythórsson, Jóhannes Gísli Jónsson, Joan Maling, Halldór Ármann Sigurðsson, and Jim Wood for assistance with the Icelandic data and many useful insights, Robert Freidin for general discussion of this material, and two anonymous reviewers for detailed comments on earlier drafts.